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, 2020 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.
30 * Copyright (c) 2019, Datto Inc. All rights reserved.
33 #include <sys/zfs_context.h>
34 #include <sys/fm/fs/zfs.h>
36 #include <sys/spa_impl.h>
37 #include <sys/bpobj.h>
39 #include <sys/dmu_tx.h>
40 #include <sys/dsl_dir.h>
41 #include <sys/vdev_impl.h>
42 #include <sys/vdev_rebuild.h>
43 #include <sys/vdev_draid.h>
44 #include <sys/uberblock_impl.h>
45 #include <sys/metaslab.h>
46 #include <sys/metaslab_impl.h>
47 #include <sys/space_map.h>
48 #include <sys/space_reftree.h>
51 #include <sys/fs/zfs.h>
54 #include <sys/dsl_scan.h>
55 #include <sys/vdev_raidz.h>
57 #include <sys/vdev_initialize.h>
58 #include <sys/vdev_trim.h>
60 #include <sys/zfs_ratelimit.h>
63 * One metaslab from each (normal-class) vdev is used by the ZIL. These are
64 * called "embedded slog metaslabs", are referenced by vdev_log_mg, and are
65 * part of the spa_embedded_log_class. The metaslab with the most free space
66 * in each vdev is selected for this purpose when the pool is opened (or a
67 * vdev is added). See vdev_metaslab_init().
69 * Log blocks can be allocated from the following locations. Each one is tried
70 * in order until the allocation succeeds:
71 * 1. dedicated log vdevs, aka "slog" (spa_log_class)
72 * 2. embedded slog metaslabs (spa_embedded_log_class)
73 * 3. other metaslabs in normal vdevs (spa_normal_class)
75 * zfs_embedded_slog_min_ms disables the embedded slog if there are fewer
76 * than this number of metaslabs in the vdev. This ensures that we don't set
77 * aside an unreasonable amount of space for the ZIL. If set to less than
78 * 1 << (spa_slop_shift + 1), on small pools the usable space may be reduced
79 * (by more than 1<<spa_slop_shift) due to the embedded slog metaslab.
81 int zfs_embedded_slog_min_ms
= 64;
83 /* default target for number of metaslabs per top-level vdev */
84 int zfs_vdev_default_ms_count
= 200;
86 /* minimum number of metaslabs per top-level vdev */
87 int zfs_vdev_min_ms_count
= 16;
89 /* practical upper limit of total metaslabs per top-level vdev */
90 int zfs_vdev_ms_count_limit
= 1ULL << 17;
92 /* lower limit for metaslab size (512M) */
93 int zfs_vdev_default_ms_shift
= 29;
95 /* upper limit for metaslab size (16G) */
96 int zfs_vdev_max_ms_shift
= 34;
98 int vdev_validate_skip
= B_FALSE
;
101 * Since the DTL space map of a vdev is not expected to have a lot of
102 * entries, we default its block size to 4K.
104 int zfs_vdev_dtl_sm_blksz
= (1 << 12);
107 * Rate limit slow IO (delay) events to this many per second.
109 unsigned int zfs_slow_io_events_per_second
= 20;
112 * Rate limit checksum events after this many checksum errors per second.
114 unsigned int zfs_checksum_events_per_second
= 20;
117 * Ignore errors during scrub/resilver. Allows to work around resilver
118 * upon import when there are pool errors.
120 int zfs_scan_ignore_errors
= 0;
123 * vdev-wide space maps that have lots of entries written to them at
124 * the end of each transaction can benefit from a higher I/O bandwidth
125 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
127 int zfs_vdev_standard_sm_blksz
= (1 << 17);
130 * Tunable parameter for debugging or performance analysis. Setting this
131 * will cause pool corruption on power loss if a volatile out-of-order
132 * write cache is enabled.
134 int zfs_nocacheflush
= 0;
136 uint64_t zfs_vdev_max_auto_ashift
= ASHIFT_MAX
;
137 uint64_t zfs_vdev_min_auto_ashift
= ASHIFT_MIN
;
141 vdev_dbgmsg(vdev_t
*vd
, const char *fmt
, ...)
147 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
150 if (vd
->vdev_path
!= NULL
) {
151 zfs_dbgmsg("%s vdev '%s': %s", vd
->vdev_ops
->vdev_op_type
,
154 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
155 vd
->vdev_ops
->vdev_op_type
,
156 (u_longlong_t
)vd
->vdev_id
,
157 (u_longlong_t
)vd
->vdev_guid
, buf
);
162 vdev_dbgmsg_print_tree(vdev_t
*vd
, int indent
)
166 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
) {
167 zfs_dbgmsg("%*svdev %u: %s", indent
, "", vd
->vdev_id
,
168 vd
->vdev_ops
->vdev_op_type
);
172 switch (vd
->vdev_state
) {
173 case VDEV_STATE_UNKNOWN
:
174 (void) snprintf(state
, sizeof (state
), "unknown");
176 case VDEV_STATE_CLOSED
:
177 (void) snprintf(state
, sizeof (state
), "closed");
179 case VDEV_STATE_OFFLINE
:
180 (void) snprintf(state
, sizeof (state
), "offline");
182 case VDEV_STATE_REMOVED
:
183 (void) snprintf(state
, sizeof (state
), "removed");
185 case VDEV_STATE_CANT_OPEN
:
186 (void) snprintf(state
, sizeof (state
), "can't open");
188 case VDEV_STATE_FAULTED
:
189 (void) snprintf(state
, sizeof (state
), "faulted");
191 case VDEV_STATE_DEGRADED
:
192 (void) snprintf(state
, sizeof (state
), "degraded");
194 case VDEV_STATE_HEALTHY
:
195 (void) snprintf(state
, sizeof (state
), "healthy");
198 (void) snprintf(state
, sizeof (state
), "<state %u>",
199 (uint_t
)vd
->vdev_state
);
202 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent
,
203 "", (int)vd
->vdev_id
, vd
->vdev_ops
->vdev_op_type
,
204 vd
->vdev_islog
? " (log)" : "",
205 (u_longlong_t
)vd
->vdev_guid
,
206 vd
->vdev_path
? vd
->vdev_path
: "N/A", state
);
208 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
209 vdev_dbgmsg_print_tree(vd
->vdev_child
[i
], indent
+ 2);
213 * Virtual device management.
216 static vdev_ops_t
*vdev_ops_table
[] = {
220 &vdev_draid_spare_ops
,
233 * Given a vdev type, return the appropriate ops vector.
236 vdev_getops(const char *type
)
238 vdev_ops_t
*ops
, **opspp
;
240 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
241 if (strcmp(ops
->vdev_op_type
, type
) == 0)
248 * Given a vdev and a metaslab class, find which metaslab group we're
249 * interested in. All vdevs may belong to two different metaslab classes.
250 * Dedicated slog devices use only the primary metaslab group, rather than a
251 * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL.
254 vdev_get_mg(vdev_t
*vd
, metaslab_class_t
*mc
)
256 if (mc
== spa_embedded_log_class(vd
->vdev_spa
) &&
257 vd
->vdev_log_mg
!= NULL
)
258 return (vd
->vdev_log_mg
);
260 return (vd
->vdev_mg
);
265 vdev_default_xlate(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
266 range_seg64_t
*physical_rs
, range_seg64_t
*remain_rs
)
268 physical_rs
->rs_start
= logical_rs
->rs_start
;
269 physical_rs
->rs_end
= logical_rs
->rs_end
;
273 * Derive the enumerated allocation bias from string input.
274 * String origin is either the per-vdev zap or zpool(8).
276 static vdev_alloc_bias_t
277 vdev_derive_alloc_bias(const char *bias
)
279 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
281 if (strcmp(bias
, VDEV_ALLOC_BIAS_LOG
) == 0)
282 alloc_bias
= VDEV_BIAS_LOG
;
283 else if (strcmp(bias
, VDEV_ALLOC_BIAS_SPECIAL
) == 0)
284 alloc_bias
= VDEV_BIAS_SPECIAL
;
285 else if (strcmp(bias
, VDEV_ALLOC_BIAS_DEDUP
) == 0)
286 alloc_bias
= VDEV_BIAS_DEDUP
;
292 * Default asize function: return the MAX of psize with the asize of
293 * all children. This is what's used by anything other than RAID-Z.
296 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
298 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
301 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
302 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
303 asize
= MAX(asize
, csize
);
310 vdev_default_min_asize(vdev_t
*vd
)
312 return (vd
->vdev_min_asize
);
316 * Get the minimum allocatable size. We define the allocatable size as
317 * the vdev's asize rounded to the nearest metaslab. This allows us to
318 * replace or attach devices which don't have the same physical size but
319 * can still satisfy the same number of allocations.
322 vdev_get_min_asize(vdev_t
*vd
)
324 vdev_t
*pvd
= vd
->vdev_parent
;
327 * If our parent is NULL (inactive spare or cache) or is the root,
328 * just return our own asize.
331 return (vd
->vdev_asize
);
334 * The top-level vdev just returns the allocatable size rounded
335 * to the nearest metaslab.
337 if (vd
== vd
->vdev_top
)
338 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
340 return (pvd
->vdev_ops
->vdev_op_min_asize(pvd
));
344 vdev_set_min_asize(vdev_t
*vd
)
346 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
348 for (int c
= 0; c
< vd
->vdev_children
; c
++)
349 vdev_set_min_asize(vd
->vdev_child
[c
]);
353 * Get the minimal allocation size for the top-level vdev.
356 vdev_get_min_alloc(vdev_t
*vd
)
358 uint64_t min_alloc
= 1ULL << vd
->vdev_ashift
;
360 if (vd
->vdev_ops
->vdev_op_min_alloc
!= NULL
)
361 min_alloc
= vd
->vdev_ops
->vdev_op_min_alloc(vd
);
367 * Get the parity level for a top-level vdev.
370 vdev_get_nparity(vdev_t
*vd
)
372 uint64_t nparity
= 0;
374 if (vd
->vdev_ops
->vdev_op_nparity
!= NULL
)
375 nparity
= vd
->vdev_ops
->vdev_op_nparity(vd
);
381 * Get the number of data disks for a top-level vdev.
384 vdev_get_ndisks(vdev_t
*vd
)
388 if (vd
->vdev_ops
->vdev_op_ndisks
!= NULL
)
389 ndisks
= vd
->vdev_ops
->vdev_op_ndisks(vd
);
395 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
397 vdev_t
*rvd
= spa
->spa_root_vdev
;
399 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
401 if (vdev
< rvd
->vdev_children
) {
402 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
403 return (rvd
->vdev_child
[vdev
]);
410 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
414 if (vd
->vdev_guid
== guid
)
417 for (int c
= 0; c
< vd
->vdev_children
; c
++)
418 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
426 vdev_count_leaves_impl(vdev_t
*vd
)
430 if (vd
->vdev_ops
->vdev_op_leaf
)
433 for (int c
= 0; c
< vd
->vdev_children
; c
++)
434 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
440 vdev_count_leaves(spa_t
*spa
)
444 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
445 rc
= vdev_count_leaves_impl(spa
->spa_root_vdev
);
446 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
452 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
454 size_t oldsize
, newsize
;
455 uint64_t id
= cvd
->vdev_id
;
458 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
459 ASSERT(cvd
->vdev_parent
== NULL
);
461 cvd
->vdev_parent
= pvd
;
466 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
468 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
469 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
470 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
472 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
473 if (pvd
->vdev_child
!= NULL
) {
474 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
475 kmem_free(pvd
->vdev_child
, oldsize
);
478 pvd
->vdev_child
= newchild
;
479 pvd
->vdev_child
[id
] = cvd
;
481 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
482 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
485 * Walk up all ancestors to update guid sum.
487 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
488 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
490 if (cvd
->vdev_ops
->vdev_op_leaf
) {
491 list_insert_head(&cvd
->vdev_spa
->spa_leaf_list
, cvd
);
492 cvd
->vdev_spa
->spa_leaf_list_gen
++;
497 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
500 uint_t id
= cvd
->vdev_id
;
502 ASSERT(cvd
->vdev_parent
== pvd
);
507 ASSERT(id
< pvd
->vdev_children
);
508 ASSERT(pvd
->vdev_child
[id
] == cvd
);
510 pvd
->vdev_child
[id
] = NULL
;
511 cvd
->vdev_parent
= NULL
;
513 for (c
= 0; c
< pvd
->vdev_children
; c
++)
514 if (pvd
->vdev_child
[c
])
517 if (c
== pvd
->vdev_children
) {
518 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
519 pvd
->vdev_child
= NULL
;
520 pvd
->vdev_children
= 0;
523 if (cvd
->vdev_ops
->vdev_op_leaf
) {
524 spa_t
*spa
= cvd
->vdev_spa
;
525 list_remove(&spa
->spa_leaf_list
, cvd
);
526 spa
->spa_leaf_list_gen
++;
530 * Walk up all ancestors to update guid sum.
532 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
533 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
537 * Remove any holes in the child array.
540 vdev_compact_children(vdev_t
*pvd
)
542 vdev_t
**newchild
, *cvd
;
543 int oldc
= pvd
->vdev_children
;
546 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
551 for (int c
= newc
= 0; c
< oldc
; c
++)
552 if (pvd
->vdev_child
[c
])
556 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
558 for (int c
= newc
= 0; c
< oldc
; c
++) {
559 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
560 newchild
[newc
] = cvd
;
561 cvd
->vdev_id
= newc
++;
568 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
569 pvd
->vdev_child
= newchild
;
570 pvd
->vdev_children
= newc
;
574 * Allocate and minimally initialize a vdev_t.
577 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
580 vdev_indirect_config_t
*vic
;
582 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
583 vic
= &vd
->vdev_indirect_config
;
585 if (spa
->spa_root_vdev
== NULL
) {
586 ASSERT(ops
== &vdev_root_ops
);
587 spa
->spa_root_vdev
= vd
;
588 spa
->spa_load_guid
= spa_generate_guid(NULL
);
591 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
592 if (spa
->spa_root_vdev
== vd
) {
594 * The root vdev's guid will also be the pool guid,
595 * which must be unique among all pools.
597 guid
= spa_generate_guid(NULL
);
600 * Any other vdev's guid must be unique within the pool.
602 guid
= spa_generate_guid(spa
);
604 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
609 vd
->vdev_guid
= guid
;
610 vd
->vdev_guid_sum
= guid
;
612 vd
->vdev_state
= VDEV_STATE_CLOSED
;
613 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
614 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
616 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
617 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
618 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, RANGE_SEG64
, NULL
,
622 * Initialize rate limit structs for events. We rate limit ZIO delay
623 * and checksum events so that we don't overwhelm ZED with thousands
624 * of events when a disk is acting up.
626 zfs_ratelimit_init(&vd
->vdev_delay_rl
, &zfs_slow_io_events_per_second
,
628 zfs_ratelimit_init(&vd
->vdev_checksum_rl
,
629 &zfs_checksum_events_per_second
, 1);
631 list_link_init(&vd
->vdev_config_dirty_node
);
632 list_link_init(&vd
->vdev_state_dirty_node
);
633 list_link_init(&vd
->vdev_initialize_node
);
634 list_link_init(&vd
->vdev_leaf_node
);
635 list_link_init(&vd
->vdev_trim_node
);
637 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
638 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
639 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
640 mutex_init(&vd
->vdev_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
642 mutex_init(&vd
->vdev_initialize_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
643 mutex_init(&vd
->vdev_initialize_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
644 cv_init(&vd
->vdev_initialize_cv
, NULL
, CV_DEFAULT
, NULL
);
645 cv_init(&vd
->vdev_initialize_io_cv
, NULL
, CV_DEFAULT
, NULL
);
647 mutex_init(&vd
->vdev_trim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
648 mutex_init(&vd
->vdev_autotrim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
649 mutex_init(&vd
->vdev_trim_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
650 cv_init(&vd
->vdev_trim_cv
, NULL
, CV_DEFAULT
, NULL
);
651 cv_init(&vd
->vdev_autotrim_cv
, NULL
, CV_DEFAULT
, NULL
);
652 cv_init(&vd
->vdev_trim_io_cv
, NULL
, CV_DEFAULT
, NULL
);
654 mutex_init(&vd
->vdev_rebuild_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
655 cv_init(&vd
->vdev_rebuild_cv
, NULL
, CV_DEFAULT
, NULL
);
657 for (int t
= 0; t
< DTL_TYPES
; t
++) {
658 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0,
662 txg_list_create(&vd
->vdev_ms_list
, spa
,
663 offsetof(struct metaslab
, ms_txg_node
));
664 txg_list_create(&vd
->vdev_dtl_list
, spa
,
665 offsetof(struct vdev
, vdev_dtl_node
));
666 vd
->vdev_stat
.vs_timestamp
= gethrtime();
674 * Allocate a new vdev. The 'alloctype' is used to control whether we are
675 * creating a new vdev or loading an existing one - the behavior is slightly
676 * different for each case.
