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 [2021] 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 const 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_top_zap
!= 0) {
401 objid
= vd
->vdev_top_zap
;
402 } else if (vd
->vdev_leaf_zap
!= 0) {
403 objid
= vd
->vdev_leaf_zap
;
408 err
= zap_lookup(mos
, objid
, vdev_prop_to_name(prop
),
409 sizeof (uint64_t), 1, value
);
412 *value
= vdev_prop_default_numeric(prop
);
418 * Get the number of data disks for a top-level vdev.
421 vdev_get_ndisks(vdev_t
*vd
)
425 if (vd
->vdev_ops
->vdev_op_ndisks
!= NULL
)
426 ndisks
= vd
->vdev_ops
->vdev_op_ndisks(vd
);
432 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
434 vdev_t
*rvd
= spa
->spa_root_vdev
;
436 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
438 if (vdev
< rvd
->vdev_children
) {
439 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
440 return (rvd
->vdev_child
[vdev
]);
447 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
451 if (vd
->vdev_guid
== guid
)
454 for (int c
= 0; c
< vd
->vdev_children
; c
++)
455 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
463 vdev_count_leaves_impl(vdev_t
*vd
)
467 if (vd
->vdev_ops
->vdev_op_leaf
)
470 for (int c
= 0; c
< vd
->vdev_children
; c
++)
471 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
477 vdev_count_leaves(spa_t
*spa
)
481 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
482 rc
= vdev_count_leaves_impl(spa
->spa_root_vdev
);
483 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
489 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
491 size_t oldsize
, newsize
;
492 uint64_t id
= cvd
->vdev_id
;
495 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
496 ASSERT(cvd
->vdev_parent
== NULL
);
498 cvd
->vdev_parent
= pvd
;
503 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
505 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
506 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
507 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
509 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
510 if (pvd
->vdev_child
!= NULL
) {
511 memcpy(newchild
, pvd
->vdev_child
, oldsize
);
512 kmem_free(pvd
->vdev_child
, oldsize
);
515 pvd
->vdev_child
= newchild
;
516 pvd
->vdev_child
[id
] = cvd
;
518 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
519 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
522 * Walk up all ancestors to update guid sum.
524 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
525 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
527 if (cvd
->vdev_ops
->vdev_op_leaf
) {
528 list_insert_head(&cvd
->vdev_spa
->spa_leaf_list
, cvd
);
529 cvd
->vdev_spa
->spa_leaf_list_gen
++;
534 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
537 uint_t id
= cvd
->vdev_id
;
539 ASSERT(cvd
->vdev_parent
== pvd
);
544 ASSERT(id
< pvd
->vdev_children
);
545 ASSERT(pvd
->vdev_child
[id
] == cvd
);
547 pvd
->vdev_child
[id
] = NULL
;
548 cvd
->vdev_parent
= NULL
;
550 for (c
= 0; c
< pvd
->vdev_children
; c
++)
551 if (pvd
->vdev_child
[c
])
554 if (c
== pvd
->vdev_children
) {
555 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
556 pvd
->vdev_child
= NULL
;
557 pvd
->vdev_children
= 0;
560 if (cvd
->vdev_ops
->vdev_op_leaf
) {
561 spa_t
*spa
= cvd
->vdev_spa
;
562 list_remove(&spa
->spa_leaf_list
, cvd
);
563 spa
->spa_leaf_list_gen
++;
567 * Walk up all ancestors to update guid sum.
569 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
570 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
574 * Remove any holes in the child array.
577 vdev_compact_children(vdev_t
*pvd
)
579 vdev_t
**newchild
, *cvd
;
580 int oldc
= pvd
->vdev_children
;
583 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
588 for (int c
= newc
= 0; c
< oldc
; c
++)
589 if (pvd
->vdev_child
[c
])
593 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
595 for (int c
= newc
= 0; c
< oldc
; c
++) {
596 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
597 newchild
[newc
] = cvd
;
598 cvd
->vdev_id
= newc
++;
605 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
606 pvd
->vdev_child
= newchild
;
607 pvd
->vdev_children
= newc
;
611 * Allocate and minimally initialize a vdev_t.
614 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
617 vdev_indirect_config_t
*vic
;
619 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
620 vic
= &vd
->vdev_indirect_config
;
622 if (spa
->spa_root_vdev
== NULL
) {
623 ASSERT(ops
== &vdev_root_ops
);
624 spa
->spa_root_vdev
= vd
;
625 spa
->spa_load_guid
= spa_generate_guid(NULL
);
628 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
629 if (spa
->spa_root_vdev
== vd
) {
631 * The root vdev's guid will also be the pool guid,
632 * which must be unique among all pools.
634 guid
= spa_generate_guid(NULL
);
637 * Any other vdev's guid must be unique within the pool.
639 guid
= spa_generate_guid(spa
);
641 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
646 vd
->vdev_guid
= guid
;
647 vd
->vdev_guid_sum
= guid
;
649 vd
->vdev_state
= VDEV_STATE_CLOSED
;
650 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
651 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
653 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
654 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
655 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, RANGE_SEG64
, NULL
,
659 * Initialize rate limit structs for events. We rate limit ZIO delay
660 * and checksum events so that we don't overwhelm ZED with thousands
661 * of events when a disk is acting up.
663 zfs_ratelimit_init(&vd
->vdev_delay_rl
, &zfs_slow_io_events_per_second
,
665 zfs_ratelimit_init(&vd
->vdev_deadman_rl
, &zfs_slow_io_events_per_second
,
667 zfs_ratelimit_init(&vd
->vdev_checksum_rl
,
668 &zfs_checksum_events_per_second
, 1);
671 * Default Thresholds for tuning ZED
673 vd
->vdev_checksum_n
= vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_N
);
674 vd
->vdev_checksum_t
= vdev_prop_default_numeric(VDEV_PROP_CHECKSUM_T
);
675 vd
->vdev_io_n
= vdev_prop_default_numeric(VDEV_PROP_IO_N
);
676 vd
->vdev_io_t
= vdev_prop_default_numeric(VDEV_PROP_IO_T
);
678 list_link_init(&vd
->vdev_config_dirty_node
);
679 list_link_init(&vd
->vdev_state_dirty_node
);
680 list_link_init(&vd
->vdev_initialize_node
);
681 list_link_init(&vd
->vdev_leaf_node
);
682 list_link_init(&vd
->vdev_trim_node
);
684 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
685 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
686 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
687 mutex_init(&vd
->vdev_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
689 mutex_init(&vd
->vdev_initialize_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
690 mutex_init(&vd
->vdev_initialize_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
691 cv_init(&vd
->vdev_initialize_cv
, NULL
, CV_DEFAULT
, NULL
);
692 cv_init(&vd
->vdev_initialize_io_cv
, NULL
, CV_DEFAULT
, NULL
);
694 mutex_init(&vd
->vdev_trim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
695 mutex_init(&vd
->vdev_autotrim_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
696 mutex_init(&vd
->vdev_trim_io_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
697 cv_init(&vd
->vdev_trim_cv
, NULL
, CV_DEFAULT
, NULL
);
698 cv_init(&vd
->vdev_autotrim_cv
, NULL
, CV_DEFAULT
, NULL
);
699 cv_init(&vd
->vdev_trim_io_cv
, NULL
, CV_DEFAULT
, NULL
);
701 mutex_init(&vd
->vdev_rebuild_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
702 cv_init(&vd
->vdev_rebuild_cv
, NULL
, CV_DEFAULT
, NULL
);
704 for (int t
= 0; t
< DTL_TYPES
; t
++) {
705 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0,
709 txg_list_create(&vd
->vdev_ms_list
, spa
,
710 offsetof(struct metaslab
, ms_txg_node
));
711 txg_list_create(&vd
->vdev_dtl_list
, spa
,
712 offsetof(struct vdev
, vdev_dtl_node
));
713 vd
->vdev_stat
.vs_timestamp
= gethrtime();
721 * Allocate a new vdev. The 'alloctype' is used to control whether we are
722 * creating a new vdev or loading an existing one - the behavior is slightly
723 * different for each case.
726 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
731 uint64_t guid
= 0, islog
;
733 vdev_indirect_config_t
*vic
;
736 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
737 boolean_t top_level
= (parent
&& !parent
->vdev_parent
);
739 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
741 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
742 return (SET_ERROR(EINVAL
));
744 if ((ops
= vdev_getops(type
)) == NULL
)
745 return (SET_ERROR(EINVAL
));
748 * If this is a load, get the vdev guid from the nvlist.
749 * Otherwise, vdev_alloc_common() will generate one for us.
751 if (alloctype
== VDEV_ALLOC_LOAD
) {
754 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
756 return (SET_ERROR(EINVAL
));
758 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
759 return (SET_ERROR(EINVAL
));
760 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
761 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
762 return (SET_ERROR(EINVAL
));
763 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
764 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
765 return (SET_ERROR(EINVAL
));
766 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
767 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
768 return (SET_ERROR(EINVAL
));
772 * The first allocated vdev must be of type 'root'.
774 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
775 return (SET_ERROR(EINVAL
));
778 * Determine whether we're a log vdev.
