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 https://opensource.org/licenses/CDDL-1.0.
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, 2021 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.
31 * Copyright (c) 2021, Klara Inc.
32 * Copyright (c) 2021, 2023 Hewlett Packard Enterprise Development LP.
35 #include <sys/zfs_context.h>
36 #include <sys/fm/fs/zfs.h>
38 #include <sys/spa_impl.h>
39 #include <sys/bpobj.h>
41 #include <sys/dmu_tx.h>
42 #include <sys/dsl_dir.h>
43 #include <sys/vdev_impl.h>
44 #include <sys/vdev_rebuild.h>
45 #include <sys/vdev_draid.h>
46 #include <sys/uberblock_impl.h>
47 #include <sys/metaslab.h>
48 #include <sys/metaslab_impl.h>
49 #include <sys/space_map.h>
50 #include <sys/space_reftree.h>
53 #include <sys/fs/zfs.h>
56 #include <sys/dsl_scan.h>
57 #include <sys/vdev_raidz.h>
59 #include <sys/vdev_initialize.h>
60 #include <sys/vdev_trim.h>
62 #include <sys/zfs_ratelimit.h>
66 * One metaslab from each (normal-class) vdev is used by the ZIL. These are
67 * called "embedded slog metaslabs", are referenced by vdev_log_mg, and are
68 * part of the spa_embedded_log_class. The metaslab with the most free space
69 * in each vdev is selected for this purpose when the pool is opened (or a
70 * vdev is added). See vdev_metaslab_init().
72 * Log blocks can be allocated from the following locations. Each one is tried
73 * in order until the allocation succeeds:
74 * 1. dedicated log vdevs, aka "slog" (spa_log_class)
75 * 2. embedded slog metaslabs (spa_embedded_log_class)
76 * 3. other metaslabs in normal vdevs (spa_normal_class)
78 * zfs_embedded_slog_min_ms disables the embedded slog if there are fewer
79 * than this number of metaslabs in the vdev. This ensures that we don't set
80 * aside an unreasonable amount of space for the ZIL. If set to less than
81 * 1 << (spa_slop_shift + 1), on small pools the usable space may be reduced
82 * (by more than 1<<spa_slop_shift) due to the embedded slog metaslab.
84 static uint_t zfs_embedded_slog_min_ms
= 64;
86 /* default target for number of metaslabs per top-level vdev */
87 static uint_t zfs_vdev_default_ms_count
= 200;
89 /* minimum number of metaslabs per top-level vdev */
90 static uint_t zfs_vdev_min_ms_count
= 16;
92 /* practical upper limit of total metaslabs per top-level vdev */
93 static uint_t zfs_vdev_ms_count_limit
= 1ULL << 17;
95 /* lower limit for metaslab size (512M) */
96 static uint_t zfs_vdev_default_ms_shift
= 29;
98 /* upper limit for metaslab size (16G) */
99 static uint_t zfs_vdev_max_ms_shift
= 34;
101 int vdev_validate_skip
= B_FALSE
;
104 * Since the DTL space map of a vdev is not expected to have a lot of
105 * entries, we default its block size to 4K.
107 int zfs_vdev_dtl_sm_blksz
= (1 << 12);
110 * Rate limit slow IO (delay) events to this many per second.
112 static unsigned int zfs_slow_io_events_per_second
= 20;
115 * Rate limit checksum events after this many checksum errors per second.
117 static unsigned int zfs_checksum_events_per_second
= 20;
120 * Ignore errors during scrub/resilver. Allows to work around resilver
121 * upon import when there are pool errors.
123 static int zfs_scan_ignore_errors
= 0;
126 * vdev-wide space maps that have lots of entries written to them at
127 * the end of each transaction can benefit from a higher I/O bandwidth
128 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
130 int zfs_vdev_standard_sm_blksz
= (1 << 17);
133 * Tunable parameter for debugging or performance analysis. Setting this
134 * will cause pool corruption on power loss if a volatile out-of-order
135 * write cache is enabled.
137 int zfs_nocacheflush
= 0;
140 * Maximum and minimum ashift values that can be automatically set based on
141 * vdev's physical ashift (disk's physical sector size). While ASHIFT_MAX
142 * is higher than the maximum value, it is intentionally limited here to not
143 * excessively impact pool space efficiency. Higher ashift values may still
144 * be forced by vdev logical ashift or by user via ashift property, but won't
145 * be set automatically as a performance optimization.
147 uint_t zfs_vdev_max_auto_ashift
= 14;
148 uint_t zfs_vdev_min_auto_ashift
= ASHIFT_MIN
;
151 vdev_dbgmsg(vdev_t
*vd
, const char *fmt
, ...)
157 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
160 if (vd
->vdev_path
!= NULL
) {
161 zfs_dbgmsg("%s vdev '%s': %s", vd
->vdev_ops
->vdev_op_type
,
164 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
165 vd
->vdev_ops
->vdev_op_type
,
166 (u_longlong_t
)vd
->vdev_id
,
167 (u_longlong_t
)vd
->vdev_guid
, buf
);
172 vdev_dbgmsg_print_tree(vdev_t
*vd
, int indent
)
176 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
) {
177 zfs_dbgmsg("%*svdev %llu: %s", indent
, "",
178 (u_longlong_t
)vd
->vdev_id
,
179 vd
->vdev_ops
->vdev_op_type
);
183 switch (vd
->vdev_state
) {
184 case VDEV_STATE_UNKNOWN
:
185 (void) snprintf(state
, sizeof (state
), "unknown");
187 case VDEV_STATE_CLOSED
:
188 (void) snprintf(state
, sizeof (state
), "closed");
190 case VDEV_STATE_OFFLINE
:
191 (void) snprintf(state
, sizeof (state
), "offline");
193 case VDEV_STATE_REMOVED
:
194 (void) snprintf(state
, sizeof (state
), "removed");
196 case VDEV_STATE_CANT_OPEN
:
197 (void) snprintf(state
, sizeof (state
), "can't open");
199 case VDEV_STATE_FAULTED
:
200 (void) snprintf(state
, sizeof (state
), "faulted");
202 case VDEV_STATE_DEGRADED
:
203 (void) snprintf(state
, sizeof (state
), "degraded");
205 case VDEV_STATE_HEALTHY
:
206 (void) snprintf(state
, sizeof (state
), "healthy");
209 (void) snprintf(state
, sizeof (state
), "<state %u>",
210 (uint_t
)vd
->vdev_state
);
213 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent
,
214 "", (int)vd
->vdev_id
, vd
->vdev_ops
->vdev_op_type
,
215 vd
->vdev_islog
? " (log)" : "",
216 (u_longlong_t
)vd
->vdev_guid
,
217 vd
->vdev_path
? vd
->vdev_path
: "N/A", state
);
219 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
220 vdev_dbgmsg_print_tree(vd
->vdev_child
[i
], indent
+ 2);
224 * Virtual device management.
227 static vdev_ops_t
*const vdev_ops_table
[] = {
231 &vdev_draid_spare_ops
,
244 * Given a vdev type, return the appropriate ops vector.
247 vdev_getops(const char *type
)
249 vdev_ops_t
*ops
, *const *opspp
;
251 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
252 if (strcmp(ops
->vdev_op_type
, type
) == 0)
259 * Given a vdev and a metaslab class, find which metaslab group we're
260 * interested in. All vdevs may belong to two different metaslab classes.
261 * Dedicated slog devices use only the primary metaslab group, rather than a
262 * separate log group. For embedded slogs, the vdev_log_mg will be non-NULL.
265 vdev_get_mg(vdev_t
*vd
, metaslab_class_t
*mc
)
267 if (mc
== spa_embedded_log_class(vd
->vdev_spa
) &&
268 vd
->vdev_log_mg
!= NULL
)
269 return (vd
->vdev_log_mg
);
271 return (vd
->vdev_mg
);
275 vdev_default_xlate(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
276 range_seg64_t
*physical_rs
, range_seg64_t
*remain_rs
)
278 (void) vd
, (void) remain_rs
;
280 physical_rs
->rs_start
= logical_rs
->rs_start
;
281 physical_rs
->rs_end
= logical_rs
->rs_end
;
285 * Derive the enumerated allocation bias from string input.
286 * String origin is either the per-vdev zap or zpool(8).
288 static vdev_alloc_bias_t
289 vdev_derive_alloc_bias(const char *bias
)
291 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
293 if (strcmp(bias
, VDEV_ALLOC_BIAS_LOG
) == 0)
294 alloc_bias
= VDEV_BIAS_LOG
;
295 else if (strcmp(bias
, VDEV_ALLOC_BIAS_SPECIAL
) == 0)
296 alloc_bias
= VDEV_BIAS_SPECIAL
;
297 else if (strcmp(bias
, VDEV_ALLOC_BIAS_DEDUP
) == 0)
298 alloc_bias
= VDEV_BIAS_DEDUP
;
304 * Default asize function: return the MAX of psize with the asize of
305 * all children. This is what's used by anything other than RAID-Z.
308 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
310 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
313 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
314 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
315 asize
= MAX(asize
, csize
);
322 vdev_default_min_asize(vdev_t
*vd
)
324 return (vd
->vdev_min_asize
);
328 * Get the minimum allocatable size. We define the allocatable size as
329 * the vdev's asize rounded to the nearest metaslab. This allows us to
330 * replace or attach devices which don't have the same physical size but
331 * can still satisfy the same number of allocations.
334 vdev_get_min_asize(vdev_t
*vd
)
336 vdev_t
*pvd
= vd
->vdev_parent
;
339 * If our parent is NULL (inactive spare or cache) or is the root,
340 * just return our own asize.
343 return (vd
->vdev_asize
);
346 * The top-level vdev just returns the allocatable size rounded
347 * to the nearest metaslab.
349 if (vd
== vd
->vdev_top
)
350 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
352 return (pvd
->vdev_ops
->vdev_op_min_asize(pvd
));
356 vdev_set_min_asize(vdev_t
*vd
)
358 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
360 for (int c
= 0; c
< vd
->vdev_children
; c
++)
361 vdev_set_min_asize(vd
->vdev_child
[c
]);
365 * Get the minimal allocation size for the top-level vdev.
368 vdev_get_min_alloc(vdev_t
*vd
)
370 uint64_t min_alloc
= 1ULL << vd
->vdev_ashift
;
372 if (vd
->vdev_ops
->vdev_op_min_alloc
!= NULL
)
373 min_alloc
= vd
->vdev_ops
->vdev_op_min_alloc(vd
);
379 * Get the parity level for a top-level vdev.
382 vdev_get_nparity(vdev_t
*vd
)
384 uint64_t nparity
= 0;
386 if (vd
->vdev_ops
->vdev_op_nparity
!= NULL
)
387 nparity
= vd
->vdev_ops
->vdev_op_nparity(vd
);
393 vdev_prop_get_int(vdev_t
*vd
, vdev_prop_t prop
, uint64_t *value
)
395 spa_t
*spa
= vd
->vdev_spa
;
396 objset_t
*mos
= spa
->spa_meta_objset
;
400 if (vd
->vdev_root_zap
!= 0) {
401 objid
= vd
->vdev_root_zap
;
402 } else if (vd
->vdev_top_zap
!= 0) {
403 objid
= vd
->vdev_top_zap
;
404 } else if (vd
->vdev_leaf_zap
!= 0) {
405 objid
= vd
->vdev_leaf_zap
;
410 err
= zap_lookup(mos
, objid
, vdev_prop_to_name(prop
),
411 sizeof (uint64_t), 1, value
);
414 *value
= vdev_prop_default_numeric(prop
);
420 * Get the number of data disks for a top-level vdev.
423 vdev_get_ndisks(vdev_t
*vd
)
427 if (vd
->vdev_ops
->vdev_op_ndisks
!= NULL
)
428 ndisks
= vd
->vdev_ops
->vdev_op_ndisks(vd
);
434 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
436 vdev_t
*rvd
= spa
->spa_root_vdev
;
438 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
440 if (vdev
< rvd
->vdev_children
) {
441 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
442 return (rvd
->vdev_child
[vdev
]);
449 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
453 if (vd
->vdev_guid
== guid
)
456 for (int c
= 0; c
< vd
->vdev_children
; c
++)
457 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
465 vdev_count_leaves_impl(vdev_t
*vd
)
469 if (vd
->vdev_ops
->vdev_op_leaf
)
472 for (int c
= 0; c
< vd
->vdev_children
; c
++)
473 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
479 vdev_count_leaves(spa_t
*spa
)
483 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
484 rc
= vdev_count_leaves_impl(spa
->spa_root_vdev
);
485 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
491 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
493 size_t oldsize
, newsize
;
494 uint64_t id
= cvd
->vdev_id
;
497 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
498 ASSERT(cvd
->vdev_parent
== NULL
);
500 cvd
->vdev_parent
= pvd
;
505 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
507 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
508 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
509 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
511 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
512 if (pvd
->vdev_child
!= NULL
) {
513 memcpy(newchild
, pvd
->vdev_child
, oldsize
);
514 kmem_free(pvd
->vdev_child
, oldsize
);
517 pvd
->vdev_child
= newchild
;
518 pvd
->vdev_child
[id
] = cvd
;
520 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
521 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
524 * Walk up all ancestors to update guid sum.
526 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
527 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
529 if (cvd
->vdev_ops
->vdev_op_leaf
) {
530 list_insert_head(&cvd
->vdev_spa
->spa_leaf_list
, cvd
);
531 cvd
->vdev_spa
->spa_leaf_list_gen
++;
536 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
539 uint_t id
= cvd
->vdev_id
;
541 ASSERT(cvd
->vdev_parent
== pvd
);
546 ASSERT(id
< pvd
->vdev_children
);
547 ASSERT(pvd
->vdev_child
[id
] == cvd
);
549 pvd
->vdev_child
[id
] = NULL
;
550 cvd
->vdev_parent
= NULL
;
552 for (c
= 0; c
< pvd
->vdev_children
; c
++)
553 if (pvd
->vdev_child
[c
])
556 if (c
== pvd
->vdev_children
) {
557 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
558 pvd
->vdev_child
= NULL
;
559 pvd
->vdev_children
= 0;
562 if (cvd
->vdev_ops
->vdev_op_leaf
) {
563 spa_t
*spa
= cvd
->vdev_spa
;
564 list_remove(&spa
->spa_leaf_list
, cvd
);
565 spa
->spa_leaf_list_gen
++;
569 * Walk up all ancestors to update guid sum.
571 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
572 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
576 * Remove any holes in the child array.
579 vdev_compact_children(vdev_t
*pvd
)
581 vdev_t
**newchild
, *cvd
;
582 int oldc
= pvd
->vdev_children
;
585 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
590 for (int c
= newc
= 0; c
< oldc
; c
++)
591 if (pvd
->vdev_child
[c
])
595 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
597 for (int c
= newc
= 0; c
< oldc
; c
++) {
598 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
599 newchild
[newc
] = cvd
;
600 cvd
->vdev_id
= newc
++;
607 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
608 pvd
->vdev_child
= newchild
;
609 pvd
->vdev_children
= newc
;
613 * Allocate and minimally initialize a vdev_t.
616 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
619 vdev_indirect_config_t
*vic
;
621 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
622 vic
= &vd
->vdev_indirect_config
;
624 if (spa
->spa_root_vdev
== NULL
) {
625 ASSERT(ops
== &vdev_root_ops
);
626 spa
->spa_root_vdev
= vd
;
627 spa
->spa_load_guid
= spa_generate_guid(NULL
);
630 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
631 if (spa
->spa_root_vdev
== vd
) {
633 * The root vdev's guid will also be the pool guid,
634 * which must be unique among all pools.
636 guid
= spa_generate_guid(NULL
);
639 * Any other vdev's guid must be unique within the pool.
641 guid
= spa_generate_guid(spa
);
643 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
648 vd
->vdev_guid
= guid
;
649 vd
->vdev_guid_sum
= guid
;
651 vd
->vdev_state
= VDEV_STATE_CLOSED
;
652 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
653 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
655 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
656 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
657 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, RANGE_SEG64
, NULL
,
661 * Initialize rate limit structs for events. We rate limit ZIO delay
662 * and checksum events so that we don't overwhelm ZED with thousands
663 * of events when a disk is acting up.
665 zfs_ratelimit_init(&vd
->vdev_delay_rl
, &zfs_slow_io_events_per_second
,
667 zfs_ratelimit_init(&vd
->vdev_deadman_rl
, &zfs_slow_io_events_per_second
,
669 zfs_ratelimit_init(&vd
->vdev_checksum_rl
,
670 &zfs_checksum_events_per_second
, 1);
673 * Default Thresholds for tuning ZED
675 vd
->vdev_checksum_n
= vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_N
);
676 vd
->vdev_checksum_t
= vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_T
);
677 vd
->vdev_io_n
= vdev_prop_default_numeric(VDEV_PROP_IO_N
);
678 vd
->vdev_io_t
= vdev_prop_default_numeric(VDEV_PROP_IO_T
);
680 list_link_init(&vd
->vdev_config_dirty_node
);
681 list_link_init(&vd
->vdev_state_dirty_node
);
682 list_link_init(&vd
->vdev_initialize_node
);
683 list_link_init(&vd
->vdev_leaf_node
);
684 list_link_init(&vd
->vdev_trim_node
);
686 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
687 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
688 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
689 mutex_init(&vd
->vdev_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
691 mutex_init(&vd
->vdev_initialize_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
692 mutex_init(&vd
->vdev_initialize_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
693 cv_init(&vd
->vdev_initialize_cv
, NULL
, CV_DEFAULT
, NULL
);
694 cv_init(&vd
->vdev_initialize_io_cv
, NULL
, CV_DEFAULT
, NULL
);
696 mutex_init(&vd
->vdev_trim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
697 mutex_init(&vd
->vdev_autotrim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
698 mutex_init(&vd
->vdev_trim_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
699 cv_init(&vd
->vdev_trim_cv
, NULL
, CV_DEFAULT
, NULL
);
700 cv_init(&vd
->vdev_autotrim_cv
, NULL
, CV_DEFAULT
, NULL
);
701 cv_init(&vd
->vdev_autotrim_kick_cv
, NULL
, CV_DEFAULT
, NULL
);
702 cv_init(&vd
->vdev_trim_io_cv
, NULL
, CV_DEFAULT
, NULL
);
704 mutex_init(&vd
->vdev_rebuild_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
705 cv_init(&vd
->vdev_rebuild_cv
, NULL
, CV_DEFAULT
, NULL
);
707 for (int t
= 0; t
< DTL_TYPES
; t
++) {
708 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0,
712 txg_list_create(&vd
->vdev_ms_list
, spa
,
713 offsetof(struct metaslab
, ms_txg_node
));
714 txg_list_create(&vd
->vdev_dtl_list
, spa
,
715 offsetof(struct vdev
, vdev_dtl_node
));
716 vd
->vdev_stat
.vs_timestamp
= gethrtime();
723 * Allocate a new vdev. The 'alloctype' is used to control whether we are
724 * creating a new vdev or loading an existing one - the behavior is slightly
725 * different for each case.
728 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
733 uint64_t guid
= 0, islog
;
735 vdev_indirect_config_t
*vic
;
736 const char *tmp
= NULL
;
738 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
739 boolean_t top_level
= (parent
&& !parent
->vdev_parent
);
741 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
743 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
744 return (SET_ERROR(EINVAL
));
746 if ((ops
= vdev_getops(type
)) == NULL
)
747 return (SET_ERROR(EINVAL
));
750 * If this is a load, get the vdev guid from the nvlist.
751 * Otherwise, vdev_alloc_common() will generate one for us.
753 if (alloctype
== VDEV_ALLOC_LOAD
) {
756 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
758 return (SET_ERROR(EINVAL
));
760 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
761 return (SET_ERROR(EINVAL
));
762 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
763 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
764 return (SET_ERROR(EINVAL
));
765 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
766 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
767 return (SET_ERROR(EINVAL
));
768 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
769 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
770 return (SET_ERROR(EINVAL
));
774 * The first allocated vdev must be of type 'root'.
776 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
777 return (SET_ERROR(EINVAL
));
780 * Determine whether we're a log vdev.
783 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
784 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
785 return (SET_ERROR(ENOTSUP
));
787 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
788 return (SET_ERROR(ENOTSUP
));
790 if (top_level
&& alloctype
== VDEV_ALLOC_ADD
) {
794 * If creating a top-level vdev, check for allocation
797 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
799 alloc_bias
= vdev_derive_alloc_bias(bias
);
801 /* spa_vdev_add() expects feature to be enabled */
802 if (spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
803 !spa_feature_is_enabled(spa
,
804 SPA_FEATURE_ALLOCATION_CLASSES
)) {
805 return (SET_ERROR(ENOTSUP
));
809 /* spa_vdev_add() expects feature to be enabled */
810 if (ops
== &vdev_draid_ops
&&
811 spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
812 !spa_feature_is_enabled(spa
, SPA_FEATURE_DRAID
)) {
813 return (SET_ERROR(ENOTSUP
));
818 * Initialize the vdev specific data. This is done before calling
819 * vdev_alloc_common() since it may fail and this simplifies the
820 * error reporting and cleanup code paths.
