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 /* default target for number of metaslabs per top-level vdev */
58 int zfs_vdev_default_ms_count
= 200;
60 /* minimum number of metaslabs per top-level vdev */
61 int zfs_vdev_min_ms_count
= 16;
63 /* practical upper limit of total metaslabs per top-level vdev */
64 int zfs_vdev_ms_count_limit
= 1ULL << 17;
66 /* lower limit for metaslab size (512M) */
67 int zfs_vdev_default_ms_shift
= 29;
69 /* upper limit for metaslab size (16G) */
70 int zfs_vdev_max_ms_shift
= 34;
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_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
538 mutex_init(&vd
->vdev_initialize_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
539 mutex_init(&vd
->vdev_initialize_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
540 cv_init(&vd
->vdev_initialize_cv
, NULL
, CV_DEFAULT
, NULL
);
541 cv_init(&vd
->vdev_initialize_io_cv
, NULL
, CV_DEFAULT
, NULL
);
543 for (int t
= 0; t
< DTL_TYPES
; t
++) {
544 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
);
546 txg_list_create(&vd
->vdev_ms_list
, spa
,
547 offsetof(struct metaslab
, ms_txg_node
));
548 txg_list_create(&vd
->vdev_dtl_list
, spa
,
549 offsetof(struct vdev
, vdev_dtl_node
));
550 vd
->vdev_stat
.vs_timestamp
= gethrtime();
558 * Allocate a new vdev. The 'alloctype' is used to control whether we are
559 * creating a new vdev or loading an existing one - the behavior is slightly
560 * different for each case.
563 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
568 uint64_t guid
= 0, islog
, nparity
;
570 vdev_indirect_config_t
*vic
;
573 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
574 boolean_t top_level
= (parent
&& !parent
->vdev_parent
);
576 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
578 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
579 return (SET_ERROR(EINVAL
));
581 if ((ops
= vdev_getops(type
)) == NULL
)
582 return (SET_ERROR(EINVAL
));
585 * If this is a load, get the vdev guid from the nvlist.
586 * Otherwise, vdev_alloc_common() will generate one for us.
588 if (alloctype
== VDEV_ALLOC_LOAD
) {
591 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
593 return (SET_ERROR(EINVAL
));
595 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
596 return (SET_ERROR(EINVAL
));
597 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
598 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
599 return (SET_ERROR(EINVAL
));
600 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
601 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
602 return (SET_ERROR(EINVAL
));
603 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
604 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
605 return (SET_ERROR(EINVAL
));
609 * The first allocated vdev must be of type 'root'.
611 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
612 return (SET_ERROR(EINVAL
));
615 * Determine whether we're a log vdev.
618 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
619 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
620 return (SET_ERROR(ENOTSUP
));
622 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
623 return (SET_ERROR(ENOTSUP
));
626 * Set the nparity property for RAID-Z vdevs.
629 if (ops
== &vdev_raidz_ops
) {
630 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
632 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
633 return (SET_ERROR(EINVAL
));
635 * Previous versions could only support 1 or 2 parity
639 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
640 return (SET_ERROR(ENOTSUP
));
642 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
643 return (SET_ERROR(ENOTSUP
));
646 * We require the parity to be specified for SPAs that
647 * support multiple parity levels.
649 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
650 return (SET_ERROR(EINVAL
));
652 * Otherwise, we default to 1 parity device for RAID-Z.
659 ASSERT(nparity
!= -1ULL);
662 * If creating a top-level vdev, check for allocation classes input
664 if (top_level
&& alloctype
== VDEV_ALLOC_ADD
) {
667 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
669 alloc_bias
= vdev_derive_alloc_bias(bias
);
671 /* spa_vdev_add() expects feature to be enabled */
672 if (spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
673 !spa_feature_is_enabled(spa
,
674 SPA_FEATURE_ALLOCATION_CLASSES
)) {
675 return (SET_ERROR(ENOTSUP
));
680 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
681 vic
= &vd
->vdev_indirect_config
;
683 vd
->vdev_islog
= islog
;
684 vd
->vdev_nparity
= nparity
;
685 if (top_level
&& alloc_bias
!= VDEV_BIAS_NONE
)
686 vd
->vdev_alloc_bias
= alloc_bias
;
688 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
689 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
692 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
693 * fault on a vdev and want it to persist across imports (like with
696 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
697 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
698 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
699 vd
->vdev_faulted
= 1;
700 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
703 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
704 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
705 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
706 &vd
->vdev_physpath
) == 0)
707 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
709 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
710 &vd
->vdev_enc_sysfs_path
) == 0)
711 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
713 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
714 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
717 * Set the whole_disk property. If it's not specified, leave the value
720 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
721 &vd
->vdev_wholedisk
) != 0)
722 vd
->vdev_wholedisk
= -1ULL;
724 ASSERT0(vic
->vic_mapping_object
);
725 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
726 &vic
->vic_mapping_object
);
727 ASSERT0(vic
->vic_births_object
);
728 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
729 &vic
->vic_births_object
);
730 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
731 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
732 &vic
->vic_prev_indirect_vdev
);
735 * Look for the 'not present' flag. This will only be set if the device
736 * was not present at the time of import.
738 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
739 &vd
->vdev_not_present
);
742 * Get the alignment requirement.
744 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
747 * Retrieve the vdev creation time.
749 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
753 * If we're a top-level vdev, try to load the allocation parameters.
756 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
757 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
759 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
761 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
763 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
765 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
768 ASSERT0(vd
->vdev_top_zap
);
771 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
772 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
773 alloctype
== VDEV_ALLOC_ADD
||
774 alloctype
== VDEV_ALLOC_SPLIT
||
775 alloctype
== VDEV_ALLOC_ROOTPOOL
);
776 /* Note: metaslab_group_create() is now deferred */
779 if (vd
->vdev_ops
->vdev_op_leaf
&&
780 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
781 (void) nvlist_lookup_uint64(nv
,
782 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
784 ASSERT0(vd
->vdev_leaf_zap
);
788 * If we're a leaf vdev, try to load the DTL object and other state.
791 if (vd
->vdev_ops
->vdev_op_leaf
&&
792 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
793 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
794 if (alloctype
== VDEV_ALLOC_LOAD
) {
795 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
796 &vd
->vdev_dtl_object
);
797 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
801 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
804 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
805 &spare
) == 0 && spare
)
809 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
812 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
813 &vd
->vdev_resilver_txg
);
815 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
816 vdev_set_deferred_resilver(spa
, vd
);
819 * In general, when importing a pool we want to ignore the
820 * persistent fault state, as the diagnosis made on another
821 * system may not be valid in the current context. The only
822 * exception is if we forced a vdev to a persistently faulted
823 * state with 'zpool offline -f'. The persistent fault will
824 * remain across imports until cleared.
826 * Local vdevs will remain in the faulted state.
828 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
829 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
830 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
832 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
834 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
837 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
841 VDEV_AUX_ERR_EXCEEDED
;
842 if (nvlist_lookup_string(nv
,
843 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
844 strcmp(aux
, "external") == 0)
845 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
847 vd
->vdev_faulted
= 0ULL;
853 * Add ourselves to the parent's list of children.
855 vdev_add_child(parent
, vd
);
863 vdev_free(vdev_t
*vd
)
865 spa_t
*spa
= vd
->vdev_spa
;
866 ASSERT3P(vd
->vdev_initialize_thread
, ==, NULL
);
869 * Scan queues are normally destroyed at the end of a scan. If the
870 * queue exists here, that implies the vdev is being removed while
871 * the scan is still running.
873 if (vd
->vdev_scan_io_queue
!= NULL
) {
874 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
875 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
876 vd
->vdev_scan_io_queue
= NULL
;
877 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
881 * vdev_free() implies closing the vdev first. This is simpler than
882 * trying to ensure complicated semantics for all callers.
886 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
887 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
892 for (int c
= 0; c
< vd
->vdev_children
; c
++)
893 vdev_free(vd
->vdev_child
[c
]);
895 ASSERT(vd
->vdev_child
== NULL
);
896 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
897 ASSERT(vd
->vdev_initialize_thread
== NULL
);
900 * Discard allocation state.
902 if (vd
->vdev_mg
!= NULL
) {
903 vdev_metaslab_fini(vd
);
904 metaslab_group_destroy(vd
->vdev_mg
);
907 ASSERT0(vd
->vdev_stat
.vs_space
);
908 ASSERT0(vd
->vdev_stat
.vs_dspace
);
909 ASSERT0(vd
->vdev_stat
.vs_alloc
);
912 * Remove this vdev from its parent's child list.
914 vdev_remove_child(vd
->vdev_parent
, vd
);
916 ASSERT(vd
->vdev_parent
== NULL
);
919 * Clean up vdev structure.
925 spa_strfree(vd
->vdev_path
);
927 spa_strfree(vd
->vdev_devid
);
928 if (vd
->vdev_physpath
)
929 spa_strfree(vd
->vdev_physpath
);
931 if (vd
->vdev_enc_sysfs_path
)
932 spa_strfree(vd
->vdev_enc_sysfs_path
);
935 spa_strfree(vd
->vdev_fru
);
937 if (vd
->vdev_isspare
)
938 spa_spare_remove(vd
);
939 if (vd
->vdev_isl2cache
)
940 spa_l2cache_remove(vd
);
942 txg_list_destroy(&vd
->vdev_ms_list
);
943 txg_list_destroy(&vd
->vdev_dtl_list
);
945 mutex_enter(&vd
->vdev_dtl_lock
);
946 space_map_close(vd
->vdev_dtl_sm
);
947 for (int t
= 0; t
< DTL_TYPES
; t
++) {
948 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
949 range_tree_destroy(vd
->vdev_dtl
[t
]);
951 mutex_exit(&vd
->vdev_dtl_lock
);
953 EQUIV(vd
->vdev_indirect_births
!= NULL
,
954 vd
->vdev_indirect_mapping
!= NULL
);
955 if (vd
->vdev_indirect_births
!= NULL
) {
956 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
957 vdev_indirect_births_close(vd
->vdev_indirect_births
);
960 if (vd
->vdev_obsolete_sm
!= NULL
) {
961 ASSERT(vd
->vdev_removing
||
962 vd
->vdev_ops
== &vdev_indirect_ops
);
963 space_map_close(vd
->vdev_obsolete_sm
);
964 vd
->vdev_obsolete_sm
= NULL
;
966 range_tree_destroy(vd
->vdev_obsolete_segments
);
967 rw_destroy(&vd
->vdev_indirect_rwlock
);
968 mutex_destroy(&vd
->vdev_obsolete_lock
);
970 mutex_destroy(&vd
->vdev_dtl_lock
);
971 mutex_destroy(&vd
->vdev_stat_lock
);
972 mutex_destroy(&vd
->vdev_probe_lock
);
973 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
974 mutex_destroy(&vd
->vdev_initialize_lock
);
975 mutex_destroy(&vd
->vdev_initialize_io_lock
);
976 cv_destroy(&vd
->vdev_initialize_io_cv
);
977 cv_destroy(&vd
->vdev_initialize_cv
);
979 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
980 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
982 if (vd
== spa
->spa_root_vdev
)
983 spa
->spa_root_vdev
= NULL
;
985 kmem_free(vd
, sizeof (vdev_t
));
989 * Transfer top-level vdev state from svd to tvd.
992 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
994 spa_t
*spa
= svd
->vdev_spa
;
999 ASSERT(tvd
== tvd
->vdev_top
);
1001 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
1002 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
1003 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
1004 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
1005 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
1007 svd
->vdev_ms_array
= 0;
1008 svd
->vdev_ms_shift
= 0;
1009 svd
->vdev_ms_count
= 0;
1010 svd
->vdev_top_zap
= 0;
1013 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
1014 tvd
->vdev_mg
= svd
->vdev_mg
;
1015 tvd
->vdev_ms
= svd
->vdev_ms
;
1017 svd
->vdev_mg
= NULL
;
1018 svd
->vdev_ms
= NULL
;
1020 if (tvd
->vdev_mg
!= NULL
)
1021 tvd
->vdev_mg
->mg_vd
= tvd
;
1023 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
1024 svd
->vdev_checkpoint_sm
= NULL
;
1026 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1027 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1029 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1030 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1031 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1033 svd
->vdev_stat
.vs_alloc
= 0;
1034 svd
->vdev_stat
.vs_space
= 0;
1035 svd
->vdev_stat
.vs_dspace
= 0;
1038 * State which may be set on a top-level vdev that's in the
1039 * process of being removed.