781 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
782 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
783 return (SET_ERROR(ENOTSUP
));
785 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
786 return (SET_ERROR(ENOTSUP
));
788 if (top_level
&& alloctype
== VDEV_ALLOC_ADD
) {
792 * If creating a top-level vdev, check for allocation
795 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
797 alloc_bias
= vdev_derive_alloc_bias(bias
);
799 /* spa_vdev_add() expects feature to be enabled */
800 if (spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
801 !spa_feature_is_enabled(spa
,
802 SPA_FEATURE_ALLOCATION_CLASSES
)) {
803 return (SET_ERROR(ENOTSUP
));
807 /* spa_vdev_add() expects feature to be enabled */
808 if (ops
== &vdev_draid_ops
&&
809 spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
810 !spa_feature_is_enabled(spa
, SPA_FEATURE_DRAID
)) {
811 return (SET_ERROR(ENOTSUP
));
816 * Initialize the vdev specific data. This is done before calling
817 * vdev_alloc_common() since it may fail and this simplifies the
818 * error reporting and cleanup code paths.
821 if (ops
->vdev_op_init
!= NULL
) {
822 rc
= ops
->vdev_op_init(spa
, nv
, &tsd
);
828 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
830 vd
->vdev_islog
= islog
;
832 if (top_level
&& alloc_bias
!= VDEV_BIAS_NONE
)
833 vd
->vdev_alloc_bias
= alloc_bias
;
835 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
836 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
839 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
840 * fault on a vdev and want it to persist across imports (like with
843 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
844 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
845 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
846 vd
->vdev_faulted
= 1;
847 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
850 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
851 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
852 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
853 &vd
->vdev_physpath
) == 0)
854 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
856 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
857 &vd
->vdev_enc_sysfs_path
) == 0)
858 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
860 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
861 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
864 * Set the whole_disk property. If it's not specified, leave the value
867 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
868 &vd
->vdev_wholedisk
) != 0)
869 vd
->vdev_wholedisk
= -1ULL;
871 vic
= &vd
->vdev_indirect_config
;
873 ASSERT0(vic
->vic_mapping_object
);
874 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
875 &vic
->vic_mapping_object
);
876 ASSERT0(vic
->vic_births_object
);
877 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
878 &vic
->vic_births_object
);
879 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
880 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
881 &vic
->vic_prev_indirect_vdev
);
884 * Look for the 'not present' flag. This will only be set if the device
885 * was not present at the time of import.
887 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
888 &vd
->vdev_not_present
);
891 * Get the alignment requirement.
893 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
896 * Retrieve the vdev creation time.
898 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
902 * If we're a top-level vdev, try to load the allocation parameters.
905 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
906 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
908 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
910 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
912 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NONALLOCATING
,
914 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
916 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
919 ASSERT0(vd
->vdev_top_zap
);
922 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
923 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
924 alloctype
== VDEV_ALLOC_ADD
||
925 alloctype
== VDEV_ALLOC_SPLIT
||
926 alloctype
== VDEV_ALLOC_ROOTPOOL
);
927 /* Note: metaslab_group_create() is now deferred */
930 if (vd
->vdev_ops
->vdev_op_leaf
&&
931 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
932 (void) nvlist_lookup_uint64(nv
,
933 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
935 ASSERT0(vd
->vdev_leaf_zap
);
939 * If we're a leaf vdev, try to load the DTL object and other state.
942 if (vd
->vdev_ops
->vdev_op_leaf
&&
943 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
944 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
945 if (alloctype
== VDEV_ALLOC_LOAD
) {
946 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
947 &vd
->vdev_dtl_object
);
948 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
952 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
955 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
956 &spare
) == 0 && spare
)
960 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
963 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
964 &vd
->vdev_resilver_txg
);
966 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REBUILD_TXG
,
967 &vd
->vdev_rebuild_txg
);
969 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
970 vdev_defer_resilver(vd
);
973 * In general, when importing a pool we want to ignore the
974 * persistent fault state, as the diagnosis made on another
975 * system may not be valid in the current context. The only
976 * exception is if we forced a vdev to a persistently faulted
977 * state with 'zpool offline -f'. The persistent fault will
978 * remain across imports until cleared.
980 * Local vdevs will remain in the faulted state.
982 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
983 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
984 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
986 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
988 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
991 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
995 VDEV_AUX_ERR_EXCEEDED
;
996 if (nvlist_lookup_string(nv
,
997 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
998 strcmp(aux
, "external") == 0)
999 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
1001 vd
->vdev_faulted
= 0ULL;
1007 * Add ourselves to the parent's list of children.
1009 vdev_add_child(parent
, vd
);
1017 vdev_free(vdev_t
*vd
)
1019 spa_t
*spa
= vd
->vdev_spa
;
1021 ASSERT3P(vd
->vdev_initialize_thread
, ==, NULL
);
1022 ASSERT3P(vd
->vdev_trim_thread
, ==, NULL
);
1023 ASSERT3P(vd
->vdev_autotrim_thread
, ==, NULL
);
1024 ASSERT3P(vd
->vdev_rebuild_thread
, ==, NULL
);
1027 * Scan queues are normally destroyed at the end of a scan. If the
1028 * queue exists here, that implies the vdev is being removed while
1029 * the scan is still running.
1031 if (vd
->vdev_scan_io_queue
!= NULL
) {
1032 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
1033 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
1034 vd
->vdev_scan_io_queue
= NULL
;
1035 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
1039 * vdev_free() implies closing the vdev first. This is simpler than
1040 * trying to ensure complicated semantics for all callers.
1044 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
1045 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
1048 * Free all children.
1050 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1051 vdev_free(vd
->vdev_child
[c
]);
1053 ASSERT(vd
->vdev_child
== NULL
);
1054 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
1056 if (vd
->vdev_ops
->vdev_op_fini
!= NULL
)
1057 vd
->vdev_ops
->vdev_op_fini(vd
);
1060 * Discard allocation state.
1062 if (vd
->vdev_mg
!= NULL
) {
1063 vdev_metaslab_fini(vd
);
1064 metaslab_group_destroy(vd
->vdev_mg
);
1067 if (vd
->vdev_log_mg
!= NULL
) {
1068 ASSERT0(vd
->vdev_ms_count
);
1069 metaslab_group_destroy(vd
->vdev_log_mg
);
1070 vd
->vdev_log_mg
= NULL
;
1073 ASSERT0(vd
->vdev_stat
.vs_space
);
1074 ASSERT0(vd
->vdev_stat
.vs_dspace
);
1075 ASSERT0(vd
->vdev_stat
.vs_alloc
);
1078 * Remove this vdev from its parent's child list.
1080 vdev_remove_child(vd
->vdev_parent
, vd
);
1082 ASSERT(vd
->vdev_parent
== NULL
);
1083 ASSERT(!list_link_active(&vd
->vdev_leaf_node
));
1086 * Clean up vdev structure.
1088 vdev_queue_fini(vd
);
1089 vdev_cache_fini(vd
);
1092 spa_strfree(vd
->vdev_path
);
1094 spa_strfree(vd
->vdev_devid
);
1095 if (vd
->vdev_physpath
)
1096 spa_strfree(vd
->vdev_physpath
);
1098 if (vd
->vdev_enc_sysfs_path
)
1099 spa_strfree(vd
->vdev_enc_sysfs_path
);
1102 spa_strfree(vd
->vdev_fru
);
1104 if (vd
->vdev_isspare
)
1105 spa_spare_remove(vd
);
1106 if (vd
->vdev_isl2cache
)
1107 spa_l2cache_remove(vd
);
1109 txg_list_destroy(&vd
->vdev_ms_list
);
1110 txg_list_destroy(&vd
->vdev_dtl_list
);
1112 mutex_enter(&vd
->vdev_dtl_lock
);
1113 space_map_close(vd
->vdev_dtl_sm
);
1114 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1115 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
1116 range_tree_destroy(vd
->vdev_dtl
[t
]);
1118 mutex_exit(&vd
->vdev_dtl_lock
);
1120 EQUIV(vd
->vdev_indirect_births
!= NULL
,
1121 vd
->vdev_indirect_mapping
!= NULL
);
1122 if (vd
->vdev_indirect_births
!= NULL
) {
1123 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
1124 vdev_indirect_births_close(vd
->vdev_indirect_births
);
1127 if (vd
->vdev_obsolete_sm
!= NULL
) {
1128 ASSERT(vd
->vdev_removing
||
1129 vd
->vdev_ops
== &vdev_indirect_ops
);
1130 space_map_close(vd
->vdev_obsolete_sm
);
1131 vd
->vdev_obsolete_sm
= NULL
;
1133 range_tree_destroy(vd
->vdev_obsolete_segments
);
1134 rw_destroy(&vd
->vdev_indirect_rwlock
);
1135 mutex_destroy(&vd
->vdev_obsolete_lock
);
1137 mutex_destroy(&vd
->vdev_dtl_lock
);
1138 mutex_destroy(&vd
->vdev_stat_lock
);
1139 mutex_destroy(&vd
->vdev_probe_lock
);
1140 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
1142 mutex_destroy(&vd
->vdev_initialize_lock
);
1143 mutex_destroy(&vd
->vdev_initialize_io_lock
);
1144 cv_destroy(&vd
->vdev_initialize_io_cv
);
1145 cv_destroy(&vd
->vdev_initialize_cv
);
1147 mutex_destroy(&vd
->vdev_trim_lock
);
1148 mutex_destroy(&vd
->vdev_autotrim_lock
);
1149 mutex_destroy(&vd
->vdev_trim_io_lock
);
1150 cv_destroy(&vd
->vdev_trim_cv
);
1151 cv_destroy(&vd
->vdev_autotrim_cv
);
1152 cv_destroy(&vd
->vdev_trim_io_cv
);
1154 mutex_destroy(&vd
->vdev_rebuild_lock
);
1155 cv_destroy(&vd
->vdev_rebuild_cv
);
1157 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
1158 zfs_ratelimit_fini(&vd
->vdev_deadman_rl
);
1159 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
1161 if (vd
== spa
->spa_root_vdev
)
1162 spa
->spa_root_vdev
= NULL
;
1164 kmem_free(vd
, sizeof (vdev_t
));
1168 * Transfer top-level vdev state from svd to tvd.