679 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
684 uint64_t guid
= 0, islog
;
686 vdev_indirect_config_t
*vic
;
689 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
690 boolean_t top_level
= (parent
&& !parent
->vdev_parent
);
692 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
694 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
695 return (SET_ERROR(EINVAL
));
697 if ((ops
= vdev_getops(type
)) == NULL
)
698 return (SET_ERROR(EINVAL
));
701 * If this is a load, get the vdev guid from the nvlist.
702 * Otherwise, vdev_alloc_common() will generate one for us.
704 if (alloctype
== VDEV_ALLOC_LOAD
) {
707 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
709 return (SET_ERROR(EINVAL
));
711 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
712 return (SET_ERROR(EINVAL
));
713 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
714 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
715 return (SET_ERROR(EINVAL
));
716 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
717 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
718 return (SET_ERROR(EINVAL
));
719 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
720 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
721 return (SET_ERROR(EINVAL
));
725 * The first allocated vdev must be of type 'root'.
727 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
728 return (SET_ERROR(EINVAL
));
731 * Determine whether we're a log vdev.
734 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
735 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
736 return (SET_ERROR(ENOTSUP
));
738 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
739 return (SET_ERROR(ENOTSUP
));
741 if (top_level
&& alloctype
== VDEV_ALLOC_ADD
) {
745 * If creating a top-level vdev, check for allocation
748 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
750 alloc_bias
= vdev_derive_alloc_bias(bias
);
752 /* spa_vdev_add() expects feature to be enabled */
753 if (spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
754 !spa_feature_is_enabled(spa
,
755 SPA_FEATURE_ALLOCATION_CLASSES
)) {
756 return (SET_ERROR(ENOTSUP
));
760 /* spa_vdev_add() expects feature to be enabled */
761 if (ops
== &vdev_draid_ops
&&
762 spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
763 !spa_feature_is_enabled(spa
, SPA_FEATURE_DRAID
)) {
764 return (SET_ERROR(ENOTSUP
));
769 * Initialize the vdev specific data. This is done before calling
770 * vdev_alloc_common() since it may fail and this simplifies the
771 * error reporting and cleanup code paths.
774 if (ops
->vdev_op_init
!= NULL
) {
775 rc
= ops
->vdev_op_init(spa
, nv
, &tsd
);
781 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
783 vd
->vdev_islog
= islog
;
785 if (top_level
&& alloc_bias
!= VDEV_BIAS_NONE
)
786 vd
->vdev_alloc_bias
= alloc_bias
;
788 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
789 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
792 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
793 * fault on a vdev and want it to persist across imports (like with
796 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
797 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
798 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
799 vd
->vdev_faulted
= 1;
800 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
803 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
804 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
805 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
806 &vd
->vdev_physpath
) == 0)
807 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
809 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
810 &vd
->vdev_enc_sysfs_path
) == 0)
811 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
813 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
814 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
817 * Set the whole_disk property. If it's not specified, leave the value
820 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
821 &vd
->vdev_wholedisk
) != 0)
822 vd
->vdev_wholedisk
= -1ULL;
824 vic
= &vd
->vdev_indirect_config
;
826 ASSERT0(vic
->vic_mapping_object
);
827 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
828 &vic
->vic_mapping_object
);
829 ASSERT0(vic
->vic_births_object
);
830 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
831 &vic
->vic_births_object
);
832 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
833 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
834 &vic
->vic_prev_indirect_vdev
);
837 * Look for the 'not present' flag. This will only be set if the device
838 * was not present at the time of import.
840 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
841 &vd
->vdev_not_present
);
844 * Get the alignment requirement.
846 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
849 * Retrieve the vdev creation time.
851 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
855 * If we're a top-level vdev, try to load the allocation parameters.
858 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
859 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
861 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
863 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
865 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
867 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
870 ASSERT0(vd
->vdev_top_zap
);
873 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
874 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
875 alloctype
== VDEV_ALLOC_ADD
||
876 alloctype
== VDEV_ALLOC_SPLIT
||
877 alloctype
== VDEV_ALLOC_ROOTPOOL
);
878 /* Note: metaslab_group_create() is now deferred */
881 if (vd
->vdev_ops
->vdev_op_leaf
&&
882 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
883 (void) nvlist_lookup_uint64(nv
,
884 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
886 ASSERT0(vd
->vdev_leaf_zap
);
890 * If we're a leaf vdev, try to load the DTL object and other state.
893 if (vd
->vdev_ops
->vdev_op_leaf
&&
894 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
895 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
896 if (alloctype
== VDEV_ALLOC_LOAD
) {
897 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
898 &vd
->vdev_dtl_object
);
899 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
903 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
906 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
907 &spare
) == 0 && spare
)
911 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
914 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
915 &vd
->vdev_resilver_txg
);
917 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REBUILD_TXG
,
918 &vd
->vdev_rebuild_txg
);
920 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
921 vdev_defer_resilver(vd
);
924 * In general, when importing a pool we want to ignore the
925 * persistent fault state, as the diagnosis made on another
926 * system may not be valid in the current context. The only
927 * exception is if we forced a vdev to a persistently faulted
928 * state with 'zpool offline -f'. The persistent fault will
929 * remain across imports until cleared.
931 * Local vdevs will remain in the faulted state.
933 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
934 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
935 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
937 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
939 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
942 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
946 VDEV_AUX_ERR_EXCEEDED
;
947 if (nvlist_lookup_string(nv
,
948 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
949 strcmp(aux
, "external") == 0)
950 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
952 vd
->vdev_faulted
= 0ULL;
958 * Add ourselves to the parent's list of children.
960 vdev_add_child(parent
, vd
);
968 vdev_free(vdev_t
*vd
)
970 spa_t
*spa
= vd
->vdev_spa
;
972 ASSERT3P(vd
->vdev_initialize_thread
, ==, NULL
);
973 ASSERT3P(vd
->vdev_trim_thread
, ==, NULL
);
974 ASSERT3P(vd
->vdev_autotrim_thread
, ==, NULL
);
975 ASSERT3P(vd
->vdev_rebuild_thread
, ==, NULL
);
978 * Scan queues are normally destroyed at the end of a scan. If the
979 * queue exists here, that implies the vdev is being removed while
980 * the scan is still running.
982 if (vd
->vdev_scan_io_queue
!= NULL
) {
983 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
984 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
985 vd
->vdev_scan_io_queue
= NULL
;
986 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
990 * vdev_free() implies closing the vdev first. This is simpler than
991 * trying to ensure complicated semantics for all callers.
995 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
996 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
1001 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1002 vdev_free(vd
->vdev_child
[c
]);
1004 ASSERT(vd
->vdev_child
== NULL
);
1005 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
1007 if (vd
->vdev_ops
->vdev_op_fini
!= NULL
)
1008 vd
->vdev_ops
->vdev_op_fini(vd
);
1011 * Discard allocation state.
1013 if (vd
->vdev_mg
!= NULL
) {
1014 vdev_metaslab_fini(vd
);
1015 metaslab_group_destroy(vd
->vdev_mg
);
1018 if (vd
->vdev_log_mg
!= NULL
) {
1019 ASSERT0(vd
->vdev_ms_count
);
1020 metaslab_group_destroy(vd
->vdev_log_mg
);
1021 vd
->vdev_log_mg
= NULL
;
1024 ASSERT0(vd
->vdev_stat
.vs_space
);
1025 ASSERT0(vd
->vdev_stat
.vs_dspace
);
1026 ASSERT0(vd
->vdev_stat
.vs_alloc
);
1029 * Remove this vdev from its parent's child list.
1031 vdev_remove_child(vd
->vdev_parent
, vd
);
1033 ASSERT(vd
->vdev_parent
== NULL
);
1034 ASSERT(!list_link_active(&vd
->vdev_leaf_node
));
1037 * Clean up vdev structure.
1039 vdev_queue_fini(vd
);
1040 vdev_cache_fini(vd
);
1043 spa_strfree(vd
->vdev_path
);
1045 spa_strfree(vd
->vdev_devid
);
1046 if (vd
->vdev_physpath
)
1047 spa_strfree(vd
->vdev_physpath
);
1049 if (vd
->vdev_enc_sysfs_path
)
1050 spa_strfree(vd
->vdev_enc_sysfs_path
);
1053 spa_strfree(vd
->vdev_fru
);
1055 if (vd
->vdev_isspare
)
1056 spa_spare_remove(vd
);
1057 if (vd
->vdev_isl2cache
)
1058 spa_l2cache_remove(vd
);
1060 txg_list_destroy(&vd
->vdev_ms_list
);
1061 txg_list_destroy(&vd
->vdev_dtl_list
);
1063 mutex_enter(&vd
->vdev_dtl_lock
);
1064 space_map_close(vd
->vdev_dtl_sm
);
1065 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1066 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
1067 range_tree_destroy(vd
->vdev_dtl
[t
]);
1069 mutex_exit(&vd
->vdev_dtl_lock
);
1071 EQUIV(vd
->vdev_indirect_births
!= NULL
,
1072 vd
->vdev_indirect_mapping
!= NULL
);
1073 if (vd
->vdev_indirect_births
!= NULL
) {
1074 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
1075 vdev_indirect_births_close(vd
->vdev_indirect_births
);
1078 if (vd
->vdev_obsolete_sm
!= NULL
) {
1079 ASSERT(vd
->vdev_removing
||
1080 vd
->vdev_ops
== &vdev_indirect_ops
);
1081 space_map_close(vd
->vdev_obsolete_sm
);
1082 vd
->vdev_obsolete_sm
= NULL
;
1084 range_tree_destroy(vd
->vdev_obsolete_segments
);
1085 rw_destroy(&vd
->vdev_indirect_rwlock
);
1086 mutex_destroy(&vd
->vdev_obsolete_lock
);
1088 mutex_destroy(&vd
->vdev_dtl_lock
);
1089 mutex_destroy(&vd
->vdev_stat_lock
);
1090 mutex_destroy(&vd
->vdev_probe_lock
);
1091 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
1093 mutex_destroy(&vd
->vdev_initialize_lock
);
1094 mutex_destroy(&vd
->vdev_initialize_io_lock
);
1095 cv_destroy(&vd
->vdev_initialize_io_cv
);
1096 cv_destroy(&vd
->vdev_initialize_cv
);
1098 mutex_destroy(&vd
->vdev_trim_lock
);
1099 mutex_destroy(&vd
->vdev_autotrim_lock
);
1100 mutex_destroy(&vd
->vdev_trim_io_lock
);
1101 cv_destroy(&vd
->vdev_trim_cv
);
1102 cv_destroy(&vd
->vdev_autotrim_cv
);
1103 cv_destroy(&vd
->vdev_trim_io_cv
);
1105 mutex_destroy(&vd
->vdev_rebuild_lock
);
1106 cv_destroy(&vd
->vdev_rebuild_cv
);
1108 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
1109 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
1111 if (vd
== spa
->spa_root_vdev
)
1112 spa
->spa_root_vdev
= NULL
;
1114 kmem_free(vd
, sizeof (vdev_t
));
1118 * Transfer top-level vdev state from svd to tvd.
1121 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
1123 spa_t
*spa
= svd
->vdev_spa
;
1128 ASSERT(tvd
== tvd
->vdev_top
);
1130 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
1131 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
1132 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
1133 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
1134 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
1136 svd
->vdev_ms_array
= 0;
1137 svd
->vdev_ms_shift
= 0;
1138 svd
->vdev_ms_count
= 0;
1139 svd
->vdev_top_zap
= 0;
1142 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
1143 if (tvd
->vdev_log_mg
)
1144 ASSERT3P(tvd
->vdev_log_mg
, ==, svd
->vdev_log_mg
);
1145 tvd
->vdev_mg
= svd
->vdev_mg
;
1146 tvd
->vdev_log_mg
= svd
->vdev_log_mg
;
1147 tvd
->vdev_ms
= svd
->vdev_ms
;
1149 svd
->vdev_mg
= NULL
;
1150 svd
->vdev_log_mg
= NULL
;
1151 svd
->vdev_ms
= NULL
;
1153 if (tvd
->vdev_mg
!= NULL
)
1154 tvd
->vdev_mg
->mg_vd
= tvd
;
1155 if (tvd
->vdev_log_mg
!= NULL
)
1156 tvd
->vdev_log_mg
->mg_vd
= tvd
;
1158 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
1159 svd
->vdev_checkpoint_sm
= NULL
;
1161 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1162 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1164 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1165 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1166 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1168 svd
->vdev_stat
.vs_alloc
= 0;
1169 svd
->vdev_stat
.vs_space
= 0;
1170 svd
->vdev_stat
.vs_dspace
= 0;
1173 * State which may be set on a top-level vdev that's in the
1174 * process of being removed.
1176 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1177 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1178 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1179 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1180 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1181 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1182 ASSERT0(tvd
->vdev_removing
);
1183 ASSERT0(tvd
->vdev_rebuilding
);
1184 tvd
->vdev_removing
= svd
->vdev_removing
;
1185 tvd
->vdev_rebuilding
= svd
->vdev_rebuilding
;
1186 tvd
->vdev_rebuild_config
= svd
->vdev_rebuild_config
;
1187 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1188 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1189 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1190 range_tree_swap(&svd
->vdev_obsolete_segments
,
1191 &tvd
->vdev_obsolete_segments
);
1192 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1193 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1194 svd
->vdev_indirect_config
.vic_births_object
= 0;
1195 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1196 svd
->vdev_indirect_mapping
= NULL
;
1197 svd
->vdev_indirect_births
= NULL
;
1198 svd
->vdev_obsolete_sm
= NULL
;
1199 svd
->vdev_removing
= 0;
1200 svd
->vdev_rebuilding
= 0;
1202 for (t
= 0; t
< TXG_SIZE
; t
++) {
1203 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1204 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1205 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1206 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1207 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1208 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1211 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1212 vdev_config_clean(svd
);
1213 vdev_config_dirty(tvd
);
1216 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1217 vdev_state_clean(svd
);
1218 vdev_state_dirty(tvd
);
1221 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1222 svd
->vdev_deflate_ratio
= 0;
1224 tvd
->vdev_islog
= svd
->vdev_islog
;
1225 svd
->vdev_islog
= 0;
1227 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1231 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1238 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1239 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1243 * Add a mirror/replacing vdev above an existing vdev. There is no need to
1244 * call .vdev_op_init() since mirror/replacing vdevs do not have private state.
1247 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1249 spa_t
*spa
= cvd
->vdev_spa
;
1250 vdev_t
*pvd
= cvd
->vdev_parent
;
1253 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1255 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1257 mvd
->vdev_asize
= cvd
->vdev_asize
;
1258 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1259 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1260 mvd
->vdev_psize
= cvd
->vdev_psize
;
1261 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1262 mvd
->vdev_logical_ashift
= cvd
->vdev_logical_ashift
;
1263 mvd
->vdev_physical_ashift
= cvd
->vdev_physical_ashift
;
1264 mvd
->vdev_state
= cvd
->vdev_state
;
1265 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1267 vdev_remove_child(pvd
, cvd
);
1268 vdev_add_child(pvd
, mvd
);
1269 cvd
->vdev_id
= mvd
->vdev_children
;
1270 vdev_add_child(mvd
, cvd
);
1271 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1273 if (mvd
== mvd
->vdev_top
)
1274 vdev_top_transfer(cvd
, mvd
);
1280 * Remove a 1-way mirror/replacing vdev from the tree.
1283 vdev_remove_parent(vdev_t
*cvd
)
1285 vdev_t
*mvd
= cvd
->vdev_parent
;
1286 vdev_t
*pvd
= mvd
->vdev_parent
;
1288 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1290 ASSERT(mvd
->vdev_children
== 1);
1291 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1292 mvd
->vdev_ops
== &vdev_replacing_ops
||
1293 mvd
->vdev_ops
== &vdev_spare_ops
);
1294 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1295 cvd
->vdev_logical_ashift
= mvd
->vdev_logical_ashift
;
1296 cvd
->vdev_physical_ashift
= mvd
->vdev_physical_ashift
;
1297 vdev_remove_child(mvd
, cvd
);
1298 vdev_remove_child(pvd
, mvd
);
1301 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1302 * Otherwise, we could have detached an offline device, and when we
1303 * go to import the pool we'll think we have two top-level vdevs,
1304 * instead of a different version of the same top-level vdev.
1306 if (mvd
->vdev_top
== mvd
) {
1307 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1308 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1309 cvd
->vdev_guid
+= guid_delta
;
1310 cvd
->vdev_guid_sum
+= guid_delta
;
1313 * If pool not set for autoexpand, we need to also preserve
1314 * mvd's asize to prevent automatic expansion of cvd.
1315 * Otherwise if we are adjusting the mirror by attaching and
1316 * detaching children of non-uniform sizes, the mirror could
1317 * autoexpand, unexpectedly requiring larger devices to
1318 * re-establish the mirror.