823 if (ops
->vdev_op_init
!= NULL
) {
824 rc
= ops
->vdev_op_init(spa
, nv
, &tsd
);
830 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
832 vd
->vdev_islog
= islog
;
834 if (top_level
&& alloc_bias
!= VDEV_BIAS_NONE
)
835 vd
->vdev_alloc_bias
= alloc_bias
;
837 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &tmp
) == 0)
838 vd
->vdev_path
= spa_strdup(tmp
);
841 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
842 * fault on a vdev and want it to persist across imports (like with
845 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
846 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
847 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
848 vd
->vdev_faulted
= 1;
849 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
852 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &tmp
) == 0)
853 vd
->vdev_devid
= spa_strdup(tmp
);
854 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
, &tmp
) == 0)
855 vd
->vdev_physpath
= spa_strdup(tmp
);
857 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
859 vd
->vdev_enc_sysfs_path
= spa_strdup(tmp
);
861 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &tmp
) == 0)
862 vd
->vdev_fru
= spa_strdup(tmp
);
865 * Set the whole_disk property. If it's not specified, leave the value
868 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
869 &vd
->vdev_wholedisk
) != 0)
870 vd
->vdev_wholedisk
= -1ULL;
872 vic
= &vd
->vdev_indirect_config
;
874 ASSERT0(vic
->vic_mapping_object
);
875 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
876 &vic
->vic_mapping_object
);
877 ASSERT0(vic
->vic_births_object
);
878 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
879 &vic
->vic_births_object
);
880 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
881 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
882 &vic
->vic_prev_indirect_vdev
);
885 * Look for the 'not present' flag. This will only be set if the device
886 * was not present at the time of import.
888 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
889 &vd
->vdev_not_present
);
892 * Get the alignment requirement.
894 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
897 * Retrieve the vdev creation time.
899 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
902 if (vd
->vdev_ops
== &vdev_root_ops
&&
903 (alloctype
== VDEV_ALLOC_LOAD
||
904 alloctype
== VDEV_ALLOC_SPLIT
||
905 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
906 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_ROOT_ZAP
,
911 * If we're a top-level vdev, try to load the allocation parameters.
914 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
915 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
917 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
919 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
921 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NONALLOCATING
,
923 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
925 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
928 ASSERT0(vd
->vdev_top_zap
);
931 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
932 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
933 alloctype
== VDEV_ALLOC_ADD
||
934 alloctype
== VDEV_ALLOC_SPLIT
||
935 alloctype
== VDEV_ALLOC_ROOTPOOL
);
936 /* Note: metaslab_group_create() is now deferred */
939 if (vd
->vdev_ops
->vdev_op_leaf
&&
940 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
941 (void) nvlist_lookup_uint64(nv
,
942 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
944 ASSERT0(vd
->vdev_leaf_zap
);
948 * If we're a leaf vdev, try to load the DTL object and other state.
951 if (vd
->vdev_ops
->vdev_op_leaf
&&
952 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
953 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
954 if (alloctype
== VDEV_ALLOC_LOAD
) {
955 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
956 &vd
->vdev_dtl_object
);
957 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
961 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
964 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
965 &spare
) == 0 && spare
)
969 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
972 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
973 &vd
->vdev_resilver_txg
);
975 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REBUILD_TXG
,
976 &vd
->vdev_rebuild_txg
);
978 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
979 vdev_defer_resilver(vd
);
982 * In general, when importing a pool we want to ignore the
983 * persistent fault state, as the diagnosis made on another
984 * system may not be valid in the current context. The only
985 * exception is if we forced a vdev to a persistently faulted
986 * state with 'zpool offline -f'. The persistent fault will
987 * remain across imports until cleared.
989 * Local vdevs will remain in the faulted state.
991 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
992 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
993 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
995 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
997 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
1000 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
1003 vd
->vdev_label_aux
=
1004 VDEV_AUX_ERR_EXCEEDED
;
1005 if (nvlist_lookup_string(nv
,
1006 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
1007 strcmp(aux
, "external") == 0)
1008 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
1010 vd
->vdev_faulted
= 0ULL;
1016 * Add ourselves to the parent's list of children.
1018 vdev_add_child(parent
, vd
);
1026 vdev_free(vdev_t
*vd
)
1028 spa_t
*spa
= vd
->vdev_spa
;
1030 ASSERT3P(vd
->vdev_initialize_thread
, ==, NULL
);
1031 ASSERT3P(vd
->vdev_trim_thread
, ==, NULL
);
1032 ASSERT3P(vd
->vdev_autotrim_thread
, ==, NULL
);
1033 ASSERT3P(vd
->vdev_rebuild_thread
, ==, NULL
);
1036 * Scan queues are normally destroyed at the end of a scan. If the
1037 * queue exists here, that implies the vdev is being removed while
1038 * the scan is still running.
1040 if (vd
->vdev_scan_io_queue
!= NULL
) {
1041 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
1042 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
1043 vd
->vdev_scan_io_queue
= NULL
;
1044 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
1048 * vdev_free() implies closing the vdev first. This is simpler than
1049 * trying to ensure complicated semantics for all callers.
1053 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
1054 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
1057 * Free all children.
1059 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1060 vdev_free(vd
->vdev_child
[c
]);
1062 ASSERT(vd
->vdev_child
== NULL
);
1063 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
1065 if (vd
->vdev_ops
->vdev_op_fini
!= NULL
)
1066 vd
->vdev_ops
->vdev_op_fini(vd
);
1069 * Discard allocation state.
1071 if (vd
->vdev_mg
!= NULL
) {
1072 vdev_metaslab_fini(vd
);
1073 metaslab_group_destroy(vd
->vdev_mg
);
1076 if (vd
->vdev_log_mg
!= NULL
) {
1077 ASSERT0(vd
->vdev_ms_count
);
1078 metaslab_group_destroy(vd
->vdev_log_mg
);
1079 vd
->vdev_log_mg
= NULL
;
1082 ASSERT0(vd
->vdev_stat
.vs_space
);
1083 ASSERT0(vd
->vdev_stat
.vs_dspace
);
1084 ASSERT0(vd
->vdev_stat
.vs_alloc
);
1087 * Remove this vdev from its parent's child list.
1089 vdev_remove_child(vd
->vdev_parent
, vd
);
1091 ASSERT(vd
->vdev_parent
== NULL
);
1092 ASSERT(!list_link_active(&vd
->vdev_leaf_node
));
1095 * Clean up vdev structure.
1097 vdev_queue_fini(vd
);
1100 spa_strfree(vd
->vdev_path
);
1102 spa_strfree(vd
->vdev_devid
);
1103 if (vd
->vdev_physpath
)
1104 spa_strfree(vd
->vdev_physpath
);
1106 if (vd
->vdev_enc_sysfs_path
)
1107 spa_strfree(vd
->vdev_enc_sysfs_path
);
1110 spa_strfree(vd
->vdev_fru
);
1112 if (vd
->vdev_isspare
)
1113 spa_spare_remove(vd
);
1114 if (vd
->vdev_isl2cache
)
1115 spa_l2cache_remove(vd
);
1117 txg_list_destroy(&vd
->vdev_ms_list
);
1118 txg_list_destroy(&vd
->vdev_dtl_list
);
1120 mutex_enter(&vd
->vdev_dtl_lock
);
1121 space_map_close(vd
->vdev_dtl_sm
);
1122 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1123 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
1124 range_tree_destroy(vd
->vdev_dtl
[t
]);
1126 mutex_exit(&vd
->vdev_dtl_lock
);
1128 EQUIV(vd
->vdev_indirect_births
!= NULL
,
1129 vd
->vdev_indirect_mapping
!= NULL
);
1130 if (vd
->vdev_indirect_births
!= NULL
) {
1131 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
1132 vdev_indirect_births_close(vd
->vdev_indirect_births
);
1135 if (vd
->vdev_obsolete_sm
!= NULL
) {
1136 ASSERT(vd
->vdev_removing
||
1137 vd
->vdev_ops
== &vdev_indirect_ops
);
1138 space_map_close(vd
->vdev_obsolete_sm
);
1139 vd
->vdev_obsolete_sm
= NULL
;
1141 range_tree_destroy(vd
->vdev_obsolete_segments
);
1142 rw_destroy(&vd
->vdev_indirect_rwlock
);
1143 mutex_destroy(&vd
->vdev_obsolete_lock
);
1145 mutex_destroy(&vd
->vdev_dtl_lock
);
1146 mutex_destroy(&vd
->vdev_stat_lock
);
1147 mutex_destroy(&vd
->vdev_probe_lock
);
1148 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
1150 mutex_destroy(&vd
->vdev_initialize_lock
);
1151 mutex_destroy(&vd
->vdev_initialize_io_lock
);
1152 cv_destroy(&vd
->vdev_initialize_io_cv
);
1153 cv_destroy(&vd
->vdev_initialize_cv
);
1155 mutex_destroy(&vd
->vdev_trim_lock
);
1156 mutex_destroy(&vd
->vdev_autotrim_lock
);
1157 mutex_destroy(&vd
->vdev_trim_io_lock
);
1158 cv_destroy(&vd
->vdev_trim_cv
);
1159 cv_destroy(&vd
->vdev_autotrim_cv
);
1160 cv_destroy(&vd
->vdev_autotrim_kick_cv
);
1161 cv_destroy(&vd
->vdev_trim_io_cv
);
1163 mutex_destroy(&vd
->vdev_rebuild_lock
);
1164 cv_destroy(&vd
->vdev_rebuild_cv
);
1166 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
1167 zfs_ratelimit_fini(&vd
->vdev_deadman_rl
);
1168 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
1170 if (vd
== spa
->spa_root_vdev
)
1171 spa
->spa_root_vdev
= NULL
;
1173 kmem_free(vd
, sizeof (vdev_t
));
1177 * Transfer top-level vdev state from svd to tvd.
1180 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
1182 spa_t
*spa
= svd
->vdev_spa
;
1187 ASSERT(tvd
== tvd
->vdev_top
);
1189 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
1190 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
1191 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
1192 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
1193 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
1195 svd
->vdev_ms_array
= 0;
1196 svd
->vdev_ms_shift
= 0;
1197 svd
->vdev_ms_count
= 0;
1198 svd
->vdev_top_zap
= 0;
1201 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
1202 if (tvd
->vdev_log_mg
)
1203 ASSERT3P(tvd
->vdev_log_mg
, ==, svd
->vdev_log_mg
);
1204 tvd
->vdev_mg
= svd
->vdev_mg
;
1205 tvd
->vdev_log_mg
= svd
->vdev_log_mg
;
1206 tvd
->vdev_ms
= svd
->vdev_ms
;
1208 svd
->vdev_mg
= NULL
;
1209 svd
->vdev_log_mg
= NULL
;
1210 svd
->vdev_ms
= NULL
;
1212 if (tvd
->vdev_mg
!= NULL
)
1213 tvd
->vdev_mg
->mg_vd
= tvd
;
1214 if (tvd
->vdev_log_mg
!= NULL
)
1215 tvd
->vdev_log_mg
->mg_vd
= tvd
;
1217 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
1218 svd
->vdev_checkpoint_sm
= NULL
;
1220 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1221 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1223 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1224 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1225 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1227 svd
->vdev_stat
.vs_alloc
= 0;
1228 svd
->vdev_stat
.vs_space
= 0;
1229 svd
->vdev_stat
.vs_dspace
= 0;
1232 * State which may be set on a top-level vdev that's in the
1233 * process of being removed.
1235 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1236 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1237 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1238 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1239 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1240 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1241 ASSERT0(tvd
->vdev_noalloc
);
1242 ASSERT0(tvd
->vdev_removing
);
1243 ASSERT0(tvd
->vdev_rebuilding
);
1244 tvd
->vdev_noalloc
= svd
->vdev_noalloc
;
1245 tvd
->vdev_removing
= svd
->vdev_removing
;
1246 tvd
->vdev_rebuilding
= svd
->vdev_rebuilding
;
1247 tvd
->vdev_rebuild_config
= svd
->vdev_rebuild_config
;
1248 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1249 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1250 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1251 range_tree_swap(&svd
->vdev_obsolete_segments
,
1252 &tvd
->vdev_obsolete_segments
);
1253 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1254 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1255 svd
->vdev_indirect_config
.vic_births_object
= 0;
1256 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1257 svd
->vdev_indirect_mapping
= NULL
;
1258 svd
->vdev_indirect_births
= NULL
;
1259 svd
->vdev_obsolete_sm
= NULL
;
1260 svd
->vdev_noalloc
= 0;
1261 svd
->vdev_removing
= 0;
1262 svd
->vdev_rebuilding
= 0;
1264 for (t
= 0; t
< TXG_SIZE
; t
++) {
1265 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1266 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1267 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1268 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1269 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1270 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1273 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1274 vdev_config_clean(svd
);
1275 vdev_config_dirty(tvd
);
1278 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1279 vdev_state_clean(svd
);
1280 vdev_state_dirty(tvd
);
1283 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1284 svd
->vdev_deflate_ratio
= 0;
1286 tvd
->vdev_islog
= svd
->vdev_islog
;
1287 svd
->vdev_islog
= 0;
1289 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1293 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1300 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1301 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1305 * Add a mirror/replacing vdev above an existing vdev. There is no need to
1306 * call .vdev_op_init() since mirror/replacing vdevs do not have private state.
1309 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1311 spa_t
*spa
= cvd
->vdev_spa
;
1312 vdev_t
*pvd
= cvd
->vdev_parent
;
1315 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1317 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1319 mvd
->vdev_asize
= cvd
->vdev_asize
;
1320 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1321 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1322 mvd
->vdev_psize
= cvd
->vdev_psize
;
1323 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1324 mvd
->vdev_logical_ashift
= cvd
->vdev_logical_ashift
;
1325 mvd
->vdev_physical_ashift
= cvd
->vdev_physical_ashift
;
1326 mvd
->vdev_state
= cvd
->vdev_state
;
1327 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1329 vdev_remove_child(pvd
, cvd
);
1330 vdev_add_child(pvd
, mvd
);
1331 cvd
->vdev_id
= mvd
->vdev_children
;
1332 vdev_add_child(mvd
, cvd
);
1333 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1335 if (mvd
== mvd
->vdev_top
)
1336 vdev_top_transfer(cvd
, mvd
);
1342 * Remove a 1-way mirror/replacing vdev from the tree.
1345 vdev_remove_parent(vdev_t
*cvd
)
1347 vdev_t
*mvd
= cvd
->vdev_parent
;
1348 vdev_t
*pvd
= mvd
->vdev_parent
;
1350 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1352 ASSERT(mvd
->vdev_children
== 1);
1353 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1354 mvd
->vdev_ops
== &vdev_replacing_ops
||
1355 mvd
->vdev_ops
== &vdev_spare_ops
);
1356 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1357 cvd
->vdev_logical_ashift
= mvd
->vdev_logical_ashift
;
1358 cvd
->vdev_physical_ashift
= mvd
->vdev_physical_ashift
;
1359 vdev_remove_child(mvd
, cvd
);
1360 vdev_remove_child(pvd
, mvd
);
1363 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1364 * Otherwise, we could have detached an offline device, and when we
1365 * go to import the pool we'll think we have two top-level vdevs,
1366 * instead of a different version of the same top-level vdev.
1368 if (mvd
->vdev_top
== mvd
) {
1369 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1370 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1371 cvd
->vdev_guid
+= guid_delta
;
1372 cvd
->vdev_guid_sum
+= guid_delta
;
1375 * If pool not set for autoexpand, we need to also preserve
1376 * mvd's asize to prevent automatic expansion of cvd.
1377 * Otherwise if we are adjusting the mirror by attaching and
1378 * detaching children of non-uniform sizes, the mirror could
1379 * autoexpand, unexpectedly requiring larger devices to
1380 * re-establish the mirror.
1382 if (!cvd
->vdev_spa
->spa_autoexpand
)
1383 cvd
->vdev_asize
= mvd
->vdev_asize
;
1385 cvd
->vdev_id
= mvd
->vdev_id
;
1386 vdev_add_child(pvd
, cvd
);
1387 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1389 if (cvd
== cvd
->vdev_top
)
1390 vdev_top_transfer(mvd
, cvd
);
1392 ASSERT(mvd
->vdev_children
== 0);
1397 vdev_metaslab_group_create(vdev_t
*vd
)
1399 spa_t
*spa
= vd
->vdev_spa
;
1402 * metaslab_group_create was delayed until allocation bias was available
1404 if (vd
->vdev_mg
== NULL
) {
1405 metaslab_class_t
*mc
;
1407 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1408 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1410 ASSERT3U(vd
->vdev_islog
, ==,
1411 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1413 switch (vd
->vdev_alloc_bias
) {
1415 mc
= spa_log_class(spa
);
1417 case VDEV_BIAS_SPECIAL
:
1418 mc
= spa_special_class(spa
);
1420 case VDEV_BIAS_DEDUP
:
1421 mc
= spa_dedup_class(spa
);
1424 mc
= spa_normal_class(spa
);
1427 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1428 spa
->spa_alloc_count
);
1430 if (!vd
->vdev_islog
) {
1431 vd
->vdev_log_mg
= metaslab_group_create(
1432 spa_embedded_log_class(spa
), vd
, 1);
1436 * The spa ashift min/max only apply for the normal metaslab
1437 * class. Class destination is late binding so ashift boundary
1438 * setting had to wait until now.
1440 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1441 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1442 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1443 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1444 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1445 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1447 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
1448 if (min_alloc
< spa
->spa_min_alloc
)
1449 spa
->spa_min_alloc
= min_alloc
;
1455 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1457 spa_t
*spa
= vd
->vdev_spa
;
1458 uint64_t oldc
= vd
->vdev_ms_count
;
1459 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1462 boolean_t expanding
= (oldc
!= 0);
1464 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1467 * This vdev is not being allocated from yet or is a hole.
1469 if (vd
->vdev_ms_shift
== 0)
1472 ASSERT(!vd
->vdev_ishole
);
1474 ASSERT(oldc
<= newc
);
1476 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1479 memcpy(mspp
, vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1480 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1484 vd
->vdev_ms_count
= newc
;
1486 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1487 uint64_t object
= 0;
1489 * vdev_ms_array may be 0 if we are creating the "fake"
1490 * metaslabs for an indirect vdev for zdb's leak detection.
1491 * See zdb_leak_init().
1493 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1494 error
= dmu_read(spa
->spa_meta_objset
,
1496 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1499 vdev_dbgmsg(vd
, "unable to read the metaslab "
1500 "array [error=%d]", error
);
1505 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1508 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1515 * Find the emptiest metaslab on the vdev and mark it for use for
1516 * embedded slog by moving it from the regular to the log metaslab
1519 if (vd
->vdev_mg
->mg_class
== spa_normal_class(spa
) &&
1520 vd
->vdev_ms_count
> zfs_embedded_slog_min_ms
&&
1521 avl_is_empty(&vd
->vdev_log_mg
->mg_metaslab_tree
)) {
1522 uint64_t slog_msid
= 0;
1523 uint64_t smallest
= UINT64_MAX
;
1526 * Note, we only search the new metaslabs, because the old
1527 * (pre-existing) ones may be active (e.g. have non-empty
1528 * range_tree's), and we don't move them to the new
1531 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1533 space_map_allocated(vd
->vdev_ms
[m
]->ms_sm
);
1534 if (alloc
< smallest
) {
1539 metaslab_t
*slog_ms
= vd
->vdev_ms
[slog_msid
];
1541 * The metaslab was marked as dirty at the end of
1542 * metaslab_init(). Remove it from the dirty list so that we
1543 * can uninitialize and reinitialize it to the new class.
1546 (void) txg_list_remove_this(&vd
->vdev_ms_list
,
1549 uint64_t sm_obj
= space_map_object(slog_ms
->ms_sm
);
1550 metaslab_fini(slog_ms
);
1551 VERIFY0(metaslab_init(vd
->vdev_log_mg
, slog_msid
, sm_obj
, txg
,
1552 &vd
->vdev_ms
[slog_msid
]));
1556 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1559 * If the vdev is marked as non-allocating then don't
1560 * activate the metaslabs since we want to ensure that
1561 * no allocations are performed on this device.
1563 if (vd
->vdev_noalloc
) {
1564 /* track non-allocating vdev space */
1565 spa
->spa_nonallocating_dspace
+= spa_deflate(spa
) ?
1566 vd
->vdev_stat
.vs_dspace
: vd
->vdev_stat
.vs_space
;
1567 } else if (!expanding
) {
1568 metaslab_group_activate(vd
->vdev_mg
);
1569 if (vd
->vdev_log_mg
!= NULL
)
1570 metaslab_group_activate(vd
->vdev_log_mg
);
1574 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1580 vdev_metaslab_fini(vdev_t
*vd
)
1582 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1583 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1584 SPA_FEATURE_POOL_CHECKPOINT
));
1585 space_map_close(vd
->vdev_checkpoint_sm
);
1587 * Even though we close the space map, we need to set its
1588 * pointer to NULL. The reason is that vdev_metaslab_fini()
1589 * may be called multiple times for certain operations
1590 * (i.e. when destroying a pool) so we need to ensure that
1591 * this clause never executes twice. This logic is similar
1592 * to the one used for the vdev_ms clause below.