1171 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
1173 spa_t
*spa
= svd
->vdev_spa
;
1178 ASSERT(tvd
== tvd
->vdev_top
);
1180 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
1181 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
1182 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
1183 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
1184 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
1186 svd
->vdev_ms_array
= 0;
1187 svd
->vdev_ms_shift
= 0;
1188 svd
->vdev_ms_count
= 0;
1189 svd
->vdev_top_zap
= 0;
1192 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
1193 if (tvd
->vdev_log_mg
)
1194 ASSERT3P(tvd
->vdev_log_mg
, ==, svd
->vdev_log_mg
);
1195 tvd
->vdev_mg
= svd
->vdev_mg
;
1196 tvd
->vdev_log_mg
= svd
->vdev_log_mg
;
1197 tvd
->vdev_ms
= svd
->vdev_ms
;
1199 svd
->vdev_mg
= NULL
;
1200 svd
->vdev_log_mg
= NULL
;
1201 svd
->vdev_ms
= NULL
;
1203 if (tvd
->vdev_mg
!= NULL
)
1204 tvd
->vdev_mg
->mg_vd
= tvd
;
1205 if (tvd
->vdev_log_mg
!= NULL
)
1206 tvd
->vdev_log_mg
->mg_vd
= tvd
;
1208 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
1209 svd
->vdev_checkpoint_sm
= NULL
;
1211 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1212 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1214 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1215 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1216 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1218 svd
->vdev_stat
.vs_alloc
= 0;
1219 svd
->vdev_stat
.vs_space
= 0;
1220 svd
->vdev_stat
.vs_dspace
= 0;
1223 * State which may be set on a top-level vdev that's in the
1224 * process of being removed.
1226 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1227 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1228 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1229 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1230 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1231 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1232 ASSERT0(tvd
->vdev_noalloc
);
1233 ASSERT0(tvd
->vdev_removing
);
1234 ASSERT0(tvd
->vdev_rebuilding
);
1235 tvd
->vdev_noalloc
= svd
->vdev_noalloc
;
1236 tvd
->vdev_removing
= svd
->vdev_removing
;
1237 tvd
->vdev_rebuilding
= svd
->vdev_rebuilding
;
1238 tvd
->vdev_rebuild_config
= svd
->vdev_rebuild_config
;
1239 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1240 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1241 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1242 range_tree_swap(&svd
->vdev_obsolete_segments
,
1243 &tvd
->vdev_obsolete_segments
);
1244 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1245 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1246 svd
->vdev_indirect_config
.vic_births_object
= 0;
1247 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1248 svd
->vdev_indirect_mapping
= NULL
;
1249 svd
->vdev_indirect_births
= NULL
;
1250 svd
->vdev_obsolete_sm
= NULL
;
1251 svd
->vdev_noalloc
= 0;
1252 svd
->vdev_removing
= 0;
1253 svd
->vdev_rebuilding
= 0;
1255 for (t
= 0; t
< TXG_SIZE
; t
++) {
1256 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1257 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1258 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1259 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1260 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1261 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1264 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1265 vdev_config_clean(svd
);
1266 vdev_config_dirty(tvd
);
1269 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1270 vdev_state_clean(svd
);
1271 vdev_state_dirty(tvd
);
1274 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1275 svd
->vdev_deflate_ratio
= 0;
1277 tvd
->vdev_islog
= svd
->vdev_islog
;
1278 svd
->vdev_islog
= 0;
1280 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1284 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1291 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1292 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1296 * Add a mirror/replacing vdev above an existing vdev. There is no need to
1297 * call .vdev_op_init() since mirror/replacing vdevs do not have private state.
1300 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1302 spa_t
*spa
= cvd
->vdev_spa
;
1303 vdev_t
*pvd
= cvd
->vdev_parent
;
1306 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1308 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1310 mvd
->vdev_asize
= cvd
->vdev_asize
;
1311 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1312 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1313 mvd
->vdev_psize
= cvd
->vdev_psize
;
1314 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1315 mvd
->vdev_logical_ashift
= cvd
->vdev_logical_ashift
;
1316 mvd
->vdev_physical_ashift
= cvd
->vdev_physical_ashift
;
1317 mvd
->vdev_state
= cvd
->vdev_state
;
1318 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1320 vdev_remove_child(pvd
, cvd
);
1321 vdev_add_child(pvd
, mvd
);
1322 cvd
->vdev_id
= mvd
->vdev_children
;
1323 vdev_add_child(mvd
, cvd
);
1324 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1326 if (mvd
== mvd
->vdev_top
)
1327 vdev_top_transfer(cvd
, mvd
);
1333 * Remove a 1-way mirror/replacing vdev from the tree.
1336 vdev_remove_parent(vdev_t
*cvd
)
1338 vdev_t
*mvd
= cvd
->vdev_parent
;
1339 vdev_t
*pvd
= mvd
->vdev_parent
;
1341 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1343 ASSERT(mvd
->vdev_children
== 1);
1344 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1345 mvd
->vdev_ops
== &vdev_replacing_ops
||
1346 mvd
->vdev_ops
== &vdev_spare_ops
);
1347 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1348 cvd
->vdev_logical_ashift
= mvd
->vdev_logical_ashift
;
1349 cvd
->vdev_physical_ashift
= mvd
->vdev_physical_ashift
;
1350 vdev_remove_child(mvd
, cvd
);
1351 vdev_remove_child(pvd
, mvd
);
1354 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1355 * Otherwise, we could have detached an offline device, and when we
1356 * go to import the pool we'll think we have two top-level vdevs,
1357 * instead of a different version of the same top-level vdev.
1359 if (mvd
->vdev_top
== mvd
) {
1360 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1361 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1362 cvd
->vdev_guid
+= guid_delta
;
1363 cvd
->vdev_guid_sum
+= guid_delta
;
1366 * If pool not set for autoexpand, we need to also preserve
1367 * mvd's asize to prevent automatic expansion of cvd.
1368 * Otherwise if we are adjusting the mirror by attaching and
1369 * detaching children of non-uniform sizes, the mirror could
1370 * autoexpand, unexpectedly requiring larger devices to
1371 * re-establish the mirror.
1373 if (!cvd
->vdev_spa
->spa_autoexpand
)
1374 cvd
->vdev_asize
= mvd
->vdev_asize
;
1376 cvd
->vdev_id
= mvd
->vdev_id
;
1377 vdev_add_child(pvd
, cvd
);
1378 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1380 if (cvd
== cvd
->vdev_top
)
1381 vdev_top_transfer(mvd
, cvd
);
1383 ASSERT(mvd
->vdev_children
== 0);
1388 vdev_metaslab_group_create(vdev_t
*vd
)
1390 spa_t
*spa
= vd
->vdev_spa
;
1393 * metaslab_group_create was delayed until allocation bias was available
1395 if (vd
->vdev_mg
== NULL
) {
1396 metaslab_class_t
*mc
;
1398 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1399 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1401 ASSERT3U(vd
->vdev_islog
, ==,
1402 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1404 switch (vd
->vdev_alloc_bias
) {
1406 mc
= spa_log_class(spa
);
1408 case VDEV_BIAS_SPECIAL
:
1409 mc
= spa_special_class(spa
);
1411 case VDEV_BIAS_DEDUP
:
1412 mc
= spa_dedup_class(spa
);
1415 mc
= spa_normal_class(spa
);
1418 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1419 spa
->spa_alloc_count
);
1421 if (!vd
->vdev_islog
) {
1422 vd
->vdev_log_mg
= metaslab_group_create(
1423 spa_embedded_log_class(spa
), vd
, 1);
1427 * The spa ashift min/max only apply for the normal metaslab
1428 * class. Class destination is late binding so ashift boundary
1429 * setting had to wait until now.
1431 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1432 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1433 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1434 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1435 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1436 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1438 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
1439 if (min_alloc
< spa
->spa_min_alloc
)
1440 spa
->spa_min_alloc
= min_alloc
;
1446 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1448 spa_t
*spa
= vd
->vdev_spa
;
1449 uint64_t oldc
= vd
->vdev_ms_count
;
1450 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1453 boolean_t expanding
= (oldc
!= 0);
1455 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1458 * This vdev is not being allocated from yet or is a hole.
1460 if (vd
->vdev_ms_shift
== 0)
1463 ASSERT(!vd
->vdev_ishole
);
1465 ASSERT(oldc
<= newc
);
1467 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1470 memcpy(mspp
, vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1471 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1475 vd
->vdev_ms_count
= newc
;
1477 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1478 uint64_t object
= 0;
1480 * vdev_ms_array may be 0 if we are creating the "fake"
1481 * metaslabs for an indirect vdev for zdb's leak detection.
1482 * See zdb_leak_init().