1041 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1042 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1043 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1044 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1045 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1046 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1047 ASSERT0(tvd
->vdev_removing
);
1048 tvd
->vdev_removing
= svd
->vdev_removing
;
1049 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1050 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1051 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1052 range_tree_swap(&svd
->vdev_obsolete_segments
,
1053 &tvd
->vdev_obsolete_segments
);
1054 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1055 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1056 svd
->vdev_indirect_config
.vic_births_object
= 0;
1057 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1058 svd
->vdev_indirect_mapping
= NULL
;
1059 svd
->vdev_indirect_births
= NULL
;
1060 svd
->vdev_obsolete_sm
= NULL
;
1061 svd
->vdev_removing
= 0;
1063 for (t
= 0; t
< TXG_SIZE
; t
++) {
1064 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1065 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1066 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1067 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1068 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1069 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1072 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1073 vdev_config_clean(svd
);
1074 vdev_config_dirty(tvd
);
1077 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1078 vdev_state_clean(svd
);
1079 vdev_state_dirty(tvd
);
1082 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1083 svd
->vdev_deflate_ratio
= 0;
1085 tvd
->vdev_islog
= svd
->vdev_islog
;
1086 svd
->vdev_islog
= 0;
1088 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1092 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1099 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1100 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1104 * Add a mirror/replacing vdev above an existing vdev.
1107 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1109 spa_t
*spa
= cvd
->vdev_spa
;
1110 vdev_t
*pvd
= cvd
->vdev_parent
;
1113 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1115 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1117 mvd
->vdev_asize
= cvd
->vdev_asize
;
1118 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1119 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1120 mvd
->vdev_psize
= cvd
->vdev_psize
;
1121 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1122 mvd
->vdev_state
= cvd
->vdev_state
;
1123 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1125 vdev_remove_child(pvd
, cvd
);
1126 vdev_add_child(pvd
, mvd
);
1127 cvd
->vdev_id
= mvd
->vdev_children
;
1128 vdev_add_child(mvd
, cvd
);
1129 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1131 if (mvd
== mvd
->vdev_top
)
1132 vdev_top_transfer(cvd
, mvd
);
1138 * Remove a 1-way mirror/replacing vdev from the tree.
1141 vdev_remove_parent(vdev_t
*cvd
)
1143 vdev_t
*mvd
= cvd
->vdev_parent
;
1144 vdev_t
*pvd
= mvd
->vdev_parent
;
1146 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1148 ASSERT(mvd
->vdev_children
== 1);
1149 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1150 mvd
->vdev_ops
== &vdev_replacing_ops
||
1151 mvd
->vdev_ops
== &vdev_spare_ops
);
1152 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1154 vdev_remove_child(mvd
, cvd
);
1155 vdev_remove_child(pvd
, mvd
);
1158 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1159 * Otherwise, we could have detached an offline device, and when we
1160 * go to import the pool we'll think we have two top-level vdevs,
1161 * instead of a different version of the same top-level vdev.
1163 if (mvd
->vdev_top
== mvd
) {
1164 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1165 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1166 cvd
->vdev_guid
+= guid_delta
;
1167 cvd
->vdev_guid_sum
+= guid_delta
;
1170 * If pool not set for autoexpand, we need to also preserve
1171 * mvd's asize to prevent automatic expansion of cvd.
1172 * Otherwise if we are adjusting the mirror by attaching and
1173 * detaching children of non-uniform sizes, the mirror could
1174 * autoexpand, unexpectedly requiring larger devices to
1175 * re-establish the mirror.
1177 if (!cvd
->vdev_spa
->spa_autoexpand
)
1178 cvd
->vdev_asize
= mvd
->vdev_asize
;
1180 cvd
->vdev_id
= mvd
->vdev_id
;
1181 vdev_add_child(pvd
, cvd
);
1182 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1184 if (cvd
== cvd
->vdev_top
)
1185 vdev_top_transfer(mvd
, cvd
);
1187 ASSERT(mvd
->vdev_children
== 0);
1192 vdev_metaslab_group_create(vdev_t
*vd
)
1194 spa_t
*spa
= vd
->vdev_spa
;
1197 * metaslab_group_create was delayed until allocation bias was available
1199 if (vd
->vdev_mg
== NULL
) {
1200 metaslab_class_t
*mc
;
1202 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1203 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1205 ASSERT3U(vd
->vdev_islog
, ==,
1206 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1208 switch (vd
->vdev_alloc_bias
) {
1210 mc
= spa_log_class(spa
);
1212 case VDEV_BIAS_SPECIAL
:
1213 mc
= spa_special_class(spa
);
1215 case VDEV_BIAS_DEDUP
:
1216 mc
= spa_dedup_class(spa
);
1219 mc
= spa_normal_class(spa
);
1222 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1223 spa
->spa_alloc_count
);
1226 * The spa ashift values currently only reflect the
1227 * general vdev classes. Class destination is late
1228 * binding so ashift checking had to wait until now
1230 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1231 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1232 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1233 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1234 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1235 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1241 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1243 spa_t
*spa
= vd
->vdev_spa
;
1244 objset_t
*mos
= spa
->spa_meta_objset
;
1246 uint64_t oldc
= vd
->vdev_ms_count
;
1247 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1250 boolean_t expanding
= (oldc
!= 0);
1252 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1255 * This vdev is not being allocated from yet or is a hole.
1257 if (vd
->vdev_ms_shift
== 0)
1260 ASSERT(!vd
->vdev_ishole
);
1262 ASSERT(oldc
<= newc
);
1264 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1267 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
1268 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1272 vd
->vdev_ms_count
= newc
;
1273 for (m
= oldc
; m
< newc
; m
++) {
1274 uint64_t object
= 0;
1277 * vdev_ms_array may be 0 if we are creating the "fake"
1278 * metaslabs for an indirect vdev for zdb's leak detection.
1279 * See zdb_leak_init().
1281 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1282 error
= dmu_read(mos
, vd
->vdev_ms_array
,
1283 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1286 vdev_dbgmsg(vd
, "unable to read the metaslab "
1287 "array [error=%d]", error
);
1294 * To accomodate zdb_leak_init() fake indirect
1295 * metaslabs, we allocate a metaslab group for
1296 * indirect vdevs which normally don't have one.
1298 if (vd
->vdev_mg
== NULL
) {
1299 ASSERT0(vdev_is_concrete(vd
));
1300 vdev_metaslab_group_create(vd
);
1303 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1306 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1313 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1316 * If the vdev is being removed we don't activate
1317 * the metaslabs since we want to ensure that no new
1318 * allocations are performed on this device.
1320 if (!expanding
&& !vd
->vdev_removing
) {
1321 metaslab_group_activate(vd
->vdev_mg
);
1325 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1331 vdev_metaslab_fini(vdev_t
*vd
)
1333 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1334 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1335 SPA_FEATURE_POOL_CHECKPOINT
));
1336 space_map_close(vd
->vdev_checkpoint_sm
);
1338 * Even though we close the space map, we need to set its
1339 * pointer to NULL. The reason is that vdev_metaslab_fini()
1340 * may be called multiple times for certain operations
1341 * (i.e. when destroying a pool) so we need to ensure that
1342 * this clause never executes twice. This logic is similar
1343 * to the one used for the vdev_ms clause below.
1345 vd
->vdev_checkpoint_sm
= NULL
;
1348 if (vd
->vdev_ms
!= NULL
) {
1349 uint64_t count
= vd
->vdev_ms_count
;
1351 metaslab_group_passivate(vd
->vdev_mg
);
1352 for (uint64_t m
= 0; m
< count
; m
++) {
1353 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1358 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1361 vd
->vdev_ms_count
= 0;
1363 ASSERT0(vd
->vdev_ms_count
);
1364 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1367 typedef struct vdev_probe_stats
{
1368 boolean_t vps_readable
;
1369 boolean_t vps_writeable
;
1371 } vdev_probe_stats_t
;
1374 vdev_probe_done(zio_t
*zio
)
1376 spa_t
*spa
= zio
->io_spa
;
1377 vdev_t
*vd
= zio
->io_vd
;
1378 vdev_probe_stats_t
*vps
= zio
->io_private
;
1380 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1382 if (zio
->io_type
== ZIO_TYPE_READ
) {
1383 if (zio
->io_error
== 0)
1384 vps
->vps_readable
= 1;
1385 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1386 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1387 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1388 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1389 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1391 abd_free(zio
->io_abd
);
1393 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1394 if (zio
->io_error
== 0)
1395 vps
->vps_writeable
= 1;
1396 abd_free(zio
->io_abd
);
1397 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1401 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1402 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1404 if (vdev_readable(vd
) &&
1405 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1408 ASSERT(zio
->io_error
!= 0);
1409 vdev_dbgmsg(vd
, "failed probe");
1410 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1411 spa
, vd
, NULL
, NULL
, 0, 0);
1412 zio
->io_error
= SET_ERROR(ENXIO
);
1415 mutex_enter(&vd
->vdev_probe_lock
);
1416 ASSERT(vd
->vdev_probe_zio
== zio
);
1417 vd
->vdev_probe_zio
= NULL
;
1418 mutex_exit(&vd
->vdev_probe_lock
);
1421 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1422 if (!vdev_accessible(vd
, pio
))
1423 pio
->io_error
= SET_ERROR(ENXIO
);
1425 kmem_free(vps
, sizeof (*vps
));
1430 * Determine whether this device is accessible.
1432 * Read and write to several known locations: the pad regions of each
1433 * vdev label but the first, which we leave alone in case it contains
1437 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1439 spa_t
*spa
= vd
->vdev_spa
;
1440 vdev_probe_stats_t
*vps
= NULL
;
1443 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1446 * Don't probe the probe.
1448 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1452 * To prevent 'probe storms' when a device fails, we create
1453 * just one probe i/o at a time. All zios that want to probe
1454 * this vdev will become parents of the probe io.
1456 mutex_enter(&vd
->vdev_probe_lock
);
1458 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1459 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1461 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1462 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1465 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1467 * vdev_cant_read and vdev_cant_write can only
1468 * transition from TRUE to FALSE when we have the
1469 * SCL_ZIO lock as writer; otherwise they can only
1470 * transition from FALSE to TRUE. This ensures that
1471 * any zio looking at these values can assume that
1472 * failures persist for the life of the I/O. That's
1473 * important because when a device has intermittent
1474 * connectivity problems, we want to ensure that
1475 * they're ascribed to the device (ENXIO) and not
1478 * Since we hold SCL_ZIO as writer here, clear both
1479 * values so the probe can reevaluate from first
1482 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1483 vd
->vdev_cant_read
= B_FALSE
;
1484 vd
->vdev_cant_write
= B_FALSE
;
1487 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1488 vdev_probe_done
, vps
,
1489 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1492 * We can't change the vdev state in this context, so we
1493 * kick off an async task to do it on our behalf.