1594 vd
->vdev_checkpoint_sm
= NULL
;
1597 if (vd
->vdev_ms
!= NULL
) {
1598 metaslab_group_t
*mg
= vd
->vdev_mg
;
1600 metaslab_group_passivate(mg
);
1601 if (vd
->vdev_log_mg
!= NULL
) {
1602 ASSERT(!vd
->vdev_islog
);
1603 metaslab_group_passivate(vd
->vdev_log_mg
);
1606 uint64_t count
= vd
->vdev_ms_count
;
1607 for (uint64_t m
= 0; m
< count
; m
++) {
1608 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1612 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1614 vd
->vdev_ms_count
= 0;
1616 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
1617 ASSERT0(mg
->mg_histogram
[i
]);
1618 if (vd
->vdev_log_mg
!= NULL
)
1619 ASSERT0(vd
->vdev_log_mg
->mg_histogram
[i
]);
1622 ASSERT0(vd
->vdev_ms_count
);
1623 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1626 typedef struct vdev_probe_stats
{
1627 boolean_t vps_readable
;
1628 boolean_t vps_writeable
;
1630 } vdev_probe_stats_t
;
1633 vdev_probe_done(zio_t
*zio
)
1635 spa_t
*spa
= zio
->io_spa
;
1636 vdev_t
*vd
= zio
->io_vd
;
1637 vdev_probe_stats_t
*vps
= zio
->io_private
;
1639 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1641 if (zio
->io_type
== ZIO_TYPE_READ
) {
1642 if (zio
->io_error
== 0)
1643 vps
->vps_readable
= 1;
1644 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1645 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1646 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1647 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1648 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1650 abd_free(zio
->io_abd
);
1652 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1653 if (zio
->io_error
== 0)
1654 vps
->vps_writeable
= 1;
1655 abd_free(zio
->io_abd
);
1656 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1660 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1661 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1663 if (vdev_readable(vd
) &&
1664 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1667 ASSERT(zio
->io_error
!= 0);
1668 vdev_dbgmsg(vd
, "failed probe");
1669 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1670 spa
, vd
, NULL
, NULL
, 0);
1671 zio
->io_error
= SET_ERROR(ENXIO
);
1674 mutex_enter(&vd
->vdev_probe_lock
);
1675 ASSERT(vd
->vdev_probe_zio
== zio
);
1676 vd
->vdev_probe_zio
= NULL
;
1677 mutex_exit(&vd
->vdev_probe_lock
);
1680 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1681 if (!vdev_accessible(vd
, pio
))
1682 pio
->io_error
= SET_ERROR(ENXIO
);
1684 kmem_free(vps
, sizeof (*vps
));
1689 * Determine whether this device is accessible.
1691 * Read and write to several known locations: the pad regions of each
1692 * vdev label but the first, which we leave alone in case it contains
1696 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1698 spa_t
*spa
= vd
->vdev_spa
;
1699 vdev_probe_stats_t
*vps
= NULL
;
1702 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1705 * Don't probe the probe.
1707 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1711 * To prevent 'probe storms' when a device fails, we create
1712 * just one probe i/o at a time. All zios that want to probe
1713 * this vdev will become parents of the probe io.
1715 mutex_enter(&vd
->vdev_probe_lock
);
1717 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1718 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1720 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1721 ZIO_FLAG_DONT_AGGREGATE
| ZIO_FLAG_TRYHARD
;
1723 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1725 * vdev_cant_read and vdev_cant_write can only
1726 * transition from TRUE to FALSE when we have the
1727 * SCL_ZIO lock as writer; otherwise they can only
1728 * transition from FALSE to TRUE. This ensures that
1729 * any zio looking at these values can assume that
1730 * failures persist for the life of the I/O. That's
1731 * important because when a device has intermittent
1732 * connectivity problems, we want to ensure that
1733 * they're ascribed to the device (ENXIO) and not
1736 * Since we hold SCL_ZIO as writer here, clear both
1737 * values so the probe can reevaluate from first
1740 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1741 vd
->vdev_cant_read
= B_FALSE
;
1742 vd
->vdev_cant_write
= B_FALSE
;
1745 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1746 vdev_probe_done
, vps
,
1747 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1750 * We can't change the vdev state in this context, so we
1751 * kick off an async task to do it on our behalf.
1754 vd
->vdev_probe_wanted
= B_TRUE
;
1755 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1760 zio_add_child(zio
, pio
);
1762 mutex_exit(&vd
->vdev_probe_lock
);
1765 ASSERT(zio
!= NULL
);
1769 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1770 zio_nowait(zio_read_phys(pio
, vd
,
1771 vdev_label_offset(vd
->vdev_psize
, l
,
1772 offsetof(vdev_label_t
, vl_be
)), VDEV_PAD_SIZE
,
1773 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1774 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1775 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1786 vdev_load_child(void *arg
)
1790 vd
->vdev_load_error
= vdev_load(vd
);
1794 vdev_open_child(void *arg
)
1798 vd
->vdev_open_thread
= curthread
;
1799 vd
->vdev_open_error
= vdev_open(vd
);
1800 vd
->vdev_open_thread
= NULL
;
1804 vdev_uses_zvols(vdev_t
*vd
)
1807 if (zvol_is_zvol(vd
->vdev_path
))
1811 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1812 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1819 * Returns B_TRUE if the passed child should be opened.
1822 vdev_default_open_children_func(vdev_t
*vd
)
1829 * Open the requested child vdevs. If any of the leaf vdevs are using
1830 * a ZFS volume then do the opens in a single thread. This avoids a
1831 * deadlock when the current thread is holding the spa_namespace_lock.
1834 vdev_open_children_impl(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1836 int children
= vd
->vdev_children
;
1838 taskq_t
*tq
= taskq_create("vdev_open", children
, minclsyspri
,
1839 children
, children
, TASKQ_PREPOPULATE
);
1840 vd
->vdev_nonrot
= B_TRUE
;
1842 for (int c
= 0; c
< children
; c
++) {
1843 vdev_t
*cvd
= vd
->vdev_child
[c
];
1845 if (open_func(cvd
) == B_FALSE
)
1848 if (tq
== NULL
|| vdev_uses_zvols(vd
)) {
1849 cvd
->vdev_open_error
= vdev_open(cvd
);
1851 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1852 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
1855 vd
->vdev_nonrot
&= cvd
->vdev_nonrot
;
1865 * Open all child vdevs.
1868 vdev_open_children(vdev_t
*vd
)
1870 vdev_open_children_impl(vd
, vdev_default_open_children_func
);
1874 * Conditionally open a subset of child vdevs.
1877 vdev_open_children_subset(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1879 vdev_open_children_impl(vd
, open_func
);
1883 * Compute the raidz-deflation ratio. Note, we hard-code
1884 * in 128k (1 << 17) because it is the "typical" blocksize.
1885 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1886 * otherwise it would inconsistently account for existing bp's.
1889 vdev_set_deflate_ratio(vdev_t
*vd
)
1891 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1892 vd
->vdev_deflate_ratio
= (1 << 17) /
1893 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1898 * Choose the best of two ashifts, preferring one between logical ashift
1899 * (absolute minimum) and administrator defined maximum, otherwise take
1900 * the biggest of the two.
1903 vdev_best_ashift(uint64_t logical
, uint64_t a
, uint64_t b
)
1905 if (a
> logical
&& a
<= zfs_vdev_max_auto_ashift
) {
1906 if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1910 } else if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1916 * Maximize performance by inflating the configured ashift for top level
1917 * vdevs to be as close to the physical ashift as possible while maintaining
1918 * administrator defined limits and ensuring it doesn't go below the
1922 vdev_ashift_optimize(vdev_t
*vd
)
1924 ASSERT(vd
== vd
->vdev_top
);
1926 if (vd
->vdev_ashift
< vd
->vdev_physical_ashift
&&
1927 vd
->vdev_physical_ashift
<= zfs_vdev_max_auto_ashift
) {
1928 vd
->vdev_ashift
= MIN(
1929 MAX(zfs_vdev_max_auto_ashift
, vd
->vdev_ashift
),
1930 MAX(zfs_vdev_min_auto_ashift
,
1931 vd
->vdev_physical_ashift
));
1934 * If the logical and physical ashifts are the same, then
1935 * we ensure that the top-level vdev's ashift is not smaller
1936 * than our minimum ashift value. For the unusual case
1937 * where logical ashift > physical ashift, we can't cap
1938 * the calculated ashift based on max ashift as that
1939 * would cause failures.
1940 * We still check if we need to increase it to match
1943 vd
->vdev_ashift
= MAX(zfs_vdev_min_auto_ashift
,
1949 * Prepare a virtual device for access.
1952 vdev_open(vdev_t
*vd
)
1954 spa_t
*spa
= vd
->vdev_spa
;
1957 uint64_t max_osize
= 0;
1958 uint64_t asize
, max_asize
, psize
;
1959 uint64_t logical_ashift
= 0;
1960 uint64_t physical_ashift
= 0;
1962 ASSERT(vd
->vdev_open_thread
== curthread
||
1963 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1964 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1965 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1966 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1968 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1969 vd
->vdev_cant_read
= B_FALSE
;
1970 vd
->vdev_cant_write
= B_FALSE
;
1971 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1974 * If this vdev is not removed, check its fault status. If it's
1975 * faulted, bail out of the open.
1977 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1978 ASSERT(vd
->vdev_children
== 0);
1979 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1980 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1981 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1982 vd
->vdev_label_aux
);
1983 return (SET_ERROR(ENXIO
));
1984 } else if (vd
->vdev_offline
) {
1985 ASSERT(vd
->vdev_children
== 0);
1986 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1987 return (SET_ERROR(ENXIO
));
1990 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
,
1991 &logical_ashift
, &physical_ashift
);
1993 /* Keep the device in removed state if unplugged */
1994 if (error
== ENOENT
&& vd
->vdev_removed
) {
1995 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_REMOVED
,
2001 * Physical volume size should never be larger than its max size, unless
2002 * the disk has shrunk while we were reading it or the device is buggy
2003 * or damaged: either way it's not safe for use, bail out of the open.
2005 if (osize
> max_osize
) {
2006 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2007 VDEV_AUX_OPEN_FAILED
);
2008 return (SET_ERROR(ENXIO
));
2012 * Reset the vdev_reopening flag so that we actually close
2013 * the vdev on error.
2015 vd
->vdev_reopening
= B_FALSE
;
2016 if (zio_injection_enabled
&& error
== 0)
2017 error
= zio_handle_device_injection(vd
, NULL
, SET_ERROR(ENXIO
));
2020 if (vd
->vdev_removed
&&
2021 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
2022 vd
->vdev_removed
= B_FALSE
;
2024 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
2025 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
2026 vd
->vdev_stat
.vs_aux
);
2028 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2029 vd
->vdev_stat
.vs_aux
);
2034 vd
->vdev_removed
= B_FALSE
;
2037 * Recheck the faulted flag now that we have confirmed that
2038 * the vdev is accessible. If we're faulted, bail.
2040 if (vd
->vdev_faulted
) {
2041 ASSERT(vd
->vdev_children
== 0);
2042 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
2043 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
2044 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2045 vd
->vdev_label_aux
);
2046 return (SET_ERROR(ENXIO
));
2049 if (vd
->vdev_degraded
) {
2050 ASSERT(vd
->vdev_children
== 0);
2051 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2052 VDEV_AUX_ERR_EXCEEDED
);
2054 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
2058 * For hole or missing vdevs we just return success.
2060 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
2063 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2064 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
2065 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2071 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
2072 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
2074 if (vd
->vdev_children
== 0) {
2075 if (osize
< SPA_MINDEVSIZE
) {
2076 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2077 VDEV_AUX_TOO_SMALL
);
2078 return (SET_ERROR(EOVERFLOW
));
2081 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
2082 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
2083 VDEV_LABEL_END_SIZE
);
2085 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
2086 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
2087 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2088 VDEV_AUX_TOO_SMALL
);
2089 return (SET_ERROR(EOVERFLOW
));
2093 max_asize
= max_osize
;
2097 * If the vdev was expanded, record this so that we can re-create the
2098 * uberblock rings in labels {2,3}, during the next sync.
2100 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
2101 vd
->vdev_copy_uberblocks
= B_TRUE
;
2103 vd
->vdev_psize
= psize
;
2106 * Make sure the allocatable size hasn't shrunk too much.
2108 if (asize
< vd
->vdev_min_asize
) {
2109 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2110 VDEV_AUX_BAD_LABEL
);
2111 return (SET_ERROR(EINVAL
));
2115 * We can always set the logical/physical ashift members since
2116 * their values are only used to calculate the vdev_ashift when
2117 * the device is first added to the config. These values should
2118 * not be used for anything else since they may change whenever
2119 * the device is reopened and we don't store them in the label.
2121 vd
->vdev_physical_ashift
=
2122 MAX(physical_ashift
, vd
->vdev_physical_ashift
);
2123 vd
->vdev_logical_ashift
= MAX(logical_ashift
,
2124 vd
->vdev_logical_ashift
);
2126 if (vd
->vdev_asize
== 0) {
2128 * This is the first-ever open, so use the computed values.
2129 * For compatibility, a different ashift can be requested.
2131 vd
->vdev_asize
= asize
;
2132 vd
->vdev_max_asize
= max_asize
;
2135 * If the vdev_ashift was not overridden at creation time,
2136 * then set it the logical ashift and optimize the ashift.
2138 if (vd
->vdev_ashift
== 0) {
2139 vd
->vdev_ashift
= vd
->vdev_logical_ashift
;
2141 if (vd
->vdev_logical_ashift
> ASHIFT_MAX
) {
2142 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2143 VDEV_AUX_ASHIFT_TOO_BIG
);
2144 return (SET_ERROR(EDOM
));
2147 if (vd
->vdev_top
== vd
) {
2148 vdev_ashift_optimize(vd
);
2151 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
2152 vd
->vdev_ashift
> ASHIFT_MAX
)) {
2153 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2154 VDEV_AUX_BAD_ASHIFT
);
2155 return (SET_ERROR(EDOM
));
2159 * Make sure the alignment required hasn't increased.
2161 if (vd
->vdev_ashift
> vd
->vdev_top
->vdev_ashift
&&
2162 vd
->vdev_ops
->vdev_op_leaf
) {
2163 (void) zfs_ereport_post(
2164 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
2165 spa
, vd
, NULL
, NULL
, 0);
2166 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2167 VDEV_AUX_BAD_LABEL
);
2168 return (SET_ERROR(EDOM
));
2170 vd
->vdev_max_asize
= max_asize
;
2174 * If all children are healthy we update asize if either:
2175 * The asize has increased, due to a device expansion caused by dynamic
2176 * LUN growth or vdev replacement, and automatic expansion is enabled;
2177 * making the additional space available.
2179 * The asize has decreased, due to a device shrink usually caused by a
2180 * vdev replace with a smaller device. This ensures that calculations
2181 * based of max_asize and asize e.g. esize are always valid. It's safe
2182 * to do this as we've already validated that asize is greater than
2185 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
2186 ((asize
> vd
->vdev_asize
&&
2187 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
2188 (asize
< vd
->vdev_asize
)))
2189 vd
->vdev_asize
= asize
;
2191 vdev_set_min_asize(vd
);
2194 * Ensure we can issue some IO before declaring the
2195 * vdev open for business.
2197 if (vd
->vdev_ops
->vdev_op_leaf
&&
2198 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
2199 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2200 VDEV_AUX_ERR_EXCEEDED
);
2205 * Track the minimum allocation size.
2207 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
2208 vd
->vdev_islog
== 0 && vd
->vdev_aux
== NULL
) {
2209 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
2210 if (min_alloc
< spa
->spa_min_alloc
)
2211 spa
->spa_min_alloc
= min_alloc
;
2215 * If this is a leaf vdev, assess whether a resilver is needed.
2216 * But don't do this if we are doing a reopen for a scrub, since
2217 * this would just restart the scrub we are already doing.
2219 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
)
2220 dsl_scan_assess_vdev(spa
->spa_dsl_pool
, vd
);
2226 vdev_validate_child(void *arg
)
2230 vd
->vdev_validate_thread
= curthread
;
2231 vd
->vdev_validate_error
= vdev_validate(vd
);
2232 vd
->vdev_validate_thread
= NULL
;
2236 * Called once the vdevs are all opened, this routine validates the label
2237 * contents. This needs to be done before vdev_load() so that we don't
2238 * inadvertently do repair I/Os to the wrong device.
2240 * This function will only return failure if one of the vdevs indicates that it
2241 * has since been destroyed or exported. This is only possible if
2242 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
2243 * will be updated but the function will return 0.
2246 vdev_validate(vdev_t
*vd
)
2248 spa_t
*spa
= vd
->vdev_spa
;
2251 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
2255 int children
= vd
->vdev_children
;
2257 if (vdev_validate_skip
)
2261 tq
= taskq_create("vdev_validate", children
, minclsyspri
,
2262 children
, children
, TASKQ_PREPOPULATE
);
2265 for (uint64_t c
= 0; c
< children
; c
++) {
2266 vdev_t
*cvd
= vd
->vdev_child
[c
];
2268 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
2269 vdev_validate_child(cvd
);
2271 VERIFY(taskq_dispatch(tq
, vdev_validate_child
, cvd
,
2272 TQ_SLEEP
) != TASKQID_INVALID
);
2279 for (int c
= 0; c
< children
; c
++) {
2280 int error
= vd
->vdev_child
[c
]->vdev_validate_error
;
2283 return (SET_ERROR(EBADF
));
2288 * If the device has already failed, or was marked offline, don't do
2289 * any further validation. Otherwise, label I/O will fail and we will
2290 * overwrite the previous state.
2292 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
2296 * If we are performing an extreme rewind, we allow for a label that
2297 * was modified at a point after the current txg.
2298 * If config lock is not held do not check for the txg. spa_sync could
2299 * be updating the vdev's label before updating spa_last_synced_txg.
2301 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
2302 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
2305 txg
= spa_last_synced_txg(spa
);
2307 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
2308 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2309 VDEV_AUX_BAD_LABEL
);
2310 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
2311 "txg %llu", (u_longlong_t
)txg
);
2316 * Determine if this vdev has been split off into another
2317 * pool. If so, then refuse to open it.
2319 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
2320 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
2321 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2322 VDEV_AUX_SPLIT_POOL
);
2324 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
2328 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
2329 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2330 VDEV_AUX_CORRUPT_DATA
);
2332 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2333 ZPOOL_CONFIG_POOL_GUID
);
2338 * If config is not trusted then ignore the spa guid check. This is
2339 * necessary because if the machine crashed during a re-guid the new
2340 * guid might have been written to all of the vdev labels, but not the
2341 * cached config. The check will be performed again once we have the
2342 * trusted config from the MOS.
2344 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
2345 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2346 VDEV_AUX_CORRUPT_DATA
);
2348 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
2349 "match config (%llu != %llu)", (u_longlong_t
)guid
,
2350 (u_longlong_t
)spa_guid(spa
));
2354 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
2355 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
2359 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
2360 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2361 VDEV_AUX_CORRUPT_DATA
);
2363 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2368 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
2370 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2371 VDEV_AUX_CORRUPT_DATA
);
2373 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2374 ZPOOL_CONFIG_TOP_GUID
);
2379 * If this vdev just became a top-level vdev because its sibling was
2380 * detached, it will have adopted the parent's vdev guid -- but the
2381 * label may or may not be on disk yet. Fortunately, either version
2382 * of the label will have the same top guid, so if we're a top-level
2383 * vdev, we can safely compare to that instead.
2384 * However, if the config comes from a cachefile that failed to update
2385 * after the detach, a top-level vdev will appear as a non top-level
2386 * vdev in the config. Also relax the constraints if we perform an
2389 * If we split this vdev off instead, then we also check the
2390 * original pool's guid. We don't want to consider the vdev
2391 * corrupt if it is partway through a split operation.
2393 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
2394 boolean_t mismatch
= B_FALSE
;
2395 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
2396 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
2399 if (vd
->vdev_guid
!= top_guid
&&
2400 vd
->vdev_top
->vdev_guid
!= guid
)
2405 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2406 VDEV_AUX_CORRUPT_DATA
);
2408 vdev_dbgmsg(vd
, "vdev_validate: config guid "
2409 "doesn't match label guid");
2410 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
2411 (u_longlong_t
)vd
->vdev_guid
,
2412 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
2413 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
2414 "aux_guid %llu", (u_longlong_t
)guid
,
2415 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
2420 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
2422 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2423 VDEV_AUX_CORRUPT_DATA
);
2425 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2426 ZPOOL_CONFIG_POOL_STATE
);
2433 * If this is a verbatim import, no need to check the
2434 * state of the pool.