1484 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1485 error
= dmu_read(spa
->spa_meta_objset
,
1487 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1490 vdev_dbgmsg(vd
, "unable to read the metaslab "
1491 "array [error=%d]", error
);
1496 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1499 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1506 * Find the emptiest metaslab on the vdev and mark it for use for
1507 * embedded slog by moving it from the regular to the log metaslab
1510 if (vd
->vdev_mg
->mg_class
== spa_normal_class(spa
) &&
1511 vd
->vdev_ms_count
> zfs_embedded_slog_min_ms
&&
1512 avl_is_empty(&vd
->vdev_log_mg
->mg_metaslab_tree
)) {
1513 uint64_t slog_msid
= 0;
1514 uint64_t smallest
= UINT64_MAX
;
1517 * Note, we only search the new metaslabs, because the old
1518 * (pre-existing) ones may be active (e.g. have non-empty
1519 * range_tree's), and we don't move them to the new
1522 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1524 space_map_allocated(vd
->vdev_ms
[m
]->ms_sm
);
1525 if (alloc
< smallest
) {
1530 metaslab_t
*slog_ms
= vd
->vdev_ms
[slog_msid
];
1532 * The metaslab was marked as dirty at the end of
1533 * metaslab_init(). Remove it from the dirty list so that we
1534 * can uninitialize and reinitialize it to the new class.
1537 (void) txg_list_remove_this(&vd
->vdev_ms_list
,
1540 uint64_t sm_obj
= space_map_object(slog_ms
->ms_sm
);
1541 metaslab_fini(slog_ms
);
1542 VERIFY0(metaslab_init(vd
->vdev_log_mg
, slog_msid
, sm_obj
, txg
,
1543 &vd
->vdev_ms
[slog_msid
]));
1547 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1550 * If the vdev is marked as non-allocating then don't
1551 * activate the metaslabs since we want to ensure that
1552 * no allocations are performed on this device.
1554 if (vd
->vdev_noalloc
) {
1555 /* track non-allocating vdev space */
1556 spa
->spa_nonallocating_dspace
+= spa_deflate(spa
) ?
1557 vd
->vdev_stat
.vs_dspace
: vd
->vdev_stat
.vs_space
;
1558 } else if (!expanding
) {
1559 metaslab_group_activate(vd
->vdev_mg
);
1560 if (vd
->vdev_log_mg
!= NULL
)
1561 metaslab_group_activate(vd
->vdev_log_mg
);
1565 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1571 vdev_metaslab_fini(vdev_t
*vd
)
1573 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1574 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1575 SPA_FEATURE_POOL_CHECKPOINT
));
1576 space_map_close(vd
->vdev_checkpoint_sm
);
1578 * Even though we close the space map, we need to set its
1579 * pointer to NULL. The reason is that vdev_metaslab_fini()
1580 * may be called multiple times for certain operations
1581 * (i.e. when destroying a pool) so we need to ensure that
1582 * this clause never executes twice. This logic is similar
1583 * to the one used for the vdev_ms clause below.
1585 vd
->vdev_checkpoint_sm
= NULL
;
1588 if (vd
->vdev_ms
!= NULL
) {
1589 metaslab_group_t
*mg
= vd
->vdev_mg
;
1591 metaslab_group_passivate(mg
);
1592 if (vd
->vdev_log_mg
!= NULL
) {
1593 ASSERT(!vd
->vdev_islog
);
1594 metaslab_group_passivate(vd
->vdev_log_mg
);
1597 uint64_t count
= vd
->vdev_ms_count
;
1598 for (uint64_t m
= 0; m
< count
; m
++) {
1599 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1603 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1605 vd
->vdev_ms_count
= 0;
1607 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
1608 ASSERT0(mg
->mg_histogram
[i
]);
1609 if (vd
->vdev_log_mg
!= NULL
)
1610 ASSERT0(vd
->vdev_log_mg
->mg_histogram
[i
]);
1613 ASSERT0(vd
->vdev_ms_count
);
1614 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1617 typedef struct vdev_probe_stats
{
1618 boolean_t vps_readable
;
1619 boolean_t vps_writeable
;
1621 } vdev_probe_stats_t
;
1624 vdev_probe_done(zio_t
*zio
)
1626 spa_t
*spa
= zio
->io_spa
;
1627 vdev_t
*vd
= zio
->io_vd
;
1628 vdev_probe_stats_t
*vps
= zio
->io_private
;
1630 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1632 if (zio
->io_type
== ZIO_TYPE_READ
) {
1633 if (zio
->io_error
== 0)
1634 vps
->vps_readable
= 1;
1635 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1636 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1637 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1638 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1639 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1641 abd_free(zio
->io_abd
);
1643 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1644 if (zio
->io_error
== 0)
1645 vps
->vps_writeable
= 1;
1646 abd_free(zio
->io_abd
);
1647 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1651 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1652 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1654 if (vdev_readable(vd
) &&
1655 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1658 ASSERT(zio
->io_error
!= 0);
1659 vdev_dbgmsg(vd
, "failed probe");
1660 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1661 spa
, vd
, NULL
, NULL
, 0);
1662 zio
->io_error
= SET_ERROR(ENXIO
);
1665 mutex_enter(&vd
->vdev_probe_lock
);
1666 ASSERT(vd
->vdev_probe_zio
== zio
);
1667 vd
->vdev_probe_zio
= NULL
;
1668 mutex_exit(&vd
->vdev_probe_lock
);
1671 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1672 if (!vdev_accessible(vd
, pio
))
1673 pio
->io_error
= SET_ERROR(ENXIO
);
1675 kmem_free(vps
, sizeof (*vps
));
1680 * Determine whether this device is accessible.
1682 * Read and write to several known locations: the pad regions of each
1683 * vdev label but the first, which we leave alone in case it contains
1687 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1689 spa_t
*spa
= vd
->vdev_spa
;
1690 vdev_probe_stats_t
*vps
= NULL
;
1693 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1696 * Don't probe the probe.
1698 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1702 * To prevent 'probe storms' when a device fails, we create
1703 * just one probe i/o at a time. All zios that want to probe
1704 * this vdev will become parents of the probe io.
1706 mutex_enter(&vd
->vdev_probe_lock
);
1708 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1709 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1711 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1712 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1715 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1717 * vdev_cant_read and vdev_cant_write can only
1718 * transition from TRUE to FALSE when we have the
1719 * SCL_ZIO lock as writer; otherwise they can only
1720 * transition from FALSE to TRUE. This ensures that
1721 * any zio looking at these values can assume that
1722 * failures persist for the life of the I/O. That's
1723 * important because when a device has intermittent
1724 * connectivity problems, we want to ensure that
1725 * they're ascribed to the device (ENXIO) and not
1728 * Since we hold SCL_ZIO as writer here, clear both
1729 * values so the probe can reevaluate from first
1732 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1733 vd
->vdev_cant_read
= B_FALSE
;
1734 vd
->vdev_cant_write
= B_FALSE
;
1737 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1738 vdev_probe_done
, vps
,
1739 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1742 * We can't change the vdev state in this context, so we
1743 * kick off an async task to do it on our behalf.
1746 vd
->vdev_probe_wanted
= B_TRUE
;
1747 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1752 zio_add_child(zio
, pio
);
1754 mutex_exit(&vd
->vdev_probe_lock
);
1757 ASSERT(zio
!= NULL
);
1761 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1762 zio_nowait(zio_read_phys(pio
, vd
,
1763 vdev_label_offset(vd
->vdev_psize
, l
,
1764 offsetof(vdev_label_t
, vl_be
)), VDEV_PAD_SIZE
,
1765 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1766 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1767 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1778 vdev_load_child(void *arg
)
1782 vd
->vdev_load_error
= vdev_load(vd
);
1786 vdev_open_child(void *arg
)
1790 vd
->vdev_open_thread
= curthread
;
1791 vd
->vdev_open_error
= vdev_open(vd
);
1792 vd
->vdev_open_thread
= NULL
;
1796 vdev_uses_zvols(vdev_t
*vd
)
1799 if (zvol_is_zvol(vd
->vdev_path
))
1803 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1804 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1811 * Returns B_TRUE if the passed child should be opened.
1814 vdev_default_open_children_func(vdev_t
*vd
)
1821 * Open the requested child vdevs. If any of the leaf vdevs are using
1822 * a ZFS volume then do the opens in a single thread. This avoids a
1823 * deadlock when the current thread is holding the spa_namespace_lock.
1826 vdev_open_children_impl(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1828 int children
= vd
->vdev_children
;
1830 taskq_t
*tq
= taskq_create("vdev_open", children
, minclsyspri
,
1831 children
, children
, TASKQ_PREPOPULATE
);
1832 vd
->vdev_nonrot
= B_TRUE
;
1834 for (int c
= 0; c
< children
; c
++) {
1835 vdev_t
*cvd
= vd
->vdev_child
[c
];
1837 if (open_func(cvd
) == B_FALSE
)
1840 if (tq
== NULL
|| vdev_uses_zvols(vd
)) {
1841 cvd
->vdev_open_error
= vdev_open(cvd
);
1843 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1844 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
1847 vd
->vdev_nonrot
&= cvd
->vdev_nonrot
;
1857 * Open all child vdevs.
1860 vdev_open_children(vdev_t
*vd
)
1862 vdev_open_children_impl(vd
, vdev_default_open_children_func
);
1866 * Conditionally open a subset of child vdevs.
1869 vdev_open_children_subset(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1871 vdev_open_children_impl(vd
, open_func
);
1875 * Compute the raidz-deflation ratio. Note, we hard-code
1876 * in 128k (1 << 17) because it is the "typical" blocksize.
1877 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1878 * otherwise it would inconsistently account for existing bp's.
1881 vdev_set_deflate_ratio(vdev_t
*vd
)
1883 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1884 vd
->vdev_deflate_ratio
= (1 << 17) /
1885 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1890 * Choose the best of two ashifts, preferring one between logical ashift
1891 * (absolute minimum) and administrator defined maximum, otherwise take
1892 * the biggest of the two.