1496 vd
->vdev_probe_wanted
= B_TRUE
;
1497 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1502 zio_add_child(zio
, pio
);
1504 mutex_exit(&vd
->vdev_probe_lock
);
1507 ASSERT(zio
!= NULL
);
1511 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1512 zio_nowait(zio_read_phys(pio
, vd
,
1513 vdev_label_offset(vd
->vdev_psize
, l
,
1514 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1515 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1516 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1517 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1528 vdev_open_child(void *arg
)
1532 vd
->vdev_open_thread
= curthread
;
1533 vd
->vdev_open_error
= vdev_open(vd
);
1534 vd
->vdev_open_thread
= NULL
;
1538 vdev_uses_zvols(vdev_t
*vd
)
1541 if (zvol_is_zvol(vd
->vdev_path
))
1545 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1546 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1553 vdev_open_children(vdev_t
*vd
)
1556 int children
= vd
->vdev_children
;
1559 * in order to handle pools on top of zvols, do the opens
1560 * in a single thread so that the same thread holds the
1561 * spa_namespace_lock
1563 if (vdev_uses_zvols(vd
)) {
1565 for (int c
= 0; c
< children
; c
++)
1566 vd
->vdev_child
[c
]->vdev_open_error
=
1567 vdev_open(vd
->vdev_child
[c
]);
1569 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1570 children
, children
, TASKQ_PREPOPULATE
);
1574 for (int c
= 0; c
< children
; c
++)
1575 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1576 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1581 vd
->vdev_nonrot
= B_TRUE
;
1583 for (int c
= 0; c
< children
; c
++)
1584 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1588 * Compute the raidz-deflation ratio. Note, we hard-code
1589 * in 128k (1 << 17) because it is the "typical" blocksize.
1590 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1591 * otherwise it would inconsistently account for existing bp's.
1594 vdev_set_deflate_ratio(vdev_t
*vd
)
1596 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1597 vd
->vdev_deflate_ratio
= (1 << 17) /
1598 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1603 * Prepare a virtual device for access.
1606 vdev_open(vdev_t
*vd
)
1608 spa_t
*spa
= vd
->vdev_spa
;
1611 uint64_t max_osize
= 0;
1612 uint64_t asize
, max_asize
, psize
;
1613 uint64_t ashift
= 0;
1615 ASSERT(vd
->vdev_open_thread
== curthread
||
1616 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1617 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1618 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1619 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1621 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1622 vd
->vdev_cant_read
= B_FALSE
;
1623 vd
->vdev_cant_write
= B_FALSE
;
1624 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1627 * If this vdev is not removed, check its fault status. If it's
1628 * faulted, bail out of the open.
1630 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1631 ASSERT(vd
->vdev_children
== 0);
1632 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1633 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1634 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1635 vd
->vdev_label_aux
);
1636 return (SET_ERROR(ENXIO
));
1637 } else if (vd
->vdev_offline
) {
1638 ASSERT(vd
->vdev_children
== 0);
1639 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1640 return (SET_ERROR(ENXIO
));
1643 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1646 * Reset the vdev_reopening flag so that we actually close
1647 * the vdev on error.
1649 vd
->vdev_reopening
= B_FALSE
;
1650 if (zio_injection_enabled
&& error
== 0)
1651 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1654 if (vd
->vdev_removed
&&
1655 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1656 vd
->vdev_removed
= B_FALSE
;
1658 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
1659 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
1660 vd
->vdev_stat
.vs_aux
);
1662 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1663 vd
->vdev_stat
.vs_aux
);
1668 vd
->vdev_removed
= B_FALSE
;
1671 * Recheck the faulted flag now that we have confirmed that
1672 * the vdev is accessible. If we're faulted, bail.
1674 if (vd
->vdev_faulted
) {
1675 ASSERT(vd
->vdev_children
== 0);
1676 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1677 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1678 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1679 vd
->vdev_label_aux
);
1680 return (SET_ERROR(ENXIO
));
1683 if (vd
->vdev_degraded
) {
1684 ASSERT(vd
->vdev_children
== 0);
1685 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1686 VDEV_AUX_ERR_EXCEEDED
);
1688 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1692 * For hole or missing vdevs we just return success.
1694 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1697 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1698 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1699 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1705 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1706 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1708 if (vd
->vdev_children
== 0) {
1709 if (osize
< SPA_MINDEVSIZE
) {
1710 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1711 VDEV_AUX_TOO_SMALL
);
1712 return (SET_ERROR(EOVERFLOW
));
1715 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1716 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1717 VDEV_LABEL_END_SIZE
);
1719 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1720 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1721 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1722 VDEV_AUX_TOO_SMALL
);
1723 return (SET_ERROR(EOVERFLOW
));
1727 max_asize
= max_osize
;
1731 * If the vdev was expanded, record this so that we can re-create the
1732 * uberblock rings in labels {2,3}, during the next sync.
1734 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
1735 vd
->vdev_copy_uberblocks
= B_TRUE
;
1737 vd
->vdev_psize
= psize
;
1740 * Make sure the allocatable size hasn't shrunk too much.
1742 if (asize
< vd
->vdev_min_asize
) {
1743 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1744 VDEV_AUX_BAD_LABEL
);
1745 return (SET_ERROR(EINVAL
));
1748 if (vd
->vdev_asize
== 0) {
1750 * This is the first-ever open, so use the computed values.
1751 * For compatibility, a different ashift can be requested.
1753 vd
->vdev_asize
= asize
;
1754 vd
->vdev_max_asize
= max_asize
;
1755 if (vd
->vdev_ashift
== 0) {
1756 vd
->vdev_ashift
= ashift
; /* use detected value */
1758 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1759 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1760 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1761 VDEV_AUX_BAD_ASHIFT
);
1762 return (SET_ERROR(EDOM
));
1766 * Detect if the alignment requirement has increased.
1767 * We don't want to make the pool unavailable, just
1768 * post an event instead.
1770 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1771 vd
->vdev_ops
->vdev_op_leaf
) {
1772 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1773 spa
, vd
, NULL
, NULL
, 0, 0);
1776 vd
->vdev_max_asize
= max_asize
;
1780 * If all children are healthy we update asize if either:
1781 * The asize has increased, due to a device expansion caused by dynamic
1782 * LUN growth or vdev replacement, and automatic expansion is enabled;
1783 * making the additional space available.
1785 * The asize has decreased, due to a device shrink usually caused by a
1786 * vdev replace with a smaller device. This ensures that calculations
1787 * based of max_asize and asize e.g. esize are always valid. It's safe
1788 * to do this as we've already validated that asize is greater than
1791 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1792 ((asize
> vd
->vdev_asize
&&
1793 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1794 (asize
< vd
->vdev_asize
)))
1795 vd
->vdev_asize
= asize
;
1797 vdev_set_min_asize(vd
);
1800 * Ensure we can issue some IO before declaring the
1801 * vdev open for business.
1803 if (vd
->vdev_ops
->vdev_op_leaf
&&
1804 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1805 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1806 VDEV_AUX_ERR_EXCEEDED
);
1811 * Track the min and max ashift values for normal data devices.
1813 * DJB - TBD these should perhaps be tracked per allocation class
1814 * (e.g. spa_min_ashift is used to round up post compression buffers)
1816 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1817 vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
&&
1818 vd
->vdev_aux
== NULL
) {
1819 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1820 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1821 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1822 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1826 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1827 * resilver. But don't do this if we are doing a reopen for a scrub,
1828 * since this would just restart the scrub we are already doing.
1830 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1831 vdev_resilver_needed(vd
, NULL
, NULL
)) {
1832 if (dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
1833 spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
1834 vdev_set_deferred_resilver(spa
, vd
);
1836 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1843 * Called once the vdevs are all opened, this routine validates the label
1844 * contents. This needs to be done before vdev_load() so that we don't
1845 * inadvertently do repair I/Os to the wrong device.
1847 * This function will only return failure if one of the vdevs indicates that it
1848 * has since been destroyed or exported. This is only possible if
1849 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1850 * will be updated but the function will return 0.
1853 vdev_validate(vdev_t
*vd
)
1855 spa_t
*spa
= vd
->vdev_spa
;
1857 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
1862 if (vdev_validate_skip
)
1865 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1866 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
1867 return (SET_ERROR(EBADF
));
1870 * If the device has already failed, or was marked offline, don't do
1871 * any further validation. Otherwise, label I/O will fail and we will
1872 * overwrite the previous state.
1874 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
1878 * If we are performing an extreme rewind, we allow for a label that
1879 * was modified at a point after the current txg.
1880 * If config lock is not held do not check for the txg. spa_sync could
1881 * be updating the vdev's label before updating spa_last_synced_txg.
1883 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
1884 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
1887 txg
= spa_last_synced_txg(spa
);
1889 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1890 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1891 VDEV_AUX_BAD_LABEL
);
1892 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
1893 "txg %llu", (u_longlong_t
)txg
);
1898 * Determine if this vdev has been split off into another
1899 * pool. If so, then refuse to open it.
1901 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1902 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1903 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1904 VDEV_AUX_SPLIT_POOL
);
1906 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
1910 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
1911 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1912 VDEV_AUX_CORRUPT_DATA
);
1914 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1915 ZPOOL_CONFIG_POOL_GUID
);
1920 * If config is not trusted then ignore the spa guid check. This is
1921 * necessary because if the machine crashed during a re-guid the new
1922 * guid might have been written to all of the vdev labels, but not the
1923 * cached config. The check will be performed again once we have the
1924 * trusted config from the MOS.
1926 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
1927 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1928 VDEV_AUX_CORRUPT_DATA
);
1930 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
1931 "match config (%llu != %llu)", (u_longlong_t
)guid
,
1932 (u_longlong_t
)spa_guid(spa
));
1936 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1937 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1941 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
1942 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1943 VDEV_AUX_CORRUPT_DATA
);
1945 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1950 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
1952 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1953 VDEV_AUX_CORRUPT_DATA
);
1955 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1956 ZPOOL_CONFIG_TOP_GUID
);
1961 * If this vdev just became a top-level vdev because its sibling was
1962 * detached, it will have adopted the parent's vdev guid -- but the
1963 * label may or may not be on disk yet. Fortunately, either version
1964 * of the label will have the same top guid, so if we're a top-level
1965 * vdev, we can safely compare to that instead.
1966 * However, if the config comes from a cachefile that failed to update
1967 * after the detach, a top-level vdev will appear as a non top-level
1968 * vdev in the config. Also relax the constraints if we perform an
1971 * If we split this vdev off instead, then we also check the
1972 * original pool's guid. We don't want to consider the vdev
1973 * corrupt if it is partway through a split operation.
1975 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
1976 boolean_t mismatch
= B_FALSE
;
1977 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
1978 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
1981 if (vd
->vdev_guid
!= top_guid
&&
1982 vd
->vdev_top
->vdev_guid
!= guid
)
1987 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1988 VDEV_AUX_CORRUPT_DATA
);
1990 vdev_dbgmsg(vd
, "vdev_validate: config guid "
1991 "doesn't match label guid");
1992 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
1993 (u_longlong_t
)vd
->vdev_guid
,
1994 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
1995 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
1996 "aux_guid %llu", (u_longlong_t
)guid
,
1997 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
2002 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
2004 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2005 VDEV_AUX_CORRUPT_DATA
);
2007 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2008 ZPOOL_CONFIG_POOL_STATE
);
2015 * If this is a verbatim import, no need to check the
2016 * state of the pool.
2018 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
2019 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
2020 state
!= POOL_STATE_ACTIVE
) {
2021 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
2022 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
2023 return (SET_ERROR(EBADF
));
2027 * If we were able to open and validate a vdev that was
2028 * previously marked permanently unavailable, clear that state
2031 if (vd
->vdev_not_present
)
2032 vd
->vdev_not_present
= 0;
2038 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2040 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2041 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2042 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2043 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2044 dvd
->vdev_path
, svd
->vdev_path
);
2045 spa_strfree(dvd
->vdev_path
);
2046 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2048 } else if (svd
->vdev_path
!= NULL
) {
2049 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2050 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2051 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2056 * Recursively copy vdev paths from one vdev to another. Source and destination
2057 * vdev trees must have same geometry otherwise return error. Intended to copy
2058 * paths from userland config into MOS config.