2436 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
2437 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
2438 state
!= POOL_STATE_ACTIVE
) {
2439 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
2440 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
2441 return (SET_ERROR(EBADF
));
2445 * If we were able to open and validate a vdev that was
2446 * previously marked permanently unavailable, clear that state
2449 if (vd
->vdev_not_present
)
2450 vd
->vdev_not_present
= 0;
2456 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2459 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2460 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2461 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2462 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2463 dvd
->vdev_path
, svd
->vdev_path
);
2464 spa_strfree(dvd
->vdev_path
);
2465 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2467 } else if (svd
->vdev_path
!= NULL
) {
2468 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2469 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2470 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2474 * Our enclosure sysfs path may have changed between imports
2476 old
= dvd
->vdev_enc_sysfs_path
;
2477 new = svd
->vdev_enc_sysfs_path
;
2478 if ((old
!= NULL
&& new == NULL
) ||
2479 (old
== NULL
&& new != NULL
) ||
2480 ((old
!= NULL
&& new != NULL
) && strcmp(new, old
) != 0)) {
2481 zfs_dbgmsg("vdev_copy_path: vdev %llu: vdev_enc_sysfs_path "
2482 "changed from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2485 if (dvd
->vdev_enc_sysfs_path
)
2486 spa_strfree(dvd
->vdev_enc_sysfs_path
);
2488 if (svd
->vdev_enc_sysfs_path
) {
2489 dvd
->vdev_enc_sysfs_path
= spa_strdup(
2490 svd
->vdev_enc_sysfs_path
);
2492 dvd
->vdev_enc_sysfs_path
= NULL
;
2498 * Recursively copy vdev paths from one vdev to another. Source and destination
2499 * vdev trees must have same geometry otherwise return error. Intended to copy
2500 * paths from userland config into MOS config.
2503 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2505 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2506 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2507 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2510 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2511 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2512 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2513 return (SET_ERROR(EINVAL
));
2516 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2517 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2518 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2519 (u_longlong_t
)dvd
->vdev_guid
);
2520 return (SET_ERROR(EINVAL
));
2523 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2524 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2525 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2526 (u_longlong_t
)dvd
->vdev_children
);
2527 return (SET_ERROR(EINVAL
));
2530 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2531 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2532 dvd
->vdev_child
[i
]);
2537 if (svd
->vdev_ops
->vdev_op_leaf
)
2538 vdev_copy_path_impl(svd
, dvd
);
2544 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2546 ASSERT(stvd
->vdev_top
== stvd
);
2547 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2549 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2550 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2553 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2557 * The idea here is that while a vdev can shift positions within
2558 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2559 * step outside of it.
2561 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2563 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2566 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2568 vdev_copy_path_impl(vd
, dvd
);
2572 * Recursively copy vdev paths from one root vdev to another. Source and
2573 * destination vdev trees may differ in geometry. For each destination leaf
2574 * vdev, search a vdev with the same guid and top vdev id in the source.
2575 * Intended to copy paths from userland config into MOS config.
2578 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2580 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2581 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2582 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2584 for (uint64_t i
= 0; i
< children
; i
++) {
2585 vdev_copy_path_search(srvd
->vdev_child
[i
],
2586 drvd
->vdev_child
[i
]);
2591 * Close a virtual device.
2594 vdev_close(vdev_t
*vd
)
2596 vdev_t
*pvd
= vd
->vdev_parent
;
2597 spa_t
*spa __maybe_unused
= vd
->vdev_spa
;
2600 ASSERT(vd
->vdev_open_thread
== curthread
||
2601 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2604 * If our parent is reopening, then we are as well, unless we are
2607 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2608 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2610 vd
->vdev_ops
->vdev_op_close(vd
);
2613 * We record the previous state before we close it, so that if we are
2614 * doing a reopen(), we don't generate FMA ereports if we notice that
2615 * it's still faulted.
2617 vd
->vdev_prevstate
= vd
->vdev_state
;
2619 if (vd
->vdev_offline
)
2620 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2622 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2623 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2627 vdev_hold(vdev_t
*vd
)
2629 spa_t
*spa
= vd
->vdev_spa
;
2631 ASSERT(spa_is_root(spa
));
2632 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2635 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2636 vdev_hold(vd
->vdev_child
[c
]);
2638 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_hold
!= NULL
)
2639 vd
->vdev_ops
->vdev_op_hold(vd
);
2643 vdev_rele(vdev_t
*vd
)
2645 ASSERT(spa_is_root(vd
->vdev_spa
));
2646 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2647 vdev_rele(vd
->vdev_child
[c
]);
2649 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_rele
!= NULL
)
2650 vd
->vdev_ops
->vdev_op_rele(vd
);
2654 * Reopen all interior vdevs and any unopened leaves. We don't actually
2655 * reopen leaf vdevs which had previously been opened as they might deadlock
2656 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2657 * If the leaf has never been opened then open it, as usual.
2660 vdev_reopen(vdev_t
*vd
)
2662 spa_t
*spa
= vd
->vdev_spa
;
2664 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2666 /* set the reopening flag unless we're taking the vdev offline */
2667 vd
->vdev_reopening
= !vd
->vdev_offline
;
2669 (void) vdev_open(vd
);
2672 * Call vdev_validate() here to make sure we have the same device.
2673 * Otherwise, a device with an invalid label could be successfully
2674 * opened in response to vdev_reopen().
2677 (void) vdev_validate_aux(vd
);
2678 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2679 vd
->vdev_aux
== &spa
->spa_l2cache
) {
2681 * In case the vdev is present we should evict all ARC
2682 * buffers and pointers to log blocks and reclaim their
2683 * space before restoring its contents to L2ARC.
2685 if (l2arc_vdev_present(vd
)) {
2686 l2arc_rebuild_vdev(vd
, B_TRUE
);
2688 l2arc_add_vdev(spa
, vd
);
2690 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
2691 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
2694 (void) vdev_validate(vd
);
2698 * Recheck if resilver is still needed and cancel any
2699 * scheduled resilver if resilver is unneeded.
2701 if (!vdev_resilver_needed(spa
->spa_root_vdev
, NULL
, NULL
) &&
2702 spa
->spa_async_tasks
& SPA_ASYNC_RESILVER
) {
2703 mutex_enter(&spa
->spa_async_lock
);
2704 spa
->spa_async_tasks
&= ~SPA_ASYNC_RESILVER
;
2705 mutex_exit(&spa
->spa_async_lock
);
2709 * Reassess parent vdev's health.
2711 vdev_propagate_state(vd
);
2715 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2720 * Normally, partial opens (e.g. of a mirror) are allowed.
2721 * For a create, however, we want to fail the request if
2722 * there are any components we can't open.
2724 error
= vdev_open(vd
);
2726 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2728 return (error
? error
: SET_ERROR(ENXIO
));
2732 * Recursively load DTLs and initialize all labels.
2734 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2735 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2736 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2745 vdev_metaslab_set_size(vdev_t
*vd
)
2747 uint64_t asize
= vd
->vdev_asize
;
2748 uint64_t ms_count
= asize
>> zfs_vdev_default_ms_shift
;
2752 * There are two dimensions to the metaslab sizing calculation:
2753 * the size of the metaslab and the count of metaslabs per vdev.
2755 * The default values used below are a good balance between memory
2756 * usage (larger metaslab size means more memory needed for loaded
2757 * metaslabs; more metaslabs means more memory needed for the
2758 * metaslab_t structs), metaslab load time (larger metaslabs take
2759 * longer to load), and metaslab sync time (more metaslabs means
2760 * more time spent syncing all of them).
2762 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2763 * The range of the dimensions are as follows:
2765 * 2^29 <= ms_size <= 2^34
2766 * 16 <= ms_count <= 131,072
2768 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2769 * at least 512MB (2^29) to minimize fragmentation effects when
2770 * testing with smaller devices. However, the count constraint
2771 * of at least 16 metaslabs will override this minimum size goal.
2773 * On the upper end of vdev sizes, we aim for a maximum metaslab
2774 * size of 16GB. However, we will cap the total count to 2^17
2775 * metaslabs to keep our memory footprint in check and let the
2776 * metaslab size grow from there if that limit is hit.
2778 * The net effect of applying above constrains is summarized below.
2780 * vdev size metaslab count
2781 * --------------|-----------------
2783 * 8GB - 100GB one per 512MB
2785 * 3TB - 2PB one per 16GB
2787 * --------------------------------
2789 * Finally, note that all of the above calculate the initial
2790 * number of metaslabs. Expanding a top-level vdev will result
2791 * in additional metaslabs being allocated making it possible
2792 * to exceed the zfs_vdev_ms_count_limit.
2795 if (ms_count
< zfs_vdev_min_ms_count
)
2796 ms_shift
= highbit64(asize
/ zfs_vdev_min_ms_count
);
2797 else if (ms_count
> zfs_vdev_default_ms_count
)
2798 ms_shift
= highbit64(asize
/ zfs_vdev_default_ms_count
);
2800 ms_shift
= zfs_vdev_default_ms_shift
;
2802 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2803 ms_shift
= SPA_MAXBLOCKSHIFT
;
2804 } else if (ms_shift
> zfs_vdev_max_ms_shift
) {
2805 ms_shift
= zfs_vdev_max_ms_shift
;
2806 /* cap the total count to constrain memory footprint */
2807 if ((asize
>> ms_shift
) > zfs_vdev_ms_count_limit
)
2808 ms_shift
= highbit64(asize
/ zfs_vdev_ms_count_limit
);
2811 vd
->vdev_ms_shift
= ms_shift
;
2812 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2816 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2818 ASSERT(vd
== vd
->vdev_top
);
2819 /* indirect vdevs don't have metaslabs or dtls */
2820 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2821 ASSERT(ISP2(flags
));
2822 ASSERT(spa_writeable(vd
->vdev_spa
));
2824 if (flags
& VDD_METASLAB
)
2825 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2827 if (flags
& VDD_DTL
)
2828 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2830 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2834 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2836 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2837 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2839 if (vd
->vdev_ops
->vdev_op_leaf
)
2840 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2846 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2847 * the vdev has less than perfect replication. There are four kinds of DTL:
2849 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2851 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2853 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2854 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2855 * txgs that was scrubbed.
2857 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2858 * persistent errors or just some device being offline.
2859 * Unlike the other three, the DTL_OUTAGE map is not generally
2860 * maintained; it's only computed when needed, typically to
2861 * determine whether a device can be detached.
2863 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2864 * either has the data or it doesn't.
2866 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2867 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2868 * if any child is less than fully replicated, then so is its parent.
2869 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2870 * comprising only those txgs which appear in 'maxfaults' or more children;
2871 * those are the txgs we don't have enough replication to read. For example,
2872 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2873 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2874 * two child DTL_MISSING maps.
2876 * It should be clear from the above that to compute the DTLs and outage maps
2877 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2878 * Therefore, that is all we keep on disk. When loading the pool, or after
2879 * a configuration change, we generate all other DTLs from first principles.
2882 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2884 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2886 ASSERT(t
< DTL_TYPES
);
2887 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2888 ASSERT(spa_writeable(vd
->vdev_spa
));
2890 mutex_enter(&vd
->vdev_dtl_lock
);
2891 if (!range_tree_contains(rt
, txg
, size
))
2892 range_tree_add(rt
, txg
, size
);
2893 mutex_exit(&vd
->vdev_dtl_lock
);
2897 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2899 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2900 boolean_t dirty
= B_FALSE
;
2902 ASSERT(t
< DTL_TYPES
);
2903 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2906 * While we are loading the pool, the DTLs have not been loaded yet.
2907 * This isn't a problem but it can result in devices being tried
2908 * which are known to not have the data. In which case, the import
2909 * is relying on the checksum to ensure that we get the right data.
2910 * Note that while importing we are only reading the MOS, which is
2911 * always checksummed.
2913 mutex_enter(&vd
->vdev_dtl_lock
);
2914 if (!range_tree_is_empty(rt
))
2915 dirty
= range_tree_contains(rt
, txg
, size
);
2916 mutex_exit(&vd
->vdev_dtl_lock
);
2922 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2924 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2927 mutex_enter(&vd
->vdev_dtl_lock
);
2928 empty
= range_tree_is_empty(rt
);
2929 mutex_exit(&vd
->vdev_dtl_lock
);
2935 * Check if the txg falls within the range which must be
2936 * resilvered. DVAs outside this range can always be skipped.
2939 vdev_default_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2940 uint64_t phys_birth
)
2942 (void) dva
, (void) psize
;
2944 /* Set by sequential resilver. */
2945 if (phys_birth
== TXG_UNKNOWN
)
2948 return (vdev_dtl_contains(vd
, DTL_PARTIAL
, phys_birth
, 1));
2952 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered.
2955 vdev_dtl_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2956 uint64_t phys_birth
)
2958 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2960 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2961 vd
->vdev_ops
->vdev_op_leaf
)
2964 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, dva
, psize
,
2969 * Returns the lowest txg in the DTL range.
2972 vdev_dtl_min(vdev_t
*vd
)
2974 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2975 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2976 ASSERT0(vd
->vdev_children
);
2978 return (range_tree_min(vd
->vdev_dtl
[DTL_MISSING
]) - 1);
2982 * Returns the highest txg in the DTL.
2985 vdev_dtl_max(vdev_t
*vd
)
2987 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2988 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2989 ASSERT0(vd
->vdev_children
);
2991 return (range_tree_max(vd
->vdev_dtl
[DTL_MISSING
]));
2995 * Determine if a resilvering vdev should remove any DTL entries from
2996 * its range. If the vdev was resilvering for the entire duration of the
2997 * scan then it should excise that range from its DTLs. Otherwise, this
2998 * vdev is considered partially resilvered and should leave its DTL
2999 * entries intact. The comment in vdev_dtl_reassess() describes how we
3003 vdev_dtl_should_excise(vdev_t
*vd
, boolean_t rebuild_done
)
3005 ASSERT0(vd
->vdev_children
);
3007 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
3010 if (vd
->vdev_resilver_deferred
)
3013 if (range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
3017 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3018 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
3020 /* Rebuild not initiated by attach */
3021 if (vd
->vdev_rebuild_txg
== 0)
3025 * When a rebuild completes without error then all missing data
3026 * up to the rebuild max txg has been reconstructed and the DTL
3027 * is eligible for excision.
3029 if (vrp
->vrp_rebuild_state
== VDEV_REBUILD_COMPLETE
&&
3030 vdev_dtl_max(vd
) <= vrp
->vrp_max_txg
) {
3031 ASSERT3U(vrp
->vrp_min_txg
, <=, vdev_dtl_min(vd
));
3032 ASSERT3U(vrp
->vrp_min_txg
, <, vd
->vdev_rebuild_txg
);
3033 ASSERT3U(vd
->vdev_rebuild_txg
, <=, vrp
->vrp_max_txg
);
3037 dsl_scan_t
*scn
= vd
->vdev_spa
->spa_dsl_pool
->dp_scan
;
3038 dsl_scan_phys_t
*scnp __maybe_unused
= &scn
->scn_phys
;
3040 /* Resilver not initiated by attach */
3041 if (vd
->vdev_resilver_txg
== 0)
3045 * When a resilver is initiated the scan will assign the
3046 * scn_max_txg value to the highest txg value that exists
3047 * in all DTLs. If this device's max DTL is not part of this
3048 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
3049 * then it is not eligible for excision.
3051 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
3052 ASSERT3U(scnp
->scn_min_txg
, <=, vdev_dtl_min(vd
));
3053 ASSERT3U(scnp
->scn_min_txg
, <, vd
->vdev_resilver_txg
);
3054 ASSERT3U(vd
->vdev_resilver_txg
, <=, scnp
->scn_max_txg
);
3063 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
3064 * write operations will be issued to the pool.
3067 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
,
3068 boolean_t scrub_done
, boolean_t rebuild_done
)
3070 spa_t
*spa
= vd
->vdev_spa
;
3074 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
3076 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3077 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
3078 scrub_txg
, scrub_done
, rebuild_done
);
3080 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
3083 if (vd
->vdev_ops
->vdev_op_leaf
) {
3084 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
3085 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3086 boolean_t check_excise
= B_FALSE
;
3087 boolean_t wasempty
= B_TRUE
;
3089 mutex_enter(&vd
->vdev_dtl_lock
);
3092 * If requested, pretend the scan or rebuild completed cleanly.
3094 if (zfs_scan_ignore_errors
) {
3096 scn
->scn_phys
.scn_errors
= 0;
3098 vr
->vr_rebuild_phys
.vrp_errors
= 0;
3101 if (scrub_txg
!= 0 &&
3102 !range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3104 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
3105 "dtl:%llu/%llu errors:%llu",
3106 (u_longlong_t
)vd
->vdev_guid
, (u_longlong_t
)txg
,
3107 (u_longlong_t
)scrub_txg
, spa
->spa_scrub_started
,
3108 (u_longlong_t
)vdev_dtl_min(vd
),
3109 (u_longlong_t
)vdev_dtl_max(vd
),
3110 (u_longlong_t
)(scn
? scn
->scn_phys
.scn_errors
: 0));
3114 * If we've completed a scrub/resilver or a rebuild cleanly
3115 * then determine if this vdev should remove any DTLs. We
3116 * only want to excise regions on vdevs that were available
3117 * during the entire duration of this scan.
3120 vr
!= NULL
&& vr
->vr_rebuild_phys
.vrp_errors
== 0) {
3121 check_excise
= B_TRUE
;
3123 if (spa
->spa_scrub_started
||
3124 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) {
3125 check_excise
= B_TRUE
;
3129 if (scrub_txg
&& check_excise
&&
3130 vdev_dtl_should_excise(vd
, rebuild_done
)) {
3132 * We completed a scrub, resilver or rebuild up to
3133 * scrub_txg. If we did it without rebooting, then
3134 * the scrub dtl will be valid, so excise the old
3135 * region and fold in the scrub dtl. Otherwise,
3136 * leave the dtl as-is if there was an error.
3138 * There's little trick here: to excise the beginning
3139 * of the DTL_MISSING map, we put it into a reference
3140 * tree and then add a segment with refcnt -1 that
3141 * covers the range [0, scrub_txg). This means
3142 * that each txg in that range has refcnt -1 or 0.
3143 * We then add DTL_SCRUB with a refcnt of 2, so that
3144 * entries in the range [0, scrub_txg) will have a
3145 * positive refcnt -- either 1 or 2. We then convert
3146 * the reference tree into the new DTL_MISSING map.
3148 space_reftree_create(&reftree
);
3149 space_reftree_add_map(&reftree
,
3150 vd
->vdev_dtl
[DTL_MISSING
], 1);
3151 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
3152 space_reftree_add_map(&reftree
,
3153 vd
->vdev_dtl
[DTL_SCRUB
], 2);
3154 space_reftree_generate_map(&reftree
,
3155 vd
->vdev_dtl
[DTL_MISSING
], 1);
3156 space_reftree_destroy(&reftree
);
3158 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3159 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
3160 (u_longlong_t
)vdev_dtl_min(vd
),
3161 (u_longlong_t
)vdev_dtl_max(vd
));
3162 } else if (!wasempty
) {
3163 zfs_dbgmsg("DTL_MISSING is now empty");
3166 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
3167 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3168 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
3170 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
3171 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
3172 if (!vdev_readable(vd
))
3173 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
3175 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3176 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
3179 * If the vdev was resilvering or rebuilding and no longer
3180 * has any DTLs then reset the appropriate flag and dirty
3181 * the top level so that we persist the change.
3184 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3185 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
3186 if (vd
->vdev_rebuild_txg
!= 0) {
3187 vd
->vdev_rebuild_txg
= 0;
3188 vdev_config_dirty(vd
->vdev_top
);
3189 } else if (vd
->vdev_resilver_txg
!= 0) {
3190 vd
->vdev_resilver_txg
= 0;
3191 vdev_config_dirty(vd
->vdev_top
);
3195 mutex_exit(&vd
->vdev_dtl_lock
);
3198 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
3202 mutex_enter(&vd
->vdev_dtl_lock
);
3203 for (int t
= 0; t
< DTL_TYPES
; t
++) {
3204 /* account for child's outage in parent's missing map */
3205 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
3207 continue; /* leaf vdevs only */
3208 if (t
== DTL_PARTIAL
)
3209 minref
= 1; /* i.e. non-zero */
3210 else if (vdev_get_nparity(vd
) != 0)
3211 minref
= vdev_get_nparity(vd
) + 1; /* RAID-Z, dRAID */
3213 minref
= vd
->vdev_children
; /* any kind of mirror */
3214 space_reftree_create(&reftree
);
3215 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3216 vdev_t
*cvd
= vd
->vdev_child
[c
];
3217 mutex_enter(&cvd
->vdev_dtl_lock
);
3218 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
3219 mutex_exit(&cvd
->vdev_dtl_lock
);
3221 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
3222 space_reftree_destroy(&reftree
);
3224 mutex_exit(&vd
->vdev_dtl_lock
);
3228 * Iterate over all the vdevs except spare, and post kobj events
3231 vdev_post_kobj_evt(vdev_t
*vd
)
3233 if (vd
->vdev_ops
->vdev_op_kobj_evt_post
&&
3234 vd
->vdev_kobj_flag
== B_FALSE
) {
3235 vd
->vdev_kobj_flag
= B_TRUE
;
3236 vd
->vdev_ops
->vdev_op_kobj_evt_post(vd
);
3239 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3240 vdev_post_kobj_evt(vd
->vdev_child
[c
]);
3244 * Iterate over all the vdevs except spare, and clear kobj events
3247 vdev_clear_kobj_evt(vdev_t
*vd
)
3249 vd
->vdev_kobj_flag
= B_FALSE
;
3251 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3252 vdev_clear_kobj_evt(vd
->vdev_child
[c
]);
3256 vdev_dtl_load(vdev_t
*vd
)
3258 spa_t
*spa
= vd
->vdev_spa
;
3259 objset_t
*mos
= spa
->spa_meta_objset
;
3263 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
3264 ASSERT(vdev_is_concrete(vd
));
3267 * If the dtl cannot be sync'd there is no need to open it.