1895 vdev_best_ashift(uint64_t logical
, uint64_t a
, uint64_t b
)
1897 if (a
> logical
&& a
<= zfs_vdev_max_auto_ashift
) {
1898 if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1902 } else if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1908 * Maximize performance by inflating the configured ashift for top level
1909 * vdevs to be as close to the physical ashift as possible while maintaining
1910 * administrator defined limits and ensuring it doesn't go below the
1914 vdev_ashift_optimize(vdev_t
*vd
)
1916 ASSERT(vd
== vd
->vdev_top
);
1918 if (vd
->vdev_ashift
< vd
->vdev_physical_ashift
&&
1919 vd
->vdev_physical_ashift
<= zfs_vdev_max_auto_ashift
) {
1920 vd
->vdev_ashift
= MIN(
1921 MAX(zfs_vdev_max_auto_ashift
, vd
->vdev_ashift
),
1922 MAX(zfs_vdev_min_auto_ashift
,
1923 vd
->vdev_physical_ashift
));
1926 * If the logical and physical ashifts are the same, then
1927 * we ensure that the top-level vdev's ashift is not smaller
1928 * than our minimum ashift value. For the unusual case
1929 * where logical ashift > physical ashift, we can't cap
1930 * the calculated ashift based on max ashift as that
1931 * would cause failures.
1932 * We still check if we need to increase it to match
1935 vd
->vdev_ashift
= MAX(zfs_vdev_min_auto_ashift
,
1941 * Prepare a virtual device for access.
1944 vdev_open(vdev_t
*vd
)
1946 spa_t
*spa
= vd
->vdev_spa
;
1949 uint64_t max_osize
= 0;
1950 uint64_t asize
, max_asize
, psize
;
1951 uint64_t logical_ashift
= 0;
1952 uint64_t physical_ashift
= 0;
1954 ASSERT(vd
->vdev_open_thread
== curthread
||
1955 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1956 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1957 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1958 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1960 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1961 vd
->vdev_cant_read
= B_FALSE
;
1962 vd
->vdev_cant_write
= B_FALSE
;
1963 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1966 * If this vdev is not removed, check its fault status. If it's
1967 * faulted, bail out of the open.
1969 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1970 ASSERT(vd
->vdev_children
== 0);
1971 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1972 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1973 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1974 vd
->vdev_label_aux
);
1975 return (SET_ERROR(ENXIO
));
1976 } else if (vd
->vdev_offline
) {
1977 ASSERT(vd
->vdev_children
== 0);
1978 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1979 return (SET_ERROR(ENXIO
));
1982 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
,
1983 &logical_ashift
, &physical_ashift
);
1985 /* Keep the device in removed state if unplugged */
1986 if (error
== ENOENT
&& vd
->vdev_removed
) {
1987 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_REMOVED
,
1993 * Physical volume size should never be larger than its max size, unless
1994 * the disk has shrunk while we were reading it or the device is buggy
1995 * or damaged: either way it's not safe for use, bail out of the open.
1997 if (osize
> max_osize
) {
1998 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1999 VDEV_AUX_OPEN_FAILED
);
2000 return (SET_ERROR(ENXIO
));
2004 * Reset the vdev_reopening flag so that we actually close
2005 * the vdev on error.
2007 vd
->vdev_reopening
= B_FALSE
;
2008 if (zio_injection_enabled
&& error
== 0)
2009 error
= zio_handle_device_injection(vd
, NULL
, SET_ERROR(ENXIO
));
2012 if (vd
->vdev_removed
&&
2013 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
2014 vd
->vdev_removed
= B_FALSE
;
2016 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
2017 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
2018 vd
->vdev_stat
.vs_aux
);
2020 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2021 vd
->vdev_stat
.vs_aux
);
2026 vd
->vdev_removed
= B_FALSE
;
2029 * Recheck the faulted flag now that we have confirmed that
2030 * the vdev is accessible. If we're faulted, bail.
2032 if (vd
->vdev_faulted
) {
2033 ASSERT(vd
->vdev_children
== 0);
2034 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
2035 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
2036 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2037 vd
->vdev_label_aux
);
2038 return (SET_ERROR(ENXIO
));
2041 if (vd
->vdev_degraded
) {
2042 ASSERT(vd
->vdev_children
== 0);
2043 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2044 VDEV_AUX_ERR_EXCEEDED
);
2046 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
2050 * For hole or missing vdevs we just return success.
2052 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
2055 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2056 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
2057 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2063 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
2064 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
2066 if (vd
->vdev_children
== 0) {
2067 if (osize
< SPA_MINDEVSIZE
) {
2068 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2069 VDEV_AUX_TOO_SMALL
);
2070 return (SET_ERROR(EOVERFLOW
));
2073 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
2074 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
2075 VDEV_LABEL_END_SIZE
);
2077 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
2078 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
2079 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2080 VDEV_AUX_TOO_SMALL
);
2081 return (SET_ERROR(EOVERFLOW
));
2085 max_asize
= max_osize
;
2089 * If the vdev was expanded, record this so that we can re-create the
2090 * uberblock rings in labels {2,3}, during the next sync.
2092 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
2093 vd
->vdev_copy_uberblocks
= B_TRUE
;
2095 vd
->vdev_psize
= psize
;
2098 * Make sure the allocatable size hasn't shrunk too much.
2100 if (asize
< vd
->vdev_min_asize
) {
2101 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2102 VDEV_AUX_BAD_LABEL
);
2103 return (SET_ERROR(EINVAL
));
2107 * We can always set the logical/physical ashift members since
2108 * their values are only used to calculate the vdev_ashift when
2109 * the device is first added to the config. These values should
2110 * not be used for anything else since they may change whenever
2111 * the device is reopened and we don't store them in the label.
2113 vd
->vdev_physical_ashift
=
2114 MAX(physical_ashift
, vd
->vdev_physical_ashift
);
2115 vd
->vdev_logical_ashift
= MAX(logical_ashift
,
2116 vd
->vdev_logical_ashift
);
2118 if (vd
->vdev_asize
== 0) {
2120 * This is the first-ever open, so use the computed values.
2121 * For compatibility, a different ashift can be requested.
2123 vd
->vdev_asize
= asize
;
2124 vd
->vdev_max_asize
= max_asize
;
2127 * If the vdev_ashift was not overridden at creation time,
2128 * then set it the logical ashift and optimize the ashift.
2130 if (vd
->vdev_ashift
== 0) {
2131 vd
->vdev_ashift
= vd
->vdev_logical_ashift
;
2133 if (vd
->vdev_logical_ashift
> ASHIFT_MAX
) {
2134 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2135 VDEV_AUX_ASHIFT_TOO_BIG
);
2136 return (SET_ERROR(EDOM
));
2139 if (vd
->vdev_top
== vd
) {
2140 vdev_ashift_optimize(vd
);
2143 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
2144 vd
->vdev_ashift
> ASHIFT_MAX
)) {
2145 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2146 VDEV_AUX_BAD_ASHIFT
);
2147 return (SET_ERROR(EDOM
));
2151 * Make sure the alignment required hasn't increased.
2153 if (vd
->vdev_ashift
> vd
->vdev_top
->vdev_ashift
&&
2154 vd
->vdev_ops
->vdev_op_leaf
) {
2155 (void) zfs_ereport_post(
2156 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
2157 spa
, vd
, NULL
, NULL
, 0);
2158 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2159 VDEV_AUX_BAD_LABEL
);
2160 return (SET_ERROR(EDOM
));
2162 vd
->vdev_max_asize
= max_asize
;
2166 * If all children are healthy we update asize if either:
2167 * The asize has increased, due to a device expansion caused by dynamic
2168 * LUN growth or vdev replacement, and automatic expansion is enabled;
2169 * making the additional space available.
2171 * The asize has decreased, due to a device shrink usually caused by a
2172 * vdev replace with a smaller device. This ensures that calculations
2173 * based of max_asize and asize e.g. esize are always valid. It's safe
2174 * to do this as we've already validated that asize is greater than
2177 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
2178 ((asize
> vd
->vdev_asize
&&
2179 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
2180 (asize
< vd
->vdev_asize
)))
2181 vd
->vdev_asize
= asize
;
2183 vdev_set_min_asize(vd
);
2186 * Ensure we can issue some IO before declaring the
2187 * vdev open for business.
2189 if (vd
->vdev_ops
->vdev_op_leaf
&&
2190 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
2191 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2192 VDEV_AUX_ERR_EXCEEDED
);
2197 * Track the minimum allocation size.
2199 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
2200 vd
->vdev_islog
== 0 && vd
->vdev_aux
== NULL
) {
2201 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
2202 if (min_alloc
< spa
->spa_min_alloc
)
2203 spa
->spa_min_alloc
= min_alloc
;
2207 * If this is a leaf vdev, assess whether a resilver is needed.
2208 * But don't do this if we are doing a reopen for a scrub, since
2209 * this would just restart the scrub we are already doing.
2211 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
)
2212 dsl_scan_assess_vdev(spa
->spa_dsl_pool
, vd
);
2218 vdev_validate_child(void *arg
)
2222 vd
->vdev_validate_thread
= curthread
;
2223 vd
->vdev_validate_error
= vdev_validate(vd
);
2224 vd
->vdev_validate_thread
= NULL
;
2228 * Called once the vdevs are all opened, this routine validates the label
2229 * contents. This needs to be done before vdev_load() so that we don't
2230 * inadvertently do repair I/Os to the wrong device.