2061 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2063 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2064 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2065 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2068 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2069 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2070 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2071 return (SET_ERROR(EINVAL
));
2074 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2075 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2076 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2077 (u_longlong_t
)dvd
->vdev_guid
);
2078 return (SET_ERROR(EINVAL
));
2081 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2082 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2083 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2084 (u_longlong_t
)dvd
->vdev_children
);
2085 return (SET_ERROR(EINVAL
));
2088 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2089 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2090 dvd
->vdev_child
[i
]);
2095 if (svd
->vdev_ops
->vdev_op_leaf
)
2096 vdev_copy_path_impl(svd
, dvd
);
2102 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2104 ASSERT(stvd
->vdev_top
== stvd
);
2105 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2107 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2108 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2111 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2115 * The idea here is that while a vdev can shift positions within
2116 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2117 * step outside of it.
2119 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2121 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2124 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2126 vdev_copy_path_impl(vd
, dvd
);
2130 * Recursively copy vdev paths from one root vdev to another. Source and
2131 * destination vdev trees may differ in geometry. For each destination leaf
2132 * vdev, search a vdev with the same guid and top vdev id in the source.
2133 * Intended to copy paths from userland config into MOS config.
2136 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2138 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2139 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2140 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2142 for (uint64_t i
= 0; i
< children
; i
++) {
2143 vdev_copy_path_search(srvd
->vdev_child
[i
],
2144 drvd
->vdev_child
[i
]);
2149 * Close a virtual device.
2152 vdev_close(vdev_t
*vd
)
2154 vdev_t
*pvd
= vd
->vdev_parent
;
2155 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2157 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2160 * If our parent is reopening, then we are as well, unless we are
2163 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2164 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2166 vd
->vdev_ops
->vdev_op_close(vd
);
2168 vdev_cache_purge(vd
);
2171 * We record the previous state before we close it, so that if we are
2172 * doing a reopen(), we don't generate FMA ereports if we notice that
2173 * it's still faulted.
2175 vd
->vdev_prevstate
= vd
->vdev_state
;
2177 if (vd
->vdev_offline
)
2178 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2180 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2181 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2185 vdev_hold(vdev_t
*vd
)
2187 spa_t
*spa
= vd
->vdev_spa
;
2189 ASSERT(spa_is_root(spa
));
2190 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2193 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2194 vdev_hold(vd
->vdev_child
[c
]);
2196 if (vd
->vdev_ops
->vdev_op_leaf
)
2197 vd
->vdev_ops
->vdev_op_hold(vd
);
2201 vdev_rele(vdev_t
*vd
)
2203 ASSERT(spa_is_root(vd
->vdev_spa
));
2204 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2205 vdev_rele(vd
->vdev_child
[c
]);
2207 if (vd
->vdev_ops
->vdev_op_leaf
)
2208 vd
->vdev_ops
->vdev_op_rele(vd
);
2212 * Reopen all interior vdevs and any unopened leaves. We don't actually
2213 * reopen leaf vdevs which had previously been opened as they might deadlock
2214 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2215 * If the leaf has never been opened then open it, as usual.
2218 vdev_reopen(vdev_t
*vd
)
2220 spa_t
*spa
= vd
->vdev_spa
;
2222 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2224 /* set the reopening flag unless we're taking the vdev offline */
2225 vd
->vdev_reopening
= !vd
->vdev_offline
;
2227 (void) vdev_open(vd
);
2230 * Call vdev_validate() here to make sure we have the same device.
2231 * Otherwise, a device with an invalid label could be successfully
2232 * opened in response to vdev_reopen().
2235 (void) vdev_validate_aux(vd
);
2236 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2237 vd
->vdev_aux
== &spa
->spa_l2cache
&&
2238 !l2arc_vdev_present(vd
))
2239 l2arc_add_vdev(spa
, vd
);
2241 (void) vdev_validate(vd
);
2245 * Reassess parent vdev's health.
2247 vdev_propagate_state(vd
);
2251 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2256 * Normally, partial opens (e.g. of a mirror) are allowed.
2257 * For a create, however, we want to fail the request if
2258 * there are any components we can't open.
2260 error
= vdev_open(vd
);
2262 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2264 return (error
? error
: ENXIO
);
2268 * Recursively load DTLs and initialize all labels.
2270 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2271 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2272 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2281 vdev_metaslab_set_size(vdev_t
*vd
)
2283 uint64_t asize
= vd
->vdev_asize
;
2284 uint64_t ms_count
= asize
>> zfs_vdev_default_ms_shift
;
2288 * There are two dimensions to the metaslab sizing calculation:
2289 * the size of the metaslab and the count of metaslabs per vdev.
2291 * The default values used below are a good balance between memory
2292 * usage (larger metaslab size means more memory needed for loaded
2293 * metaslabs; more metaslabs means more memory needed for the
2294 * metaslab_t structs), metaslab load time (larger metaslabs take
2295 * longer to load), and metaslab sync time (more metaslabs means
2296 * more time spent syncing all of them).
2298 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2299 * The range of the dimensions are as follows:
2301 * 2^29 <= ms_size <= 2^34
2302 * 16 <= ms_count <= 131,072
2304 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2305 * at least 512MB (2^29) to minimize fragmentation effects when
2306 * testing with smaller devices. However, the count constraint
2307 * of at least 16 metaslabs will override this minimum size goal.
2309 * On the upper end of vdev sizes, we aim for a maximum metaslab
2310 * size of 16GB. However, we will cap the total count to 2^17
2311 * metaslabs to keep our memory footprint in check and let the
2312 * metaslab size grow from there if that limit is hit.
2314 * The net effect of applying above constrains is summarized below.
2316 * vdev size metaslab count
2317 * --------------|-----------------
2319 * 8GB - 100GB one per 512MB
2321 * 3TB - 2PB one per 16GB
2323 * --------------------------------
2325 * Finally, note that all of the above calculate the initial
2326 * number of metaslabs. Expanding a top-level vdev will result
2327 * in additional metaslabs being allocated making it possible
2328 * to exceed the zfs_vdev_ms_count_limit.
2331 if (ms_count
< zfs_vdev_min_ms_count
)
2332 ms_shift
= highbit64(asize
/ zfs_vdev_min_ms_count
);
2333 else if (ms_count
> zfs_vdev_default_ms_count
)
2334 ms_shift
= highbit64(asize
/ zfs_vdev_default_ms_count
);
2336 ms_shift
= zfs_vdev_default_ms_shift
;
2338 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2339 ms_shift
= SPA_MAXBLOCKSHIFT
;
2340 } else if (ms_shift
> zfs_vdev_max_ms_shift
) {
2341 ms_shift
= zfs_vdev_max_ms_shift
;
2342 /* cap the total count to constrain memory footprint */
2343 if ((asize
>> ms_shift
) > zfs_vdev_ms_count_limit
)
2344 ms_shift
= highbit64(asize
/ zfs_vdev_ms_count_limit
);
2347 vd
->vdev_ms_shift
= ms_shift
;
2348 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2352 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2354 ASSERT(vd
== vd
->vdev_top
);
2355 /* indirect vdevs don't have metaslabs or dtls */
2356 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2357 ASSERT(ISP2(flags
));
2358 ASSERT(spa_writeable(vd
->vdev_spa
));
2360 if (flags
& VDD_METASLAB
)
2361 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2363 if (flags
& VDD_DTL
)
2364 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2366 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2370 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2372 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2373 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2375 if (vd
->vdev_ops
->vdev_op_leaf
)
2376 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2382 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2383 * the vdev has less than perfect replication. There are four kinds of DTL:
2385 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2387 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2389 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2390 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2391 * txgs that was scrubbed.
2393 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2394 * persistent errors or just some device being offline.
2395 * Unlike the other three, the DTL_OUTAGE map is not generally
2396 * maintained; it's only computed when needed, typically to
2397 * determine whether a device can be detached.
2399 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2400 * either has the data or it doesn't.
2402 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2403 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2404 * if any child is less than fully replicated, then so is its parent.
2405 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2406 * comprising only those txgs which appear in 'maxfaults' or more children;
2407 * those are the txgs we don't have enough replication to read. For example,
2408 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2409 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2410 * two child DTL_MISSING maps.
2412 * It should be clear from the above that to compute the DTLs and outage maps
2413 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2414 * Therefore, that is all we keep on disk. When loading the pool, or after
2415 * a configuration change, we generate all other DTLs from first principles.
2418 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2420 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2422 ASSERT(t
< DTL_TYPES
);
2423 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2424 ASSERT(spa_writeable(vd
->vdev_spa
));
2426 mutex_enter(&vd
->vdev_dtl_lock
);
2427 if (!range_tree_contains(rt
, txg
, size
))
2428 range_tree_add(rt
, txg
, size
);
2429 mutex_exit(&vd
->vdev_dtl_lock
);
2433 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2435 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2436 boolean_t dirty
= B_FALSE
;
2438 ASSERT(t
< DTL_TYPES
);
2439 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2442 * While we are loading the pool, the DTLs have not been loaded yet.
2443 * Ignore the DTLs and try all devices. This avoids a recursive
2444 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2445 * when loading the pool (relying on the checksum to ensure that
2446 * we get the right data -- note that we while loading, we are
2447 * only reading the MOS, which is always checksummed).
2449 if (vd
->vdev_spa
->spa_load_state
!= SPA_LOAD_NONE
)
2452 mutex_enter(&vd
->vdev_dtl_lock
);
2453 if (!range_tree_is_empty(rt
))
2454 dirty
= range_tree_contains(rt
, txg
, size
);
2455 mutex_exit(&vd
->vdev_dtl_lock
);
2461 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2463 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2466 mutex_enter(&vd
->vdev_dtl_lock
);
2467 empty
= range_tree_is_empty(rt
);
2468 mutex_exit(&vd
->vdev_dtl_lock
);
2474 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2477 vdev_dtl_need_resilver(vdev_t
*vd
, uint64_t offset
, size_t psize
)
2479 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2481 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2482 vd
->vdev_ops
->vdev_op_leaf
)
2485 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, offset
, psize
));
2489 * Returns the lowest txg in the DTL range.
2492 vdev_dtl_min(vdev_t
*vd
)
2496 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2497 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2498 ASSERT0(vd
->vdev_children
);
2500 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2501 return (rs
->rs_start
- 1);
2505 * Returns the highest txg in the DTL.
2508 vdev_dtl_max(vdev_t
*vd
)
2512 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2513 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2514 ASSERT0(vd
->vdev_children
);
2516 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2517 return (rs
->rs_end
);
2521 * Determine if a resilvering vdev should remove any DTL entries from
2522 * its range. If the vdev was resilvering for the entire duration of the
2523 * scan then it should excise that range from its DTLs. Otherwise, this
2524 * vdev is considered partially resilvered and should leave its DTL
2525 * entries intact. The comment in vdev_dtl_reassess() describes how we
2529 vdev_dtl_should_excise(vdev_t
*vd
)
2531 spa_t
*spa
= vd
->vdev_spa
;
2532 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2534 ASSERT0(scn
->scn_phys
.scn_errors
);
2535 ASSERT0(vd
->vdev_children
);
2537 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2540 if (vd
->vdev_resilver_deferred
)
2543 if (vd
->vdev_resilver_txg
== 0 ||
2544 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
2548 * When a resilver is initiated the scan will assign the scn_max_txg
2549 * value to the highest txg value that exists in all DTLs. If this
2550 * device's max DTL is not part of this scan (i.e. it is not in
2551 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2554 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
2555 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
2556 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
2557 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
2564 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2565 * write operations will be issued to the pool.
2568 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
2570 spa_t
*spa
= vd
->vdev_spa
;
2574 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2576 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2577 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
2578 scrub_txg
, scrub_done
);
2580 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
2583 if (vd
->vdev_ops
->vdev_op_leaf
) {
2584 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2586 mutex_enter(&vd
->vdev_dtl_lock
);
2589 * If requested, pretend the scan completed cleanly.