1320 if (!cvd
->vdev_spa
->spa_autoexpand
)
1321 cvd
->vdev_asize
= mvd
->vdev_asize
;
1323 cvd
->vdev_id
= mvd
->vdev_id
;
1324 vdev_add_child(pvd
, cvd
);
1325 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1327 if (cvd
== cvd
->vdev_top
)
1328 vdev_top_transfer(mvd
, cvd
);
1330 ASSERT(mvd
->vdev_children
== 0);
1335 vdev_metaslab_group_create(vdev_t
*vd
)
1337 spa_t
*spa
= vd
->vdev_spa
;
1340 * metaslab_group_create was delayed until allocation bias was available
1342 if (vd
->vdev_mg
== NULL
) {
1343 metaslab_class_t
*mc
;
1345 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1346 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1348 ASSERT3U(vd
->vdev_islog
, ==,
1349 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1351 switch (vd
->vdev_alloc_bias
) {
1353 mc
= spa_log_class(spa
);
1355 case VDEV_BIAS_SPECIAL
:
1356 mc
= spa_special_class(spa
);
1358 case VDEV_BIAS_DEDUP
:
1359 mc
= spa_dedup_class(spa
);
1362 mc
= spa_normal_class(spa
);
1365 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1366 spa
->spa_alloc_count
);
1368 if (!vd
->vdev_islog
) {
1369 vd
->vdev_log_mg
= metaslab_group_create(
1370 spa_embedded_log_class(spa
), vd
, 1);
1374 * The spa ashift min/max only apply for the normal metaslab
1375 * class. Class destination is late binding so ashift boundry
1376 * setting had to wait until now.
1378 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1379 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1380 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1381 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1382 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1383 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1385 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
1386 if (min_alloc
< spa
->spa_min_alloc
)
1387 spa
->spa_min_alloc
= min_alloc
;
1393 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1395 spa_t
*spa
= vd
->vdev_spa
;
1396 uint64_t oldc
= vd
->vdev_ms_count
;
1397 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1400 boolean_t expanding
= (oldc
!= 0);
1402 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1405 * This vdev is not being allocated from yet or is a hole.
1407 if (vd
->vdev_ms_shift
== 0)
1410 ASSERT(!vd
->vdev_ishole
);
1412 ASSERT(oldc
<= newc
);
1414 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1417 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
1418 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1422 vd
->vdev_ms_count
= newc
;
1424 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1425 uint64_t object
= 0;
1427 * vdev_ms_array may be 0 if we are creating the "fake"
1428 * metaslabs for an indirect vdev for zdb's leak detection.
1429 * See zdb_leak_init().
1431 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1432 error
= dmu_read(spa
->spa_meta_objset
,
1434 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1437 vdev_dbgmsg(vd
, "unable to read the metaslab "
1438 "array [error=%d]", error
);
1443 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1446 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1453 * Find the emptiest metaslab on the vdev and mark it for use for
1454 * embedded slog by moving it from the regular to the log metaslab
1457 if (vd
->vdev_mg
->mg_class
== spa_normal_class(spa
) &&
1458 vd
->vdev_ms_count
> zfs_embedded_slog_min_ms
&&
1459 avl_is_empty(&vd
->vdev_log_mg
->mg_metaslab_tree
)) {
1460 uint64_t slog_msid
= 0;
1461 uint64_t smallest
= UINT64_MAX
;
1464 * Note, we only search the new metaslabs, because the old
1465 * (pre-existing) ones may be active (e.g. have non-empty
1466 * range_tree's), and we don't move them to the new
1469 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1471 space_map_allocated(vd
->vdev_ms
[m
]->ms_sm
);
1472 if (alloc
< smallest
) {
1477 metaslab_t
*slog_ms
= vd
->vdev_ms
[slog_msid
];
1479 * The metaslab was marked as dirty at the end of
1480 * metaslab_init(). Remove it from the dirty list so that we
1481 * can uninitialize and reinitialize it to the new class.
1484 (void) txg_list_remove_this(&vd
->vdev_ms_list
,
1487 uint64_t sm_obj
= space_map_object(slog_ms
->ms_sm
);
1488 metaslab_fini(slog_ms
);
1489 VERIFY0(metaslab_init(vd
->vdev_log_mg
, slog_msid
, sm_obj
, txg
,
1490 &vd
->vdev_ms
[slog_msid
]));
1494 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1497 * If the vdev is being removed we don't activate
1498 * the metaslabs since we want to ensure that no new
1499 * allocations are performed on this device.
1501 if (!expanding
&& !vd
->vdev_removing
) {
1502 metaslab_group_activate(vd
->vdev_mg
);
1503 if (vd
->vdev_log_mg
!= NULL
)
1504 metaslab_group_activate(vd
->vdev_log_mg
);
1508 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1511 * Regardless whether this vdev was just added or it is being
1512 * expanded, the metaslab count has changed. Recalculate the
1515 spa_log_sm_set_blocklimit(spa
);
1521 vdev_metaslab_fini(vdev_t
*vd
)
1523 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1524 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1525 SPA_FEATURE_POOL_CHECKPOINT
));
1526 space_map_close(vd
->vdev_checkpoint_sm
);
1528 * Even though we close the space map, we need to set its
1529 * pointer to NULL. The reason is that vdev_metaslab_fini()
1530 * may be called multiple times for certain operations
1531 * (i.e. when destroying a pool) so we need to ensure that
1532 * this clause never executes twice. This logic is similar
1533 * to the one used for the vdev_ms clause below.
1535 vd
->vdev_checkpoint_sm
= NULL
;
1538 if (vd
->vdev_ms
!= NULL
) {
1539 metaslab_group_t
*mg
= vd
->vdev_mg
;
1541 metaslab_group_passivate(mg
);
1542 if (vd
->vdev_log_mg
!= NULL
) {
1543 ASSERT(!vd
->vdev_islog
);
1544 metaslab_group_passivate(vd
->vdev_log_mg
);
1547 uint64_t count
= vd
->vdev_ms_count
;
1548 for (uint64_t m
= 0; m
< count
; m
++) {
1549 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1553 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1555 vd
->vdev_ms_count
= 0;
1557 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
1558 ASSERT0(mg
->mg_histogram
[i
]);
1559 if (vd
->vdev_log_mg
!= NULL
)
1560 ASSERT0(vd
->vdev_log_mg
->mg_histogram
[i
]);
1563 ASSERT0(vd
->vdev_ms_count
);
1564 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1567 typedef struct vdev_probe_stats
{
1568 boolean_t vps_readable
;
1569 boolean_t vps_writeable
;
1571 } vdev_probe_stats_t
;
1574 vdev_probe_done(zio_t
*zio
)
1576 spa_t
*spa
= zio
->io_spa
;
1577 vdev_t
*vd
= zio
->io_vd
;
1578 vdev_probe_stats_t
*vps
= zio
->io_private
;
1580 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1582 if (zio
->io_type
== ZIO_TYPE_READ
) {
1583 if (zio
->io_error
== 0)
1584 vps
->vps_readable
= 1;
1585 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1586 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1587 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1588 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1589 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1591 abd_free(zio
->io_abd
);
1593 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1594 if (zio
->io_error
== 0)
1595 vps
->vps_writeable
= 1;
1596 abd_free(zio
->io_abd
);
1597 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1601 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1602 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1604 if (vdev_readable(vd
) &&
1605 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1608 ASSERT(zio
->io_error
!= 0);
1609 vdev_dbgmsg(vd
, "failed probe");
1610 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1611 spa
, vd
, NULL
, NULL
, 0);
1612 zio
->io_error
= SET_ERROR(ENXIO
);
1615 mutex_enter(&vd
->vdev_probe_lock
);
1616 ASSERT(vd
->vdev_probe_zio
== zio
);
1617 vd
->vdev_probe_zio
= NULL
;
1618 mutex_exit(&vd
->vdev_probe_lock
);
1621 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1622 if (!vdev_accessible(vd
, pio
))
1623 pio
->io_error
= SET_ERROR(ENXIO
);
1625 kmem_free(vps
, sizeof (*vps
));
1630 * Determine whether this device is accessible.
1632 * Read and write to several known locations: the pad regions of each
1633 * vdev label but the first, which we leave alone in case it contains
1637 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1639 spa_t
*spa
= vd
->vdev_spa
;
1640 vdev_probe_stats_t
*vps
= NULL
;
1643 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1646 * Don't probe the probe.
1648 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1652 * To prevent 'probe storms' when a device fails, we create
1653 * just one probe i/o at a time. All zios that want to probe
1654 * this vdev will become parents of the probe io.
1656 mutex_enter(&vd
->vdev_probe_lock
);
1658 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1659 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1661 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1662 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1665 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1667 * vdev_cant_read and vdev_cant_write can only
1668 * transition from TRUE to FALSE when we have the
1669 * SCL_ZIO lock as writer; otherwise they can only
1670 * transition from FALSE to TRUE. This ensures that
1671 * any zio looking at these values can assume that
1672 * failures persist for the life of the I/O. That's
1673 * important because when a device has intermittent
1674 * connectivity problems, we want to ensure that
1675 * they're ascribed to the device (ENXIO) and not
1678 * Since we hold SCL_ZIO as writer here, clear both
1679 * values so the probe can reevaluate from first
1682 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1683 vd
->vdev_cant_read
= B_FALSE
;
1684 vd
->vdev_cant_write
= B_FALSE
;
1687 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1688 vdev_probe_done
, vps
,
1689 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1692 * We can't change the vdev state in this context, so we
1693 * kick off an async task to do it on our behalf.
1696 vd
->vdev_probe_wanted
= B_TRUE
;
1697 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1702 zio_add_child(zio
, pio
);
1704 mutex_exit(&vd
->vdev_probe_lock
);
1707 ASSERT(zio
!= NULL
);
1711 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1712 zio_nowait(zio_read_phys(pio
, vd
,
1713 vdev_label_offset(vd
->vdev_psize
, l
,
1714 offsetof(vdev_label_t
, vl_be
)), VDEV_PAD_SIZE
,
1715 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1716 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1717 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1728 vdev_load_child(void *arg
)
1732 vd
->vdev_load_error
= vdev_load(vd
);
1736 vdev_open_child(void *arg
)
1740 vd
->vdev_open_thread
= curthread
;
1741 vd
->vdev_open_error
= vdev_open(vd
);
1742 vd
->vdev_open_thread
= NULL
;
1746 vdev_uses_zvols(vdev_t
*vd
)
1749 if (zvol_is_zvol(vd
->vdev_path
))
1753 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1754 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1761 * Returns B_TRUE if the passed child should be opened.
1764 vdev_default_open_children_func(vdev_t
*vd
)
1770 * Open the requested child vdevs. If any of the leaf vdevs are using
1771 * a ZFS volume then do the opens in a single thread. This avoids a
1772 * deadlock when the current thread is holding the spa_namespace_lock.
1775 vdev_open_children_impl(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1777 int children
= vd
->vdev_children
;
1779 taskq_t
*tq
= taskq_create("vdev_open", children
, minclsyspri
,
1780 children
, children
, TASKQ_PREPOPULATE
);
1781 vd
->vdev_nonrot
= B_TRUE
;
1783 for (int c
= 0; c
< children
; c
++) {
1784 vdev_t
*cvd
= vd
->vdev_child
[c
];
1786 if (open_func(cvd
) == B_FALSE
)
1789 if (tq
== NULL
|| vdev_uses_zvols(vd
)) {
1790 cvd
->vdev_open_error
= vdev_open(cvd
);
1792 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1793 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
1796 vd
->vdev_nonrot
&= cvd
->vdev_nonrot
;
1806 * Open all child vdevs.
1809 vdev_open_children(vdev_t
*vd
)
1811 vdev_open_children_impl(vd
, vdev_default_open_children_func
);
1815 * Conditionally open a subset of child vdevs.
1818 vdev_open_children_subset(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1820 vdev_open_children_impl(vd
, open_func
);
1824 * Compute the raidz-deflation ratio. Note, we hard-code
1825 * in 128k (1 << 17) because it is the "typical" blocksize.
1826 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1827 * otherwise it would inconsistently account for existing bp's.
1830 vdev_set_deflate_ratio(vdev_t
*vd
)
1832 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1833 vd
->vdev_deflate_ratio
= (1 << 17) /
1834 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1839 * Maximize performance by inflating the configured ashift for top level
1840 * vdevs to be as close to the physical ashift as possible while maintaining
1841 * administrator defined limits and ensuring it doesn't go below the
1845 vdev_ashift_optimize(vdev_t
*vd
)
1847 ASSERT(vd
== vd
->vdev_top
);
1849 if (vd
->vdev_ashift
< vd
->vdev_physical_ashift
) {
1850 vd
->vdev_ashift
= MIN(
1851 MAX(zfs_vdev_max_auto_ashift
, vd
->vdev_ashift
),
1852 MAX(zfs_vdev_min_auto_ashift
,
1853 vd
->vdev_physical_ashift
));
1856 * If the logical and physical ashifts are the same, then
1857 * we ensure that the top-level vdev's ashift is not smaller
1858 * than our minimum ashift value. For the unusual case
1859 * where logical ashift > physical ashift, we can't cap
1860 * the calculated ashift based on max ashift as that
1861 * would cause failures.
1862 * We still check if we need to increase it to match
1865 vd
->vdev_ashift
= MAX(zfs_vdev_min_auto_ashift
,
1871 * Prepare a virtual device for access.
1874 vdev_open(vdev_t
*vd
)
1876 spa_t
*spa
= vd
->vdev_spa
;
1879 uint64_t max_osize
= 0;
1880 uint64_t asize
, max_asize
, psize
;
1881 uint64_t logical_ashift
= 0;
1882 uint64_t physical_ashift
= 0;
1884 ASSERT(vd
->vdev_open_thread
== curthread
||
1885 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1886 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1887 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1888 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1890 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1891 vd
->vdev_cant_read
= B_FALSE
;
1892 vd
->vdev_cant_write
= B_FALSE
;
1893 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1896 * If this vdev is not removed, check its fault status. If it's
1897 * faulted, bail out of the open.
1899 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1900 ASSERT(vd
->vdev_children
== 0);
1901 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1902 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1903 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1904 vd
->vdev_label_aux
);
1905 return (SET_ERROR(ENXIO
));
1906 } else if (vd
->vdev_offline
) {
1907 ASSERT(vd
->vdev_children
== 0);
1908 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1909 return (SET_ERROR(ENXIO
));
1912 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
,
1913 &logical_ashift
, &physical_ashift
);
1915 * Physical volume size should never be larger than its max size, unless
1916 * the disk has shrunk while we were reading it or the device is buggy
1917 * or damaged: either way it's not safe for use, bail out of the open.
1919 if (osize
> max_osize
) {
1920 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1921 VDEV_AUX_OPEN_FAILED
);
1922 return (SET_ERROR(ENXIO
));
1926 * Reset the vdev_reopening flag so that we actually close
1927 * the vdev on error.
1929 vd
->vdev_reopening
= B_FALSE
;
1930 if (zio_injection_enabled
&& error
== 0)
1931 error
= zio_handle_device_injection(vd
, NULL
, SET_ERROR(ENXIO
));
1934 if (vd
->vdev_removed
&&
1935 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1936 vd
->vdev_removed
= B_FALSE
;
1938 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
1939 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
1940 vd
->vdev_stat
.vs_aux
);
1942 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1943 vd
->vdev_stat
.vs_aux
);
1948 vd
->vdev_removed
= B_FALSE
;
1951 * Recheck the faulted flag now that we have confirmed that
1952 * the vdev is accessible. If we're faulted, bail.
1954 if (vd
->vdev_faulted
) {
1955 ASSERT(vd
->vdev_children
== 0);
1956 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1957 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1958 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1959 vd
->vdev_label_aux
);
1960 return (SET_ERROR(ENXIO
));
1963 if (vd
->vdev_degraded
) {
1964 ASSERT(vd
->vdev_children
== 0);
1965 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1966 VDEV_AUX_ERR_EXCEEDED
);
1968 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1972 * For hole or missing vdevs we just return success.
1974 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1977 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1978 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1979 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1985 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1986 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1988 if (vd
->vdev_children
== 0) {
1989 if (osize
< SPA_MINDEVSIZE
) {
1990 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1991 VDEV_AUX_TOO_SMALL
);
1992 return (SET_ERROR(EOVERFLOW
));
1995 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1996 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1997 VDEV_LABEL_END_SIZE
);
1999 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
2000 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
2001 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2002 VDEV_AUX_TOO_SMALL
);
2003 return (SET_ERROR(EOVERFLOW
));
2007 max_asize
= max_osize
;
2011 * If the vdev was expanded, record this so that we can re-create the
2012 * uberblock rings in labels {2,3}, during the next sync.
2014 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
2015 vd
->vdev_copy_uberblocks
= B_TRUE
;
2017 vd
->vdev_psize
= psize
;
2020 * Make sure the allocatable size hasn't shrunk too much.
2022 if (asize
< vd
->vdev_min_asize
) {
2023 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2024 VDEV_AUX_BAD_LABEL
);
2025 return (SET_ERROR(EINVAL
));
2029 * We can always set the logical/physical ashift members since
2030 * their values are only used to calculate the vdev_ashift when
2031 * the device is first added to the config. These values should
2032 * not be used for anything else since they may change whenever
2033 * the device is reopened and we don't store them in the label.