2232 * This function will only return failure if one of the vdevs indicates that it
2233 * has since been destroyed or exported. This is only possible if
2234 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
2235 * will be updated but the function will return 0.
2238 vdev_validate(vdev_t
*vd
)
2240 spa_t
*spa
= vd
->vdev_spa
;
2243 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
2247 int children
= vd
->vdev_children
;
2249 if (vdev_validate_skip
)
2253 tq
= taskq_create("vdev_validate", children
, minclsyspri
,
2254 children
, children
, TASKQ_PREPOPULATE
);
2257 for (uint64_t c
= 0; c
< children
; c
++) {
2258 vdev_t
*cvd
= vd
->vdev_child
[c
];
2260 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
2261 vdev_validate_child(cvd
);
2263 VERIFY(taskq_dispatch(tq
, vdev_validate_child
, cvd
,
2264 TQ_SLEEP
) != TASKQID_INVALID
);
2271 for (int c
= 0; c
< children
; c
++) {
2272 int error
= vd
->vdev_child
[c
]->vdev_validate_error
;
2275 return (SET_ERROR(EBADF
));
2280 * If the device has already failed, or was marked offline, don't do
2281 * any further validation. Otherwise, label I/O will fail and we will
2282 * overwrite the previous state.
2284 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
2288 * If we are performing an extreme rewind, we allow for a label that
2289 * was modified at a point after the current txg.
2290 * If config lock is not held do not check for the txg. spa_sync could
2291 * be updating the vdev's label before updating spa_last_synced_txg.
2293 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
2294 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
2297 txg
= spa_last_synced_txg(spa
);
2299 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
2300 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2301 VDEV_AUX_BAD_LABEL
);
2302 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
2303 "txg %llu", (u_longlong_t
)txg
);
2308 * Determine if this vdev has been split off into another
2309 * pool. If so, then refuse to open it.
2311 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
2312 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
2313 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2314 VDEV_AUX_SPLIT_POOL
);
2316 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
2320 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
2321 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2322 VDEV_AUX_CORRUPT_DATA
);
2324 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2325 ZPOOL_CONFIG_POOL_GUID
);
2330 * If config is not trusted then ignore the spa guid check. This is
2331 * necessary because if the machine crashed during a re-guid the new
2332 * guid might have been written to all of the vdev labels, but not the
2333 * cached config. The check will be performed again once we have the
2334 * trusted config from the MOS.
2336 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
2337 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2338 VDEV_AUX_CORRUPT_DATA
);
2340 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
2341 "match config (%llu != %llu)", (u_longlong_t
)guid
,
2342 (u_longlong_t
)spa_guid(spa
));
2346 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
2347 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
2351 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
2352 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2353 VDEV_AUX_CORRUPT_DATA
);
2355 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2360 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
2362 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2363 VDEV_AUX_CORRUPT_DATA
);
2365 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2366 ZPOOL_CONFIG_TOP_GUID
);
2371 * If this vdev just became a top-level vdev because its sibling was
2372 * detached, it will have adopted the parent's vdev guid -- but the
2373 * label may or may not be on disk yet. Fortunately, either version
2374 * of the label will have the same top guid, so if we're a top-level
2375 * vdev, we can safely compare to that instead.
2376 * However, if the config comes from a cachefile that failed to update
2377 * after the detach, a top-level vdev will appear as a non top-level
2378 * vdev in the config. Also relax the constraints if we perform an
2381 * If we split this vdev off instead, then we also check the
2382 * original pool's guid. We don't want to consider the vdev
2383 * corrupt if it is partway through a split operation.
2385 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
2386 boolean_t mismatch
= B_FALSE
;
2387 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
2388 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
2391 if (vd
->vdev_guid
!= top_guid
&&
2392 vd
->vdev_top
->vdev_guid
!= guid
)
2397 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2398 VDEV_AUX_CORRUPT_DATA
);
2400 vdev_dbgmsg(vd
, "vdev_validate: config guid "
2401 "doesn't match label guid");
2402 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
2403 (u_longlong_t
)vd
->vdev_guid
,
2404 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
2405 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
2406 "aux_guid %llu", (u_longlong_t
)guid
,
2407 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
2412 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
2414 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2415 VDEV_AUX_CORRUPT_DATA
);
2417 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2418 ZPOOL_CONFIG_POOL_STATE
);
2425 * If this is a verbatim import, no need to check the
2426 * state of the pool.
2428 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
2429 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
2430 state
!= POOL_STATE_ACTIVE
) {
2431 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
2432 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
2433 return (SET_ERROR(EBADF
));
2437 * If we were able to open and validate a vdev that was
2438 * previously marked permanently unavailable, clear that state
2441 if (vd
->vdev_not_present
)
2442 vd
->vdev_not_present
= 0;
2448 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2451 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2452 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2453 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2454 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2455 dvd
->vdev_path
, svd
->vdev_path
);
2456 spa_strfree(dvd
->vdev_path
);
2457 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2459 } else if (svd
->vdev_path
!= NULL
) {
2460 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2461 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2462 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2466 * Our enclosure sysfs path may have changed between imports
2468 old
= dvd
->vdev_enc_sysfs_path
;
2469 new = svd
->vdev_enc_sysfs_path
;
2470 if ((old
!= NULL
&& new == NULL
) ||
2471 (old
== NULL
&& new != NULL
) ||
2472 ((old
!= NULL
&& new != NULL
) && strcmp(new, old
) != 0)) {
2473 zfs_dbgmsg("vdev_copy_path: vdev %llu: vdev_enc_sysfs_path "
2474 "changed from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2477 if (dvd
->vdev_enc_sysfs_path
)
2478 spa_strfree(dvd
->vdev_enc_sysfs_path
);
2480 if (svd
->vdev_enc_sysfs_path
) {
2481 dvd
->vdev_enc_sysfs_path
= spa_strdup(
2482 svd
->vdev_enc_sysfs_path
);
2484 dvd
->vdev_enc_sysfs_path
= NULL
;
2490 * Recursively copy vdev paths from one vdev to another. Source and destination
2491 * vdev trees must have same geometry otherwise return error. Intended to copy
2492 * paths from userland config into MOS config.
2495 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2497 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2498 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2499 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2502 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2503 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2504 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2505 return (SET_ERROR(EINVAL
));
2508 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2509 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2510 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2511 (u_longlong_t
)dvd
->vdev_guid
);
2512 return (SET_ERROR(EINVAL
));
2515 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2516 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2517 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2518 (u_longlong_t
)dvd
->vdev_children
);
2519 return (SET_ERROR(EINVAL
));
2522 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2523 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2524 dvd
->vdev_child
[i
]);
2529 if (svd
->vdev_ops
->vdev_op_leaf
)
2530 vdev_copy_path_impl(svd
, dvd
);
2536 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2538 ASSERT(stvd
->vdev_top
== stvd
);
2539 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2541 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2542 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2545 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2549 * The idea here is that while a vdev can shift positions within
2550 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2551 * step outside of it.
2553 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2555 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2558 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2560 vdev_copy_path_impl(vd
, dvd
);
2564 * Recursively copy vdev paths from one root vdev to another. Source and
2565 * destination vdev trees may differ in geometry. For each destination leaf
2566 * vdev, search a vdev with the same guid and top vdev id in the source.
2567 * Intended to copy paths from userland config into MOS config.
2570 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2572 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2573 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2574 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2576 for (uint64_t i
= 0; i
< children
; i
++) {
2577 vdev_copy_path_search(srvd
->vdev_child
[i
],
2578 drvd
->vdev_child
[i
]);
2583 * Close a virtual device.
2586 vdev_close(vdev_t
*vd
)
2588 vdev_t
*pvd
= vd
->vdev_parent
;
2589 spa_t
*spa __maybe_unused
= vd
->vdev_spa
;
2592 ASSERT(vd
->vdev_open_thread
== curthread
||
2593 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2596 * If our parent is reopening, then we are as well, unless we are
2599 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2600 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2602 vd
->vdev_ops
->vdev_op_close(vd
);
2604 vdev_cache_purge(vd
);
2607 * We record the previous state before we close it, so that if we are
2608 * doing a reopen(), we don't generate FMA ereports if we notice that
2609 * it's still faulted.
2611 vd
->vdev_prevstate
= vd
->vdev_state
;
2613 if (vd
->vdev_offline
)
2614 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2616 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2617 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2621 vdev_hold(vdev_t
*vd
)
2623 spa_t
*spa
= vd
->vdev_spa
;
2625 ASSERT(spa_is_root(spa
));
2626 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2629 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2630 vdev_hold(vd
->vdev_child
[c
]);
2632 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_hold
!= NULL
)
2633 vd
->vdev_ops
->vdev_op_hold(vd
);
2637 vdev_rele(vdev_t
*vd
)
2639 ASSERT(spa_is_root(vd
->vdev_spa
));
2640 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2641 vdev_rele(vd
->vdev_child
[c
]);
2643 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_rele
!= NULL
)
2644 vd
->vdev_ops
->vdev_op_rele(vd
);
2648 * Reopen all interior vdevs and any unopened leaves. We don't actually
2649 * reopen leaf vdevs which had previously been opened as they might deadlock
2650 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2651 * If the leaf has never been opened then open it, as usual.
2654 vdev_reopen(vdev_t
*vd
)
2656 spa_t
*spa
= vd
->vdev_spa
;
2658 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2660 /* set the reopening flag unless we're taking the vdev offline */
2661 vd
->vdev_reopening
= !vd
->vdev_offline
;
2663 (void) vdev_open(vd
);
2666 * Call vdev_validate() here to make sure we have the same device.