2591 if (zfs_scan_ignore_errors
&& scn
)
2592 scn
->scn_phys
.scn_errors
= 0;
2595 * If we've completed a scan cleanly then determine
2596 * if this vdev should remove any DTLs. We only want to
2597 * excise regions on vdevs that were available during
2598 * the entire duration of this scan.
2600 if (scrub_txg
!= 0 &&
2601 (spa
->spa_scrub_started
||
2602 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
2603 vdev_dtl_should_excise(vd
)) {
2605 * We completed a scrub up to scrub_txg. If we
2606 * did it without rebooting, then the scrub dtl
2607 * will be valid, so excise the old region and
2608 * fold in the scrub dtl. Otherwise, leave the
2609 * dtl as-is if there was an error.
2611 * There's little trick here: to excise the beginning
2612 * of the DTL_MISSING map, we put it into a reference
2613 * tree and then add a segment with refcnt -1 that
2614 * covers the range [0, scrub_txg). This means
2615 * that each txg in that range has refcnt -1 or 0.
2616 * We then add DTL_SCRUB with a refcnt of 2, so that
2617 * entries in the range [0, scrub_txg) will have a
2618 * positive refcnt -- either 1 or 2. We then convert
2619 * the reference tree into the new DTL_MISSING map.
2621 space_reftree_create(&reftree
);
2622 space_reftree_add_map(&reftree
,
2623 vd
->vdev_dtl
[DTL_MISSING
], 1);
2624 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
2625 space_reftree_add_map(&reftree
,
2626 vd
->vdev_dtl
[DTL_SCRUB
], 2);
2627 space_reftree_generate_map(&reftree
,
2628 vd
->vdev_dtl
[DTL_MISSING
], 1);
2629 space_reftree_destroy(&reftree
);
2631 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
2632 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2633 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
2635 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
2636 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
2637 if (!vdev_readable(vd
))
2638 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
2640 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2641 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2644 * If the vdev was resilvering and no longer has any
2645 * DTLs then reset its resilvering flag and dirty
2646 * the top level so that we persist the change.
2648 if (txg
!= 0 && vd
->vdev_resilver_txg
!= 0 &&
2649 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2650 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
2651 vd
->vdev_resilver_txg
= 0;
2652 vdev_config_dirty(vd
->vdev_top
);
2655 mutex_exit(&vd
->vdev_dtl_lock
);
2658 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2662 mutex_enter(&vd
->vdev_dtl_lock
);
2663 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2664 /* account for child's outage in parent's missing map */
2665 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2667 continue; /* leaf vdevs only */
2668 if (t
== DTL_PARTIAL
)
2669 minref
= 1; /* i.e. non-zero */
2670 else if (vd
->vdev_nparity
!= 0)
2671 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2673 minref
= vd
->vdev_children
; /* any kind of mirror */
2674 space_reftree_create(&reftree
);
2675 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2676 vdev_t
*cvd
= vd
->vdev_child
[c
];
2677 mutex_enter(&cvd
->vdev_dtl_lock
);
2678 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2679 mutex_exit(&cvd
->vdev_dtl_lock
);
2681 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2682 space_reftree_destroy(&reftree
);
2684 mutex_exit(&vd
->vdev_dtl_lock
);
2688 vdev_dtl_load(vdev_t
*vd
)
2690 spa_t
*spa
= vd
->vdev_spa
;
2691 objset_t
*mos
= spa
->spa_meta_objset
;
2694 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2695 ASSERT(vdev_is_concrete(vd
));
2697 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2698 vd
->vdev_dtl_object
, 0, -1ULL, 0);
2701 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2703 mutex_enter(&vd
->vdev_dtl_lock
);
2704 error
= space_map_load(vd
->vdev_dtl_sm
,
2705 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2706 mutex_exit(&vd
->vdev_dtl_lock
);
2711 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2712 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2721 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
2723 spa_t
*spa
= vd
->vdev_spa
;
2724 objset_t
*mos
= spa
->spa_meta_objset
;
2725 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
2728 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
2731 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
2732 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
2733 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
2735 ASSERT(string
!= NULL
);
2736 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
2737 1, strlen(string
) + 1, string
, tx
));
2739 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
2740 spa_activate_allocation_classes(spa
, tx
);
2745 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2747 spa_t
*spa
= vd
->vdev_spa
;
2749 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2750 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2755 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2757 spa_t
*spa
= vd
->vdev_spa
;
2758 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2759 DMU_OT_NONE
, 0, tx
);
2762 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2769 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2771 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2772 vd
->vdev_ops
!= &vdev_missing_ops
&&
2773 vd
->vdev_ops
!= &vdev_root_ops
&&
2774 !vd
->vdev_top
->vdev_removing
) {
2775 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2776 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2778 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2779 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2780 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
2781 vdev_zap_allocation_data(vd
, tx
);
2785 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2786 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2791 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2793 spa_t
*spa
= vd
->vdev_spa
;
2794 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2795 objset_t
*mos
= spa
->spa_meta_objset
;
2796 range_tree_t
*rtsync
;
2798 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2800 ASSERT(vdev_is_concrete(vd
));
2801 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2803 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2805 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2806 mutex_enter(&vd
->vdev_dtl_lock
);
2807 space_map_free(vd
->vdev_dtl_sm
, tx
);
2808 space_map_close(vd
->vdev_dtl_sm
);
2809 vd
->vdev_dtl_sm
= NULL
;
2810 mutex_exit(&vd
->vdev_dtl_lock
);
2813 * We only destroy the leaf ZAP for detached leaves or for
2814 * removed log devices. Removed data devices handle leaf ZAP
2815 * cleanup later, once cancellation is no longer possible.
2817 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2818 vd
->vdev_top
->vdev_islog
)) {
2819 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2820 vd
->vdev_leaf_zap
= 0;
2827 if (vd
->vdev_dtl_sm
== NULL
) {
2828 uint64_t new_object
;
2830 new_object
= space_map_alloc(mos
, vdev_dtl_sm_blksz
, tx
);
2831 VERIFY3U(new_object
, !=, 0);
2833 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2835 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2838 rtsync
= range_tree_create(NULL
, NULL
);
2840 mutex_enter(&vd
->vdev_dtl_lock
);
2841 range_tree_walk(rt
, range_tree_add
, rtsync
);
2842 mutex_exit(&vd
->vdev_dtl_lock
);
2844 space_map_truncate(vd
->vdev_dtl_sm
, vdev_dtl_sm_blksz
, tx
);
2845 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
2846 range_tree_vacate(rtsync
, NULL
, NULL
);
2848 range_tree_destroy(rtsync
);
2851 * If the object for the space map has changed then dirty
2852 * the top level so that we update the config.
2854 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2855 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
2856 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
2857 (u_longlong_t
)object
,
2858 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
2859 vdev_config_dirty(vd
->vdev_top
);
2866 * Determine whether the specified vdev can be offlined/detached/removed
2867 * without losing data.
2870 vdev_dtl_required(vdev_t
*vd
)
2872 spa_t
*spa
= vd
->vdev_spa
;
2873 vdev_t
*tvd
= vd
->vdev_top
;
2874 uint8_t cant_read
= vd
->vdev_cant_read
;
2877 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2879 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2883 * Temporarily mark the device as unreadable, and then determine
2884 * whether this results in any DTL outages in the top-level vdev.
2885 * If not, we can safely offline/detach/remove the device.
2887 vd
->vdev_cant_read
= B_TRUE
;
2888 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2889 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2890 vd
->vdev_cant_read
= cant_read
;
2891 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2893 if (!required
&& zio_injection_enabled
)
2894 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2900 * Determine if resilver is needed, and if so the txg range.
2903 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2905 boolean_t needed
= B_FALSE
;
2906 uint64_t thismin
= UINT64_MAX
;
2907 uint64_t thismax
= 0;
2909 if (vd
->vdev_children
== 0) {
2910 mutex_enter(&vd
->vdev_dtl_lock
);
2911 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2912 vdev_writeable(vd
)) {
2914 thismin
= vdev_dtl_min(vd
);
2915 thismax
= vdev_dtl_max(vd
);
2918 mutex_exit(&vd
->vdev_dtl_lock
);
2920 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2921 vdev_t
*cvd
= vd
->vdev_child
[c
];
2922 uint64_t cmin
, cmax
;
2924 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2925 thismin
= MIN(thismin
, cmin
);
2926 thismax
= MAX(thismax
, cmax
);
2932 if (needed
&& minp
) {
2940 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
2941 * will contain either the checkpoint spacemap object or zero if none exists.
2942 * All other errors are returned to the caller.
2945 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
2947 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
2949 if (vd
->vdev_top_zap
== 0) {
2954 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
2955 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
2956 if (error
== ENOENT
) {
2965 vdev_load(vdev_t
*vd
)
2970 * Recursively load all children.
2972 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2973 error
= vdev_load(vd
->vdev_child
[c
]);
2979 vdev_set_deflate_ratio(vd
);
2982 * On spa_load path, grab the allocation bias from our zap
2984 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
2985 spa_t
*spa
= vd
->vdev_spa
;
2988 if (zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
2989 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
2991 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
2992 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
2997 * If this is a top-level vdev, initialize its metaslabs.
2999 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
3000 vdev_metaslab_group_create(vd
);
3002 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
3003 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3004 VDEV_AUX_CORRUPT_DATA
);
3005 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
3006 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
3007 (u_longlong_t
)vd
->vdev_asize
);
3008 return (SET_ERROR(ENXIO
));
3009 } else if ((error
= vdev_metaslab_init(vd
, 0)) != 0) {
3010 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
3011 "[error=%d]", error
);
3012 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3013 VDEV_AUX_CORRUPT_DATA
);
3017 uint64_t checkpoint_sm_obj
;
3018 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
3019 if (error
== 0 && checkpoint_sm_obj
!= 0) {
3020 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3021 ASSERT(vd
->vdev_asize
!= 0);
3022 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
3024 if ((error
= space_map_open(&vd
->vdev_checkpoint_sm
,
3025 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
3026 vd
->vdev_ashift
))) {
3027 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3028 "failed for checkpoint spacemap (obj %llu) "
3030 (u_longlong_t
)checkpoint_sm_obj
, error
);
3033 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3036 * Since the checkpoint_sm contains free entries
3037 * exclusively we can use space_map_allocated() to
3038 * indicate the cumulative checkpointed space that
3041 vd
->vdev_stat
.vs_checkpoint_space
=
3042 -space_map_allocated(vd
->vdev_checkpoint_sm
);
3043 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3044 vd
->vdev_stat
.vs_checkpoint_space
;
3045 } else if (error
!= 0) {
3046 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3047 "checkpoint space map object from vdev ZAP "
3048 "[error=%d]", error
);
3054 * If this is a leaf vdev, load its DTL.
3056 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3057 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3058 VDEV_AUX_CORRUPT_DATA
);
3059 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3060 "[error=%d]", error
);
3064 uint64_t obsolete_sm_object
;
3065 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3066 if (error
== 0 && obsolete_sm_object
!= 0) {
3067 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3068 ASSERT(vd
->vdev_asize
!= 0);
3069 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3071 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3072 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3073 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3074 VDEV_AUX_CORRUPT_DATA
);
3075 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3076 "obsolete spacemap (obj %llu) [error=%d]",
3077 (u_longlong_t
)obsolete_sm_object
, error
);
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 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3172 vdev_destroy_spacemaps(vd
, tx
);
3173 if (vd
->vdev_top_zap
!= 0) {
3174 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3175 vd
->vdev_top_zap
= 0;
3182 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3185 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3187 ASSERT(vdev_is_concrete(vd
));
3189 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
3191 metaslab_sync_done(msp
, txg
);
3194 metaslab_sync_reassess(vd
->vdev_mg
);
3198 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3200 spa_t
*spa
= vd
->vdev_spa
;
3204 ASSERT3U(txg
, ==, spa
->spa_syncing_txg
);
3205 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3206 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3207 ASSERT(vd
->vdev_removing
||
3208 vd
->vdev_ops
== &vdev_indirect_ops
);
3210 vdev_indirect_sync_obsolete(vd
, tx
);
3213 * If the vdev is indirect, it can't have dirty
3214 * metaslabs or DTLs.