2035 vd
->vdev_physical_ashift
=
2036 MAX(physical_ashift
, vd
->vdev_physical_ashift
);
2037 vd
->vdev_logical_ashift
= MAX(logical_ashift
,
2038 vd
->vdev_logical_ashift
);
2040 if (vd
->vdev_asize
== 0) {
2042 * This is the first-ever open, so use the computed values.
2043 * For compatibility, a different ashift can be requested.
2045 vd
->vdev_asize
= asize
;
2046 vd
->vdev_max_asize
= max_asize
;
2049 * If the vdev_ashift was not overriden at creation time,
2050 * then set it the logical ashift and optimize the ashift.
2052 if (vd
->vdev_ashift
== 0) {
2053 vd
->vdev_ashift
= vd
->vdev_logical_ashift
;
2055 if (vd
->vdev_logical_ashift
> ASHIFT_MAX
) {
2056 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2057 VDEV_AUX_ASHIFT_TOO_BIG
);
2058 return (SET_ERROR(EDOM
));
2061 if (vd
->vdev_top
== vd
) {
2062 vdev_ashift_optimize(vd
);
2065 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
2066 vd
->vdev_ashift
> ASHIFT_MAX
)) {
2067 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2068 VDEV_AUX_BAD_ASHIFT
);
2069 return (SET_ERROR(EDOM
));
2073 * Make sure the alignment required hasn't increased.
2075 if (vd
->vdev_ashift
> vd
->vdev_top
->vdev_ashift
&&
2076 vd
->vdev_ops
->vdev_op_leaf
) {
2077 (void) zfs_ereport_post(
2078 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
2079 spa
, vd
, NULL
, NULL
, 0);
2080 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2081 VDEV_AUX_BAD_LABEL
);
2082 return (SET_ERROR(EDOM
));
2084 vd
->vdev_max_asize
= max_asize
;
2088 * If all children are healthy we update asize if either:
2089 * The asize has increased, due to a device expansion caused by dynamic
2090 * LUN growth or vdev replacement, and automatic expansion is enabled;
2091 * making the additional space available.
2093 * The asize has decreased, due to a device shrink usually caused by a
2094 * vdev replace with a smaller device. This ensures that calculations
2095 * based of max_asize and asize e.g. esize are always valid. It's safe
2096 * to do this as we've already validated that asize is greater than
2099 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
2100 ((asize
> vd
->vdev_asize
&&
2101 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
2102 (asize
< vd
->vdev_asize
)))
2103 vd
->vdev_asize
= asize
;
2105 vdev_set_min_asize(vd
);
2108 * Ensure we can issue some IO before declaring the
2109 * vdev open for business.
2111 if (vd
->vdev_ops
->vdev_op_leaf
&&
2112 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
2113 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2114 VDEV_AUX_ERR_EXCEEDED
);
2119 * Track the the minimum allocation size.
2121 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
2122 vd
->vdev_islog
== 0 && vd
->vdev_aux
== NULL
) {
2123 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
2124 if (min_alloc
< spa
->spa_min_alloc
)
2125 spa
->spa_min_alloc
= min_alloc
;
2129 * If this is a leaf vdev, assess whether a resilver is needed.
2130 * But don't do this if we are doing a reopen for a scrub, since
2131 * this would just restart the scrub we are already doing.
2133 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
)
2134 dsl_scan_assess_vdev(spa
->spa_dsl_pool
, vd
);
2140 * Called once the vdevs are all opened, this routine validates the label
2141 * contents. This needs to be done before vdev_load() so that we don't
2142 * inadvertently do repair I/Os to the wrong device.
2144 * This function will only return failure if one of the vdevs indicates that it
2145 * has since been destroyed or exported. This is only possible if
2146 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
2147 * will be updated but the function will return 0.
2150 vdev_validate(vdev_t
*vd
)
2152 spa_t
*spa
= vd
->vdev_spa
;
2154 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
2159 if (vdev_validate_skip
)
2162 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
2163 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
2164 return (SET_ERROR(EBADF
));
2167 * If the device has already failed, or was marked offline, don't do
2168 * any further validation. Otherwise, label I/O will fail and we will
2169 * overwrite the previous state.
2171 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
2175 * If we are performing an extreme rewind, we allow for a label that
2176 * was modified at a point after the current txg.
2177 * If config lock is not held do not check for the txg. spa_sync could
2178 * be updating the vdev's label before updating spa_last_synced_txg.
2180 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
2181 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
2184 txg
= spa_last_synced_txg(spa
);
2186 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
2187 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2188 VDEV_AUX_BAD_LABEL
);
2189 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
2190 "txg %llu", (u_longlong_t
)txg
);
2195 * Determine if this vdev has been split off into another
2196 * pool. If so, then refuse to open it.
2198 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
2199 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
2200 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2201 VDEV_AUX_SPLIT_POOL
);
2203 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
2207 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
2208 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2209 VDEV_AUX_CORRUPT_DATA
);
2211 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2212 ZPOOL_CONFIG_POOL_GUID
);
2217 * If config is not trusted then ignore the spa guid check. This is
2218 * necessary because if the machine crashed during a re-guid the new
2219 * guid might have been written to all of the vdev labels, but not the
2220 * cached config. The check will be performed again once we have the
2221 * trusted config from the MOS.
2223 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
2224 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2225 VDEV_AUX_CORRUPT_DATA
);
2227 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
2228 "match config (%llu != %llu)", (u_longlong_t
)guid
,
2229 (u_longlong_t
)spa_guid(spa
));
2233 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
2234 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
2238 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
2239 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2240 VDEV_AUX_CORRUPT_DATA
);
2242 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2247 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
2249 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2250 VDEV_AUX_CORRUPT_DATA
);
2252 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2253 ZPOOL_CONFIG_TOP_GUID
);
2258 * If this vdev just became a top-level vdev because its sibling was
2259 * detached, it will have adopted the parent's vdev guid -- but the
2260 * label may or may not be on disk yet. Fortunately, either version
2261 * of the label will have the same top guid, so if we're a top-level
2262 * vdev, we can safely compare to that instead.
2263 * However, if the config comes from a cachefile that failed to update
2264 * after the detach, a top-level vdev will appear as a non top-level
2265 * vdev in the config. Also relax the constraints if we perform an
2268 * If we split this vdev off instead, then we also check the
2269 * original pool's guid. We don't want to consider the vdev
2270 * corrupt if it is partway through a split operation.
2272 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
2273 boolean_t mismatch
= B_FALSE
;
2274 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
2275 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
2278 if (vd
->vdev_guid
!= top_guid
&&
2279 vd
->vdev_top
->vdev_guid
!= guid
)
2284 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2285 VDEV_AUX_CORRUPT_DATA
);
2287 vdev_dbgmsg(vd
, "vdev_validate: config guid "
2288 "doesn't match label guid");
2289 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
2290 (u_longlong_t
)vd
->vdev_guid
,
2291 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
2292 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
2293 "aux_guid %llu", (u_longlong_t
)guid
,
2294 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
2299 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
2301 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2302 VDEV_AUX_CORRUPT_DATA
);
2304 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2305 ZPOOL_CONFIG_POOL_STATE
);
2312 * If this is a verbatim import, no need to check the
2313 * state of the pool.
2315 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
2316 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
2317 state
!= POOL_STATE_ACTIVE
) {
2318 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
2319 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
2320 return (SET_ERROR(EBADF
));
2324 * If we were able to open and validate a vdev that was
2325 * previously marked permanently unavailable, clear that state
2328 if (vd
->vdev_not_present
)
2329 vd
->vdev_not_present
= 0;
2335 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2337 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2338 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2339 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2340 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2341 dvd
->vdev_path
, svd
->vdev_path
);
2342 spa_strfree(dvd
->vdev_path
);
2343 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2345 } else if (svd
->vdev_path
!= NULL
) {
2346 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2347 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2348 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2353 * Recursively copy vdev paths from one vdev to another. Source and destination
2354 * vdev trees must have same geometry otherwise return error. Intended to copy
2355 * paths from userland config into MOS config.
2358 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2360 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2361 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2362 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2365 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2366 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2367 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2368 return (SET_ERROR(EINVAL
));
2371 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2372 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2373 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2374 (u_longlong_t
)dvd
->vdev_guid
);
2375 return (SET_ERROR(EINVAL
));
2378 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2379 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2380 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2381 (u_longlong_t
)dvd
->vdev_children
);
2382 return (SET_ERROR(EINVAL
));
2385 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2386 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2387 dvd
->vdev_child
[i
]);
2392 if (svd
->vdev_ops
->vdev_op_leaf
)
2393 vdev_copy_path_impl(svd
, dvd
);
2399 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2401 ASSERT(stvd
->vdev_top
== stvd
);
2402 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2404 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2405 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2408 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2412 * The idea here is that while a vdev can shift positions within
2413 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2414 * step outside of it.
2416 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2418 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2421 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2423 vdev_copy_path_impl(vd
, dvd
);
2427 * Recursively copy vdev paths from one root vdev to another. Source and
2428 * destination vdev trees may differ in geometry. For each destination leaf
2429 * vdev, search a vdev with the same guid and top vdev id in the source.
2430 * Intended to copy paths from userland config into MOS config.
2433 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2435 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2436 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2437 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2439 for (uint64_t i
= 0; i
< children
; i
++) {
2440 vdev_copy_path_search(srvd
->vdev_child
[i
],
2441 drvd
->vdev_child
[i
]);
2446 * Close a virtual device.
2449 vdev_close(vdev_t
*vd
)
2451 vdev_t
*pvd
= vd
->vdev_parent
;
2452 spa_t
*spa __maybe_unused
= vd
->vdev_spa
;
2455 ASSERT(vd
->vdev_open_thread
== curthread
||
2456 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2459 * If our parent is reopening, then we are as well, unless we are
2462 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2463 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2465 vd
->vdev_ops
->vdev_op_close(vd
);
2467 vdev_cache_purge(vd
);
2470 * We record the previous state before we close it, so that if we are
2471 * doing a reopen(), we don't generate FMA ereports if we notice that
2472 * it's still faulted.
2474 vd
->vdev_prevstate
= vd
->vdev_state
;
2476 if (vd
->vdev_offline
)
2477 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2479 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2480 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2484 vdev_hold(vdev_t
*vd
)
2486 spa_t
*spa
= vd
->vdev_spa
;
2488 ASSERT(spa_is_root(spa
));
2489 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2492 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2493 vdev_hold(vd
->vdev_child
[c
]);
2495 if (vd
->vdev_ops
->vdev_op_leaf
)
2496 vd
->vdev_ops
->vdev_op_hold(vd
);
2500 vdev_rele(vdev_t
*vd
)
2502 ASSERT(spa_is_root(vd
->vdev_spa
));
2503 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2504 vdev_rele(vd
->vdev_child
[c
]);
2506 if (vd
->vdev_ops
->vdev_op_leaf
)
2507 vd
->vdev_ops
->vdev_op_rele(vd
);
2511 * Reopen all interior vdevs and any unopened leaves. We don't actually
2512 * reopen leaf vdevs which had previously been opened as they might deadlock
2513 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2514 * If the leaf has never been opened then open it, as usual.
2517 vdev_reopen(vdev_t
*vd
)
2519 spa_t
*spa
= vd
->vdev_spa
;
2521 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2523 /* set the reopening flag unless we're taking the vdev offline */
2524 vd
->vdev_reopening
= !vd
->vdev_offline
;
2526 (void) vdev_open(vd
);
2529 * Call vdev_validate() here to make sure we have the same device.
2530 * Otherwise, a device with an invalid label could be successfully
2531 * opened in response to vdev_reopen().
2534 (void) vdev_validate_aux(vd
);
2535 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2536 vd
->vdev_aux
== &spa
->spa_l2cache
) {
2538 * In case the vdev is present we should evict all ARC
2539 * buffers and pointers to log blocks and reclaim their
2540 * space before restoring its contents to L2ARC.
2542 if (l2arc_vdev_present(vd
)) {
2543 l2arc_rebuild_vdev(vd
, B_TRUE
);
2545 l2arc_add_vdev(spa
, vd
);
2547 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
2548 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
2551 (void) vdev_validate(vd
);
2555 * Reassess parent vdev's health.
2557 vdev_propagate_state(vd
);
2561 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2566 * Normally, partial opens (e.g. of a mirror) are allowed.
2567 * For a create, however, we want to fail the request if
2568 * there are any components we can't open.
2570 error
= vdev_open(vd
);
2572 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2574 return (error
? error
: SET_ERROR(ENXIO
));
2578 * Recursively load DTLs and initialize all labels.
2580 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2581 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2582 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2591 vdev_metaslab_set_size(vdev_t
*vd
)
2593 uint64_t asize
= vd
->vdev_asize
;
2594 uint64_t ms_count
= asize
>> zfs_vdev_default_ms_shift
;
2598 * There are two dimensions to the metaslab sizing calculation:
2599 * the size of the metaslab and the count of metaslabs per vdev.
2601 * The default values used below are a good balance between memory
2602 * usage (larger metaslab size means more memory needed for loaded
2603 * metaslabs; more metaslabs means more memory needed for the
2604 * metaslab_t structs), metaslab load time (larger metaslabs take
2605 * longer to load), and metaslab sync time (more metaslabs means
2606 * more time spent syncing all of them).
2608 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2609 * The range of the dimensions are as follows:
2611 * 2^29 <= ms_size <= 2^34
2612 * 16 <= ms_count <= 131,072
2614 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2615 * at least 512MB (2^29) to minimize fragmentation effects when
2616 * testing with smaller devices. However, the count constraint
2617 * of at least 16 metaslabs will override this minimum size goal.
2619 * On the upper end of vdev sizes, we aim for a maximum metaslab
2620 * size of 16GB. However, we will cap the total count to 2^17
2621 * metaslabs to keep our memory footprint in check and let the
2622 * metaslab size grow from there if that limit is hit.
2624 * The net effect of applying above constrains is summarized below.
2626 * vdev size metaslab count
2627 * --------------|-----------------
2629 * 8GB - 100GB one per 512MB
2631 * 3TB - 2PB one per 16GB
2633 * --------------------------------
2635 * Finally, note that all of the above calculate the initial
2636 * number of metaslabs. Expanding a top-level vdev will result
2637 * in additional metaslabs being allocated making it possible
2638 * to exceed the zfs_vdev_ms_count_limit.
2641 if (ms_count
< zfs_vdev_min_ms_count
)
2642 ms_shift
= highbit64(asize
/ zfs_vdev_min_ms_count
);
2643 else if (ms_count
> zfs_vdev_default_ms_count
)
2644 ms_shift
= highbit64(asize
/ zfs_vdev_default_ms_count
);
2646 ms_shift
= zfs_vdev_default_ms_shift
;
2648 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2649 ms_shift
= SPA_MAXBLOCKSHIFT
;
2650 } else if (ms_shift
> zfs_vdev_max_ms_shift
) {
2651 ms_shift
= zfs_vdev_max_ms_shift
;
2652 /* cap the total count to constrain memory footprint */
2653 if ((asize
>> ms_shift
) > zfs_vdev_ms_count_limit
)
2654 ms_shift
= highbit64(asize
/ zfs_vdev_ms_count_limit
);
2657 vd
->vdev_ms_shift
= ms_shift
;
2658 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2662 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2664 ASSERT(vd
== vd
->vdev_top
);
2665 /* indirect vdevs don't have metaslabs or dtls */
2666 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2667 ASSERT(ISP2(flags
));
2668 ASSERT(spa_writeable(vd
->vdev_spa
));
2670 if (flags
& VDD_METASLAB
)
2671 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2673 if (flags
& VDD_DTL
)
2674 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2676 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2680 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2682 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2683 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2685 if (vd
->vdev_ops
->vdev_op_leaf
)
2686 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2692 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2693 * the vdev has less than perfect replication. There are four kinds of DTL:
2695 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2697 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2699 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2700 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2701 * txgs that was scrubbed.
2703 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2704 * persistent errors or just some device being offline.
2705 * Unlike the other three, the DTL_OUTAGE map is not generally
2706 * maintained; it's only computed when needed, typically to
2707 * determine whether a device can be detached.
2709 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2710 * either has the data or it doesn't.
2712 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2713 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2714 * if any child is less than fully replicated, then so is its parent.
2715 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2716 * comprising only those txgs which appear in 'maxfaults' or more children;
2717 * those are the txgs we don't have enough replication to read. For example,
2718 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2719 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2720 * two child DTL_MISSING maps.
2722 * It should be clear from the above that to compute the DTLs and outage maps
2723 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2724 * Therefore, that is all we keep on disk. When loading the pool, or after
2725 * a configuration change, we generate all other DTLs from first principles.
2728 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2730 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2732 ASSERT(t
< DTL_TYPES
);
2733 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2734 ASSERT(spa_writeable(vd
->vdev_spa
));
2736 mutex_enter(&vd
->vdev_dtl_lock
);
2737 if (!range_tree_contains(rt
, txg
, size
))
2738 range_tree_add(rt
, txg
, size
);
2739 mutex_exit(&vd
->vdev_dtl_lock
);
2743 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2745 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2746 boolean_t dirty
= B_FALSE
;
2748 ASSERT(t
< DTL_TYPES
);
2749 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2752 * While we are loading the pool, the DTLs have not been loaded yet.