3269 if (spa
->spa_mode
== SPA_MODE_READ
&& !spa
->spa_read_spacemaps
)
3272 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
3273 vd
->vdev_dtl_object
, 0, -1ULL, 0);
3276 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3278 rt
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3279 error
= space_map_load(vd
->vdev_dtl_sm
, rt
, SM_ALLOC
);
3281 mutex_enter(&vd
->vdev_dtl_lock
);
3282 range_tree_walk(rt
, range_tree_add
,
3283 vd
->vdev_dtl
[DTL_MISSING
]);
3284 mutex_exit(&vd
->vdev_dtl_lock
);
3287 range_tree_vacate(rt
, NULL
, NULL
);
3288 range_tree_destroy(rt
);
3293 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3294 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
3303 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3305 spa_t
*spa
= vd
->vdev_spa
;
3306 objset_t
*mos
= spa
->spa_meta_objset
;
3307 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
3310 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
3313 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
3314 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
3315 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
3317 ASSERT(string
!= NULL
);
3318 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
3319 1, strlen(string
) + 1, string
, tx
));
3321 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
3322 spa_activate_allocation_classes(spa
, tx
);
3327 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
3329 spa_t
*spa
= vd
->vdev_spa
;
3331 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
3332 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3337 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
3339 spa_t
*spa
= vd
->vdev_spa
;
3340 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
3341 DMU_OT_NONE
, 0, tx
);
3344 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3351 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3353 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
3354 vd
->vdev_ops
!= &vdev_missing_ops
&&
3355 vd
->vdev_ops
!= &vdev_root_ops
&&
3356 !vd
->vdev_top
->vdev_removing
) {
3357 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
3358 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
3360 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
3361 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
3362 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
3363 vdev_zap_allocation_data(vd
, tx
);
3366 if (vd
->vdev_ops
== &vdev_root_ops
&& vd
->vdev_root_zap
== 0 &&
3367 spa_feature_is_enabled(vd
->vdev_spa
, SPA_FEATURE_AVZ_V2
)) {
3368 if (!spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_AVZ_V2
))
3369 spa_feature_incr(vd
->vdev_spa
, SPA_FEATURE_AVZ_V2
, tx
);
3370 vd
->vdev_root_zap
= vdev_create_link_zap(vd
, tx
);
3373 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3374 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
3379 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
3381 spa_t
*spa
= vd
->vdev_spa
;
3382 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
3383 objset_t
*mos
= spa
->spa_meta_objset
;
3384 range_tree_t
*rtsync
;
3386 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
3388 ASSERT(vdev_is_concrete(vd
));
3389 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3391 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3393 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
3394 mutex_enter(&vd
->vdev_dtl_lock
);
3395 space_map_free(vd
->vdev_dtl_sm
, tx
);
3396 space_map_close(vd
->vdev_dtl_sm
);
3397 vd
->vdev_dtl_sm
= NULL
;
3398 mutex_exit(&vd
->vdev_dtl_lock
);
3401 * We only destroy the leaf ZAP for detached leaves or for
3402 * removed log devices. Removed data devices handle leaf ZAP
3403 * cleanup later, once cancellation is no longer possible.
3405 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
3406 vd
->vdev_top
->vdev_islog
)) {
3407 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
3408 vd
->vdev_leaf_zap
= 0;
3415 if (vd
->vdev_dtl_sm
== NULL
) {
3416 uint64_t new_object
;
3418 new_object
= space_map_alloc(mos
, zfs_vdev_dtl_sm_blksz
, tx
);
3419 VERIFY3U(new_object
, !=, 0);
3421 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
3423 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3426 rtsync
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3428 mutex_enter(&vd
->vdev_dtl_lock
);
3429 range_tree_walk(rt
, range_tree_add
, rtsync
);
3430 mutex_exit(&vd
->vdev_dtl_lock
);
3432 space_map_truncate(vd
->vdev_dtl_sm
, zfs_vdev_dtl_sm_blksz
, tx
);
3433 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
3434 range_tree_vacate(rtsync
, NULL
, NULL
);
3436 range_tree_destroy(rtsync
);
3439 * If the object for the space map has changed then dirty
3440 * the top level so that we update the config.
3442 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
3443 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
3444 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
3445 (u_longlong_t
)object
,
3446 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
3447 vdev_config_dirty(vd
->vdev_top
);
3454 * Determine whether the specified vdev can be offlined/detached/removed
3455 * without losing data.
3458 vdev_dtl_required(vdev_t
*vd
)
3460 spa_t
*spa
= vd
->vdev_spa
;
3461 vdev_t
*tvd
= vd
->vdev_top
;
3462 uint8_t cant_read
= vd
->vdev_cant_read
;
3465 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3467 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
3471 * Temporarily mark the device as unreadable, and then determine
3472 * whether this results in any DTL outages in the top-level vdev.
3473 * If not, we can safely offline/detach/remove the device.
3475 vd
->vdev_cant_read
= B_TRUE
;
3476 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3477 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
3478 vd
->vdev_cant_read
= cant_read
;
3479 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3481 if (!required
&& zio_injection_enabled
) {
3482 required
= !!zio_handle_device_injection(vd
, NULL
,
3490 * Determine if resilver is needed, and if so the txg range.
3493 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
3495 boolean_t needed
= B_FALSE
;
3496 uint64_t thismin
= UINT64_MAX
;
3497 uint64_t thismax
= 0;
3499 if (vd
->vdev_children
== 0) {
3500 mutex_enter(&vd
->vdev_dtl_lock
);
3501 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3502 vdev_writeable(vd
)) {
3504 thismin
= vdev_dtl_min(vd
);
3505 thismax
= vdev_dtl_max(vd
);
3508 mutex_exit(&vd
->vdev_dtl_lock
);
3510 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3511 vdev_t
*cvd
= vd
->vdev_child
[c
];
3512 uint64_t cmin
, cmax
;
3514 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
3515 thismin
= MIN(thismin
, cmin
);
3516 thismax
= MAX(thismax
, cmax
);
3522 if (needed
&& minp
) {
3530 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3531 * will contain either the checkpoint spacemap object or zero if none exists.
3532 * All other errors are returned to the caller.
3535 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
3537 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
3539 if (vd
->vdev_top_zap
== 0) {
3544 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
3545 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
3546 if (error
== ENOENT
) {
3555 vdev_load(vdev_t
*vd
)
3557 int children
= vd
->vdev_children
;
3562 * It's only worthwhile to use the taskq for the root vdev, because the
3563 * slow part is metaslab_init, and that only happens for top-level
3566 if (vd
->vdev_ops
== &vdev_root_ops
&& vd
->vdev_children
> 0) {
3567 tq
= taskq_create("vdev_load", children
, minclsyspri
,
3568 children
, children
, TASKQ_PREPOPULATE
);
3572 * Recursively load all children.
3574 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3575 vdev_t
*cvd
= vd
->vdev_child
[c
];
3577 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
3578 cvd
->vdev_load_error
= vdev_load(cvd
);
3580 VERIFY(taskq_dispatch(tq
, vdev_load_child
,
3581 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
3590 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3591 int error
= vd
->vdev_child
[c
]->vdev_load_error
;
3597 vdev_set_deflate_ratio(vd
);
3600 * On spa_load path, grab the allocation bias from our zap
3602 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3603 spa_t
*spa
= vd
->vdev_spa
;
3606 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3607 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
3610 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
3611 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
3612 } else if (error
!= ENOENT
) {
3613 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3614 VDEV_AUX_CORRUPT_DATA
);
3615 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(top_zap=%llu) "
3616 "failed [error=%d]",
3617 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3622 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3623 spa_t
*spa
= vd
->vdev_spa
;
3626 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3627 vdev_prop_to_name(VDEV_PROP_FAILFAST
), sizeof (failfast
),
3630 vd
->vdev_failfast
= failfast
& 1;
3631 } else if (error
== ENOENT
) {
3632 vd
->vdev_failfast
= vdev_prop_default_numeric(
3633 VDEV_PROP_FAILFAST
);
3636 "vdev_load: zap_lookup(top_zap=%llu) "
3637 "failed [error=%d]",
3638 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3643 * Load any rebuild state from the top-level vdev zap.
3645 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3646 error
= vdev_rebuild_load(vd
);
3647 if (error
&& error
!= ENOTSUP
) {
3648 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3649 VDEV_AUX_CORRUPT_DATA
);
3650 vdev_dbgmsg(vd
, "vdev_load: vdev_rebuild_load "
3651 "failed [error=%d]", error
);
3656 if (vd
->vdev_top_zap
!= 0 || vd
->vdev_leaf_zap
!= 0) {
3659 if (vd
->vdev_top_zap
!= 0)
3660 zapobj
= vd
->vdev_top_zap
;
3662 zapobj
= vd
->vdev_leaf_zap
;
3664 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_N
,
3665 &vd
->vdev_checksum_n
);
3666 if (error
&& error
!= ENOENT
)
3667 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3668 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3670 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_T
,
3671 &vd
->vdev_checksum_t
);
3672 if (error
&& error
!= ENOENT
)
3673 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3674 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3676 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_N
,
3678 if (error
&& error
!= ENOENT
)
3679 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3680 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3682 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_T
,
3684 if (error
&& error
!= ENOENT
)
3685 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3686 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3690 * If this is a top-level vdev, initialize its metaslabs.
3692 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
3693 vdev_metaslab_group_create(vd
);
3695 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
3696 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3697 VDEV_AUX_CORRUPT_DATA
);
3698 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
3699 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
3700 (u_longlong_t
)vd
->vdev_asize
);
3701 return (SET_ERROR(ENXIO
));
3704 error
= vdev_metaslab_init(vd
, 0);
3706 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
3707 "[error=%d]", error
);
3708 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3709 VDEV_AUX_CORRUPT_DATA
);
3713 uint64_t checkpoint_sm_obj
;
3714 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
3715 if (error
== 0 && checkpoint_sm_obj
!= 0) {
3716 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3717 ASSERT(vd
->vdev_asize
!= 0);
3718 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
3720 error
= space_map_open(&vd
->vdev_checkpoint_sm
,
3721 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
3724 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3725 "failed for checkpoint spacemap (obj %llu) "
3727 (u_longlong_t
)checkpoint_sm_obj
, error
);
3730 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3733 * Since the checkpoint_sm contains free entries
3734 * exclusively we can use space_map_allocated() to
3735 * indicate the cumulative checkpointed space that
3738 vd
->vdev_stat
.vs_checkpoint_space
=
3739 -space_map_allocated(vd
->vdev_checkpoint_sm
);
3740 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3741 vd
->vdev_stat
.vs_checkpoint_space
;
3742 } else if (error
!= 0) {
3743 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3744 "checkpoint space map object from vdev ZAP "
3745 "[error=%d]", error
);
3751 * If this is a leaf vdev, load its DTL.
3753 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3754 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3755 VDEV_AUX_CORRUPT_DATA
);
3756 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3757 "[error=%d]", error
);
3761 uint64_t obsolete_sm_object
;
3762 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3763 if (error
== 0 && obsolete_sm_object
!= 0) {
3764 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3765 ASSERT(vd
->vdev_asize
!= 0);
3766 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3768 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3769 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3770 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3771 VDEV_AUX_CORRUPT_DATA
);
3772 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3773 "obsolete spacemap (obj %llu) [error=%d]",
3774 (u_longlong_t
)obsolete_sm_object
, error
);
3777 } else if (error
!= 0) {
3778 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve obsolete "
3779 "space map object from vdev ZAP [error=%d]", error
);
3787 * The special vdev case is used for hot spares and l2cache devices. Its
3788 * sole purpose it to set the vdev state for the associated vdev. To do this,
3789 * we make sure that we can open the underlying device, then try to read the
3790 * label, and make sure that the label is sane and that it hasn't been
3791 * repurposed to another pool.
3794 vdev_validate_aux(vdev_t
*vd
)
3797 uint64_t guid
, version
;
3800 if (!vdev_readable(vd
))
3803 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
3804 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3805 VDEV_AUX_CORRUPT_DATA
);
3809 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
3810 !SPA_VERSION_IS_SUPPORTED(version
) ||
3811 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
3812 guid
!= vd
->vdev_guid
||
3813 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
3814 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3815 VDEV_AUX_CORRUPT_DATA
);
3821 * We don't actually check the pool state here. If it's in fact in
3822 * use by another pool, we update this fact on the fly when requested.
3829 vdev_destroy_ms_flush_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3831 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3833 if (vd
->vdev_top_zap
== 0)
3836 uint64_t object
= 0;
3837 int err
= zap_lookup(mos
, vd
->vdev_top_zap
,
3838 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, sizeof (uint64_t), 1, &object
);
3843 VERIFY0(dmu_object_free(mos
, object
, tx
));
3844 VERIFY0(zap_remove(mos
, vd
->vdev_top_zap
,
3845 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, tx
));
3849 * Free the objects used to store this vdev's spacemaps, and the array
3850 * that points to them.
3853 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3855 if (vd
->vdev_ms_array
== 0)
3858 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3859 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
3860 size_t array_bytes
= array_count
* sizeof (uint64_t);
3861 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
3862 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
3863 array_bytes
, smobj_array
, 0));
3865 for (uint64_t i
= 0; i
< array_count
; i
++) {
3866 uint64_t smobj
= smobj_array
[i
];
3870 space_map_free_obj(mos
, smobj
, tx
);
3873 kmem_free(smobj_array
, array_bytes
);
3874 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
3875 vdev_destroy_ms_flush_data(vd
, tx
);
3876 vd
->vdev_ms_array
= 0;
3880 vdev_remove_empty_log(vdev_t
*vd
, uint64_t txg
)
3882 spa_t
*spa
= vd
->vdev_spa
;
3884 ASSERT(vd
->vdev_islog
);
3885 ASSERT(vd
== vd
->vdev_top
);
3886 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
3888 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3890 vdev_destroy_spacemaps(vd
, tx
);
3891 if (vd
->vdev_top_zap
!= 0) {
3892 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3893 vd
->vdev_top_zap
= 0;
3900 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3903 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3905 ASSERT(vdev_is_concrete(vd
));
3907 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
3909 metaslab_sync_done(msp
, txg
);
3912 metaslab_sync_reassess(vd
->vdev_mg
);
3913 if (vd
->vdev_log_mg
!= NULL
)
3914 metaslab_sync_reassess(vd
->vdev_log_mg
);
3919 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3921 spa_t
*spa
= vd
->vdev_spa
;
3925 ASSERT3U(txg
, ==, spa
->spa_syncing_txg
);
3926 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3927 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3928 ASSERT(vd
->vdev_removing
||
3929 vd
->vdev_ops
== &vdev_indirect_ops
);
3931 vdev_indirect_sync_obsolete(vd
, tx
);
3934 * If the vdev is indirect, it can't have dirty
3935 * metaslabs or DTLs.
3937 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3938 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3939 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3945 ASSERT(vdev_is_concrete(vd
));
3947 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3948 !vd
->vdev_removing
) {
3949 ASSERT(vd
== vd
->vdev_top
);
3950 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3951 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3952 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3953 ASSERT(vd
->vdev_ms_array
!= 0);
3954 vdev_config_dirty(vd
);
3957 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3958 metaslab_sync(msp
, txg
);
3959 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3962 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3963 vdev_dtl_sync(lvd
, txg
);
3966 * If this is an empty log device being removed, destroy the
3967 * metadata associated with it.
3969 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3970 vdev_remove_empty_log(vd
, txg
);
3972 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3977 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3979 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3983 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3984 * not be opened, and no I/O is attempted.
3987 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3991 spa_vdev_state_enter(spa
, SCL_NONE
);
3993 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3994 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3996 if (!vd
->vdev_ops
->vdev_op_leaf
)
3997 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4002 * If user did a 'zpool offline -f' then make the fault persist across
4005 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
4007 * There are two kinds of forced faults: temporary and
4008 * persistent. Temporary faults go away at pool import, while
4009 * persistent faults stay set. Both types of faults can be
4010 * cleared with a zpool clear.
4012 * We tell if a vdev is persistently faulted by looking at the
4013 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
4014 * import then it's a persistent fault. Otherwise, it's
4015 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
4016 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
4017 * tells vdev_config_generate() (which gets run later) to set
4018 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
4020 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
4021 vd
->vdev_tmpoffline
= B_FALSE
;
4022 aux
= VDEV_AUX_EXTERNAL
;
4024 vd
->vdev_tmpoffline
= B_TRUE
;
4028 * We don't directly use the aux state here, but if we do a
4029 * vdev_reopen(), we need this value to be present to remember why we
4032 vd
->vdev_label_aux
= aux
;
4035 * Faulted state takes precedence over degraded.
4037 vd
->vdev_delayed_close
= B_FALSE
;
4038 vd
->vdev_faulted
= 1ULL;
4039 vd
->vdev_degraded
= 0ULL;
4040 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
4043 * If this device has the only valid copy of the data, then
4044 * back off and simply mark the vdev as degraded instead.
4046 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
4047 vd
->vdev_degraded
= 1ULL;
4048 vd
->vdev_faulted
= 0ULL;
4051 * If we reopen the device and it's not dead, only then do we
4056 if (vdev_readable(vd
))
4057 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
4060 return (spa_vdev_state_exit(spa
, vd
, 0));
4064 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
4065 * user that something is wrong. The vdev continues to operate as normal as far
4066 * as I/O is concerned.
4069 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
4073 spa_vdev_state_enter(spa
, SCL_NONE
);
4075 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4076 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4078 if (!vd
->vdev_ops
->vdev_op_leaf
)
4079 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4082 * If the vdev is already faulted, then don't do anything.
4084 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
4085 return (spa_vdev_state_exit(spa
, NULL
, 0));
4087 vd
->vdev_degraded
= 1ULL;
4088 if (!vdev_is_dead(vd
))
4089 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
4092 return (spa_vdev_state_exit(spa
, vd
, 0));
4096 vdev_remove_wanted(spa_t
*spa
, uint64_t guid
)
4100 spa_vdev_state_enter(spa
, SCL_NONE
);
4102 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4103 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4106 * If the vdev is already removed, or expanding which can trigger
4107 * repartition add/remove events, then don't do anything.
4109 if (vd
->vdev_removed
|| vd
->vdev_expanding
)
4110 return (spa_vdev_state_exit(spa
, NULL
, 0));
4113 * Confirm the vdev has been removed, otherwise don't do anything.
4115 if (vd
->vdev_ops
->vdev_op_leaf
&& !zio_wait(vdev_probe(vd
, NULL
)))
4116 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(EEXIST
)));
4118 vd
->vdev_remove_wanted
= B_TRUE
;
4119 spa_async_request(spa
, SPA_ASYNC_REMOVE
);
4121 return (spa_vdev_state_exit(spa
, vd
, 0));
4126 * Online the given vdev.
4128 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
4129 * spare device should be detached when the device finishes resilvering.
4130 * Second, the online should be treated like a 'test' online case, so no FMA
4131 * events are generated if the device fails to open.
4134 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
4136 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
4137 boolean_t wasoffline
;
4138 vdev_state_t oldstate
;
4140 spa_vdev_state_enter(spa
, SCL_NONE
);
4142 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4143 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4145 if (!vd
->vdev_ops
->vdev_op_leaf
)
4146 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4148 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
4149 oldstate
= vd
->vdev_state
;
4152 vd
->vdev_offline
= B_FALSE
;
4153 vd
->vdev_tmpoffline
= B_FALSE
;
4154 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
4155 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
4157 /* XXX - L2ARC 1.0 does not support expansion */
4158 if (!vd
->vdev_aux
) {
4159 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4160 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
4161 spa
->spa_autoexpand
);
4162 vd
->vdev_expansion_time
= gethrestime_sec();
4166 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
4168 if (!vd
->vdev_aux
) {
4169 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4170 pvd
->vdev_expanding
= B_FALSE
;
4174 *newstate
= vd
->vdev_state
;
4175 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
4176 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
4177 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4178 vd
->vdev_parent
->vdev_child
[0] == vd
)
4179 vd
->vdev_unspare
= B_TRUE
;
4181 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
4183 /* XXX - L2ARC 1.0 does not support expansion */
4185 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
4186 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4189 /* Restart initializing if necessary */
4190 mutex_enter(&vd
->vdev_initialize_lock
);
4191 if (vdev_writeable(vd
) &&
4192 vd
->vdev_initialize_thread
== NULL
&&
4193 vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) {
4194 (void) vdev_initialize(vd
);
4196 mutex_exit(&vd
->vdev_initialize_lock
);
4199 * Restart trimming if necessary. We do not restart trimming for cache
4200 * devices here. This is triggered by l2arc_rebuild_vdev()
4201 * asynchronously for the whole device or in l2arc_evict() as it evicts
4202 * space for upcoming writes.