3216 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3217 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3218 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3224 ASSERT(vdev_is_concrete(vd
));
3226 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3227 !vd
->vdev_removing
) {
3228 ASSERT(vd
== vd
->vdev_top
);
3229 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3230 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3231 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3232 ASSERT(vd
->vdev_ms_array
!= 0);
3233 vdev_config_dirty(vd
);
3236 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3237 metaslab_sync(msp
, txg
);
3238 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3241 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3242 vdev_dtl_sync(lvd
, txg
);
3245 * If this is an empty log device being removed, destroy the
3246 * metadata associated with it.
3248 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3249 vdev_remove_empty_log(vd
, txg
);
3251 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3256 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3258 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3262 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3263 * not be opened, and no I/O is attempted.
3266 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3270 spa_vdev_state_enter(spa
, SCL_NONE
);
3272 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3273 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3275 if (!vd
->vdev_ops
->vdev_op_leaf
)
3276 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3281 * If user did a 'zpool offline -f' then make the fault persist across
3284 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3286 * There are two kinds of forced faults: temporary and
3287 * persistent. Temporary faults go away at pool import, while
3288 * persistent faults stay set. Both types of faults can be
3289 * cleared with a zpool clear.
3291 * We tell if a vdev is persistently faulted by looking at the
3292 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3293 * import then it's a persistent fault. Otherwise, it's
3294 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3295 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3296 * tells vdev_config_generate() (which gets run later) to set
3297 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3299 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3300 vd
->vdev_tmpoffline
= B_FALSE
;
3301 aux
= VDEV_AUX_EXTERNAL
;
3303 vd
->vdev_tmpoffline
= B_TRUE
;
3307 * We don't directly use the aux state here, but if we do a
3308 * vdev_reopen(), we need this value to be present to remember why we
3311 vd
->vdev_label_aux
= aux
;
3314 * Faulted state takes precedence over degraded.
3316 vd
->vdev_delayed_close
= B_FALSE
;
3317 vd
->vdev_faulted
= 1ULL;
3318 vd
->vdev_degraded
= 0ULL;
3319 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
3322 * If this device has the only valid copy of the data, then
3323 * back off and simply mark the vdev as degraded instead.
3325 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
3326 vd
->vdev_degraded
= 1ULL;
3327 vd
->vdev_faulted
= 0ULL;
3330 * If we reopen the device and it's not dead, only then do we
3335 if (vdev_readable(vd
))
3336 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
3339 return (spa_vdev_state_exit(spa
, vd
, 0));
3343 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3344 * user that something is wrong. The vdev continues to operate as normal as far
3345 * as I/O is concerned.
3348 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3352 spa_vdev_state_enter(spa
, SCL_NONE
);
3354 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3355 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3357 if (!vd
->vdev_ops
->vdev_op_leaf
)
3358 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3361 * If the vdev is already faulted, then don't do anything.
3363 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
3364 return (spa_vdev_state_exit(spa
, NULL
, 0));
3366 vd
->vdev_degraded
= 1ULL;
3367 if (!vdev_is_dead(vd
))
3368 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
3371 return (spa_vdev_state_exit(spa
, vd
, 0));
3375 * Online the given vdev.
3377 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3378 * spare device should be detached when the device finishes resilvering.
3379 * Second, the online should be treated like a 'test' online case, so no FMA
3380 * events are generated if the device fails to open.
3383 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
3385 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
3386 boolean_t wasoffline
;
3387 vdev_state_t oldstate
;
3389 spa_vdev_state_enter(spa
, SCL_NONE
);
3391 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3392 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3394 if (!vd
->vdev_ops
->vdev_op_leaf
)
3395 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3397 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
3398 oldstate
= vd
->vdev_state
;
3401 vd
->vdev_offline
= B_FALSE
;
3402 vd
->vdev_tmpoffline
= B_FALSE
;
3403 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
3404 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
3406 /* XXX - L2ARC 1.0 does not support expansion */
3407 if (!vd
->vdev_aux
) {
3408 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3409 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
3410 spa
->spa_autoexpand
);
3411 vd
->vdev_expansion_time
= gethrestime_sec();
3415 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
3417 if (!vd
->vdev_aux
) {
3418 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3419 pvd
->vdev_expanding
= B_FALSE
;
3423 *newstate
= vd
->vdev_state
;
3424 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
3425 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
3426 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3427 vd
->vdev_parent
->vdev_child
[0] == vd
)
3428 vd
->vdev_unspare
= B_TRUE
;
3430 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
3432 /* XXX - L2ARC 1.0 does not support expansion */
3434 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
3435 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3438 /* Restart initializing if necessary */
3439 mutex_enter(&vd
->vdev_initialize_lock
);
3440 if (vdev_writeable(vd
) &&
3441 vd
->vdev_initialize_thread
== NULL
&&
3442 vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) {
3443 (void) vdev_initialize(vd
);
3445 mutex_exit(&vd
->vdev_initialize_lock
);
3448 (oldstate
< VDEV_STATE_DEGRADED
&&
3449 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
3450 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
3452 return (spa_vdev_state_exit(spa
, vd
, 0));
3456 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3460 uint64_t generation
;
3461 metaslab_group_t
*mg
;
3464 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3466 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3467 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3469 if (!vd
->vdev_ops
->vdev_op_leaf
)
3470 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3474 generation
= spa
->spa_config_generation
+ 1;
3477 * If the device isn't already offline, try to offline it.
3479 if (!vd
->vdev_offline
) {
3481 * If this device has the only valid copy of some data,
3482 * don't allow it to be offlined. Log devices are always
3485 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3486 vdev_dtl_required(vd
))
3487 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3490 * If the top-level is a slog and it has had allocations
3491 * then proceed. We check that the vdev's metaslab group
3492 * is not NULL since it's possible that we may have just
3493 * added this vdev but not yet initialized its metaslabs.
3495 if (tvd
->vdev_islog
&& mg
!= NULL
) {
3497 * Prevent any future allocations.
3499 metaslab_group_passivate(mg
);
3500 (void) spa_vdev_state_exit(spa
, vd
, 0);
3502 error
= spa_reset_logs(spa
);
3505 * If the log device was successfully reset but has
3506 * checkpointed data, do not offline it.
3509 tvd
->vdev_checkpoint_sm
!= NULL
) {
3510 ASSERT3U(space_map_allocated(
3511 tvd
->vdev_checkpoint_sm
), !=, 0);
3512 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
3515 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3518 * Check to see if the config has changed.
3520 if (error
|| generation
!= spa
->spa_config_generation
) {
3521 metaslab_group_activate(mg
);
3523 return (spa_vdev_state_exit(spa
,
3525 (void) spa_vdev_state_exit(spa
, vd
, 0);
3528 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
3532 * Offline this device and reopen its top-level vdev.
3533 * If the top-level vdev is a log device then just offline
3534 * it. Otherwise, if this action results in the top-level
3535 * vdev becoming unusable, undo it and fail the request.
3537 vd
->vdev_offline
= B_TRUE
;
3540 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3541 vdev_is_dead(tvd
)) {
3542 vd
->vdev_offline
= B_FALSE
;
3544 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3548 * Add the device back into the metaslab rotor so that
3549 * once we online the device it's open for business.
3551 if (tvd
->vdev_islog
&& mg
!= NULL
)
3552 metaslab_group_activate(mg
);
3555 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
3557 return (spa_vdev_state_exit(spa
, vd
, 0));
3561 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3565 mutex_enter(&spa
->spa_vdev_top_lock
);
3566 error
= vdev_offline_locked(spa
, guid
, flags
);
3567 mutex_exit(&spa
->spa_vdev_top_lock
);
3573 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3574 * vdev_offline(), we assume the spa config is locked. We also clear all
3575 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3578 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
3580 vdev_t
*rvd
= spa
->spa_root_vdev
;
3582 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3587 vd
->vdev_stat
.vs_read_errors
= 0;
3588 vd
->vdev_stat
.vs_write_errors
= 0;
3589 vd
->vdev_stat
.vs_checksum_errors
= 0;
3590 vd
->vdev_stat
.vs_slow_ios
= 0;
3592 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3593 vdev_clear(spa
, vd
->vdev_child
[c
]);
3596 * It makes no sense to "clear" an indirect vdev.
3598 if (!vdev_is_concrete(vd
))
3602 * If we're in the FAULTED state or have experienced failed I/O, then
3603 * clear the persistent state and attempt to reopen the device. We
3604 * also mark the vdev config dirty, so that the new faulted state is
3605 * written out to disk.
3607 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
3608 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
3610 * When reopening in response to a clear event, it may be due to
3611 * a fmadm repair request. In this case, if the device is
3612 * still broken, we want to still post the ereport again.
3614 vd
->vdev_forcefault
= B_TRUE
;
3616 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
3617 vd
->vdev_cant_read
= B_FALSE
;
3618 vd
->vdev_cant_write
= B_FALSE
;
3619 vd
->vdev_stat
.vs_aux
= 0;
3621 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
3623 vd
->vdev_forcefault
= B_FALSE
;
3625 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
3626 vdev_state_dirty(vd
->vdev_top
);
3628 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
)) {
3629 if (dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3630 spa_feature_is_enabled(spa
,
3631 SPA_FEATURE_RESILVER_DEFER
))
3632 vdev_set_deferred_resilver(spa
, vd
);
3634 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3637 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
3641 * When clearing a FMA-diagnosed fault, we always want to
3642 * unspare the device, as we assume that the original spare was
3643 * done in response to the FMA fault.
3645 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
3646 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3647 vd
->vdev_parent
->vdev_child
[0] == vd
)
3648 vd
->vdev_unspare
= B_TRUE
;
3652 vdev_is_dead(vdev_t
*vd
)
3655 * Holes and missing devices are always considered "dead".
3656 * This simplifies the code since we don't have to check for
3657 * these types of devices in the various code paths.
3658 * Instead we rely on the fact that we skip over dead devices
3659 * before issuing I/O to them.
3661 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
3662 vd
->vdev_ops
== &vdev_hole_ops
||
3663 vd
->vdev_ops
== &vdev_missing_ops
);
3667 vdev_readable(vdev_t
*vd
)
3669 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
3673 vdev_writeable(vdev_t
*vd
)
3675 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
3676 vdev_is_concrete(vd
));
3680 vdev_allocatable(vdev_t
*vd
)
3682 uint64_t state
= vd
->vdev_state
;
3685 * We currently allow allocations from vdevs which may be in the
3686 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3687 * fails to reopen then we'll catch it later when we're holding
3688 * the proper locks. Note that we have to get the vdev state
3689 * in a local variable because although it changes atomically,
3690 * we're asking two separate questions about it.
3692 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
3693 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
3694 vd
->vdev_mg
->mg_initialized
);
3698 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
3700 ASSERT(zio
->io_vd
== vd
);
3702 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
3705 if (zio
->io_type
== ZIO_TYPE_READ
)
3706 return (!vd
->vdev_cant_read
);
3708 if (zio
->io_type
== ZIO_TYPE_WRITE
)
3709 return (!vd
->vdev_cant_write
);
3715 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
3718 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3719 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
3720 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
3723 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
3727 * Get extended stats
3730 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
3733 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3734 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
3735 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
3737 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
3738 vsx
->vsx_total_histo
[t
][b
] +=
3739 cvsx
->vsx_total_histo
[t
][b
];
3743 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
3744 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
3745 vsx
->vsx_queue_histo
[t
][b
] +=
3746 cvsx
->vsx_queue_histo
[t
][b
];
3748 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
3749 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
3751 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
3752 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
3754 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
3755 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
3761 vdev_is_spacemap_addressable(vdev_t
*vd
)
3763 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_SPACEMAP_V2
))
3767 * If double-word space map entries are not enabled we assume
3768 * 47 bits of the space map entry are dedicated to the entry's
3769 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
3770 * to calculate the maximum address that can be described by a
3771 * space map entry for the given device.