2753 * This isn't a problem but it can result in devices being tried
2754 * which are known to not have the data. In which case, the import
2755 * is relying on the checksum to ensure that we get the right data.
2756 * Note that while importing we are only reading the MOS, which is
2757 * always checksummed.
2759 mutex_enter(&vd
->vdev_dtl_lock
);
2760 if (!range_tree_is_empty(rt
))
2761 dirty
= range_tree_contains(rt
, txg
, size
);
2762 mutex_exit(&vd
->vdev_dtl_lock
);
2768 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2770 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2773 mutex_enter(&vd
->vdev_dtl_lock
);
2774 empty
= range_tree_is_empty(rt
);
2775 mutex_exit(&vd
->vdev_dtl_lock
);
2781 * Check if the txg falls within the range which must be
2782 * resilvered. DVAs outside this range can always be skipped.
2785 vdev_default_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2786 uint64_t phys_birth
)
2788 /* Set by sequential resilver. */
2789 if (phys_birth
== TXG_UNKNOWN
)
2792 return (vdev_dtl_contains(vd
, DTL_PARTIAL
, phys_birth
, 1));
2796 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered.
2799 vdev_dtl_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2800 uint64_t phys_birth
)
2802 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2804 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2805 vd
->vdev_ops
->vdev_op_leaf
)
2808 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, dva
, psize
,
2813 * Returns the lowest txg in the DTL range.
2816 vdev_dtl_min(vdev_t
*vd
)
2818 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2819 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2820 ASSERT0(vd
->vdev_children
);
2822 return (range_tree_min(vd
->vdev_dtl
[DTL_MISSING
]) - 1);
2826 * Returns the highest txg in the DTL.
2829 vdev_dtl_max(vdev_t
*vd
)
2831 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2832 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2833 ASSERT0(vd
->vdev_children
);
2835 return (range_tree_max(vd
->vdev_dtl
[DTL_MISSING
]));
2839 * Determine if a resilvering vdev should remove any DTL entries from
2840 * its range. If the vdev was resilvering for the entire duration of the
2841 * scan then it should excise that range from its DTLs. Otherwise, this
2842 * vdev is considered partially resilvered and should leave its DTL
2843 * entries intact. The comment in vdev_dtl_reassess() describes how we
2847 vdev_dtl_should_excise(vdev_t
*vd
, boolean_t rebuild_done
)
2849 ASSERT0(vd
->vdev_children
);
2851 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2854 if (vd
->vdev_resilver_deferred
)
2857 if (range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
2861 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
2862 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
2864 /* Rebuild not initiated by attach */
2865 if (vd
->vdev_rebuild_txg
== 0)
2869 * When a rebuild completes without error then all missing data
2870 * up to the rebuild max txg has been reconstructed and the DTL
2871 * is eligible for excision.
2873 if (vrp
->vrp_rebuild_state
== VDEV_REBUILD_COMPLETE
&&
2874 vdev_dtl_max(vd
) <= vrp
->vrp_max_txg
) {
2875 ASSERT3U(vrp
->vrp_min_txg
, <=, vdev_dtl_min(vd
));
2876 ASSERT3U(vrp
->vrp_min_txg
, <, vd
->vdev_rebuild_txg
);
2877 ASSERT3U(vd
->vdev_rebuild_txg
, <=, vrp
->vrp_max_txg
);
2881 dsl_scan_t
*scn
= vd
->vdev_spa
->spa_dsl_pool
->dp_scan
;
2882 dsl_scan_phys_t
*scnp __maybe_unused
= &scn
->scn_phys
;
2884 /* Resilver not initiated by attach */
2885 if (vd
->vdev_resilver_txg
== 0)
2889 * When a resilver is initiated the scan will assign the
2890 * scn_max_txg value to the highest txg value that exists
2891 * in all DTLs. If this device's max DTL is not part of this
2892 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
2893 * then it is not eligible for excision.
2895 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
2896 ASSERT3U(scnp
->scn_min_txg
, <=, vdev_dtl_min(vd
));
2897 ASSERT3U(scnp
->scn_min_txg
, <, vd
->vdev_resilver_txg
);
2898 ASSERT3U(vd
->vdev_resilver_txg
, <=, scnp
->scn_max_txg
);
2907 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2908 * write operations will be issued to the pool.
2911 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
,
2912 boolean_t scrub_done
, boolean_t rebuild_done
)
2914 spa_t
*spa
= vd
->vdev_spa
;
2918 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2920 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2921 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
2922 scrub_txg
, scrub_done
, rebuild_done
);
2924 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
2927 if (vd
->vdev_ops
->vdev_op_leaf
) {
2928 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2929 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
2930 boolean_t check_excise
= B_FALSE
;
2931 boolean_t wasempty
= B_TRUE
;
2933 mutex_enter(&vd
->vdev_dtl_lock
);
2936 * If requested, pretend the scan or rebuild completed cleanly.
2938 if (zfs_scan_ignore_errors
) {
2940 scn
->scn_phys
.scn_errors
= 0;
2942 vr
->vr_rebuild_phys
.vrp_errors
= 0;
2945 if (scrub_txg
!= 0 &&
2946 !range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
2948 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
2949 "dtl:%llu/%llu errors:%llu",
2950 (u_longlong_t
)vd
->vdev_guid
, (u_longlong_t
)txg
,
2951 (u_longlong_t
)scrub_txg
, spa
->spa_scrub_started
,
2952 (u_longlong_t
)vdev_dtl_min(vd
),
2953 (u_longlong_t
)vdev_dtl_max(vd
),
2954 (u_longlong_t
)(scn
? scn
->scn_phys
.scn_errors
: 0));
2958 * If we've completed a scrub/resilver or a rebuild cleanly
2959 * then determine if this vdev should remove any DTLs. We
2960 * only want to excise regions on vdevs that were available
2961 * during the entire duration of this scan.
2964 vr
!= NULL
&& vr
->vr_rebuild_phys
.vrp_errors
== 0) {
2965 check_excise
= B_TRUE
;
2967 if (spa
->spa_scrub_started
||
2968 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) {
2969 check_excise
= B_TRUE
;
2973 if (scrub_txg
&& check_excise
&&
2974 vdev_dtl_should_excise(vd
, rebuild_done
)) {
2976 * We completed a scrub, resilver or rebuild up to
2977 * scrub_txg. If we did it without rebooting, then
2978 * the scrub dtl will be valid, so excise the old
2979 * region and fold in the scrub dtl. Otherwise,
2980 * leave the dtl as-is if there was an error.
2982 * There's little trick here: to excise the beginning
2983 * of the DTL_MISSING map, we put it into a reference
2984 * tree and then add a segment with refcnt -1 that
2985 * covers the range [0, scrub_txg). This means
2986 * that each txg in that range has refcnt -1 or 0.
2987 * We then add DTL_SCRUB with a refcnt of 2, so that
2988 * entries in the range [0, scrub_txg) will have a
2989 * positive refcnt -- either 1 or 2. We then convert
2990 * the reference tree into the new DTL_MISSING map.
2992 space_reftree_create(&reftree
);
2993 space_reftree_add_map(&reftree
,
2994 vd
->vdev_dtl
[DTL_MISSING
], 1);
2995 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
2996 space_reftree_add_map(&reftree
,
2997 vd
->vdev_dtl
[DTL_SCRUB
], 2);
2998 space_reftree_generate_map(&reftree
,
2999 vd
->vdev_dtl
[DTL_MISSING
], 1);
3000 space_reftree_destroy(&reftree
);
3002 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3003 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
3004 (u_longlong_t
)vdev_dtl_min(vd
),
3005 (u_longlong_t
)vdev_dtl_max(vd
));
3006 } else if (!wasempty
) {
3007 zfs_dbgmsg("DTL_MISSING is now empty");
3010 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
3011 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3012 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
3014 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
3015 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
3016 if (!vdev_readable(vd
))
3017 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
3019 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3020 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
3023 * If the vdev was resilvering or rebuilding and no longer
3024 * has any DTLs then reset the appropriate flag and dirty
3025 * the top level so that we persist the change.
3028 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3029 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
3030 if (vd
->vdev_rebuild_txg
!= 0) {
3031 vd
->vdev_rebuild_txg
= 0;
3032 vdev_config_dirty(vd
->vdev_top
);
3033 } else if (vd
->vdev_resilver_txg
!= 0) {
3034 vd
->vdev_resilver_txg
= 0;
3035 vdev_config_dirty(vd
->vdev_top
);
3039 mutex_exit(&vd
->vdev_dtl_lock
);
3042 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
3046 mutex_enter(&vd
->vdev_dtl_lock
);
3047 for (int t
= 0; t
< DTL_TYPES
; t
++) {
3048 /* account for child's outage in parent's missing map */
3049 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
3051 continue; /* leaf vdevs only */
3052 if (t
== DTL_PARTIAL
)
3053 minref
= 1; /* i.e. non-zero */
3054 else if (vdev_get_nparity(vd
) != 0)
3055 minref
= vdev_get_nparity(vd
) + 1; /* RAID-Z, dRAID */
3057 minref
= vd
->vdev_children
; /* any kind of mirror */
3058 space_reftree_create(&reftree
);
3059 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3060 vdev_t
*cvd
= vd
->vdev_child
[c
];
3061 mutex_enter(&cvd
->vdev_dtl_lock
);
3062 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
3063 mutex_exit(&cvd
->vdev_dtl_lock
);
3065 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
3066 space_reftree_destroy(&reftree
);
3068 mutex_exit(&vd
->vdev_dtl_lock
);
3072 vdev_dtl_load(vdev_t
*vd
)
3074 spa_t
*spa
= vd
->vdev_spa
;
3075 objset_t
*mos
= spa
->spa_meta_objset
;
3079 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
3080 ASSERT(vdev_is_concrete(vd
));
3082 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
3083 vd
->vdev_dtl_object
, 0, -1ULL, 0);
3086 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3088 rt
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3089 error
= space_map_load(vd
->vdev_dtl_sm
, rt
, SM_ALLOC
);
3091 mutex_enter(&vd
->vdev_dtl_lock
);
3092 range_tree_walk(rt
, range_tree_add
,
3093 vd
->vdev_dtl
[DTL_MISSING
]);
3094 mutex_exit(&vd
->vdev_dtl_lock
);
3097 range_tree_vacate(rt
, NULL
, NULL
);
3098 range_tree_destroy(rt
);
3103 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3104 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
3113 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3115 spa_t
*spa
= vd
->vdev_spa
;
3116 objset_t
*mos
= spa
->spa_meta_objset
;
3117 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
3120 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
3123 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
3124 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
3125 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
3127 ASSERT(string
!= NULL
);
3128 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
3129 1, strlen(string
) + 1, string
, tx
));
3131 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
3132 spa_activate_allocation_classes(spa
, tx
);
3137 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
3139 spa_t
*spa
= vd
->vdev_spa
;
3141 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
3142 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3147 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
3149 spa_t
*spa
= vd
->vdev_spa
;
3150 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
3151 DMU_OT_NONE
, 0, tx
);
3154 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3161 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3163 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
3164 vd
->vdev_ops
!= &vdev_missing_ops
&&
3165 vd
->vdev_ops
!= &vdev_root_ops
&&
3166 !vd
->vdev_top
->vdev_removing
) {
3167 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
3168 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
3170 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
3171 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
3172 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
3173 vdev_zap_allocation_data(vd
, tx
);
3177 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3178 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
3183 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
3185 spa_t
*spa
= vd
->vdev_spa
;
3186 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
3187 objset_t
*mos
= spa
->spa_meta_objset
;
3188 range_tree_t
*rtsync
;
3190 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
3192 ASSERT(vdev_is_concrete(vd
));
3193 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3195 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3197 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
3198 mutex_enter(&vd
->vdev_dtl_lock
);
3199 space_map_free(vd
->vdev_dtl_sm
, tx
);
3200 space_map_close(vd
->vdev_dtl_sm
);
3201 vd
->vdev_dtl_sm
= NULL
;
3202 mutex_exit(&vd
->vdev_dtl_lock
);
3205 * We only destroy the leaf ZAP for detached leaves or for
3206 * removed log devices. Removed data devices handle leaf ZAP
3207 * cleanup later, once cancellation is no longer possible.
3209 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
3210 vd
->vdev_top
->vdev_islog
)) {
3211 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
3212 vd
->vdev_leaf_zap
= 0;
3219 if (vd
->vdev_dtl_sm
== NULL
) {
3220 uint64_t new_object
;
3222 new_object
= space_map_alloc(mos
, zfs_vdev_dtl_sm_blksz
, tx
);
3223 VERIFY3U(new_object
, !=, 0);
3225 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
3227 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3230 rtsync
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3232 mutex_enter(&vd
->vdev_dtl_lock
);
3233 range_tree_walk(rt
, range_tree_add
, rtsync
);
3234 mutex_exit(&vd
->vdev_dtl_lock
);
3236 space_map_truncate(vd
->vdev_dtl_sm
, zfs_vdev_dtl_sm_blksz
, tx
);
3237 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
3238 range_tree_vacate(rtsync
, NULL
, NULL
);
3240 range_tree_destroy(rtsync
);
3243 * If the object for the space map has changed then dirty
3244 * the top level so that we update the config.
3246 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
3247 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
3248 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
3249 (u_longlong_t
)object
,
3250 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
3251 vdev_config_dirty(vd
->vdev_top
);
3258 * Determine whether the specified vdev can be offlined/detached/removed
3259 * without losing data.
3262 vdev_dtl_required(vdev_t
*vd
)
3264 spa_t
*spa
= vd
->vdev_spa
;
3265 vdev_t
*tvd
= vd
->vdev_top
;
3266 uint8_t cant_read
= vd
->vdev_cant_read
;
3269 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3271 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
3275 * Temporarily mark the device as unreadable, and then determine
3276 * whether this results in any DTL outages in the top-level vdev.
3277 * If not, we can safely offline/detach/remove the device.
3279 vd
->vdev_cant_read
= B_TRUE
;
3280 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3281 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
3282 vd
->vdev_cant_read
= cant_read
;
3283 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3285 if (!required
&& zio_injection_enabled
) {
3286 required
= !!zio_handle_device_injection(vd
, NULL
,
3294 * Determine if resilver is needed, and if so the txg range.
3297 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
3299 boolean_t needed
= B_FALSE
;
3300 uint64_t thismin
= UINT64_MAX
;
3301 uint64_t thismax
= 0;
3303 if (vd
->vdev_children
== 0) {
3304 mutex_enter(&vd
->vdev_dtl_lock
);
3305 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3306 vdev_writeable(vd
)) {
3308 thismin
= vdev_dtl_min(vd
);
3309 thismax
= vdev_dtl_max(vd
);
3312 mutex_exit(&vd
->vdev_dtl_lock
);
3314 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3315 vdev_t
*cvd
= vd
->vdev_child
[c
];
3316 uint64_t cmin
, cmax
;
3318 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
3319 thismin
= MIN(thismin
, cmin
);
3320 thismax
= MAX(thismax
, cmax
);
3326 if (needed
&& minp
) {
3334 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3335 * will contain either the checkpoint spacemap object or zero if none exists.
3336 * All other errors are returned to the caller.
3339 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
3341 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
3343 if (vd
->vdev_top_zap
== 0) {
3348 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
3349 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
3350 if (error
== ENOENT
) {
3359 vdev_load(vdev_t
*vd
)
3361 int children
= vd
->vdev_children
;
3366 * It's only worthwhile to use the taskq for the root vdev, because the
3367 * slow part is metaslab_init, and that only happens for top-level
3370 if (vd
->vdev_ops
== &vdev_root_ops
&& vd
->vdev_children
> 0) {
3371 tq
= taskq_create("vdev_load", children
, minclsyspri
,
3372 children
, children
, TASKQ_PREPOPULATE
);
3376 * Recursively load all children.
3378 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3379 vdev_t
*cvd
= vd
->vdev_child
[c
];
3381 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
3382 cvd
->vdev_load_error
= vdev_load(cvd
);
3384 VERIFY(taskq_dispatch(tq
, vdev_load_child
,
3385 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
3394 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3395 int error
= vd
->vdev_child
[c
]->vdev_load_error
;
3401 vdev_set_deflate_ratio(vd
);
3404 * On spa_load path, grab the allocation bias from our zap
3406 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3407 spa_t
*spa
= vd
->vdev_spa
;
3410 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3411 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
3414 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
3415 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
3416 } else if (error
!= ENOENT
) {
3417 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3418 VDEV_AUX_CORRUPT_DATA
);
3419 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(top_zap=%llu) "
3420 "failed [error=%d]", vd
->vdev_top_zap
, error
);
3426 * Load any rebuild state from the top-level vdev zap.
3428 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3429 error
= vdev_rebuild_load(vd
);
3430 if (error
&& error
!= ENOTSUP
) {
3431 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3432 VDEV_AUX_CORRUPT_DATA
);
3433 vdev_dbgmsg(vd
, "vdev_load: vdev_rebuild_load "
3434 "failed [error=%d]", error
);
3440 * If this is a top-level vdev, initialize its metaslabs.