4204 mutex_enter(&vd
->vdev_trim_lock
);
4205 if (vdev_writeable(vd
) && !vd
->vdev_isl2cache
&&
4206 vd
->vdev_trim_thread
== NULL
&&
4207 vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
) {
4208 (void) vdev_trim(vd
, vd
->vdev_trim_rate
, vd
->vdev_trim_partial
,
4209 vd
->vdev_trim_secure
);
4211 mutex_exit(&vd
->vdev_trim_lock
);
4214 (oldstate
< VDEV_STATE_DEGRADED
&&
4215 vd
->vdev_state
>= VDEV_STATE_DEGRADED
)) {
4216 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
4219 * Asynchronously detach spare vdev if resilver or
4220 * rebuild is not required
4222 if (vd
->vdev_unspare
&&
4223 !dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4224 !dsl_scan_resilver_scheduled(spa
->spa_dsl_pool
) &&
4225 !vdev_rebuild_active(tvd
))
4226 spa_async_request(spa
, SPA_ASYNC_DETACH_SPARE
);
4228 return (spa_vdev_state_exit(spa
, vd
, 0));
4232 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4236 uint64_t generation
;
4237 metaslab_group_t
*mg
;
4240 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4242 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4243 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4245 if (!vd
->vdev_ops
->vdev_op_leaf
)
4246 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4248 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
4249 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
4253 generation
= spa
->spa_config_generation
+ 1;
4256 * If the device isn't already offline, try to offline it.
4258 if (!vd
->vdev_offline
) {
4260 * If this device has the only valid copy of some data,
4261 * don't allow it to be offlined. Log devices are always
4264 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4265 vdev_dtl_required(vd
))
4266 return (spa_vdev_state_exit(spa
, NULL
,
4270 * If the top-level is a slog and it has had allocations
4271 * then proceed. We check that the vdev's metaslab group
4272 * is not NULL since it's possible that we may have just
4273 * added this vdev but not yet initialized its metaslabs.
4275 if (tvd
->vdev_islog
&& mg
!= NULL
) {
4277 * Prevent any future allocations.
4279 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
4280 metaslab_group_passivate(mg
);
4281 (void) spa_vdev_state_exit(spa
, vd
, 0);
4283 error
= spa_reset_logs(spa
);
4286 * If the log device was successfully reset but has
4287 * checkpointed data, do not offline it.
4290 tvd
->vdev_checkpoint_sm
!= NULL
) {
4291 ASSERT3U(space_map_allocated(
4292 tvd
->vdev_checkpoint_sm
), !=, 0);
4293 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
4296 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4299 * Check to see if the config has changed.
4301 if (error
|| generation
!= spa
->spa_config_generation
) {
4302 metaslab_group_activate(mg
);
4304 return (spa_vdev_state_exit(spa
,
4306 (void) spa_vdev_state_exit(spa
, vd
, 0);
4309 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
4313 * Offline this device and reopen its top-level vdev.
4314 * If the top-level vdev is a log device then just offline
4315 * it. Otherwise, if this action results in the top-level
4316 * vdev becoming unusable, undo it and fail the request.
4318 vd
->vdev_offline
= B_TRUE
;
4321 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4322 vdev_is_dead(tvd
)) {
4323 vd
->vdev_offline
= B_FALSE
;
4325 return (spa_vdev_state_exit(spa
, NULL
,
4330 * Add the device back into the metaslab rotor so that
4331 * once we online the device it's open for business.
4333 if (tvd
->vdev_islog
&& mg
!= NULL
)
4334 metaslab_group_activate(mg
);
4337 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
4339 return (spa_vdev_state_exit(spa
, vd
, 0));
4343 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4347 mutex_enter(&spa
->spa_vdev_top_lock
);
4348 error
= vdev_offline_locked(spa
, guid
, flags
);
4349 mutex_exit(&spa
->spa_vdev_top_lock
);
4355 * Clear the error counts associated with this vdev. Unlike vdev_online() and
4356 * vdev_offline(), we assume the spa config is locked. We also clear all
4357 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
4360 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
4362 vdev_t
*rvd
= spa
->spa_root_vdev
;
4364 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
4369 vd
->vdev_stat
.vs_read_errors
= 0;
4370 vd
->vdev_stat
.vs_write_errors
= 0;
4371 vd
->vdev_stat
.vs_checksum_errors
= 0;
4372 vd
->vdev_stat
.vs_slow_ios
= 0;
4374 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4375 vdev_clear(spa
, vd
->vdev_child
[c
]);
4378 * It makes no sense to "clear" an indirect or removed vdev.
4380 if (!vdev_is_concrete(vd
) || vd
->vdev_removed
)
4384 * If we're in the FAULTED state or have experienced failed I/O, then
4385 * clear the persistent state and attempt to reopen the device. We
4386 * also mark the vdev config dirty, so that the new faulted state is
4387 * written out to disk.
4389 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
4390 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
4392 * When reopening in response to a clear event, it may be due to
4393 * a fmadm repair request. In this case, if the device is
4394 * still broken, we want to still post the ereport again.
4396 vd
->vdev_forcefault
= B_TRUE
;
4398 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
4399 vd
->vdev_cant_read
= B_FALSE
;
4400 vd
->vdev_cant_write
= B_FALSE
;
4401 vd
->vdev_stat
.vs_aux
= 0;
4403 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
4405 vd
->vdev_forcefault
= B_FALSE
;
4407 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
4408 vdev_state_dirty(vd
->vdev_top
);
4410 /* If a resilver isn't required, check if vdevs can be culled */
4411 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
) &&
4412 !dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4413 !dsl_scan_resilver_scheduled(spa
->spa_dsl_pool
))
4414 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
4416 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
4420 * When clearing a FMA-diagnosed fault, we always want to
4421 * unspare the device, as we assume that the original spare was
4422 * done in response to the FMA fault.
4424 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
4425 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4426 vd
->vdev_parent
->vdev_child
[0] == vd
)
4427 vd
->vdev_unspare
= B_TRUE
;
4429 /* Clear recent error events cache (i.e. duplicate events tracking) */
4430 zfs_ereport_clear(spa
, vd
);
4434 vdev_is_dead(vdev_t
*vd
)
4437 * Holes and missing devices are always considered "dead".
4438 * This simplifies the code since we don't have to check for
4439 * these types of devices in the various code paths.
4440 * Instead we rely on the fact that we skip over dead devices
4441 * before issuing I/O to them.
4443 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
4444 vd
->vdev_ops
== &vdev_hole_ops
||
4445 vd
->vdev_ops
== &vdev_missing_ops
);
4449 vdev_readable(vdev_t
*vd
)
4451 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
4455 vdev_writeable(vdev_t
*vd
)
4457 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
4458 vdev_is_concrete(vd
));
4462 vdev_allocatable(vdev_t
*vd
)
4464 uint64_t state
= vd
->vdev_state
;
4467 * We currently allow allocations from vdevs which may be in the
4468 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
4469 * fails to reopen then we'll catch it later when we're holding
4470 * the proper locks. Note that we have to get the vdev state
4471 * in a local variable because although it changes atomically,
4472 * we're asking two separate questions about it.
4474 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
4475 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
4476 vd
->vdev_mg
->mg_initialized
);
4480 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
4482 ASSERT(zio
->io_vd
== vd
);
4484 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
4487 if (zio
->io_type
== ZIO_TYPE_READ
)
4488 return (!vd
->vdev_cant_read
);
4490 if (zio
->io_type
== ZIO_TYPE_WRITE
)
4491 return (!vd
->vdev_cant_write
);
4497 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
4500 * Exclude the dRAID spare when aggregating to avoid double counting
4501 * the ops and bytes. These IOs are counted by the physical leaves.
4503 if (cvd
->vdev_ops
== &vdev_draid_spare_ops
)
4506 for (int t
= 0; t
< VS_ZIO_TYPES
; t
++) {
4507 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
4508 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
4511 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
4515 * Get extended stats
4518 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
4523 for (t
= 0; t
< ZIO_TYPES
; t
++) {
4524 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
4525 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
4527 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
4528 vsx
->vsx_total_histo
[t
][b
] +=
4529 cvsx
->vsx_total_histo
[t
][b
];
4533 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
4534 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
4535 vsx
->vsx_queue_histo
[t
][b
] +=
4536 cvsx
->vsx_queue_histo
[t
][b
];
4538 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
4539 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
4541 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
4542 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
4544 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
4545 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
4551 vdev_is_spacemap_addressable(vdev_t
*vd
)
4553 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_SPACEMAP_V2
))
4557 * If double-word space map entries are not enabled we assume
4558 * 47 bits of the space map entry are dedicated to the entry's
4559 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4560 * to calculate the maximum address that can be described by a
4561 * space map entry for the given device.
4563 uint64_t shift
= vd
->vdev_ashift
+ SM_OFFSET_BITS
;
4565 if (shift
>= 63) /* detect potential overflow */
4568 return (vd
->vdev_asize
< (1ULL << shift
));
4572 * Get statistics for the given vdev.
4575 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4579 * If we're getting stats on the root vdev, aggregate the I/O counts
4580 * over all top-level vdevs (i.e. the direct children of the root).
4582 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4584 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
4585 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
4588 memset(vsx
, 0, sizeof (*vsx
));
4590 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4591 vdev_t
*cvd
= vd
->vdev_child
[c
];
4592 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
4593 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
4595 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
4597 vdev_get_child_stat(cvd
, vs
, cvs
);
4599 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
4603 * We're a leaf. Just copy our ZIO active queue stats in. The
4604 * other leaf stats are updated in vdev_stat_update().
4609 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
4611 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
4612 vsx
->vsx_active_queue
[t
] =
4613 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
4614 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
4615 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
4621 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4623 vdev_t
*tvd
= vd
->vdev_top
;
4624 mutex_enter(&vd
->vdev_stat_lock
);
4626 memcpy(vs
, &vd
->vdev_stat
, sizeof (*vs
));
4627 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
4628 vs
->vs_state
= vd
->vdev_state
;
4629 vs
->vs_rsize
= vdev_get_min_asize(vd
);
4631 if (vd
->vdev_ops
->vdev_op_leaf
) {
4632 vs
->vs_pspace
= vd
->vdev_psize
;
4633 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
4634 VDEV_LABEL_END_SIZE
;
4636 * Report initializing progress. Since we don't
4637 * have the initializing locks held, this is only
4638 * an estimate (although a fairly accurate one).
4640 vs
->vs_initialize_bytes_done
=
4641 vd
->vdev_initialize_bytes_done
;
4642 vs
->vs_initialize_bytes_est
=
4643 vd
->vdev_initialize_bytes_est
;
4644 vs
->vs_initialize_state
= vd
->vdev_initialize_state
;
4645 vs
->vs_initialize_action_time
=
4646 vd
->vdev_initialize_action_time
;
4649 * Report manual TRIM progress. Since we don't have
4650 * the manual TRIM locks held, this is only an
4651 * estimate (although fairly accurate one).
4653 vs
->vs_trim_notsup
= !vd
->vdev_has_trim
;
4654 vs
->vs_trim_bytes_done
= vd
->vdev_trim_bytes_done
;
4655 vs
->vs_trim_bytes_est
= vd
->vdev_trim_bytes_est
;
4656 vs
->vs_trim_state
= vd
->vdev_trim_state
;
4657 vs
->vs_trim_action_time
= vd
->vdev_trim_action_time
;
4659 /* Set when there is a deferred resilver. */
4660 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
4664 * Report expandable space on top-level, non-auxiliary devices
4665 * only. The expandable space is reported in terms of metaslab
4666 * sized units since that determines how much space the pool
4669 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
4670 vs
->vs_esize
= P2ALIGN(
4671 vd
->vdev_max_asize
- vd
->vdev_asize
,
4672 1ULL << tvd
->vdev_ms_shift
);
4675 vs
->vs_configured_ashift
= vd
->vdev_top
!= NULL
4676 ? vd
->vdev_top
->vdev_ashift
: vd
->vdev_ashift
;
4677 vs
->vs_logical_ashift
= vd
->vdev_logical_ashift
;
4678 if (vd
->vdev_physical_ashift
<= ASHIFT_MAX
)
4679 vs
->vs_physical_ashift
= vd
->vdev_physical_ashift
;
4681 vs
->vs_physical_ashift
= 0;
4684 * Report fragmentation and rebuild progress for top-level,
4685 * non-auxiliary, concrete devices.
4687 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
4688 vdev_is_concrete(vd
)) {
4690 * The vdev fragmentation rating doesn't take into
4691 * account the embedded slog metaslab (vdev_log_mg).
4692 * Since it's only one metaslab, it would have a tiny
4693 * impact on the overall fragmentation.
4695 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
4696 vd
->vdev_mg
->mg_fragmentation
: 0;
4698 vs
->vs_noalloc
= MAX(vd
->vdev_noalloc
,
4699 tvd
? tvd
->vdev_noalloc
: 0);
4702 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
4703 mutex_exit(&vd
->vdev_stat_lock
);
4707 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
4709 return (vdev_get_stats_ex(vd
, vs
, NULL
));
4713 vdev_clear_stats(vdev_t
*vd
)
4715 mutex_enter(&vd
->vdev_stat_lock
);
4716 vd
->vdev_stat
.vs_space
= 0;
4717 vd
->vdev_stat
.vs_dspace
= 0;
4718 vd
->vdev_stat
.vs_alloc
= 0;
4719 mutex_exit(&vd
->vdev_stat_lock
);
4723 vdev_scan_stat_init(vdev_t
*vd
)
4725 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4727 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4728 vdev_scan_stat_init(vd
->vdev_child
[c
]);
4730 mutex_enter(&vd
->vdev_stat_lock
);
4731 vs
->vs_scan_processed
= 0;
4732 mutex_exit(&vd
->vdev_stat_lock
);
4736 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
4738 spa_t
*spa
= zio
->io_spa
;
4739 vdev_t
*rvd
= spa
->spa_root_vdev
;
4740 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
4742 uint64_t txg
= zio
->io_txg
;
4743 /* Suppress ASAN false positive */
4744 #ifdef __SANITIZE_ADDRESS__
4745 vdev_stat_t
*vs
= vd
? &vd
->vdev_stat
: NULL
;
4746 vdev_stat_ex_t
*vsx
= vd
? &vd
->vdev_stat_ex
: NULL
;
4748 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4749 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
4751 zio_type_t type
= zio
->io_type
;
4752 int flags
= zio
->io_flags
;
4755 * If this i/o is a gang leader, it didn't do any actual work.
4757 if (zio
->io_gang_tree
)
4760 if (zio
->io_error
== 0) {
4762 * If this is a root i/o, don't count it -- we've already
4763 * counted the top-level vdevs, and vdev_get_stats() will
4764 * aggregate them when asked. This reduces contention on
4765 * the root vdev_stat_lock and implicitly handles blocks
4766 * that compress away to holes, for which there is no i/o.
4767 * (Holes never create vdev children, so all the counters
4768 * remain zero, which is what we want.)
4770 * Note: this only applies to successful i/o (io_error == 0)
4771 * because unlike i/o counts, errors are not additive.
4772 * When reading a ditto block, for example, failure of
4773 * one top-level vdev does not imply a root-level error.
4778 ASSERT(vd
== zio
->io_vd
);
4780 if (flags
& ZIO_FLAG_IO_BYPASS
)
4783 mutex_enter(&vd
->vdev_stat_lock
);
4785 if (flags
& ZIO_FLAG_IO_REPAIR
) {
4787 * Repair is the result of a resilver issued by the
4788 * scan thread (spa_sync).
4790 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4791 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
4792 dsl_scan_phys_t
*scn_phys
= &scn
->scn_phys
;
4793 uint64_t *processed
= &scn_phys
->scn_processed
;
4795 if (vd
->vdev_ops
->vdev_op_leaf
)
4796 atomic_add_64(processed
, psize
);
4797 vs
->vs_scan_processed
+= psize
;
4801 * Repair is the result of a rebuild issued by the
4802 * rebuild thread (vdev_rebuild_thread). To avoid
4803 * double counting repaired bytes the virtual dRAID
4804 * spare vdev is excluded from the processed bytes.
4806 if (zio
->io_priority
== ZIO_PRIORITY_REBUILD
) {
4807 vdev_t
*tvd
= vd
->vdev_top
;
4808 vdev_rebuild_t
*vr
= &tvd
->vdev_rebuild_config
;
4809 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
4810 uint64_t *rebuilt
= &vrp
->vrp_bytes_rebuilt
;
4812 if (vd
->vdev_ops
->vdev_op_leaf
&&
4813 vd
->vdev_ops
!= &vdev_draid_spare_ops
) {
4814 atomic_add_64(rebuilt
, psize
);
4816 vs
->vs_rebuild_processed
+= psize
;
4819 if (flags
& ZIO_FLAG_SELF_HEAL
)
4820 vs
->vs_self_healed
+= psize
;
4824 * The bytes/ops/histograms are recorded at the leaf level and
4825 * aggregated into the higher level vdevs in vdev_get_stats().
4827 if (vd
->vdev_ops
->vdev_op_leaf
&&
4828 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
4829 zio_type_t vs_type
= type
;
4830 zio_priority_t priority
= zio
->io_priority
;
4833 * TRIM ops and bytes are reported to user space as
4834 * ZIO_TYPE_IOCTL. This is done to preserve the
4835 * vdev_stat_t structure layout for user space.
4837 if (type
== ZIO_TYPE_TRIM
)
4838 vs_type
= ZIO_TYPE_IOCTL
;
4841 * Solely for the purposes of 'zpool iostat -lqrw'
4842 * reporting use the priority to categorize the IO.
4843 * Only the following are reported to user space:
4845 * ZIO_PRIORITY_SYNC_READ,
4846 * ZIO_PRIORITY_SYNC_WRITE,
4847 * ZIO_PRIORITY_ASYNC_READ,
4848 * ZIO_PRIORITY_ASYNC_WRITE,
4849 * ZIO_PRIORITY_SCRUB,
4850 * ZIO_PRIORITY_TRIM,
4851 * ZIO_PRIORITY_REBUILD.
4853 if (priority
== ZIO_PRIORITY_INITIALIZING
) {
4854 ASSERT3U(type
, ==, ZIO_TYPE_WRITE
);
4855 priority
= ZIO_PRIORITY_ASYNC_WRITE
;
4856 } else if (priority
== ZIO_PRIORITY_REMOVAL
) {
4857 priority
= ((type
== ZIO_TYPE_WRITE
) ?
4858 ZIO_PRIORITY_ASYNC_WRITE
:
4859 ZIO_PRIORITY_ASYNC_READ
);
4862 vs
->vs_ops
[vs_type
]++;
4863 vs
->vs_bytes
[vs_type
] += psize
;
4865 if (flags
& ZIO_FLAG_DELEGATED
) {
4866 vsx
->vsx_agg_histo
[priority
]
4867 [RQ_HISTO(zio
->io_size
)]++;
4869 vsx
->vsx_ind_histo
[priority
]
4870 [RQ_HISTO(zio
->io_size
)]++;
4873 if (zio
->io_delta
&& zio
->io_delay
) {
4874 vsx
->vsx_queue_histo
[priority
]
4875 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
4876 vsx
->vsx_disk_histo
[type
]
4877 [L_HISTO(zio
->io_delay
)]++;
4878 vsx
->vsx_total_histo
[type
]
4879 [L_HISTO(zio
->io_delta
)]++;
4883 mutex_exit(&vd
->vdev_stat_lock
);
4887 if (flags
& ZIO_FLAG_SPECULATIVE
)
4891 * If this is an I/O error that is going to be retried, then ignore the
4892 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4893 * hard errors, when in reality they can happen for any number of
4894 * innocuous reasons (bus resets, MPxIO link failure, etc).
4896 if (zio
->io_error
== EIO
&&
4897 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4901 * Intent logs writes won't propagate their error to the root
4902 * I/O so don't mark these types of failures as pool-level
4905 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4908 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4909 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4910 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4911 spa
->spa_claiming
)) {
4913 * This is either a normal write (not a repair), or it's
4914 * a repair induced by the scrub thread, or it's a repair
4915 * made by zil_claim() during spa_load() in the first txg.
4916 * In the normal case, we commit the DTL change in the same
4917 * txg as the block was born. In the scrub-induced repair
4918 * case, we know that scrubs run in first-pass syncing context,
4919 * so we commit the DTL change in spa_syncing_txg(spa).