3773 uint64_t shift
= vd
->vdev_ashift
+ SM_OFFSET_BITS
;
3775 if (shift
>= 63) /* detect potential overflow */
3778 return (vd
->vdev_asize
< (1ULL << shift
));
3782 * Get statistics for the given vdev.
3785 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3789 * If we're getting stats on the root vdev, aggregate the I/O counts
3790 * over all top-level vdevs (i.e. the direct children of the root).
3792 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3794 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
3795 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
3798 memset(vsx
, 0, sizeof (*vsx
));
3800 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3801 vdev_t
*cvd
= vd
->vdev_child
[c
];
3802 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
3803 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
3805 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
3807 vdev_get_child_stat(cvd
, vs
, cvs
);
3809 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
3814 * We're a leaf. Just copy our ZIO active queue stats in. The
3815 * other leaf stats are updated in vdev_stat_update().
3820 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
3822 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
3823 vsx
->vsx_active_queue
[t
] =
3824 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
3825 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
3826 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
3832 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3834 vdev_t
*tvd
= vd
->vdev_top
;
3835 mutex_enter(&vd
->vdev_stat_lock
);
3837 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
3838 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
3839 vs
->vs_state
= vd
->vdev_state
;
3840 vs
->vs_rsize
= vdev_get_min_asize(vd
);
3841 if (vd
->vdev_ops
->vdev_op_leaf
) {
3842 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
3843 VDEV_LABEL_END_SIZE
;
3845 * Report intializing progress. Since we don't
3846 * have the initializing locks held, this is only
3847 * an estimate (although a fairly accurate one).
3849 vs
->vs_initialize_bytes_done
=
3850 vd
->vdev_initialize_bytes_done
;
3851 vs
->vs_initialize_bytes_est
=
3852 vd
->vdev_initialize_bytes_est
;
3853 vs
->vs_initialize_state
= vd
->vdev_initialize_state
;
3854 vs
->vs_initialize_action_time
=
3855 vd
->vdev_initialize_action_time
;
3858 * Report expandable space on top-level, non-auxillary devices
3859 * only. The expandable space is reported in terms of metaslab
3860 * sized units since that determines how much space the pool
3863 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
3864 vs
->vs_esize
= P2ALIGN(
3865 vd
->vdev_max_asize
- vd
->vdev_asize
,
3866 1ULL << tvd
->vdev_ms_shift
);
3868 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
3869 vdev_is_concrete(vd
)) {
3870 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
3871 vd
->vdev_mg
->mg_fragmentation
: 0;
3873 if (vd
->vdev_ops
->vdev_op_leaf
)
3874 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
3877 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
3878 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
3879 mutex_exit(&vd
->vdev_stat_lock
);
3883 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
3885 return (vdev_get_stats_ex(vd
, vs
, NULL
));
3889 vdev_clear_stats(vdev_t
*vd
)
3891 mutex_enter(&vd
->vdev_stat_lock
);
3892 vd
->vdev_stat
.vs_space
= 0;
3893 vd
->vdev_stat
.vs_dspace
= 0;
3894 vd
->vdev_stat
.vs_alloc
= 0;
3895 mutex_exit(&vd
->vdev_stat_lock
);
3899 vdev_scan_stat_init(vdev_t
*vd
)
3901 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3903 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3904 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3906 mutex_enter(&vd
->vdev_stat_lock
);
3907 vs
->vs_scan_processed
= 0;
3908 mutex_exit(&vd
->vdev_stat_lock
);
3912 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3914 spa_t
*spa
= zio
->io_spa
;
3915 vdev_t
*rvd
= spa
->spa_root_vdev
;
3916 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3918 uint64_t txg
= zio
->io_txg
;
3919 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3920 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
3921 zio_type_t type
= zio
->io_type
;
3922 int flags
= zio
->io_flags
;
3925 * If this i/o is a gang leader, it didn't do any actual work.
3927 if (zio
->io_gang_tree
)
3930 if (zio
->io_error
== 0) {
3932 * If this is a root i/o, don't count it -- we've already
3933 * counted the top-level vdevs, and vdev_get_stats() will
3934 * aggregate them when asked. This reduces contention on
3935 * the root vdev_stat_lock and implicitly handles blocks
3936 * that compress away to holes, for which there is no i/o.
3937 * (Holes never create vdev children, so all the counters
3938 * remain zero, which is what we want.)
3940 * Note: this only applies to successful i/o (io_error == 0)
3941 * because unlike i/o counts, errors are not additive.
3942 * When reading a ditto block, for example, failure of
3943 * one top-level vdev does not imply a root-level error.
3948 ASSERT(vd
== zio
->io_vd
);
3950 if (flags
& ZIO_FLAG_IO_BYPASS
)
3953 mutex_enter(&vd
->vdev_stat_lock
);
3955 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3956 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3957 dsl_scan_phys_t
*scn_phys
=
3958 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3959 uint64_t *processed
= &scn_phys
->scn_processed
;
3962 if (vd
->vdev_ops
->vdev_op_leaf
)
3963 atomic_add_64(processed
, psize
);
3964 vs
->vs_scan_processed
+= psize
;
3967 if (flags
& ZIO_FLAG_SELF_HEAL
)
3968 vs
->vs_self_healed
+= psize
;
3972 * The bytes/ops/histograms are recorded at the leaf level and
3973 * aggregated into the higher level vdevs in vdev_get_stats().
3975 if (vd
->vdev_ops
->vdev_op_leaf
&&
3976 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3979 vs
->vs_bytes
[type
] += psize
;
3981 if (flags
& ZIO_FLAG_DELEGATED
) {
3982 vsx
->vsx_agg_histo
[zio
->io_priority
]
3983 [RQ_HISTO(zio
->io_size
)]++;
3985 vsx
->vsx_ind_histo
[zio
->io_priority
]
3986 [RQ_HISTO(zio
->io_size
)]++;
3989 if (zio
->io_delta
&& zio
->io_delay
) {
3990 vsx
->vsx_queue_histo
[zio
->io_priority
]
3991 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3992 vsx
->vsx_disk_histo
[type
]
3993 [L_HISTO(zio
->io_delay
)]++;
3994 vsx
->vsx_total_histo
[type
]
3995 [L_HISTO(zio
->io_delta
)]++;
3999 mutex_exit(&vd
->vdev_stat_lock
);
4003 if (flags
& ZIO_FLAG_SPECULATIVE
)
4007 * If this is an I/O error that is going to be retried, then ignore the
4008 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4009 * hard errors, when in reality they can happen for any number of
4010 * innocuous reasons (bus resets, MPxIO link failure, etc).
4012 if (zio
->io_error
== EIO
&&
4013 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4017 * Intent logs writes won't propagate their error to the root
4018 * I/O so don't mark these types of failures as pool-level
4021 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4024 mutex_enter(&vd
->vdev_stat_lock
);
4025 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
4026 if (zio
->io_error
== ECKSUM
)
4027 vs
->vs_checksum_errors
++;
4029 vs
->vs_read_errors
++;
4031 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
4032 vs
->vs_write_errors
++;
4033 mutex_exit(&vd
->vdev_stat_lock
);
4035 if (spa
->spa_load_state
== SPA_LOAD_NONE
&&
4036 type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4037 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4038 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4039 spa
->spa_claiming
)) {
4041 * This is either a normal write (not a repair), or it's
4042 * a repair induced by the scrub thread, or it's a repair
4043 * made by zil_claim() during spa_load() in the first txg.
4044 * In the normal case, we commit the DTL change in the same
4045 * txg as the block was born. In the scrub-induced repair
4046 * case, we know that scrubs run in first-pass syncing context,
4047 * so we commit the DTL change in spa_syncing_txg(spa).
4048 * In the zil_claim() case, we commit in spa_first_txg(spa).
4050 * We currently do not make DTL entries for failed spontaneous
4051 * self-healing writes triggered by normal (non-scrubbing)
4052 * reads, because we have no transactional context in which to
4053 * do so -- and it's not clear that it'd be desirable anyway.
4055 if (vd
->vdev_ops
->vdev_op_leaf
) {
4056 uint64_t commit_txg
= txg
;
4057 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4058 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4059 ASSERT(spa_sync_pass(spa
) == 1);
4060 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4061 commit_txg
= spa_syncing_txg(spa
);
4062 } else if (spa
->spa_claiming
) {
4063 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4064 commit_txg
= spa_first_txg(spa
);
4066 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4067 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4069 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4070 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4071 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4074 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4079 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4081 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4082 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4084 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4088 * Update the in-core space usage stats for this vdev and the root vdev.
4091 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4092 int64_t space_delta
)
4094 int64_t dspace_delta
;
4095 spa_t
*spa
= vd
->vdev_spa
;
4096 vdev_t
*rvd
= spa
->spa_root_vdev
;
4098 ASSERT(vd
== vd
->vdev_top
);
4101 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4102 * factor. We must calculate this here and not at the root vdev
4103 * because the root vdev's psize-to-asize is simply the max of its
4104 * childrens', thus not accurate enough for us.
4106 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4108 mutex_enter(&vd
->vdev_stat_lock
);
4109 /* ensure we won't underflow */
4110 if (alloc_delta
< 0) {
4111 ASSERT3U(vd
->vdev_stat
.vs_alloc
, >=, -alloc_delta
);
4114 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4115 vd
->vdev_stat
.vs_space
+= space_delta
;
4116 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4117 mutex_exit(&vd
->vdev_stat_lock
);
4119 /* every class but log contributes to root space stats */
4120 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4121 ASSERT(!vd
->vdev_isl2cache
);
4122 mutex_enter(&rvd
->vdev_stat_lock
);
4123 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4124 rvd
->vdev_stat
.vs_space
+= space_delta
;
4125 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4126 mutex_exit(&rvd
->vdev_stat_lock
);
4128 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4132 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4133 * so that it will be written out next time the vdev configuration is synced.
4134 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4137 vdev_config_dirty(vdev_t
*vd
)
4139 spa_t
*spa
= vd
->vdev_spa
;
4140 vdev_t
*rvd
= spa
->spa_root_vdev
;
4143 ASSERT(spa_writeable(spa
));
4146 * If this is an aux vdev (as with l2cache and spare devices), then we
4147 * update the vdev config manually and set the sync flag.
4149 if (vd
->vdev_aux
!= NULL
) {
4150 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
4154 for (c
= 0; c
< sav
->sav_count
; c
++) {
4155 if (sav
->sav_vdevs
[c
] == vd
)
4159 if (c
== sav
->sav_count
) {
4161 * We're being removed. There's nothing more to do.
4163 ASSERT(sav
->sav_sync
== B_TRUE
);
4167 sav
->sav_sync
= B_TRUE
;
4169 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
4170 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
4171 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
4172 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
4178 * Setting the nvlist in the middle if the array is a little
4179 * sketchy, but it will work.
4181 nvlist_free(aux
[c
]);
4182 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
4188 * The dirty list is protected by the SCL_CONFIG lock. The caller
4189 * must either hold SCL_CONFIG as writer, or must be the sync thread
4190 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4191 * so this is sufficient to ensure mutual exclusion.
4193 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4194 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4195 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4198 for (c
= 0; c
< rvd
->vdev_children
; c
++)
4199 vdev_config_dirty(rvd
->vdev_child
[c
]);
4201 ASSERT(vd
== vd
->vdev_top
);
4203 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
4204 vdev_is_concrete(vd
)) {
4205 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
4211 vdev_config_clean(vdev_t
*vd
)
4213 spa_t
*spa
= vd
->vdev_spa
;
4215 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4216 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4217 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4219 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
4220 list_remove(&spa
->spa_config_dirty_list
, vd
);
4224 * Mark a top-level vdev's state as dirty, so that the next pass of
4225 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4226 * the state changes from larger config changes because they require
4227 * much less locking, and are often needed for administrative actions.