3442 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
3443 vdev_metaslab_group_create(vd
);
3445 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
3446 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3447 VDEV_AUX_CORRUPT_DATA
);
3448 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
3449 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
3450 (u_longlong_t
)vd
->vdev_asize
);
3451 return (SET_ERROR(ENXIO
));
3454 error
= vdev_metaslab_init(vd
, 0);
3456 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
3457 "[error=%d]", error
);
3458 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3459 VDEV_AUX_CORRUPT_DATA
);
3463 uint64_t checkpoint_sm_obj
;
3464 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
3465 if (error
== 0 && checkpoint_sm_obj
!= 0) {
3466 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3467 ASSERT(vd
->vdev_asize
!= 0);
3468 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
3470 error
= space_map_open(&vd
->vdev_checkpoint_sm
,
3471 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
3474 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3475 "failed for checkpoint spacemap (obj %llu) "
3477 (u_longlong_t
)checkpoint_sm_obj
, error
);
3480 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3483 * Since the checkpoint_sm contains free entries
3484 * exclusively we can use space_map_allocated() to
3485 * indicate the cumulative checkpointed space that
3488 vd
->vdev_stat
.vs_checkpoint_space
=
3489 -space_map_allocated(vd
->vdev_checkpoint_sm
);
3490 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3491 vd
->vdev_stat
.vs_checkpoint_space
;
3492 } else if (error
!= 0) {
3493 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3494 "checkpoint space map object from vdev ZAP "
3495 "[error=%d]", error
);
3501 * If this is a leaf vdev, load its DTL.
3503 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3504 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3505 VDEV_AUX_CORRUPT_DATA
);
3506 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3507 "[error=%d]", error
);
3511 uint64_t obsolete_sm_object
;
3512 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3513 if (error
== 0 && obsolete_sm_object
!= 0) {
3514 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3515 ASSERT(vd
->vdev_asize
!= 0);
3516 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3518 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3519 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3520 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3521 VDEV_AUX_CORRUPT_DATA
);
3522 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3523 "obsolete spacemap (obj %llu) [error=%d]",
3524 (u_longlong_t
)obsolete_sm_object
, error
);
3527 } else if (error
!= 0) {
3528 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve obsolete "
3529 "space map object from vdev ZAP [error=%d]", error
);
3537 * The special vdev case is used for hot spares and l2cache devices. Its
3538 * sole purpose it to set the vdev state for the associated vdev. To do this,
3539 * we make sure that we can open the underlying device, then try to read the
3540 * label, and make sure that the label is sane and that it hasn't been
3541 * repurposed to another pool.
3544 vdev_validate_aux(vdev_t
*vd
)
3547 uint64_t guid
, version
;
3550 if (!vdev_readable(vd
))
3553 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
3554 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3555 VDEV_AUX_CORRUPT_DATA
);
3559 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
3560 !SPA_VERSION_IS_SUPPORTED(version
) ||
3561 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
3562 guid
!= vd
->vdev_guid
||
3563 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
3564 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3565 VDEV_AUX_CORRUPT_DATA
);
3571 * We don't actually check the pool state here. If it's in fact in
3572 * use by another pool, we update this fact on the fly when requested.
3579 vdev_destroy_ms_flush_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3581 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3583 if (vd
->vdev_top_zap
== 0)
3586 uint64_t object
= 0;
3587 int err
= zap_lookup(mos
, vd
->vdev_top_zap
,
3588 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, sizeof (uint64_t), 1, &object
);
3593 VERIFY0(dmu_object_free(mos
, object
, tx
));
3594 VERIFY0(zap_remove(mos
, vd
->vdev_top_zap
,
3595 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, tx
));
3599 * Free the objects used to store this vdev's spacemaps, and the array
3600 * that points to them.
3603 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3605 if (vd
->vdev_ms_array
== 0)
3608 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3609 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
3610 size_t array_bytes
= array_count
* sizeof (uint64_t);
3611 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
3612 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
3613 array_bytes
, smobj_array
, 0));
3615 for (uint64_t i
= 0; i
< array_count
; i
++) {
3616 uint64_t smobj
= smobj_array
[i
];
3620 space_map_free_obj(mos
, smobj
, tx
);
3623 kmem_free(smobj_array
, array_bytes
);
3624 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
3625 vdev_destroy_ms_flush_data(vd
, tx
);
3626 vd
->vdev_ms_array
= 0;
3630 vdev_remove_empty_log(vdev_t
*vd
, uint64_t txg
)
3632 spa_t
*spa
= vd
->vdev_spa
;
3634 ASSERT(vd
->vdev_islog
);
3635 ASSERT(vd
== vd
->vdev_top
);
3636 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
3638 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3640 vdev_destroy_spacemaps(vd
, tx
);
3641 if (vd
->vdev_top_zap
!= 0) {
3642 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3643 vd
->vdev_top_zap
= 0;
3650 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3653 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3655 ASSERT(vdev_is_concrete(vd
));
3657 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
3659 metaslab_sync_done(msp
, txg
);
3662 metaslab_sync_reassess(vd
->vdev_mg
);
3663 if (vd
->vdev_log_mg
!= NULL
)
3664 metaslab_sync_reassess(vd
->vdev_log_mg
);
3669 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3671 spa_t
*spa
= vd
->vdev_spa
;
3675 ASSERT3U(txg
, ==, spa
->spa_syncing_txg
);
3676 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3677 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3678 ASSERT(vd
->vdev_removing
||
3679 vd
->vdev_ops
== &vdev_indirect_ops
);
3681 vdev_indirect_sync_obsolete(vd
, tx
);
3684 * If the vdev is indirect, it can't have dirty
3685 * metaslabs or DTLs.
3687 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3688 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3689 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3695 ASSERT(vdev_is_concrete(vd
));
3697 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3698 !vd
->vdev_removing
) {
3699 ASSERT(vd
== vd
->vdev_top
);
3700 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3701 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3702 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3703 ASSERT(vd
->vdev_ms_array
!= 0);
3704 vdev_config_dirty(vd
);
3707 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3708 metaslab_sync(msp
, txg
);
3709 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3712 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3713 vdev_dtl_sync(lvd
, txg
);
3716 * If this is an empty log device being removed, destroy the
3717 * metadata associated with it.
3719 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3720 vdev_remove_empty_log(vd
, txg
);
3722 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3727 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3729 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3733 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3734 * not be opened, and no I/O is attempted.
3737 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3741 spa_vdev_state_enter(spa
, SCL_NONE
);
3743 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3744 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3746 if (!vd
->vdev_ops
->vdev_op_leaf
)
3747 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
3752 * If user did a 'zpool offline -f' then make the fault persist across
3755 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3757 * There are two kinds of forced faults: temporary and
3758 * persistent. Temporary faults go away at pool import, while
3759 * persistent faults stay set. Both types of faults can be
3760 * cleared with a zpool clear.
3762 * We tell if a vdev is persistently faulted by looking at the
3763 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3764 * import then it's a persistent fault. Otherwise, it's
3765 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3766 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3767 * tells vdev_config_generate() (which gets run later) to set
3768 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3770 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3771 vd
->vdev_tmpoffline
= B_FALSE
;
3772 aux
= VDEV_AUX_EXTERNAL
;
3774 vd
->vdev_tmpoffline
= B_TRUE
;
3778 * We don't directly use the aux state here, but if we do a
3779 * vdev_reopen(), we need this value to be present to remember why we
3782 vd
->vdev_label_aux
= aux
;
3785 * Faulted state takes precedence over degraded.
3787 vd
->vdev_delayed_close
= B_FALSE
;
3788 vd
->vdev_faulted
= 1ULL;
3789 vd
->vdev_degraded
= 0ULL;
3790 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
3793 * If this device has the only valid copy of the data, then
3794 * back off and simply mark the vdev as degraded instead.
3796 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
3797 vd
->vdev_degraded
= 1ULL;
3798 vd
->vdev_faulted
= 0ULL;
3801 * If we reopen the device and it's not dead, only then do we
3806 if (vdev_readable(vd
))
3807 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
3810 return (spa_vdev_state_exit(spa
, vd
, 0));
3814 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3815 * user that something is wrong. The vdev continues to operate as normal as far
3816 * as I/O is concerned.
3819 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3823 spa_vdev_state_enter(spa
, SCL_NONE
);
3825 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3826 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3828 if (!vd
->vdev_ops
->vdev_op_leaf
)
3829 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
3832 * If the vdev is already faulted, then don't do anything.
3834 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
3835 return (spa_vdev_state_exit(spa
, NULL
, 0));
3837 vd
->vdev_degraded
= 1ULL;
3838 if (!vdev_is_dead(vd
))
3839 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
3842 return (spa_vdev_state_exit(spa
, vd
, 0));
3846 * Online the given vdev.
3848 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3849 * spare device should be detached when the device finishes resilvering.
3850 * Second, the online should be treated like a 'test' online case, so no FMA
3851 * events are generated if the device fails to open.
3854 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
3856 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
3857 boolean_t wasoffline
;
3858 vdev_state_t oldstate
;
3860 spa_vdev_state_enter(spa
, SCL_NONE
);
3862 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3863 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3865 if (!vd
->vdev_ops
->vdev_op_leaf
)
3866 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
3868 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
3869 oldstate
= vd
->vdev_state
;
3872 vd
->vdev_offline
= B_FALSE
;
3873 vd
->vdev_tmpoffline
= B_FALSE
;
3874 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
3875 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
3877 /* XXX - L2ARC 1.0 does not support expansion */
3878 if (!vd
->vdev_aux
) {
3879 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3880 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
3881 spa
->spa_autoexpand
);
3882 vd
->vdev_expansion_time
= gethrestime_sec();
3886 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
3888 if (!vd
->vdev_aux
) {
3889 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3890 pvd
->vdev_expanding
= B_FALSE
;
3894 *newstate
= vd
->vdev_state
;
3895 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
3896 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
3897 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3898 vd
->vdev_parent
->vdev_child
[0] == vd
)
3899 vd
->vdev_unspare
= B_TRUE
;
3901 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
3903 /* XXX - L2ARC 1.0 does not support expansion */
3905 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
3906 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3909 /* Restart initializing if necessary */
3910 mutex_enter(&vd
->vdev_initialize_lock
);
3911 if (vdev_writeable(vd
) &&
3912 vd
->vdev_initialize_thread
== NULL
&&
3913 vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) {
3914 (void) vdev_initialize(vd
);
3916 mutex_exit(&vd
->vdev_initialize_lock
);
3919 * Restart trimming if necessary. We do not restart trimming for cache
3920 * devices here. This is triggered by l2arc_rebuild_vdev()
3921 * asynchronously for the whole device or in l2arc_evict() as it evicts
3922 * space for upcoming writes.
3924 mutex_enter(&vd
->vdev_trim_lock
);
3925 if (vdev_writeable(vd
) && !vd
->vdev_isl2cache
&&
3926 vd
->vdev_trim_thread
== NULL
&&
3927 vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
) {
3928 (void) vdev_trim(vd
, vd
->vdev_trim_rate
, vd
->vdev_trim_partial
,
3929 vd
->vdev_trim_secure
);
3931 mutex_exit(&vd
->vdev_trim_lock
);
3934 (oldstate
< VDEV_STATE_DEGRADED
&&
3935 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
3936 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
3938 return (spa_vdev_state_exit(spa
, vd
, 0));
3942 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3946 uint64_t generation
;
3947 metaslab_group_t
*mg
;
3950 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3952 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3953 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3955 if (!vd
->vdev_ops
->vdev_op_leaf
)
3956 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
3958 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
3959 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3963 generation
= spa
->spa_config_generation
+ 1;
3966 * If the device isn't already offline, try to offline it.
3968 if (!vd
->vdev_offline
) {
3970 * If this device has the only valid copy of some data,
3971 * don't allow it to be offlined. Log devices are always
3974 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3975 vdev_dtl_required(vd
))
3976 return (spa_vdev_state_exit(spa
, NULL
,
3980 * If the top-level is a slog and it has had allocations
3981 * then proceed. We check that the vdev's metaslab group
3982 * is not NULL since it's possible that we may have just
3983 * added this vdev but not yet initialized its metaslabs.
3985 if (tvd
->vdev_islog
&& mg
!= NULL
) {
3987 * Prevent any future allocations.
3989 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
3990 metaslab_group_passivate(mg
);
3991 (void) spa_vdev_state_exit(spa
, vd
, 0);
3993 error
= spa_reset_logs(spa
);
3996 * If the log device was successfully reset but has
3997 * checkpointed data, do not offline it.
4000 tvd
->vdev_checkpoint_sm
!= NULL
) {
4001 ASSERT3U(space_map_allocated(
4002 tvd
->vdev_checkpoint_sm
), !=, 0);
4003 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
4006 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4009 * Check to see if the config has changed.
4011 if (error
|| generation
!= spa
->spa_config_generation
) {
4012 metaslab_group_activate(mg
);
4014 return (spa_vdev_state_exit(spa
,
4016 (void) spa_vdev_state_exit(spa
, vd
, 0);
4019 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
4023 * Offline this device and reopen its top-level vdev.
4024 * If the top-level vdev is a log device then just offline
4025 * it. Otherwise, if this action results in the top-level
4026 * vdev becoming unusable, undo it and fail the request.
4028 vd
->vdev_offline
= B_TRUE
;
4031 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4032 vdev_is_dead(tvd
)) {
4033 vd
->vdev_offline
= B_FALSE
;
4035 return (spa_vdev_state_exit(spa
, NULL
,
4040 * Add the device back into the metaslab rotor so that
4041 * once we online the device it's open for business.
4043 if (tvd
->vdev_islog
&& mg
!= NULL
)
4044 metaslab_group_activate(mg
);
4047 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
4049 return (spa_vdev_state_exit(spa
, vd
, 0));
4053 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4057 mutex_enter(&spa
->spa_vdev_top_lock
);
4058 error
= vdev_offline_locked(spa
, guid
, flags
);
4059 mutex_exit(&spa
->spa_vdev_top_lock
);
4065 * Clear the error counts associated with this vdev. Unlike vdev_online() and
4066 * vdev_offline(), we assume the spa config is locked. We also clear all
4067 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
4070 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
4072 vdev_t
*rvd
= spa
->spa_root_vdev
;
4074 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
4079 vd
->vdev_stat
.vs_read_errors
= 0;
4080 vd
->vdev_stat
.vs_write_errors
= 0;
4081 vd
->vdev_stat
.vs_checksum_errors
= 0;
4082 vd
->vdev_stat
.vs_slow_ios
= 0;
4084 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4085 vdev_clear(spa
, vd
->vdev_child
[c
]);
4088 * It makes no sense to "clear" an indirect vdev.
4090 if (!vdev_is_concrete(vd
))
4094 * If we're in the FAULTED state or have experienced failed I/O, then
4095 * clear the persistent state and attempt to reopen the device. We
4096 * also mark the vdev config dirty, so that the new faulted state is
4097 * written out to disk.
4099 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
4100 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
4102 * When reopening in response to a clear event, it may be due to
4103 * a fmadm repair request. In this case, if the device is
4104 * still broken, we want to still post the ereport again.
4106 vd
->vdev_forcefault
= B_TRUE
;
4108 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
4109 vd
->vdev_cant_read
= B_FALSE
;
4110 vd
->vdev_cant_write
= B_FALSE
;
4111 vd
->vdev_stat
.vs_aux
= 0;
4113 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
4115 vd
->vdev_forcefault
= B_FALSE
;
4117 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
4118 vdev_state_dirty(vd
->vdev_top
);
4120 /* If a resilver isn't required, check if vdevs can be culled */
4121 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
) &&
4122 !dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4123 !dsl_scan_resilver_scheduled(spa
->spa_dsl_pool
))
4124 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
4126 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
4130 * When clearing a FMA-diagnosed fault, we always want to
4131 * unspare the device, as we assume that the original spare was
4132 * done in response to the FMA fault.
4134 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
4135 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4136 vd
->vdev_parent
->vdev_child
[0] == vd
)
4137 vd
->vdev_unspare
= B_TRUE
;
4141 vdev_is_dead(vdev_t
*vd
)
4144 * Holes and missing devices are always considered "dead".
4145 * This simplifies the code since we don't have to check for
4146 * these types of devices in the various code paths.
4147 * Instead we rely on the fact that we skip over dead devices
4148 * before issuing I/O to them.
4150 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
4151 vd
->vdev_ops
== &vdev_hole_ops
||
4152 vd
->vdev_ops
== &vdev_missing_ops
);
4156 vdev_readable(vdev_t
*vd
)
4158 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
4162 vdev_writeable(vdev_t
*vd
)
4164 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
4165 vdev_is_concrete(vd
));
4169 vdev_allocatable(vdev_t
*vd
)
4171 uint64_t state
= vd
->vdev_state
;
4174 * We currently allow allocations from vdevs which may be in the
4175 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
4176 * fails to reopen then we'll catch it later when we're holding
4177 * the proper locks. Note that we have to get the vdev state
4178 * in a local variable because although it changes atomically,
4179 * we're asking two separate questions about it.