2667 * Otherwise, a device with an invalid label could be successfully
2668 * opened in response to vdev_reopen().
2671 (void) vdev_validate_aux(vd
);
2672 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2673 vd
->vdev_aux
== &spa
->spa_l2cache
) {
2675 * In case the vdev is present we should evict all ARC
2676 * buffers and pointers to log blocks and reclaim their
2677 * space before restoring its contents to L2ARC.
2679 if (l2arc_vdev_present(vd
)) {
2680 l2arc_rebuild_vdev(vd
, B_TRUE
);
2682 l2arc_add_vdev(spa
, vd
);
2684 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
2685 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
2688 (void) vdev_validate(vd
);
2692 * Reassess parent vdev's health.
2694 vdev_propagate_state(vd
);
2698 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2703 * Normally, partial opens (e.g. of a mirror) are allowed.
2704 * For a create, however, we want to fail the request if
2705 * there are any components we can't open.
2707 error
= vdev_open(vd
);
2709 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2711 return (error
? error
: SET_ERROR(ENXIO
));
2715 * Recursively load DTLs and initialize all labels.
2717 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2718 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2719 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2728 vdev_metaslab_set_size(vdev_t
*vd
)
2730 uint64_t asize
= vd
->vdev_asize
;
2731 uint64_t ms_count
= asize
>> zfs_vdev_default_ms_shift
;
2735 * There are two dimensions to the metaslab sizing calculation:
2736 * the size of the metaslab and the count of metaslabs per vdev.
2738 * The default values used below are a good balance between memory
2739 * usage (larger metaslab size means more memory needed for loaded
2740 * metaslabs; more metaslabs means more memory needed for the
2741 * metaslab_t structs), metaslab load time (larger metaslabs take
2742 * longer to load), and metaslab sync time (more metaslabs means
2743 * more time spent syncing all of them).
2745 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2746 * The range of the dimensions are as follows:
2748 * 2^29 <= ms_size <= 2^34
2749 * 16 <= ms_count <= 131,072
2751 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2752 * at least 512MB (2^29) to minimize fragmentation effects when
2753 * testing with smaller devices. However, the count constraint
2754 * of at least 16 metaslabs will override this minimum size goal.
2756 * On the upper end of vdev sizes, we aim for a maximum metaslab
2757 * size of 16GB. However, we will cap the total count to 2^17
2758 * metaslabs to keep our memory footprint in check and let the
2759 * metaslab size grow from there if that limit is hit.
2761 * The net effect of applying above constrains is summarized below.
2763 * vdev size metaslab count
2764 * --------------|-----------------
2766 * 8GB - 100GB one per 512MB
2768 * 3TB - 2PB one per 16GB
2770 * --------------------------------
2772 * Finally, note that all of the above calculate the initial
2773 * number of metaslabs. Expanding a top-level vdev will result
2774 * in additional metaslabs being allocated making it possible
2775 * to exceed the zfs_vdev_ms_count_limit.
2778 if (ms_count
< zfs_vdev_min_ms_count
)
2779 ms_shift
= highbit64(asize
/ zfs_vdev_min_ms_count
);
2780 else if (ms_count
> zfs_vdev_default_ms_count
)
2781 ms_shift
= highbit64(asize
/ zfs_vdev_default_ms_count
);
2783 ms_shift
= zfs_vdev_default_ms_shift
;
2785 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2786 ms_shift
= SPA_MAXBLOCKSHIFT
;
2787 } else if (ms_shift
> zfs_vdev_max_ms_shift
) {
2788 ms_shift
= zfs_vdev_max_ms_shift
;
2789 /* cap the total count to constrain memory footprint */
2790 if ((asize
>> ms_shift
) > zfs_vdev_ms_count_limit
)
2791 ms_shift
= highbit64(asize
/ zfs_vdev_ms_count_limit
);
2794 vd
->vdev_ms_shift
= ms_shift
;
2795 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2799 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2801 ASSERT(vd
== vd
->vdev_top
);
2802 /* indirect vdevs don't have metaslabs or dtls */
2803 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2804 ASSERT(ISP2(flags
));
2805 ASSERT(spa_writeable(vd
->vdev_spa
));
2807 if (flags
& VDD_METASLAB
)
2808 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2810 if (flags
& VDD_DTL
)
2811 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2813 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2817 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2819 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2820 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2822 if (vd
->vdev_ops
->vdev_op_leaf
)
2823 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2829 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2830 * the vdev has less than perfect replication. There are four kinds of DTL:
2832 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2834 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2836 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2837 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2838 * txgs that was scrubbed.
2840 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2841 * persistent errors or just some device being offline.
2842 * Unlike the other three, the DTL_OUTAGE map is not generally
2843 * maintained; it's only computed when needed, typically to
2844 * determine whether a device can be detached.
2846 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2847 * either has the data or it doesn't.
2849 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2850 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2851 * if any child is less than fully replicated, then so is its parent.
2852 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2853 * comprising only those txgs which appear in 'maxfaults' or more children;
2854 * those are the txgs we don't have enough replication to read. For example,
2855 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2856 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2857 * two child DTL_MISSING maps.
2859 * It should be clear from the above that to compute the DTLs and outage maps
2860 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2861 * Therefore, that is all we keep on disk. When loading the pool, or after
2862 * a configuration change, we generate all other DTLs from first principles.
2865 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2867 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2869 ASSERT(t
< DTL_TYPES
);
2870 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2871 ASSERT(spa_writeable(vd
->vdev_spa
));
2873 mutex_enter(&vd
->vdev_dtl_lock
);
2874 if (!range_tree_contains(rt
, txg
, size
))
2875 range_tree_add(rt
, txg
, size
);
2876 mutex_exit(&vd
->vdev_dtl_lock
);
2880 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2882 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2883 boolean_t dirty
= B_FALSE
;
2885 ASSERT(t
< DTL_TYPES
);
2886 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2889 * While we are loading the pool, the DTLs have not been loaded yet.
2890 * This isn't a problem but it can result in devices being tried
2891 * which are known to not have the data. In which case, the import
2892 * is relying on the checksum to ensure that we get the right data.
2893 * Note that while importing we are only reading the MOS, which is
2894 * always checksummed.
2896 mutex_enter(&vd
->vdev_dtl_lock
);
2897 if (!range_tree_is_empty(rt
))
2898 dirty
= range_tree_contains(rt
, txg
, size
);
2899 mutex_exit(&vd
->vdev_dtl_lock
);
2905 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2907 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2910 mutex_enter(&vd
->vdev_dtl_lock
);
2911 empty
= range_tree_is_empty(rt
);
2912 mutex_exit(&vd
->vdev_dtl_lock
);
2918 * Check if the txg falls within the range which must be
2919 * resilvered. DVAs outside this range can always be skipped.
2922 vdev_default_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2923 uint64_t phys_birth
)
2925 (void) dva
, (void) psize
;
2927 /* Set by sequential resilver. */
2928 if (phys_birth
== TXG_UNKNOWN
)
2931 return (vdev_dtl_contains(vd
, DTL_PARTIAL
, phys_birth
, 1));
2935 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered.
2938 vdev_dtl_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2939 uint64_t phys_birth
)
2941 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2943 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2944 vd
->vdev_ops
->vdev_op_leaf
)
2947 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, dva
, psize
,
2952 * Returns the lowest txg in the DTL range.
2955 vdev_dtl_min(vdev_t
*vd
)
2957 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2958 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2959 ASSERT0(vd
->vdev_children
);
2961 return (range_tree_min(vd
->vdev_dtl
[DTL_MISSING
]) - 1);
2965 * Returns the highest txg in the DTL.
2968 vdev_dtl_max(vdev_t
*vd
)
2970 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2971 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2972 ASSERT0(vd
->vdev_children
);
2974 return (range_tree_max(vd
->vdev_dtl
[DTL_MISSING
]));
2978 * Determine if a resilvering vdev should remove any DTL entries from
2979 * its range. If the vdev was resilvering for the entire duration of the
2980 * scan then it should excise that range from its DTLs. Otherwise, this
2981 * vdev is considered partially resilvered and should leave its DTL
2982 * entries intact. The comment in vdev_dtl_reassess() describes how we
2986 vdev_dtl_should_excise(vdev_t
*vd
, boolean_t rebuild_done
)
2988 ASSERT0(vd
->vdev_children
);
2990 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2993 if (vd
->vdev_resilver_deferred
)
2996 if (range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
3000 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3001 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
3003 /* Rebuild not initiated by attach */
3004 if (vd
->vdev_rebuild_txg
== 0)
3008 * When a rebuild completes without error then all missing data
3009 * up to the rebuild max txg has been reconstructed and the DTL
3010 * is eligible for excision.
3012 if (vrp
->vrp_rebuild_state
== VDEV_REBUILD_COMPLETE
&&
3013 vdev_dtl_max(vd
) <= vrp
->vrp_max_txg
) {
3014 ASSERT3U(vrp
->vrp_min_txg
, <=, vdev_dtl_min(vd
));
3015 ASSERT3U(vrp
->vrp_min_txg
, <, vd
->vdev_rebuild_txg
);
3016 ASSERT3U(vd
->vdev_rebuild_txg
, <=, vrp
->vrp_max_txg
);
3020 dsl_scan_t
*scn
= vd
->vdev_spa
->spa_dsl_pool
->dp_scan
;
3021 dsl_scan_phys_t
*scnp __maybe_unused
= &scn
->scn_phys
;
3023 /* Resilver not initiated by attach */
3024 if (vd
->vdev_resilver_txg
== 0)
3028 * When a resilver is initiated the scan will assign the
3029 * scn_max_txg value to the highest txg value that exists
3030 * in all DTLs. If this device's max DTL is not part of this
3031 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
3032 * then it is not eligible for excision.