4230 vdev_state_dirty(vdev_t
*vd
)
4232 spa_t
*spa
= vd
->vdev_spa
;
4234 ASSERT(spa_writeable(spa
));
4235 ASSERT(vd
== vd
->vdev_top
);
4238 * The state list is protected by the SCL_STATE lock. The caller
4239 * must either hold SCL_STATE as writer, or must be the sync thread
4240 * (which holds SCL_STATE as reader). There's only one sync thread,
4241 * so this is sufficient to ensure mutual exclusion.
4243 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4244 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4245 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4247 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
4248 vdev_is_concrete(vd
))
4249 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
4253 vdev_state_clean(vdev_t
*vd
)
4255 spa_t
*spa
= vd
->vdev_spa
;
4257 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4258 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4259 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4261 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
4262 list_remove(&spa
->spa_state_dirty_list
, vd
);
4266 * Propagate vdev state up from children to parent.
4269 vdev_propagate_state(vdev_t
*vd
)
4271 spa_t
*spa
= vd
->vdev_spa
;
4272 vdev_t
*rvd
= spa
->spa_root_vdev
;
4273 int degraded
= 0, faulted
= 0;
4277 if (vd
->vdev_children
> 0) {
4278 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4279 child
= vd
->vdev_child
[c
];
4282 * Don't factor holes or indirect vdevs into the
4285 if (!vdev_is_concrete(child
))
4288 if (!vdev_readable(child
) ||
4289 (!vdev_writeable(child
) && spa_writeable(spa
))) {
4291 * Root special: if there is a top-level log
4292 * device, treat the root vdev as if it were
4295 if (child
->vdev_islog
&& vd
== rvd
)
4299 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
4303 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
4307 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
4310 * Root special: if there is a top-level vdev that cannot be
4311 * opened due to corrupted metadata, then propagate the root
4312 * vdev's aux state as 'corrupt' rather than 'insufficient
4315 if (corrupted
&& vd
== rvd
&&
4316 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
4317 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
4318 VDEV_AUX_CORRUPT_DATA
);
4321 if (vd
->vdev_parent
)
4322 vdev_propagate_state(vd
->vdev_parent
);
4326 * Set a vdev's state. If this is during an open, we don't update the parent
4327 * state, because we're in the process of opening children depth-first.
4328 * Otherwise, we propagate the change to the parent.
4330 * If this routine places a device in a faulted state, an appropriate ereport is
4334 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
4336 uint64_t save_state
;
4337 spa_t
*spa
= vd
->vdev_spa
;
4339 if (state
== vd
->vdev_state
) {
4341 * Since vdev_offline() code path is already in an offline
4342 * state we can miss a statechange event to OFFLINE. Check
4343 * the previous state to catch this condition.
4345 if (vd
->vdev_ops
->vdev_op_leaf
&&
4346 (state
== VDEV_STATE_OFFLINE
) &&
4347 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
4348 /* post an offline state change */
4349 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
4351 vd
->vdev_stat
.vs_aux
= aux
;
4355 save_state
= vd
->vdev_state
;
4357 vd
->vdev_state
= state
;
4358 vd
->vdev_stat
.vs_aux
= aux
;
4361 * If we are setting the vdev state to anything but an open state, then
4362 * always close the underlying device unless the device has requested
4363 * a delayed close (i.e. we're about to remove or fault the device).
4364 * Otherwise, we keep accessible but invalid devices open forever.
4365 * We don't call vdev_close() itself, because that implies some extra
4366 * checks (offline, etc) that we don't want here. This is limited to
4367 * leaf devices, because otherwise closing the device will affect other
4370 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
4371 vd
->vdev_ops
->vdev_op_leaf
)
4372 vd
->vdev_ops
->vdev_op_close(vd
);
4374 if (vd
->vdev_removed
&&
4375 state
== VDEV_STATE_CANT_OPEN
&&
4376 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
4378 * If the previous state is set to VDEV_STATE_REMOVED, then this
4379 * device was previously marked removed and someone attempted to
4380 * reopen it. If this failed due to a nonexistent device, then
4381 * keep the device in the REMOVED state. We also let this be if
4382 * it is one of our special test online cases, which is only
4383 * attempting to online the device and shouldn't generate an FMA
4386 vd
->vdev_state
= VDEV_STATE_REMOVED
;
4387 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
4388 } else if (state
== VDEV_STATE_REMOVED
) {
4389 vd
->vdev_removed
= B_TRUE
;
4390 } else if (state
== VDEV_STATE_CANT_OPEN
) {
4392 * If we fail to open a vdev during an import or recovery, we
4393 * mark it as "not available", which signifies that it was
4394 * never there to begin with. Failure to open such a device
4395 * is not considered an error.
4397 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
4398 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
4399 vd
->vdev_ops
->vdev_op_leaf
)
4400 vd
->vdev_not_present
= 1;
4403 * Post the appropriate ereport. If the 'prevstate' field is
4404 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4405 * that this is part of a vdev_reopen(). In this case, we don't
4406 * want to post the ereport if the device was already in the
4407 * CANT_OPEN state beforehand.
4409 * If the 'checkremove' flag is set, then this is an attempt to
4410 * online the device in response to an insertion event. If we
4411 * hit this case, then we have detected an insertion event for a
4412 * faulted or offline device that wasn't in the removed state.
4413 * In this scenario, we don't post an ereport because we are
4414 * about to replace the device, or attempt an online with
4415 * vdev_forcefault, which will generate the fault for us.
4417 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
4418 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
4419 vd
!= spa
->spa_root_vdev
) {
4423 case VDEV_AUX_OPEN_FAILED
:
4424 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
4426 case VDEV_AUX_CORRUPT_DATA
:
4427 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
4429 case VDEV_AUX_NO_REPLICAS
:
4430 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
4432 case VDEV_AUX_BAD_GUID_SUM
:
4433 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
4435 case VDEV_AUX_TOO_SMALL
:
4436 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
4438 case VDEV_AUX_BAD_LABEL
:
4439 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
4441 case VDEV_AUX_BAD_ASHIFT
:
4442 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
4445 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
4448 zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
4452 /* Erase any notion of persistent removed state */
4453 vd
->vdev_removed
= B_FALSE
;
4455 vd
->vdev_removed
= B_FALSE
;
4459 * Notify ZED of any significant state-change on a leaf vdev.
4462 if (vd
->vdev_ops
->vdev_op_leaf
) {
4463 /* preserve original state from a vdev_reopen() */
4464 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
4465 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
4466 (save_state
<= VDEV_STATE_CLOSED
))
4467 save_state
= vd
->vdev_prevstate
;
4469 /* filter out state change due to initial vdev_open */
4470 if (save_state
> VDEV_STATE_CLOSED
)
4471 zfs_post_state_change(spa
, vd
, save_state
);
4474 if (!isopen
&& vd
->vdev_parent
)
4475 vdev_propagate_state(vd
->vdev_parent
);
4479 vdev_children_are_offline(vdev_t
*vd
)
4481 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
4483 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
4484 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
4492 * Check the vdev configuration to ensure that it's capable of supporting
4493 * a root pool. We do not support partial configuration.
4496 vdev_is_bootable(vdev_t
*vd
)
4498 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4499 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
4501 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0 ||
4502 strcmp(vdev_type
, VDEV_TYPE_INDIRECT
) == 0) {
4507 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4508 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
4515 vdev_is_concrete(vdev_t
*vd
)
4517 vdev_ops_t
*ops
= vd
->vdev_ops
;
4518 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
4519 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
4527 * Determine if a log device has valid content. If the vdev was
4528 * removed or faulted in the MOS config then we know that
4529 * the content on the log device has already been written to the pool.
4532 vdev_log_state_valid(vdev_t
*vd
)
4534 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
4538 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4539 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
4546 * Expand a vdev if possible.
4549 vdev_expand(vdev_t
*vd
, uint64_t txg
)
4551 ASSERT(vd
->vdev_top
== vd
);
4552 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4553 ASSERT(vdev_is_concrete(vd
));
4555 vdev_set_deflate_ratio(vd
);
4557 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
4558 vdev_is_concrete(vd
)) {
4559 vdev_metaslab_group_create(vd
);
4560 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
4561 vdev_config_dirty(vd
);
4569 vdev_split(vdev_t
*vd
)
4571 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
4573 vdev_remove_child(pvd
, vd
);
4574 vdev_compact_children(pvd
);
4576 cvd
= pvd
->vdev_child
[0];
4577 if (pvd
->vdev_children
== 1) {
4578 vdev_remove_parent(cvd
);
4579 cvd
->vdev_splitting
= B_TRUE
;
4581 vdev_propagate_state(cvd
);
4585 vdev_deadman(vdev_t
*vd
, char *tag
)
4587 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4588 vdev_t
*cvd
= vd
->vdev_child
[c
];
4590 vdev_deadman(cvd
, tag
);
4593 if (vd
->vdev_ops
->vdev_op_leaf
) {
4594 vdev_queue_t
*vq
= &vd
->vdev_queue
;
4596 mutex_enter(&vq
->vq_lock
);
4597 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
4598 spa_t
*spa
= vd
->vdev_spa
;
4602 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4603 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
4606 * Look at the head of all the pending queues,
4607 * if any I/O has been outstanding for longer than
4608 * the spa_deadman_synctime invoke the deadman logic.
4610 fio
= avl_first(&vq
->vq_active_tree
);
4611 delta
= gethrtime() - fio
->io_timestamp
;
4612 if (delta
> spa_deadman_synctime(spa
))
4613 zio_deadman(fio
, tag
);
4615 mutex_exit(&vq
->vq_lock
);
4620 vdev_set_deferred_resilver(spa_t
*spa
, vdev_t
*vd
)
4622 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
4623 vdev_set_deferred_resilver(spa
, vd
->vdev_child
[i
]);
4625 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_writeable(vd
) ||
4626 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
4630 vd
->vdev_resilver_deferred
= B_TRUE
;
4631 spa
->spa_resilver_deferred
= B_TRUE
;
4634 #if defined(_KERNEL)
4635 EXPORT_SYMBOL(vdev_fault
);
4636 EXPORT_SYMBOL(vdev_degrade
);
4637 EXPORT_SYMBOL(vdev_online
);
4638 EXPORT_SYMBOL(vdev_offline
);
4639 EXPORT_SYMBOL(vdev_clear
);
4641 module_param(zfs_vdev_default_ms_count
, int, 0644);
4642 MODULE_PARM_DESC(zfs_vdev_default_ms_count
,
4643 "Target number of metaslabs per top-level vdev");
4645 module_param(zfs_vdev_min_ms_count
, int, 0644);
4646 MODULE_PARM_DESC(zfs_vdev_min_ms_count
,
4647 "Minimum number of metaslabs per top-level vdev");
4649 module_param(zfs_vdev_ms_count_limit
, int, 0644);
4650 MODULE_PARM_DESC(zfs_vdev_ms_count_limit
,
4651 "Practical upper limit of total metaslabs per top-level vdev");
4653 module_param(zfs_slow_io_events_per_second
, uint
, 0644);
4654 MODULE_PARM_DESC(zfs_slow_io_events_per_second
,
4655 "Rate limit slow IO (delay) events to this many per second");
4657 module_param(zfs_checksum_events_per_second
, uint
, 0644);
4658 MODULE_PARM_DESC(zfs_checksum_events_per_second
, "Rate limit checksum events "
4659 "to this many checksum errors per second (do not set below zed"
4662 module_param(zfs_scan_ignore_errors
, int, 0644);
4663 MODULE_PARM_DESC(zfs_scan_ignore_errors
,
4664 "Ignore errors during resilver/scrub");
4666 module_param(vdev_validate_skip
, int, 0644);
4667 MODULE_PARM_DESC(vdev_validate_skip
,
4668 "Bypass vdev_validate()");
4670 module_param(zfs_nocacheflush
, int, 0644);
4671 MODULE_PARM_DESC(zfs_nocacheflush
, "Disable cache flushes");