4181 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
4182 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
4183 vd
->vdev_mg
->mg_initialized
);
4187 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
4189 ASSERT(zio
->io_vd
== vd
);
4191 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
4194 if (zio
->io_type
== ZIO_TYPE_READ
)
4195 return (!vd
->vdev_cant_read
);
4197 if (zio
->io_type
== ZIO_TYPE_WRITE
)
4198 return (!vd
->vdev_cant_write
);
4204 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
4207 * Exclude the dRAID spare when aggregating to avoid double counting
4208 * the ops and bytes. These IOs are counted by the physical leaves.
4210 if (cvd
->vdev_ops
== &vdev_draid_spare_ops
)
4213 for (int t
= 0; t
< VS_ZIO_TYPES
; t
++) {
4214 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
4215 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
4218 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
4222 * Get extended stats
4225 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
4228 for (t
= 0; t
< ZIO_TYPES
; t
++) {
4229 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
4230 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
4232 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
4233 vsx
->vsx_total_histo
[t
][b
] +=
4234 cvsx
->vsx_total_histo
[t
][b
];
4238 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
4239 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
4240 vsx
->vsx_queue_histo
[t
][b
] +=
4241 cvsx
->vsx_queue_histo
[t
][b
];
4243 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
4244 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
4246 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
4247 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
4249 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
4250 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
4256 vdev_is_spacemap_addressable(vdev_t
*vd
)
4258 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_SPACEMAP_V2
))
4262 * If double-word space map entries are not enabled we assume
4263 * 47 bits of the space map entry are dedicated to the entry's
4264 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4265 * to calculate the maximum address that can be described by a
4266 * space map entry for the given device.
4268 uint64_t shift
= vd
->vdev_ashift
+ SM_OFFSET_BITS
;
4270 if (shift
>= 63) /* detect potential overflow */
4273 return (vd
->vdev_asize
< (1ULL << shift
));
4277 * Get statistics for the given vdev.
4280 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4284 * If we're getting stats on the root vdev, aggregate the I/O counts
4285 * over all top-level vdevs (i.e. the direct children of the root).
4287 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4289 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
4290 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
4293 memset(vsx
, 0, sizeof (*vsx
));
4295 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4296 vdev_t
*cvd
= vd
->vdev_child
[c
];
4297 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
4298 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
4300 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
4302 vdev_get_child_stat(cvd
, vs
, cvs
);
4304 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
4308 * We're a leaf. Just copy our ZIO active queue stats in. The
4309 * other leaf stats are updated in vdev_stat_update().
4314 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
4316 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
4317 vsx
->vsx_active_queue
[t
] =
4318 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
4319 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
4320 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
4326 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4328 vdev_t
*tvd
= vd
->vdev_top
;
4329 mutex_enter(&vd
->vdev_stat_lock
);
4331 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
4332 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
4333 vs
->vs_state
= vd
->vdev_state
;
4334 vs
->vs_rsize
= vdev_get_min_asize(vd
);
4336 if (vd
->vdev_ops
->vdev_op_leaf
) {
4337 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
4338 VDEV_LABEL_END_SIZE
;
4340 * Report initializing progress. Since we don't
4341 * have the initializing locks held, this is only
4342 * an estimate (although a fairly accurate one).
4344 vs
->vs_initialize_bytes_done
=
4345 vd
->vdev_initialize_bytes_done
;
4346 vs
->vs_initialize_bytes_est
=
4347 vd
->vdev_initialize_bytes_est
;
4348 vs
->vs_initialize_state
= vd
->vdev_initialize_state
;
4349 vs
->vs_initialize_action_time
=
4350 vd
->vdev_initialize_action_time
;
4353 * Report manual TRIM progress. Since we don't have
4354 * the manual TRIM locks held, this is only an
4355 * estimate (although fairly accurate one).
4357 vs
->vs_trim_notsup
= !vd
->vdev_has_trim
;
4358 vs
->vs_trim_bytes_done
= vd
->vdev_trim_bytes_done
;
4359 vs
->vs_trim_bytes_est
= vd
->vdev_trim_bytes_est
;
4360 vs
->vs_trim_state
= vd
->vdev_trim_state
;
4361 vs
->vs_trim_action_time
= vd
->vdev_trim_action_time
;
4363 /* Set when there is a deferred resilver. */
4364 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
4368 * Report expandable space on top-level, non-auxiliary devices
4369 * only. The expandable space is reported in terms of metaslab
4370 * sized units since that determines how much space the pool
4373 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
4374 vs
->vs_esize
= P2ALIGN(
4375 vd
->vdev_max_asize
- vd
->vdev_asize
,
4376 1ULL << tvd
->vdev_ms_shift
);
4379 vs
->vs_configured_ashift
= vd
->vdev_top
!= NULL
4380 ? vd
->vdev_top
->vdev_ashift
: vd
->vdev_ashift
;
4381 vs
->vs_logical_ashift
= vd
->vdev_logical_ashift
;
4382 vs
->vs_physical_ashift
= vd
->vdev_physical_ashift
;
4385 * Report fragmentation and rebuild progress for top-level,
4386 * non-auxiliary, concrete devices.
4388 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
4389 vdev_is_concrete(vd
)) {
4391 * The vdev fragmentation rating doesn't take into
4392 * account the embedded slog metaslab (vdev_log_mg).
4393 * Since it's only one metaslab, it would have a tiny
4394 * impact on the overall fragmentation.
4396 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
4397 vd
->vdev_mg
->mg_fragmentation
: 0;
4401 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
4402 mutex_exit(&vd
->vdev_stat_lock
);
4406 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
4408 return (vdev_get_stats_ex(vd
, vs
, NULL
));
4412 vdev_clear_stats(vdev_t
*vd
)
4414 mutex_enter(&vd
->vdev_stat_lock
);
4415 vd
->vdev_stat
.vs_space
= 0;
4416 vd
->vdev_stat
.vs_dspace
= 0;
4417 vd
->vdev_stat
.vs_alloc
= 0;
4418 mutex_exit(&vd
->vdev_stat_lock
);
4422 vdev_scan_stat_init(vdev_t
*vd
)
4424 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4426 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4427 vdev_scan_stat_init(vd
->vdev_child
[c
]);
4429 mutex_enter(&vd
->vdev_stat_lock
);
4430 vs
->vs_scan_processed
= 0;
4431 mutex_exit(&vd
->vdev_stat_lock
);
4435 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
4437 spa_t
*spa
= zio
->io_spa
;
4438 vdev_t
*rvd
= spa
->spa_root_vdev
;
4439 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
4441 uint64_t txg
= zio
->io_txg
;
4442 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4443 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
4444 zio_type_t type
= zio
->io_type
;
4445 int flags
= zio
->io_flags
;
4448 * If this i/o is a gang leader, it didn't do any actual work.
4450 if (zio
->io_gang_tree
)
4453 if (zio
->io_error
== 0) {
4455 * If this is a root i/o, don't count it -- we've already
4456 * counted the top-level vdevs, and vdev_get_stats() will
4457 * aggregate them when asked. This reduces contention on
4458 * the root vdev_stat_lock and implicitly handles blocks
4459 * that compress away to holes, for which there is no i/o.
4460 * (Holes never create vdev children, so all the counters
4461 * remain zero, which is what we want.)
4463 * Note: this only applies to successful i/o (io_error == 0)
4464 * because unlike i/o counts, errors are not additive.
4465 * When reading a ditto block, for example, failure of
4466 * one top-level vdev does not imply a root-level error.
4471 ASSERT(vd
== zio
->io_vd
);
4473 if (flags
& ZIO_FLAG_IO_BYPASS
)
4476 mutex_enter(&vd
->vdev_stat_lock
);
4478 if (flags
& ZIO_FLAG_IO_REPAIR
) {
4480 * Repair is the result of a resilver issued by the
4481 * scan thread (spa_sync).
4483 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4484 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
4485 dsl_scan_phys_t
*scn_phys
= &scn
->scn_phys
;
4486 uint64_t *processed
= &scn_phys
->scn_processed
;
4488 if (vd
->vdev_ops
->vdev_op_leaf
)
4489 atomic_add_64(processed
, psize
);
4490 vs
->vs_scan_processed
+= psize
;
4494 * Repair is the result of a rebuild issued by the
4495 * rebuild thread (vdev_rebuild_thread). To avoid
4496 * double counting repaired bytes the virtual dRAID
4497 * spare vdev is excluded from the processed bytes.
4499 if (zio
->io_priority
== ZIO_PRIORITY_REBUILD
) {
4500 vdev_t
*tvd
= vd
->vdev_top
;
4501 vdev_rebuild_t
*vr
= &tvd
->vdev_rebuild_config
;
4502 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
4503 uint64_t *rebuilt
= &vrp
->vrp_bytes_rebuilt
;
4505 if (vd
->vdev_ops
->vdev_op_leaf
&&
4506 vd
->vdev_ops
!= &vdev_draid_spare_ops
) {
4507 atomic_add_64(rebuilt
, psize
);
4509 vs
->vs_rebuild_processed
+= psize
;
4512 if (flags
& ZIO_FLAG_SELF_HEAL
)
4513 vs
->vs_self_healed
+= psize
;
4517 * The bytes/ops/histograms are recorded at the leaf level and
4518 * aggregated into the higher level vdevs in vdev_get_stats().
4520 if (vd
->vdev_ops
->vdev_op_leaf
&&
4521 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
4522 zio_type_t vs_type
= type
;
4523 zio_priority_t priority
= zio
->io_priority
;
4526 * TRIM ops and bytes are reported to user space as
4527 * ZIO_TYPE_IOCTL. This is done to preserve the
4528 * vdev_stat_t structure layout for user space.
4530 if (type
== ZIO_TYPE_TRIM
)
4531 vs_type
= ZIO_TYPE_IOCTL
;
4534 * Solely for the purposes of 'zpool iostat -lqrw'
4535 * reporting use the priority to catagorize the IO.
4536 * Only the following are reported to user space:
4538 * ZIO_PRIORITY_SYNC_READ,
4539 * ZIO_PRIORITY_SYNC_WRITE,
4540 * ZIO_PRIORITY_ASYNC_READ,
4541 * ZIO_PRIORITY_ASYNC_WRITE,
4542 * ZIO_PRIORITY_SCRUB,
4543 * ZIO_PRIORITY_TRIM.
4545 if (priority
== ZIO_PRIORITY_REBUILD
) {
4546 priority
= ((type
== ZIO_TYPE_WRITE
) ?
4547 ZIO_PRIORITY_ASYNC_WRITE
:
4548 ZIO_PRIORITY_SCRUB
);
4549 } else if (priority
== ZIO_PRIORITY_INITIALIZING
) {
4550 ASSERT3U(type
, ==, ZIO_TYPE_WRITE
);
4551 priority
= ZIO_PRIORITY_ASYNC_WRITE
;
4552 } else if (priority
== ZIO_PRIORITY_REMOVAL
) {
4553 priority
= ((type
== ZIO_TYPE_WRITE
) ?
4554 ZIO_PRIORITY_ASYNC_WRITE
:
4555 ZIO_PRIORITY_ASYNC_READ
);
4558 vs
->vs_ops
[vs_type
]++;
4559 vs
->vs_bytes
[vs_type
] += psize
;
4561 if (flags
& ZIO_FLAG_DELEGATED
) {
4562 vsx
->vsx_agg_histo
[priority
]
4563 [RQ_HISTO(zio
->io_size
)]++;
4565 vsx
->vsx_ind_histo
[priority
]
4566 [RQ_HISTO(zio
->io_size
)]++;
4569 if (zio
->io_delta
&& zio
->io_delay
) {
4570 vsx
->vsx_queue_histo
[priority
]
4571 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
4572 vsx
->vsx_disk_histo
[type
]
4573 [L_HISTO(zio
->io_delay
)]++;
4574 vsx
->vsx_total_histo
[type
]
4575 [L_HISTO(zio
->io_delta
)]++;
4579 mutex_exit(&vd
->vdev_stat_lock
);
4583 if (flags
& ZIO_FLAG_SPECULATIVE
)
4587 * If this is an I/O error that is going to be retried, then ignore the
4588 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4589 * hard errors, when in reality they can happen for any number of
4590 * innocuous reasons (bus resets, MPxIO link failure, etc).
4592 if (zio
->io_error
== EIO
&&
4593 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4597 * Intent logs writes won't propagate their error to the root
4598 * I/O so don't mark these types of failures as pool-level
4601 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4604 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4605 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4606 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4607 spa
->spa_claiming
)) {
4609 * This is either a normal write (not a repair), or it's
4610 * a repair induced by the scrub thread, or it's a repair
4611 * made by zil_claim() during spa_load() in the first txg.
4612 * In the normal case, we commit the DTL change in the same
4613 * txg as the block was born. In the scrub-induced repair
4614 * case, we know that scrubs run in first-pass syncing context,
4615 * so we commit the DTL change in spa_syncing_txg(spa).
4616 * In the zil_claim() case, we commit in spa_first_txg(spa).
4618 * We currently do not make DTL entries for failed spontaneous
4619 * self-healing writes triggered by normal (non-scrubbing)
4620 * reads, because we have no transactional context in which to
4621 * do so -- and it's not clear that it'd be desirable anyway.
4623 if (vd
->vdev_ops
->vdev_op_leaf
) {
4624 uint64_t commit_txg
= txg
;
4625 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4626 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4627 ASSERT(spa_sync_pass(spa
) == 1);
4628 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4629 commit_txg
= spa_syncing_txg(spa
);
4630 } else if (spa
->spa_claiming
) {
4631 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4632 commit_txg
= spa_first_txg(spa
);
4634 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4635 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4637 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4638 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4639 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4642 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4647 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4649 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4650 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4652 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4656 * Update the in-core space usage stats for this vdev, its metaslab class,
4657 * and the root vdev.
4660 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4661 int64_t space_delta
)
4663 int64_t dspace_delta
;
4664 spa_t
*spa
= vd
->vdev_spa
;
4665 vdev_t
*rvd
= spa
->spa_root_vdev
;
4667 ASSERT(vd
== vd
->vdev_top
);
4670 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4671 * factor. We must calculate this here and not at the root vdev
4672 * because the root vdev's psize-to-asize is simply the max of its
4673 * children's, thus not accurate enough for us.
4675 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4677 mutex_enter(&vd
->vdev_stat_lock
);
4678 /* ensure we won't underflow */
4679 if (alloc_delta
< 0) {
4680 ASSERT3U(vd
->vdev_stat
.vs_alloc
, >=, -alloc_delta
);
4683 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4684 vd
->vdev_stat
.vs_space
+= space_delta
;
4685 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4686 mutex_exit(&vd
->vdev_stat_lock
);
4688 /* every class but log contributes to root space stats */
4689 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4690 ASSERT(!vd
->vdev_isl2cache
);
4691 mutex_enter(&rvd
->vdev_stat_lock
);
4692 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4693 rvd
->vdev_stat
.vs_space
+= space_delta
;
4694 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4695 mutex_exit(&rvd
->vdev_stat_lock
);
4697 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4701 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4702 * so that it will be written out next time the vdev configuration is synced.
4703 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4706 vdev_config_dirty(vdev_t
*vd
)
4708 spa_t
*spa
= vd
->vdev_spa
;
4709 vdev_t
*rvd
= spa
->spa_root_vdev
;
4712 ASSERT(spa_writeable(spa
));
4715 * If this is an aux vdev (as with l2cache and spare devices), then we
4716 * update the vdev config manually and set the sync flag.
4718 if (vd
->vdev_aux
!= NULL
) {
4719 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
4723 for (c
= 0; c
< sav
->sav_count
; c
++) {
4724 if (sav
->sav_vdevs
[c
] == vd
)
4728 if (c
== sav
->sav_count
) {
4730 * We're being removed. There's nothing more to do.
4732 ASSERT(sav
->sav_sync
== B_TRUE
);
4736 sav
->sav_sync
= B_TRUE
;
4738 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
4739 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
4740 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
4741 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
4747 * Setting the nvlist in the middle if the array is a little
4748 * sketchy, but it will work.
4750 nvlist_free(aux
[c
]);
4751 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
4757 * The dirty list is protected by the SCL_CONFIG lock. The caller
4758 * must either hold SCL_CONFIG as writer, or must be the sync thread
4759 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4760 * so this is sufficient to ensure mutual exclusion.
4762 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4763 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4764 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4767 for (c
= 0; c
< rvd
->vdev_children
; c
++)
4768 vdev_config_dirty(rvd
->vdev_child
[c
]);
4770 ASSERT(vd
== vd
->vdev_top
);
4772 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
4773 vdev_is_concrete(vd
)) {
4774 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
4780 vdev_config_clean(vdev_t
*vd
)
4782 spa_t
*spa
= vd
->vdev_spa
;
4784 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4785 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4786 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4788 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
4789 list_remove(&spa
->spa_config_dirty_list
, vd
);
4793 * Mark a top-level vdev's state as dirty, so that the next pass of
4794 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4795 * the state changes from larger config changes because they require
4796 * much less locking, and are often needed for administrative actions.