4920 * In the zil_claim() case, we commit in spa_first_txg(spa).
4922 * We currently do not make DTL entries for failed spontaneous
4923 * self-healing writes triggered by normal (non-scrubbing)
4924 * reads, because we have no transactional context in which to
4925 * do so -- and it's not clear that it'd be desirable anyway.
4927 if (vd
->vdev_ops
->vdev_op_leaf
) {
4928 uint64_t commit_txg
= txg
;
4929 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4930 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4931 ASSERT(spa_sync_pass(spa
) == 1);
4932 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4933 commit_txg
= spa_syncing_txg(spa
);
4934 } else if (spa
->spa_claiming
) {
4935 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4936 commit_txg
= spa_first_txg(spa
);
4938 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4939 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4941 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4942 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4943 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4946 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4951 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4953 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4954 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4956 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4960 * Update the in-core space usage stats for this vdev, its metaslab class,
4961 * and the root vdev.
4964 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4965 int64_t space_delta
)
4968 int64_t dspace_delta
;
4969 spa_t
*spa
= vd
->vdev_spa
;
4970 vdev_t
*rvd
= spa
->spa_root_vdev
;
4972 ASSERT(vd
== vd
->vdev_top
);
4975 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4976 * factor. We must calculate this here and not at the root vdev
4977 * because the root vdev's psize-to-asize is simply the max of its
4978 * children's, thus not accurate enough for us.
4980 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4982 mutex_enter(&vd
->vdev_stat_lock
);
4983 /* ensure we won't underflow */
4984 if (alloc_delta
< 0) {
4985 ASSERT3U(vd
->vdev_stat
.vs_alloc
, >=, -alloc_delta
);
4988 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4989 vd
->vdev_stat
.vs_space
+= space_delta
;
4990 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4991 mutex_exit(&vd
->vdev_stat_lock
);
4993 /* every class but log contributes to root space stats */
4994 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4995 ASSERT(!vd
->vdev_isl2cache
);
4996 mutex_enter(&rvd
->vdev_stat_lock
);
4997 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4998 rvd
->vdev_stat
.vs_space
+= space_delta
;
4999 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
5000 mutex_exit(&rvd
->vdev_stat_lock
);
5002 /* Note: metaslab_class_space_update moved to metaslab_space_update */
5006 * Mark a top-level vdev's config as dirty, placing it on the dirty list
5007 * so that it will be written out next time the vdev configuration is synced.
5008 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
5011 vdev_config_dirty(vdev_t
*vd
)
5013 spa_t
*spa
= vd
->vdev_spa
;
5014 vdev_t
*rvd
= spa
->spa_root_vdev
;
5017 ASSERT(spa_writeable(spa
));
5020 * If this is an aux vdev (as with l2cache and spare devices), then we
5021 * update the vdev config manually and set the sync flag.
5023 if (vd
->vdev_aux
!= NULL
) {
5024 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
5028 for (c
= 0; c
< sav
->sav_count
; c
++) {
5029 if (sav
->sav_vdevs
[c
] == vd
)
5033 if (c
== sav
->sav_count
) {
5035 * We're being removed. There's nothing more to do.
5037 ASSERT(sav
->sav_sync
== B_TRUE
);
5041 sav
->sav_sync
= B_TRUE
;
5043 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
5044 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
5045 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
5046 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
5052 * Setting the nvlist in the middle if the array is a little
5053 * sketchy, but it will work.
5055 nvlist_free(aux
[c
]);
5056 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
5062 * The dirty list is protected by the SCL_CONFIG lock. The caller
5063 * must either hold SCL_CONFIG as writer, or must be the sync thread
5064 * (which holds SCL_CONFIG as reader). There's only one sync thread,
5065 * so this is sufficient to ensure mutual exclusion.
5067 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5068 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5069 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5072 for (c
= 0; c
< rvd
->vdev_children
; c
++)
5073 vdev_config_dirty(rvd
->vdev_child
[c
]);
5075 ASSERT(vd
== vd
->vdev_top
);
5077 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
5078 vdev_is_concrete(vd
)) {
5079 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
5085 vdev_config_clean(vdev_t
*vd
)
5087 spa_t
*spa
= vd
->vdev_spa
;
5089 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5090 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5091 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5093 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
5094 list_remove(&spa
->spa_config_dirty_list
, vd
);
5098 * Mark a top-level vdev's state as dirty, so that the next pass of
5099 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
5100 * the state changes from larger config changes because they require
5101 * much less locking, and are often needed for administrative actions.
5104 vdev_state_dirty(vdev_t
*vd
)
5106 spa_t
*spa
= vd
->vdev_spa
;
5108 ASSERT(spa_writeable(spa
));
5109 ASSERT(vd
== vd
->vdev_top
);
5112 * The state list is protected by the SCL_STATE lock. The caller
5113 * must either hold SCL_STATE as writer, or must be the sync thread
5114 * (which holds SCL_STATE as reader). There's only one sync thread,
5115 * so this is sufficient to ensure mutual exclusion.
5117 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5118 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5119 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5121 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
5122 vdev_is_concrete(vd
))
5123 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
5127 vdev_state_clean(vdev_t
*vd
)
5129 spa_t
*spa
= vd
->vdev_spa
;
5131 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5132 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5133 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5135 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
5136 list_remove(&spa
->spa_state_dirty_list
, vd
);
5140 * Propagate vdev state up from children to parent.
5143 vdev_propagate_state(vdev_t
*vd
)
5145 spa_t
*spa
= vd
->vdev_spa
;
5146 vdev_t
*rvd
= spa
->spa_root_vdev
;
5147 int degraded
= 0, faulted
= 0;
5151 if (vd
->vdev_children
> 0) {
5152 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5153 child
= vd
->vdev_child
[c
];
5156 * Don't factor holes or indirect vdevs into the
5159 if (!vdev_is_concrete(child
))
5162 if (!vdev_readable(child
) ||
5163 (!vdev_writeable(child
) && spa_writeable(spa
))) {
5165 * Root special: if there is a top-level log
5166 * device, treat the root vdev as if it were
5169 if (child
->vdev_islog
&& vd
== rvd
)
5173 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
5177 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
5181 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
5184 * Root special: if there is a top-level vdev that cannot be
5185 * opened due to corrupted metadata, then propagate the root
5186 * vdev's aux state as 'corrupt' rather than 'insufficient
5189 if (corrupted
&& vd
== rvd
&&
5190 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
5191 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
5192 VDEV_AUX_CORRUPT_DATA
);
5195 if (vd
->vdev_parent
)
5196 vdev_propagate_state(vd
->vdev_parent
);
5200 * Set a vdev's state. If this is during an open, we don't update the parent
5201 * state, because we're in the process of opening children depth-first.
5202 * Otherwise, we propagate the change to the parent.
5204 * If this routine places a device in a faulted state, an appropriate ereport is
5208 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
5210 uint64_t save_state
;
5211 spa_t
*spa
= vd
->vdev_spa
;
5213 if (state
== vd
->vdev_state
) {
5215 * Since vdev_offline() code path is already in an offline
5216 * state we can miss a statechange event to OFFLINE. Check
5217 * the previous state to catch this condition.
5219 if (vd
->vdev_ops
->vdev_op_leaf
&&
5220 (state
== VDEV_STATE_OFFLINE
) &&
5221 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
5222 /* post an offline state change */
5223 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
5225 vd
->vdev_stat
.vs_aux
= aux
;
5229 save_state
= vd
->vdev_state
;
5231 vd
->vdev_state
= state
;
5232 vd
->vdev_stat
.vs_aux
= aux
;
5235 * If we are setting the vdev state to anything but an open state, then
5236 * always close the underlying device unless the device has requested
5237 * a delayed close (i.e. we're about to remove or fault the device).
5238 * Otherwise, we keep accessible but invalid devices open forever.
5239 * We don't call vdev_close() itself, because that implies some extra
5240 * checks (offline, etc) that we don't want here. This is limited to
5241 * leaf devices, because otherwise closing the device will affect other
5244 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
5245 vd
->vdev_ops
->vdev_op_leaf
)
5246 vd
->vdev_ops
->vdev_op_close(vd
);
5248 if (vd
->vdev_removed
&&
5249 state
== VDEV_STATE_CANT_OPEN
&&
5250 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
5252 * If the previous state is set to VDEV_STATE_REMOVED, then this
5253 * device was previously marked removed and someone attempted to
5254 * reopen it. If this failed due to a nonexistent device, then
5255 * keep the device in the REMOVED state. We also let this be if
5256 * it is one of our special test online cases, which is only
5257 * attempting to online the device and shouldn't generate an FMA
5260 vd
->vdev_state
= VDEV_STATE_REMOVED
;
5261 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
5262 } else if (state
== VDEV_STATE_REMOVED
) {
5263 vd
->vdev_removed
= B_TRUE
;
5264 } else if (state
== VDEV_STATE_CANT_OPEN
) {
5266 * If we fail to open a vdev during an import or recovery, we
5267 * mark it as "not available", which signifies that it was
5268 * never there to begin with. Failure to open such a device
5269 * is not considered an error.
5271 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
5272 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
5273 vd
->vdev_ops
->vdev_op_leaf
)
5274 vd
->vdev_not_present
= 1;
5277 * Post the appropriate ereport. If the 'prevstate' field is
5278 * set to something other than VDEV_STATE_UNKNOWN, it indicates
5279 * that this is part of a vdev_reopen(). In this case, we don't
5280 * want to post the ereport if the device was already in the
5281 * CANT_OPEN state beforehand.
5283 * If the 'checkremove' flag is set, then this is an attempt to
5284 * online the device in response to an insertion event. If we
5285 * hit this case, then we have detected an insertion event for a
5286 * faulted or offline device that wasn't in the removed state.
5287 * In this scenario, we don't post an ereport because we are
5288 * about to replace the device, or attempt an online with
5289 * vdev_forcefault, which will generate the fault for us.
5291 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
5292 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
5293 vd
!= spa
->spa_root_vdev
) {
5297 case VDEV_AUX_OPEN_FAILED
:
5298 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
5300 case VDEV_AUX_CORRUPT_DATA
:
5301 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
5303 case VDEV_AUX_NO_REPLICAS
:
5304 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
5306 case VDEV_AUX_BAD_GUID_SUM
:
5307 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
5309 case VDEV_AUX_TOO_SMALL
:
5310 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
5312 case VDEV_AUX_BAD_LABEL
:
5313 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
5315 case VDEV_AUX_BAD_ASHIFT
:
5316 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
5319 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
5322 (void) zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
5326 /* Erase any notion of persistent removed state */
5327 vd
->vdev_removed
= B_FALSE
;
5329 vd
->vdev_removed
= B_FALSE
;
5333 * Notify ZED of any significant state-change on a leaf vdev.
5336 if (vd
->vdev_ops
->vdev_op_leaf
) {
5337 /* preserve original state from a vdev_reopen() */
5338 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
5339 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
5340 (save_state
<= VDEV_STATE_CLOSED
))
5341 save_state
= vd
->vdev_prevstate
;
5343 /* filter out state change due to initial vdev_open */
5344 if (save_state
> VDEV_STATE_CLOSED
)
5345 zfs_post_state_change(spa
, vd
, save_state
);
5348 if (!isopen
&& vd
->vdev_parent
)
5349 vdev_propagate_state(vd
->vdev_parent
);
5353 vdev_children_are_offline(vdev_t
*vd
)
5355 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
5357 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
5358 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
5366 * Check the vdev configuration to ensure that it's capable of supporting
5367 * a root pool. We do not support partial configuration.
5370 vdev_is_bootable(vdev_t
*vd
)
5372 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5373 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
5375 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
5379 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5380 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
5387 vdev_is_concrete(vdev_t
*vd
)
5389 vdev_ops_t
*ops
= vd
->vdev_ops
;
5390 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
5391 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
5399 * Determine if a log device has valid content. If the vdev was
5400 * removed or faulted in the MOS config then we know that
5401 * the content on the log device has already been written to the pool.
5404 vdev_log_state_valid(vdev_t
*vd
)
5406 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
5410 for (int c
= 0; c
< vd
->vdev_children
; c
++)
5411 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
5418 * Expand a vdev if possible.
5421 vdev_expand(vdev_t
*vd
, uint64_t txg
)
5423 ASSERT(vd
->vdev_top
== vd
);
5424 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5425 ASSERT(vdev_is_concrete(vd
));
5427 vdev_set_deflate_ratio(vd
);
5429 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
5430 vdev_is_concrete(vd
)) {
5431 vdev_metaslab_group_create(vd
);
5432 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
5433 vdev_config_dirty(vd
);
5441 vdev_split(vdev_t
*vd
)
5443 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
5445 VERIFY3U(pvd
->vdev_children
, >, 1);
5447 vdev_remove_child(pvd
, vd
);
5448 vdev_compact_children(pvd
);
5450 ASSERT3P(pvd
->vdev_child
, !=, NULL
);
5452 cvd
= pvd
->vdev_child
[0];
5453 if (pvd
->vdev_children
== 1) {
5454 vdev_remove_parent(cvd
);
5455 cvd
->vdev_splitting
= B_TRUE
;
5457 vdev_propagate_state(cvd
);
5461 vdev_deadman(vdev_t
*vd
, const char *tag
)
5463 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5464 vdev_t
*cvd
= vd
->vdev_child
[c
];
5466 vdev_deadman(cvd
, tag
);
5469 if (vd
->vdev_ops
->vdev_op_leaf
) {
5470 vdev_queue_t
*vq
= &vd
->vdev_queue
;
5472 mutex_enter(&vq
->vq_lock
);
5473 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
5474 spa_t
*spa
= vd
->vdev_spa
;
5478 zfs_dbgmsg("slow vdev: %s has %lu active IOs",
5479 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
5482 * Look at the head of all the pending queues,
5483 * if any I/O has been outstanding for longer than
5484 * the spa_deadman_synctime invoke the deadman logic.
5486 fio
= avl_first(&vq
->vq_active_tree
);
5487 delta
= gethrtime() - fio
->io_timestamp
;
5488 if (delta
> spa_deadman_synctime(spa
))
5489 zio_deadman(fio
, tag
);
5491 mutex_exit(&vq
->vq_lock
);
5496 vdev_defer_resilver(vdev_t
*vd
)
5498 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
5500 vd
->vdev_resilver_deferred
= B_TRUE
;
5501 vd
->vdev_spa
->spa_resilver_deferred
= B_TRUE
;
5505 * Clears the resilver deferred flag on all leaf devs under vd. Returns
5506 * B_TRUE if we have devices that need to be resilvered and are available to
5507 * accept resilver I/Os.
5510 vdev_clear_resilver_deferred(vdev_t
*vd
, dmu_tx_t
*tx
)
5512 boolean_t resilver_needed
= B_FALSE
;
5513 spa_t
*spa
= vd
->vdev_spa
;
5515 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5516 vdev_t
*cvd
= vd
->vdev_child
[c
];
5517 resilver_needed
|= vdev_clear_resilver_deferred(cvd
, tx
);
5520 if (vd
== spa
->spa_root_vdev
&&
5521 spa_feature_is_active(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
5522 spa_feature_decr(spa
, SPA_FEATURE_RESILVER_DEFER
, tx
);
5523 vdev_config_dirty(vd
);
5524 spa
->spa_resilver_deferred
= B_FALSE
;
5525 return (resilver_needed
);
5528 if (!vdev_is_concrete(vd
) || vd
->vdev_aux
||
5529 !vd
->vdev_ops
->vdev_op_leaf
)
5530 return (resilver_needed
);
5532 vd
->vdev_resilver_deferred
= B_FALSE
;
5534 return (!vdev_is_dead(vd
) && !vd
->vdev_offline
&&
5535 vdev_resilver_needed(vd
, NULL
, NULL
));
5539 vdev_xlate_is_empty(range_seg64_t
*rs
)
5541 return (rs
->rs_start
== rs
->rs_end
);
5545 * Translate a logical range to the first contiguous physical range for the
5546 * specified vdev_t. This function is initially called with a leaf vdev and
5547 * will walk each parent vdev until it reaches a top-level vdev. Once the
5548 * top-level is reached the physical range is initialized and the recursive
5549 * function begins to unwind. As it unwinds it calls the parent's vdev
5550 * specific translation function to do the real conversion.
5553 vdev_xlate(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5554 range_seg64_t
*physical_rs
, range_seg64_t
*remain_rs
)
5557 * Walk up the vdev tree
5559 if (vd
!= vd
->vdev_top
) {
5560 vdev_xlate(vd
->vdev_parent
, logical_rs
, physical_rs
,
5564 * We've reached the top-level vdev, initialize the physical
5565 * range to the logical range and set an empty remaining
5566 * range then start to unwind.
5568 physical_rs
->rs_start
= logical_rs
->rs_start
;
5569 physical_rs
->rs_end
= logical_rs
->rs_end
;
5571 remain_rs
->rs_start
= logical_rs
->rs_start
;
5572 remain_rs
->rs_end
= logical_rs
->rs_start
;
5577 vdev_t
*pvd
= vd
->vdev_parent
;
5578 ASSERT3P(pvd
, !=, NULL
);
5579 ASSERT3P(pvd
->vdev_ops
->vdev_op_xlate
, !=, NULL
);
5582 * As this recursive function unwinds, translate the logical
5583 * range into its physical and any remaining components by calling
5584 * the vdev specific translate function.
5586 range_seg64_t intermediate
= { 0 };
5587 pvd
->vdev_ops
->vdev_op_xlate(vd
, physical_rs
, &intermediate
, remain_rs
);
5589 physical_rs
->rs_start
= intermediate
.rs_start
;
5590 physical_rs
->rs_end
= intermediate
.rs_end
;
5594 vdev_xlate_walk(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5595 vdev_xlate_func_t
*func
, void *arg
)
5597 range_seg64_t iter_rs
= *logical_rs
;
5598 range_seg64_t physical_rs
;
5599 range_seg64_t remain_rs
;
5601 while (!vdev_xlate_is_empty(&iter_rs
)) {
5603 vdev_xlate(vd
, &iter_rs
, &physical_rs
, &remain_rs
);
5606 * With raidz and dRAID, it's possible that the logical range
5607 * does not live on this leaf vdev. Only when there is a non-
5608 * zero physical size call the provided function.
5610 if (!vdev_xlate_is_empty(&physical_rs
))
5611 func(arg
, &physical_rs
);
5613 iter_rs
= remain_rs
;
5618 vdev_name(vdev_t
*vd
, char *buf
, int buflen
)
5620 if (vd
->vdev_path
== NULL
) {
5621 if (strcmp(vd
->vdev_ops
->vdev_op_type
, "root") == 0) {
5622 strlcpy(buf
, vd
->vdev_spa
->spa_name
, buflen
);
5623 } else if (!vd
->vdev_ops
->vdev_op_leaf
) {
5624 snprintf(buf
, buflen
, "%s-%llu",
5625 vd
->vdev_ops
->vdev_op_type
,
5626 (u_longlong_t
)vd
->vdev_id
);
5629 strlcpy(buf
, vd
->vdev_path
, buflen
);
5635 * Look at the vdev tree and determine whether any devices are currently being
5639 vdev_replace_in_progress(vdev_t
*vdev
)
5641 ASSERT(spa_config_held(vdev
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
5643 if (vdev
->vdev_ops
== &vdev_replacing_ops
)
5647 * A 'spare' vdev indicates that we have a replace in progress, unless
5648 * it has exactly two children, and the second, the hot spare, has
5649 * finished being resilvered.
5651 if (vdev
->vdev_ops
== &vdev_spare_ops
&& (vdev
->vdev_children
> 2 ||
5652 !vdev_dtl_empty(vdev
->vdev_child
[1], DTL_MISSING
)))
5655 for (int i
= 0; i
< vdev
->vdev_children
; i
++) {
5656 if (vdev_replace_in_progress(vdev
->vdev_child
[i
]))
5664 * Add a (source=src, propname=propval) list to an nvlist.
5667 vdev_prop_add_list(nvlist_t
*nvl
, const char *propname
, const char *strval
,
5668 uint64_t intval
, zprop_source_t src
)
5672 propval
= fnvlist_alloc();
5673 fnvlist_add_uint64(propval
, ZPROP_SOURCE
, src
);
5676 fnvlist_add_string(propval
, ZPROP_VALUE
, strval
);
5678 fnvlist_add_uint64(propval
, ZPROP_VALUE
, intval
);
5680 fnvlist_add_nvlist(nvl
, propname
, propval
);
5681 nvlist_free(propval
);
5685 vdev_props_set_sync(void *arg
, dmu_tx_t
*tx
)
5688 nvlist_t
*nvp
= arg
;
5689 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5690 objset_t
*mos
= spa
->spa_meta_objset
;
5691 nvpair_t
*elem
= NULL
;
5695 vdev_guid
= fnvlist_lookup_uint64(nvp
, ZPOOL_VDEV_PROPS_SET_VDEV
);
5696 nvprops
= fnvlist_lookup_nvlist(nvp
, ZPOOL_VDEV_PROPS_SET_PROPS
);
5697 vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
);
5699 /* this vdev could get removed while waiting for this sync task */
5703 mutex_enter(&spa
->spa_props_lock
);
5705 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5706 uint64_t intval
, objid
= 0;
5709 const char *propname
= nvpair_name(elem
);
5710 zprop_type_t proptype
;
5713 * Set vdev property values in the vdev props mos object.
5715 if (vd
->vdev_root_zap
!= 0) {
5716 objid
= vd
->vdev_root_zap
;
5717 } else if (vd
->vdev_top_zap
!= 0) {
5718 objid
= vd
->vdev_top_zap
;
5719 } else if (vd
->vdev_leaf_zap
!= 0) {
5720 objid
= vd
->vdev_leaf_zap
;
5723 * XXX: implement vdev_props_set_check()
5725 panic("vdev not root/top/leaf");
5728 switch (prop
= vdev_name_to_prop(propname
)) {
5729 case VDEV_PROP_USERPROP
:
5730 if (vdev_prop_user(propname
)) {
5731 strval
= fnvpair_value_string(elem
);
5732 if (strlen(strval
) == 0) {
5733 /* remove the property if value == "" */
5734 (void) zap_remove(mos
, objid
, propname
,
5737 VERIFY0(zap_update(mos
, objid
, propname
,
5738 1, strlen(strval
) + 1, strval
, tx
));
5740 spa_history_log_internal(spa
, "vdev set", tx
,
5741 "vdev_guid=%llu: %s=%s",
5742 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5747 /* normalize the property name */
5748 propname
= vdev_prop_to_name(prop
);
5749 proptype
= vdev_prop_get_type(prop
);
5751 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5752 ASSERT(proptype
== PROP_TYPE_STRING
);
5753 strval
= fnvpair_value_string(elem
);
5754 VERIFY0(zap_update(mos
, objid
, propname
,
5755 1, strlen(strval
) + 1, strval
, tx
));
5756 spa_history_log_internal(spa
, "vdev set", tx
,
5757 "vdev_guid=%llu: %s=%s",
5758 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5760 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
5761 intval
= fnvpair_value_uint64(elem
);
5763 if (proptype
== PROP_TYPE_INDEX
) {
5765 VERIFY0(vdev_prop_index_to_string(
5766 prop
, intval
, &unused
));
5768 VERIFY0(zap_update(mos
, objid
, propname
,
5769 sizeof (uint64_t), 1, &intval
, tx
));
5770 spa_history_log_internal(spa
, "vdev set", tx
,
5771 "vdev_guid=%llu: %s=%lld",
5772 (u_longlong_t
)vdev_guid
,
5773 nvpair_name(elem
), (longlong_t
)intval
);
5775 panic("invalid vdev property type %u",
5782 mutex_exit(&spa
->spa_props_lock
);
5786 vdev_prop_set(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5788 spa_t
*spa
= vd
->vdev_spa
;
5789 nvpair_t
*elem
= NULL
;
5796 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_SET_VDEV
,
5798 return (SET_ERROR(EINVAL
));
5800 if (nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_SET_PROPS
,
5802 return (SET_ERROR(EINVAL
));
5804 if ((vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
)) == NULL
)
5805 return (SET_ERROR(EINVAL
));
5807 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5808 const char *propname
= nvpair_name(elem
);
5809 vdev_prop_t prop
= vdev_name_to_prop(propname
);
5810 uint64_t intval
= 0;
5811 const char *strval
= NULL
;
5813 if (prop
== VDEV_PROP_USERPROP
&& !vdev_prop_user(propname
)) {
5818 if (vdev_prop_readonly(prop
)) {
5823 /* Special Processing */
5825 case VDEV_PROP_PATH
:
5826 if (vd
->vdev_path
== NULL
) {
5830 if (nvpair_value_string(elem
, &strval
) != 0) {
5834 /* New path must start with /dev/ */
5835 if (strncmp(strval
, "/dev/", 5)) {
5839 error
= spa_vdev_setpath(spa
, vdev_guid
, strval
);
5841 case VDEV_PROP_ALLOCATING
:
5842 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5846 if (intval
!= vd
->vdev_noalloc
)
5849 error
= spa_vdev_noalloc(spa
, vdev_guid
);
5851 error
= spa_vdev_alloc(spa
, vdev_guid
);
5853 case VDEV_PROP_FAILFAST
:
5854 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5858 vd
->vdev_failfast
= intval
& 1;
5860 case VDEV_PROP_CHECKSUM_N
:
5861 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5865 vd
->vdev_checksum_n
= intval
;
5867 case VDEV_PROP_CHECKSUM_T
:
5868 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5872 vd
->vdev_checksum_t
= intval
;
5874 case VDEV_PROP_IO_N
:
5875 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5879 vd
->vdev_io_n
= intval
;
5881 case VDEV_PROP_IO_T
:
5882 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5886 vd
->vdev_io_t
= intval
;
5889 /* Most processing is done in vdev_props_set_sync */
5895 vdev_prop_add_list(outnvl
, propname
, strval
, intval
, 0);
5900 return (dsl_sync_task(spa
->spa_name
, NULL
, vdev_props_set_sync
,
5901 innvl
, 6, ZFS_SPACE_CHECK_EXTRA_RESERVED
));
5905 vdev_prop_get(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5907 spa_t
*spa
= vd
->vdev_spa
;
5908 objset_t
*mos
= spa
->spa_meta_objset
;
5912 nvpair_t
*elem
= NULL
;
5913 nvlist_t
*nvprops
= NULL
;
5914 uint64_t intval
= 0;
5915 char *strval
= NULL
;
5916 const char *propname
= NULL
;
5920 ASSERT(mos
!= NULL
);
5922 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_GET_VDEV
,
5924 return (SET_ERROR(EINVAL
));
5926 nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_GET_PROPS
, &nvprops
);
5928 if (vd
->vdev_root_zap
!= 0) {
5929 objid
= vd
->vdev_root_zap
;
5930 } else if (vd
->vdev_top_zap
!= 0) {
5931 objid
= vd
->vdev_top_zap
;
5932 } else if (vd
->vdev_leaf_zap
!= 0) {
5933 objid
= vd
->vdev_leaf_zap
;
5935 return (SET_ERROR(EINVAL
));
5939 mutex_enter(&spa
->spa_props_lock
);
5941 if (nvprops
!= NULL
) {
5942 char namebuf
[64] = { 0 };
5944 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5947 propname
= nvpair_name(elem
);
5948 prop
= vdev_name_to_prop(propname
);
5949 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
5950 uint64_t integer_size
, num_integers
;
5953 /* Special Read-only Properties */
5954 case VDEV_PROP_NAME
:
5955 strval
= vdev_name(vd
, namebuf
,
5959 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
5962 case VDEV_PROP_CAPACITY
:
5964 intval
= (vd
->vdev_stat
.vs_dspace
== 0) ? 0 :
5965 (vd
->vdev_stat
.vs_alloc
* 100 /
5966 vd
->vdev_stat
.vs_dspace
);
5967 vdev_prop_add_list(outnvl
, propname
, NULL
,
5968 intval
, ZPROP_SRC_NONE
);
5970 case VDEV_PROP_STATE
:
5971 vdev_prop_add_list(outnvl
, propname
, NULL
,
5972 vd
->vdev_state
, ZPROP_SRC_NONE
);
5974 case VDEV_PROP_GUID
:
5975 vdev_prop_add_list(outnvl
, propname
, NULL
,
5976 vd
->vdev_guid
, ZPROP_SRC_NONE
);
5978 case VDEV_PROP_ASIZE
:
5979 vdev_prop_add_list(outnvl
, propname
, NULL
,
5980 vd
->vdev_asize
, ZPROP_SRC_NONE
);
5982 case VDEV_PROP_PSIZE
:
5983 vdev_prop_add_list(outnvl
, propname
, NULL
,
5984 vd
->vdev_psize
, ZPROP_SRC_NONE
);
5986 case VDEV_PROP_ASHIFT
:
5987 vdev_prop_add_list(outnvl
, propname
, NULL
,
5988 vd
->vdev_ashift
, ZPROP_SRC_NONE
);
5990 case VDEV_PROP_SIZE
:
5991 vdev_prop_add_list(outnvl
, propname
, NULL
,
5992 vd
->vdev_stat
.vs_dspace
, ZPROP_SRC_NONE
);
5994 case VDEV_PROP_FREE
:
5995 vdev_prop_add_list(outnvl
, propname
, NULL
,
5996 vd
->vdev_stat
.vs_dspace
-
5997 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
5999 case VDEV_PROP_ALLOCATED
:
6000 vdev_prop_add_list(outnvl
, propname
, NULL
,
6001 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
6003 case VDEV_PROP_EXPANDSZ
:
6004 vdev_prop_add_list(outnvl
, propname
, NULL
,
6005 vd
->vdev_stat
.vs_esize
, ZPROP_SRC_NONE
);
6007 case VDEV_PROP_FRAGMENTATION
:
6008 vdev_prop_add_list(outnvl
, propname
, NULL
,
6009 vd
->vdev_stat
.vs_fragmentation
,
6012 case VDEV_PROP_PARITY
:
6013 vdev_prop_add_list(outnvl
, propname
, NULL
,
6014 vdev_get_nparity(vd
), ZPROP_SRC_NONE
);
6016 case VDEV_PROP_PATH
:
6017 if (vd
->vdev_path
== NULL
)
6019 vdev_prop_add_list(outnvl
, propname
,
6020 vd
->vdev_path
, 0, ZPROP_SRC_NONE
);
6022 case VDEV_PROP_DEVID
:
6023 if (vd
->vdev_devid
== NULL
)
6025 vdev_prop_add_list(outnvl
, propname
,
6026 vd
->vdev_devid
, 0, ZPROP_SRC_NONE
);
6028 case VDEV_PROP_PHYS_PATH
:
6029 if (vd
->vdev_physpath
== NULL
)
6031 vdev_prop_add_list(outnvl
, propname
,
6032 vd
->vdev_physpath
, 0, ZPROP_SRC_NONE
);
6034 case VDEV_PROP_ENC_PATH
:
6035 if (vd
->vdev_enc_sysfs_path
== NULL
)
6037 vdev_prop_add_list(outnvl
, propname
,
6038 vd
->vdev_enc_sysfs_path
, 0, ZPROP_SRC_NONE
);
6041 if (vd
->vdev_fru
== NULL
)
6043 vdev_prop_add_list(outnvl
, propname
,
6044 vd
->vdev_fru
, 0, ZPROP_SRC_NONE
);
6046 case VDEV_PROP_PARENT
:
6047 if (vd
->vdev_parent
!= NULL
) {
6048 strval
= vdev_name(vd
->vdev_parent
,
6049 namebuf
, sizeof (namebuf
));
6050 vdev_prop_add_list(outnvl
, propname
,
6051 strval
, 0, ZPROP_SRC_NONE
);
6054 case VDEV_PROP_CHILDREN
:
6055 if (vd
->vdev_children
> 0)
6056 strval
= kmem_zalloc(ZAP_MAXVALUELEN
,
6058 for (uint64_t i
= 0; i
< vd
->vdev_children
;
6062 vname
= vdev_name(vd
->vdev_child
[i
],
6063 namebuf
, sizeof (namebuf
));
6065 vname
= "(unknown)";
6066 if (strlen(strval
) > 0)
6067 strlcat(strval
, ",",
6069 strlcat(strval
, vname
, ZAP_MAXVALUELEN
);
6071 if (strval
!= NULL
) {
6072 vdev_prop_add_list(outnvl
, propname
,
6073 strval
, 0, ZPROP_SRC_NONE
);
6074 kmem_free(strval
, ZAP_MAXVALUELEN
);
6077 case VDEV_PROP_NUMCHILDREN
:
6078 vdev_prop_add_list(outnvl
, propname
, NULL
,
6079 vd
->vdev_children
, ZPROP_SRC_NONE
);
6081 case VDEV_PROP_READ_ERRORS
:
6082 vdev_prop_add_list(outnvl
, propname
, NULL
,
6083 vd
->vdev_stat
.vs_read_errors
,
6086 case VDEV_PROP_WRITE_ERRORS
:
6087 vdev_prop_add_list(outnvl
, propname
, NULL
,
6088 vd
->vdev_stat
.vs_write_errors
,
6091 case VDEV_PROP_CHECKSUM_ERRORS
:
6092 vdev_prop_add_list(outnvl
, propname
, NULL
,
6093 vd
->vdev_stat
.vs_checksum_errors
,
6096 case VDEV_PROP_INITIALIZE_ERRORS
:
6097 vdev_prop_add_list(outnvl
, propname
, NULL
,
6098 vd
->vdev_stat
.vs_initialize_errors
,
6101 case VDEV_PROP_OPS_NULL
:
6102 vdev_prop_add_list(outnvl
, propname
, NULL
,
6103 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_NULL
],
6106 case VDEV_PROP_OPS_READ
:
6107 vdev_prop_add_list(outnvl
, propname
, NULL
,
6108 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_READ
],
6111 case VDEV_PROP_OPS_WRITE
:
6112 vdev_prop_add_list(outnvl
, propname
, NULL
,
6113 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_WRITE
],
6116 case VDEV_PROP_OPS_FREE
:
6117 vdev_prop_add_list(outnvl
, propname
, NULL
,
6118 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_FREE
],
6121 case VDEV_PROP_OPS_CLAIM
:
6122 vdev_prop_add_list(outnvl
, propname
, NULL
,
6123 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_CLAIM
],
6126 case VDEV_PROP_OPS_TRIM
:
6128 * TRIM ops and bytes are reported to user
6129 * space as ZIO_TYPE_IOCTL. This is done to
6130 * preserve the vdev_stat_t structure layout
6133 vdev_prop_add_list(outnvl
, propname
, NULL
,
6134 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_IOCTL
],
6137 case VDEV_PROP_BYTES_NULL
:
6138 vdev_prop_add_list(outnvl
, propname
, NULL
,
6139 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_NULL
],
6142 case VDEV_PROP_BYTES_READ
:
6143 vdev_prop_add_list(outnvl
, propname
, NULL
,
6144 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_READ
],
6147 case VDEV_PROP_BYTES_WRITE
:
6148 vdev_prop_add_list(outnvl
, propname
, NULL
,
6149 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_WRITE
],
6152 case VDEV_PROP_BYTES_FREE
:
6153 vdev_prop_add_list(outnvl
, propname
, NULL
,
6154 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_FREE
],
6157 case VDEV_PROP_BYTES_CLAIM
:
6158 vdev_prop_add_list(outnvl
, propname
, NULL
,
6159 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_CLAIM
],
6162 case VDEV_PROP_BYTES_TRIM
:
6164 * TRIM ops and bytes are reported to user
6165 * space as ZIO_TYPE_IOCTL. This is done to
6166 * preserve the vdev_stat_t structure layout
6169 vdev_prop_add_list(outnvl
, propname
, NULL
,
6170 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_IOCTL
],
6173 case VDEV_PROP_REMOVING
:
6174 vdev_prop_add_list(outnvl
, propname
, NULL
,
6175 vd
->vdev_removing
, ZPROP_SRC_NONE
);
6177 /* Numeric Properites */
6178 case VDEV_PROP_ALLOCATING
:
6179 /* Leaf vdevs cannot have this property */
6180 if (vd
->vdev_mg
== NULL
&&
6181 vd
->vdev_top
!= NULL
) {
6182 src
= ZPROP_SRC_NONE
;
6183 intval
= ZPROP_BOOLEAN_NA
;
6185 err
= vdev_prop_get_int(vd
, prop
,
6187 if (err
&& err
!= ENOENT
)
6191 vdev_prop_default_numeric(prop
))
6192 src
= ZPROP_SRC_DEFAULT
;
6194 src
= ZPROP_SRC_LOCAL
;
6197 vdev_prop_add_list(outnvl
, propname
, NULL
,
6200 case VDEV_PROP_FAILFAST
:
6201 src
= ZPROP_SRC_LOCAL
;
6204 err
= zap_lookup(mos
, objid
, nvpair_name(elem
),
6205 sizeof (uint64_t), 1, &intval
);
6206 if (err
== ENOENT
) {
6207 intval
= vdev_prop_default_numeric(
6213 if (intval
== vdev_prop_default_numeric(prop
))
6214 src
= ZPROP_SRC_DEFAULT
;
6216 vdev_prop_add_list(outnvl
, propname
, strval
,
6219 case VDEV_PROP_CHECKSUM_N
:
6220 case VDEV_PROP_CHECKSUM_T
:
6221 case VDEV_PROP_IO_N
:
6222 case VDEV_PROP_IO_T
:
6223 err
= vdev_prop_get_int(vd
, prop
, &intval
);
6224 if (err
&& err
!= ENOENT
)
6227 if (intval
== vdev_prop_default_numeric(prop
))
6228 src
= ZPROP_SRC_DEFAULT
;
6230 src
= ZPROP_SRC_LOCAL
;
6232 vdev_prop_add_list(outnvl
, propname
, NULL
,
6235 /* Text Properties */
6236 case VDEV_PROP_COMMENT
:
6237 /* Exists in the ZAP below */
6239 case VDEV_PROP_USERPROP
:
6240 /* User Properites */
6241 src
= ZPROP_SRC_LOCAL
;
6243 err
= zap_length(mos
, objid
, nvpair_name(elem
),
6244 &integer_size
, &num_integers
);
6248 switch (integer_size
) {
6250 /* User properties cannot be integers */
6254 /* string property */
6255 strval
= kmem_alloc(num_integers
,
6257 err
= zap_lookup(mos
, objid
,
6258 nvpair_name(elem
), 1,
6259 num_integers
, strval
);
6265 vdev_prop_add_list(outnvl
, propname
,
6267 kmem_free(strval
, num_integers
);
6280 * Get all properties from the MOS vdev property object.
6284 for (zap_cursor_init(&zc
, mos
, objid
);
6285 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
6286 zap_cursor_advance(&zc
)) {
6289 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
6290 propname
= za
.za_name
;
6292 switch (za
.za_integer_length
) {
6294 /* We do not allow integer user properties */
6295 /* This is likely an internal value */
6298 /* string property */
6299 strval
= kmem_alloc(za
.za_num_integers
,
6301 err
= zap_lookup(mos
, objid
, za
.za_name
, 1,
6302 za
.za_num_integers
, strval
);
6304 kmem_free(strval
, za
.za_num_integers
);
6307 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
6309 kmem_free(strval
, za
.za_num_integers
);
6316 zap_cursor_fini(&zc
);
6319 mutex_exit(&spa
->spa_props_lock
);
6320 if (err
&& err
!= ENOENT
) {
6327 EXPORT_SYMBOL(vdev_fault
);
6328 EXPORT_SYMBOL(vdev_degrade
);
6329 EXPORT_SYMBOL(vdev_online
);
6330 EXPORT_SYMBOL(vdev_offline
);
6331 EXPORT_SYMBOL(vdev_clear
);
6333 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_count
, UINT
, ZMOD_RW
,
6334 "Target number of metaslabs per top-level vdev");
6336 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_shift
, UINT
, ZMOD_RW
,
6337 "Default lower limit for metaslab size");
6339 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, max_ms_shift
, UINT
, ZMOD_RW
,
6340 "Default upper limit for metaslab size");
6342 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, min_ms_count
, UINT
, ZMOD_RW
,
6343 "Minimum number of metaslabs per top-level vdev");
6345 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, ms_count_limit
, UINT
, ZMOD_RW
,
6346 "Practical upper limit of total metaslabs per top-level vdev");
6348 ZFS_MODULE_PARAM(zfs
, zfs_
, slow_io_events_per_second
, UINT
, ZMOD_RW
,
6349 "Rate limit slow IO (delay) events to this many per second");
6352 ZFS_MODULE_PARAM(zfs
, zfs_
, checksum_events_per_second
, UINT
, ZMOD_RW
,
6353 "Rate limit checksum events to this many checksum errors per second "
6354 "(do not set below ZED threshold).");
6357 ZFS_MODULE_PARAM(zfs
, zfs_
, scan_ignore_errors
, INT
, ZMOD_RW
,
6358 "Ignore errors during resilver/scrub");
6360 ZFS_MODULE_PARAM(zfs_vdev
, vdev_
, validate_skip
, INT
, ZMOD_RW
,
6361 "Bypass vdev_validate()");
6363 ZFS_MODULE_PARAM(zfs
, zfs_
, nocacheflush
, INT
, ZMOD_RW
,
6364 "Disable cache flushes");
6366 ZFS_MODULE_PARAM(zfs
, zfs_
, embedded_slog_min_ms
, UINT
, ZMOD_RW
,
6367 "Minimum number of metaslabs required to dedicate one for log blocks");
6370 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, min_auto_ashift
,
6371 param_set_min_auto_ashift
, param_get_uint
, ZMOD_RW
,
6372 "Minimum ashift used when creating new top-level vdevs");
6374 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, max_auto_ashift
,
6375 param_set_max_auto_ashift
, param_get_uint
, ZMOD_RW
,
6376 "Maximum ashift used when optimizing for logical -> physical sector "
6377 "size on new top-level vdevs");