4799 vdev_state_dirty(vdev_t
*vd
)
4801 spa_t
*spa
= vd
->vdev_spa
;
4803 ASSERT(spa_writeable(spa
));
4804 ASSERT(vd
== vd
->vdev_top
);
4807 * The state list is protected by the SCL_STATE lock. The caller
4808 * must either hold SCL_STATE as writer, or must be the sync thread
4809 * (which holds SCL_STATE as reader). There's only one sync thread,
4810 * so this is sufficient to ensure mutual exclusion.
4812 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4813 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4814 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4816 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
4817 vdev_is_concrete(vd
))
4818 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
4822 vdev_state_clean(vdev_t
*vd
)
4824 spa_t
*spa
= vd
->vdev_spa
;
4826 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4827 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4828 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4830 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
4831 list_remove(&spa
->spa_state_dirty_list
, vd
);
4835 * Propagate vdev state up from children to parent.
4838 vdev_propagate_state(vdev_t
*vd
)
4840 spa_t
*spa
= vd
->vdev_spa
;
4841 vdev_t
*rvd
= spa
->spa_root_vdev
;
4842 int degraded
= 0, faulted
= 0;
4846 if (vd
->vdev_children
> 0) {
4847 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4848 child
= vd
->vdev_child
[c
];
4851 * Don't factor holes or indirect vdevs into the
4854 if (!vdev_is_concrete(child
))
4857 if (!vdev_readable(child
) ||
4858 (!vdev_writeable(child
) && spa_writeable(spa
))) {
4860 * Root special: if there is a top-level log
4861 * device, treat the root vdev as if it were
4864 if (child
->vdev_islog
&& vd
== rvd
)
4868 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
4872 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
4876 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
4879 * Root special: if there is a top-level vdev that cannot be
4880 * opened due to corrupted metadata, then propagate the root
4881 * vdev's aux state as 'corrupt' rather than 'insufficient
4884 if (corrupted
&& vd
== rvd
&&
4885 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
4886 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
4887 VDEV_AUX_CORRUPT_DATA
);
4890 if (vd
->vdev_parent
)
4891 vdev_propagate_state(vd
->vdev_parent
);
4895 * Set a vdev's state. If this is during an open, we don't update the parent
4896 * state, because we're in the process of opening children depth-first.
4897 * Otherwise, we propagate the change to the parent.
4899 * If this routine places a device in a faulted state, an appropriate ereport is
4903 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
4905 uint64_t save_state
;
4906 spa_t
*spa
= vd
->vdev_spa
;
4908 if (state
== vd
->vdev_state
) {
4910 * Since vdev_offline() code path is already in an offline
4911 * state we can miss a statechange event to OFFLINE. Check
4912 * the previous state to catch this condition.
4914 if (vd
->vdev_ops
->vdev_op_leaf
&&
4915 (state
== VDEV_STATE_OFFLINE
) &&
4916 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
4917 /* post an offline state change */
4918 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
4920 vd
->vdev_stat
.vs_aux
= aux
;
4924 save_state
= vd
->vdev_state
;
4926 vd
->vdev_state
= state
;
4927 vd
->vdev_stat
.vs_aux
= aux
;
4930 * If we are setting the vdev state to anything but an open state, then
4931 * always close the underlying device unless the device has requested
4932 * a delayed close (i.e. we're about to remove or fault the device).
4933 * Otherwise, we keep accessible but invalid devices open forever.
4934 * We don't call vdev_close() itself, because that implies some extra
4935 * checks (offline, etc) that we don't want here. This is limited to
4936 * leaf devices, because otherwise closing the device will affect other
4939 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
4940 vd
->vdev_ops
->vdev_op_leaf
)
4941 vd
->vdev_ops
->vdev_op_close(vd
);
4943 if (vd
->vdev_removed
&&
4944 state
== VDEV_STATE_CANT_OPEN
&&
4945 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
4947 * If the previous state is set to VDEV_STATE_REMOVED, then this
4948 * device was previously marked removed and someone attempted to
4949 * reopen it. If this failed due to a nonexistent device, then
4950 * keep the device in the REMOVED state. We also let this be if
4951 * it is one of our special test online cases, which is only
4952 * attempting to online the device and shouldn't generate an FMA
4955 vd
->vdev_state
= VDEV_STATE_REMOVED
;
4956 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
4957 } else if (state
== VDEV_STATE_REMOVED
) {
4958 vd
->vdev_removed
= B_TRUE
;
4959 } else if (state
== VDEV_STATE_CANT_OPEN
) {
4961 * If we fail to open a vdev during an import or recovery, we
4962 * mark it as "not available", which signifies that it was
4963 * never there to begin with. Failure to open such a device
4964 * is not considered an error.
4966 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
4967 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
4968 vd
->vdev_ops
->vdev_op_leaf
)
4969 vd
->vdev_not_present
= 1;
4972 * Post the appropriate ereport. If the 'prevstate' field is
4973 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4974 * that this is part of a vdev_reopen(). In this case, we don't
4975 * want to post the ereport if the device was already in the
4976 * CANT_OPEN state beforehand.
4978 * If the 'checkremove' flag is set, then this is an attempt to
4979 * online the device in response to an insertion event. If we
4980 * hit this case, then we have detected an insertion event for a
4981 * faulted or offline device that wasn't in the removed state.
4982 * In this scenario, we don't post an ereport because we are
4983 * about to replace the device, or attempt an online with
4984 * vdev_forcefault, which will generate the fault for us.
4986 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
4987 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
4988 vd
!= spa
->spa_root_vdev
) {
4992 case VDEV_AUX_OPEN_FAILED
:
4993 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
4995 case VDEV_AUX_CORRUPT_DATA
:
4996 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
4998 case VDEV_AUX_NO_REPLICAS
:
4999 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
5001 case VDEV_AUX_BAD_GUID_SUM
:
5002 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
5004 case VDEV_AUX_TOO_SMALL
:
5005 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
5007 case VDEV_AUX_BAD_LABEL
:
5008 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
5010 case VDEV_AUX_BAD_ASHIFT
:
5011 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
5014 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
5017 (void) zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
5021 /* Erase any notion of persistent removed state */
5022 vd
->vdev_removed
= B_FALSE
;
5024 vd
->vdev_removed
= B_FALSE
;
5028 * Notify ZED of any significant state-change on a leaf vdev.
5031 if (vd
->vdev_ops
->vdev_op_leaf
) {
5032 /* preserve original state from a vdev_reopen() */
5033 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
5034 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
5035 (save_state
<= VDEV_STATE_CLOSED
))
5036 save_state
= vd
->vdev_prevstate
;
5038 /* filter out state change due to initial vdev_open */
5039 if (save_state
> VDEV_STATE_CLOSED
)
5040 zfs_post_state_change(spa
, vd
, save_state
);
5043 if (!isopen
&& vd
->vdev_parent
)
5044 vdev_propagate_state(vd
->vdev_parent
);
5048 vdev_children_are_offline(vdev_t
*vd
)
5050 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
5052 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
5053 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
5061 * Check the vdev configuration to ensure that it's capable of supporting
5062 * a root pool. We do not support partial configuration.
5065 vdev_is_bootable(vdev_t
*vd
)
5067 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5068 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
5070 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0 ||
5071 strcmp(vdev_type
, VDEV_TYPE_INDIRECT
) == 0) {
5076 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5077 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
5084 vdev_is_concrete(vdev_t
*vd
)
5086 vdev_ops_t
*ops
= vd
->vdev_ops
;
5087 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
5088 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
5096 * Determine if a log device has valid content. If the vdev was
5097 * removed or faulted in the MOS config then we know that
5098 * the content on the log device has already been written to the pool.
5101 vdev_log_state_valid(vdev_t
*vd
)
5103 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
5107 for (int c
= 0; c
< vd
->vdev_children
; c
++)
5108 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
5115 * Expand a vdev if possible.
5118 vdev_expand(vdev_t
*vd
, uint64_t txg
)
5120 ASSERT(vd
->vdev_top
== vd
);
5121 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5122 ASSERT(vdev_is_concrete(vd
));
5124 vdev_set_deflate_ratio(vd
);
5126 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
5127 vdev_is_concrete(vd
)) {
5128 vdev_metaslab_group_create(vd
);
5129 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
5130 vdev_config_dirty(vd
);
5138 vdev_split(vdev_t
*vd
)
5140 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
5142 vdev_remove_child(pvd
, vd
);
5143 vdev_compact_children(pvd
);
5145 cvd
= pvd
->vdev_child
[0];
5146 if (pvd
->vdev_children
== 1) {
5147 vdev_remove_parent(cvd
);
5148 cvd
->vdev_splitting
= B_TRUE
;
5150 vdev_propagate_state(cvd
);
5154 vdev_deadman(vdev_t
*vd
, char *tag
)
5156 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5157 vdev_t
*cvd
= vd
->vdev_child
[c
];
5159 vdev_deadman(cvd
, tag
);
5162 if (vd
->vdev_ops
->vdev_op_leaf
) {
5163 vdev_queue_t
*vq
= &vd
->vdev_queue
;
5165 mutex_enter(&vq
->vq_lock
);
5166 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
5167 spa_t
*spa
= vd
->vdev_spa
;
5171 zfs_dbgmsg("slow vdev: %s has %d active IOs",
5172 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
5175 * Look at the head of all the pending queues,
5176 * if any I/O has been outstanding for longer than
5177 * the spa_deadman_synctime invoke the deadman logic.
5179 fio
= avl_first(&vq
->vq_active_tree
);
5180 delta
= gethrtime() - fio
->io_timestamp
;
5181 if (delta
> spa_deadman_synctime(spa
))
5182 zio_deadman(fio
, tag
);
5184 mutex_exit(&vq
->vq_lock
);
5189 vdev_defer_resilver(vdev_t
*vd
)
5191 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
5193 vd
->vdev_resilver_deferred
= B_TRUE
;
5194 vd
->vdev_spa
->spa_resilver_deferred
= B_TRUE
;
5198 * Clears the resilver deferred flag on all leaf devs under vd. Returns
5199 * B_TRUE if we have devices that need to be resilvered and are available to
5200 * accept resilver I/Os.
5203 vdev_clear_resilver_deferred(vdev_t
*vd
, dmu_tx_t
*tx
)
5205 boolean_t resilver_needed
= B_FALSE
;
5206 spa_t
*spa
= vd
->vdev_spa
;
5208 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5209 vdev_t
*cvd
= vd
->vdev_child
[c
];
5210 resilver_needed
|= vdev_clear_resilver_deferred(cvd
, tx
);
5213 if (vd
== spa
->spa_root_vdev
&&
5214 spa_feature_is_active(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
5215 spa_feature_decr(spa
, SPA_FEATURE_RESILVER_DEFER
, tx
);
5216 vdev_config_dirty(vd
);
5217 spa
->spa_resilver_deferred
= B_FALSE
;
5218 return (resilver_needed
);
5221 if (!vdev_is_concrete(vd
) || vd
->vdev_aux
||
5222 !vd
->vdev_ops
->vdev_op_leaf
)
5223 return (resilver_needed
);
5225 vd
->vdev_resilver_deferred
= B_FALSE
;
5227 return (!vdev_is_dead(vd
) && !vd
->vdev_offline
&&
5228 vdev_resilver_needed(vd
, NULL
, NULL
));
5232 vdev_xlate_is_empty(range_seg64_t
*rs
)
5234 return (rs
->rs_start
== rs
->rs_end
);
5238 * Translate a logical range to the first contiguous physical range for the
5239 * specified vdev_t. This function is initially called with a leaf vdev and
5240 * will walk each parent vdev until it reaches a top-level vdev. Once the
5241 * top-level is reached the physical range is initialized and the recursive
5242 * function begins to unwind. As it unwinds it calls the parent's vdev
5243 * specific translation function to do the real conversion.
5246 vdev_xlate(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5247 range_seg64_t
*physical_rs
, range_seg64_t
*remain_rs
)
5250 * Walk up the vdev tree
5252 if (vd
!= vd
->vdev_top
) {
5253 vdev_xlate(vd
->vdev_parent
, logical_rs
, physical_rs
,
5257 * We've reached the top-level vdev, initialize the physical
5258 * range to the logical range and set an empty remaining
5259 * range then start to unwind.
5261 physical_rs
->rs_start
= logical_rs
->rs_start
;
5262 physical_rs
->rs_end
= logical_rs
->rs_end
;
5264 remain_rs
->rs_start
= logical_rs
->rs_start
;
5265 remain_rs
->rs_end
= logical_rs
->rs_start
;
5270 vdev_t
*pvd
= vd
->vdev_parent
;
5271 ASSERT3P(pvd
, !=, NULL
);
5272 ASSERT3P(pvd
->vdev_ops
->vdev_op_xlate
, !=, NULL
);
5275 * As this recursive function unwinds, translate the logical
5276 * range into its physical and any remaining components by calling
5277 * the vdev specific translate function.
5279 range_seg64_t intermediate
= { 0 };
5280 pvd
->vdev_ops
->vdev_op_xlate(vd
, physical_rs
, &intermediate
, remain_rs
);
5282 physical_rs
->rs_start
= intermediate
.rs_start
;
5283 physical_rs
->rs_end
= intermediate
.rs_end
;
5287 vdev_xlate_walk(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5288 vdev_xlate_func_t
*func
, void *arg
)
5290 range_seg64_t iter_rs
= *logical_rs
;
5291 range_seg64_t physical_rs
;
5292 range_seg64_t remain_rs
;
5294 while (!vdev_xlate_is_empty(&iter_rs
)) {
5296 vdev_xlate(vd
, &iter_rs
, &physical_rs
, &remain_rs
);
5299 * With raidz and dRAID, it's possible that the logical range
5300 * does not live on this leaf vdev. Only when there is a non-
5301 * zero physical size call the provided function.
5303 if (!vdev_xlate_is_empty(&physical_rs
))
5304 func(arg
, &physical_rs
);
5306 iter_rs
= remain_rs
;
5311 * Look at the vdev tree and determine whether any devices are currently being
5315 vdev_replace_in_progress(vdev_t
*vdev
)
5317 ASSERT(spa_config_held(vdev
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
5319 if (vdev
->vdev_ops
== &vdev_replacing_ops
)
5323 * A 'spare' vdev indicates that we have a replace in progress, unless
5324 * it has exactly two children, and the second, the hot spare, has
5325 * finished being resilvered.
5327 if (vdev
->vdev_ops
== &vdev_spare_ops
&& (vdev
->vdev_children
> 2 ||
5328 !vdev_dtl_empty(vdev
->vdev_child
[1], DTL_MISSING
)))
5331 for (int i
= 0; i
< vdev
->vdev_children
; i
++) {
5332 if (vdev_replace_in_progress(vdev
->vdev_child
[i
]))
5339 EXPORT_SYMBOL(vdev_fault
);
5340 EXPORT_SYMBOL(vdev_degrade
);
5341 EXPORT_SYMBOL(vdev_online
);
5342 EXPORT_SYMBOL(vdev_offline
);
5343 EXPORT_SYMBOL(vdev_clear
);
5346 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_count
, INT
, ZMOD_RW
,
5347 "Target number of metaslabs per top-level vdev");
5349 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_shift
, INT
, ZMOD_RW
,
5350 "Default limit for metaslab size");
5352 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, min_ms_count
, INT
, ZMOD_RW
,
5353 "Minimum number of metaslabs per top-level vdev");
5355 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, ms_count_limit
, INT
, ZMOD_RW
,
5356 "Practical upper limit of total metaslabs per top-level vdev");
5358 ZFS_MODULE_PARAM(zfs
, zfs_
, slow_io_events_per_second
, UINT
, ZMOD_RW
,
5359 "Rate limit slow IO (delay) events to this many per second");
5361 ZFS_MODULE_PARAM(zfs
, zfs_
, checksum_events_per_second
, UINT
, ZMOD_RW
,
5362 "Rate limit checksum events to this many checksum errors per second "
5363 "(do not set below zed threshold).");
5365 ZFS_MODULE_PARAM(zfs
, zfs_
, scan_ignore_errors
, INT
, ZMOD_RW
,
5366 "Ignore errors during resilver/scrub");
5368 ZFS_MODULE_PARAM(zfs_vdev
, vdev_
, validate_skip
, INT
, ZMOD_RW
,
5369 "Bypass vdev_validate()");
5371 ZFS_MODULE_PARAM(zfs
, zfs_
, nocacheflush
, INT
, ZMOD_RW
,
5372 "Disable cache flushes");
5374 ZFS_MODULE_PARAM(zfs
, zfs_
, embedded_slog_min_ms
, INT
, ZMOD_RW
,
5375 "Minimum number of metaslabs required to dedicate one for log blocks");
5377 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, min_auto_ashift
,
5378 param_set_min_auto_ashift
, param_get_ulong
, ZMOD_RW
,
5379 "Minimum ashift used when creating new top-level vdevs");
5381 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, max_auto_ashift
,
5382 param_set_max_auto_ashift
, param_get_ulong
, ZMOD_RW
,
5383 "Maximum ashift used when optimizing for logical -> physical sector "
5384 "size on new top-level vdevs");