3034 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
3035 ASSERT3U(scnp
->scn_min_txg
, <=, vdev_dtl_min(vd
));
3036 ASSERT3U(scnp
->scn_min_txg
, <, vd
->vdev_resilver_txg
);
3037 ASSERT3U(vd
->vdev_resilver_txg
, <=, scnp
->scn_max_txg
);
3046 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
3047 * write operations will be issued to the pool.
3050 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
,
3051 boolean_t scrub_done
, boolean_t rebuild_done
)
3053 spa_t
*spa
= vd
->vdev_spa
;
3057 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
3059 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3060 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
3061 scrub_txg
, scrub_done
, rebuild_done
);
3063 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
3066 if (vd
->vdev_ops
->vdev_op_leaf
) {
3067 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
3068 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3069 boolean_t check_excise
= B_FALSE
;
3070 boolean_t wasempty
= B_TRUE
;
3072 mutex_enter(&vd
->vdev_dtl_lock
);
3075 * If requested, pretend the scan or rebuild completed cleanly.
3077 if (zfs_scan_ignore_errors
) {
3079 scn
->scn_phys
.scn_errors
= 0;
3081 vr
->vr_rebuild_phys
.vrp_errors
= 0;
3084 if (scrub_txg
!= 0 &&
3085 !range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3087 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
3088 "dtl:%llu/%llu errors:%llu",
3089 (u_longlong_t
)vd
->vdev_guid
, (u_longlong_t
)txg
,
3090 (u_longlong_t
)scrub_txg
, spa
->spa_scrub_started
,
3091 (u_longlong_t
)vdev_dtl_min(vd
),
3092 (u_longlong_t
)vdev_dtl_max(vd
),
3093 (u_longlong_t
)(scn
? scn
->scn_phys
.scn_errors
: 0));
3097 * If we've completed a scrub/resilver or a rebuild cleanly
3098 * then determine if this vdev should remove any DTLs. We
3099 * only want to excise regions on vdevs that were available
3100 * during the entire duration of this scan.
3103 vr
!= NULL
&& vr
->vr_rebuild_phys
.vrp_errors
== 0) {
3104 check_excise
= B_TRUE
;
3106 if (spa
->spa_scrub_started
||
3107 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) {
3108 check_excise
= B_TRUE
;
3112 if (scrub_txg
&& check_excise
&&
3113 vdev_dtl_should_excise(vd
, rebuild_done
)) {
3115 * We completed a scrub, resilver or rebuild up to
3116 * scrub_txg. If we did it without rebooting, then
3117 * the scrub dtl will be valid, so excise the old
3118 * region and fold in the scrub dtl. Otherwise,
3119 * leave the dtl as-is if there was an error.
3121 * There's little trick here: to excise the beginning
3122 * of the DTL_MISSING map, we put it into a reference
3123 * tree and then add a segment with refcnt -1 that
3124 * covers the range [0, scrub_txg). This means
3125 * that each txg in that range has refcnt -1 or 0.
3126 * We then add DTL_SCRUB with a refcnt of 2, so that
3127 * entries in the range [0, scrub_txg) will have a
3128 * positive refcnt -- either 1 or 2. We then convert
3129 * the reference tree into the new DTL_MISSING map.
3131 space_reftree_create(&reftree
);
3132 space_reftree_add_map(&reftree
,
3133 vd
->vdev_dtl
[DTL_MISSING
], 1);
3134 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
3135 space_reftree_add_map(&reftree
,
3136 vd
->vdev_dtl
[DTL_SCRUB
], 2);
3137 space_reftree_generate_map(&reftree
,
3138 vd
->vdev_dtl
[DTL_MISSING
], 1);
3139 space_reftree_destroy(&reftree
);
3141 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3142 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
3143 (u_longlong_t
)vdev_dtl_min(vd
),
3144 (u_longlong_t
)vdev_dtl_max(vd
));
3145 } else if (!wasempty
) {
3146 zfs_dbgmsg("DTL_MISSING is now empty");
3149 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
3150 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3151 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
3153 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
3154 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
3155 if (!vdev_readable(vd
))
3156 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
3158 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3159 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
3162 * If the vdev was resilvering or rebuilding and no longer
3163 * has any DTLs then reset the appropriate flag and dirty
3164 * the top level so that we persist the change.
3167 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3168 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
3169 if (vd
->vdev_rebuild_txg
!= 0) {
3170 vd
->vdev_rebuild_txg
= 0;
3171 vdev_config_dirty(vd
->vdev_top
);
3172 } else if (vd
->vdev_resilver_txg
!= 0) {
3173 vd
->vdev_resilver_txg
= 0;
3174 vdev_config_dirty(vd
->vdev_top
);
3178 mutex_exit(&vd
->vdev_dtl_lock
);
3181 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
3185 mutex_enter(&vd
->vdev_dtl_lock
);
3186 for (int t
= 0; t
< DTL_TYPES
; t
++) {
3187 /* account for child's outage in parent's missing map */
3188 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
3190 continue; /* leaf vdevs only */
3191 if (t
== DTL_PARTIAL
)
3192 minref
= 1; /* i.e. non-zero */
3193 else if (vdev_get_nparity(vd
) != 0)
3194 minref
= vdev_get_nparity(vd
) + 1; /* RAID-Z, dRAID */
3196 minref
= vd
->vdev_children
; /* any kind of mirror */
3197 space_reftree_create(&reftree
);
3198 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3199 vdev_t
*cvd
= vd
->vdev_child
[c
];
3200 mutex_enter(&cvd
->vdev_dtl_lock
);
3201 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
3202 mutex_exit(&cvd
->vdev_dtl_lock
);
3204 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
3205 space_reftree_destroy(&reftree
);
3207 mutex_exit(&vd
->vdev_dtl_lock
);
3211 * Iterate over all the vdevs except spare, and post kobj events
3214 vdev_post_kobj_evt(vdev_t
*vd
)
3216 if (vd
->vdev_ops
->vdev_op_kobj_evt_post
&&
3217 vd
->vdev_kobj_flag
== B_FALSE
) {
3218 vd
->vdev_kobj_flag
= B_TRUE
;
3219 vd
->vdev_ops
->vdev_op_kobj_evt_post(vd
);
3222 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3223 vdev_post_kobj_evt(vd
->vdev_child
[c
]);
3227 * Iterate over all the vdevs except spare, and clear kobj events
3230 vdev_clear_kobj_evt(vdev_t
*vd
)
3232 vd
->vdev_kobj_flag
= B_FALSE
;
3234 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3235 vdev_clear_kobj_evt(vd
->vdev_child
[c
]);
3239 vdev_dtl_load(vdev_t
*vd
)
3241 spa_t
*spa
= vd
->vdev_spa
;
3242 objset_t
*mos
= spa
->spa_meta_objset
;
3246 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
3247 ASSERT(vdev_is_concrete(vd
));
3250 * If the dtl cannot be sync'd there is no need to open it.
3252 if (spa
->spa_mode
== SPA_MODE_READ
&& !spa
->spa_read_spacemaps
)
3255 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
3256 vd
->vdev_dtl_object
, 0, -1ULL, 0);
3259 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3261 rt
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3262 error
= space_map_load(vd
->vdev_dtl_sm
, rt
, SM_ALLOC
);
3264 mutex_enter(&vd
->vdev_dtl_lock
);
3265 range_tree_walk(rt
, range_tree_add
,
3266 vd
->vdev_dtl
[DTL_MISSING
]);
3267 mutex_exit(&vd
->vdev_dtl_lock
);
3270 range_tree_vacate(rt
, NULL
, NULL
);
3271 range_tree_destroy(rt
);
3276 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3277 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
3286 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3288 spa_t
*spa
= vd
->vdev_spa
;
3289 objset_t
*mos
= spa
->spa_meta_objset
;
3290 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
3293 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
3296 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
3297 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
3298 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
3300 ASSERT(string
!= NULL
);
3301 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
3302 1, strlen(string
) + 1, string
, tx
));
3304 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
3305 spa_activate_allocation_classes(spa
, tx
);
3310 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
3312 spa_t
*spa
= vd
->vdev_spa
;
3314 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
3315 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3320 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
3322 spa_t
*spa
= vd
->vdev_spa
;
3323 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
3324 DMU_OT_NONE
, 0, tx
);
3327 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3334 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3336 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
3337 vd
->vdev_ops
!= &vdev_missing_ops
&&
3338 vd
->vdev_ops
!= &vdev_root_ops
&&
3339 !vd
->vdev_top
->vdev_removing
) {
3340 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
3341 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
3343 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
3344 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
3345 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
3346 vdev_zap_allocation_data(vd
, tx
);
3350 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3351 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
3356 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
3358 spa_t
*spa
= vd
->vdev_spa
;
3359 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
3360 objset_t
*mos
= spa
->spa_meta_objset
;
3361 range_tree_t
*rtsync
;
3363 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
3365 ASSERT(vdev_is_concrete(vd
));
3366 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3368 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3370 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
3371 mutex_enter(&vd
->vdev_dtl_lock
);
3372 space_map_free(vd
->vdev_dtl_sm
, tx
);
3373 space_map_close(vd
->vdev_dtl_sm
);
3374 vd
->vdev_dtl_sm
= NULL
;
3375 mutex_exit(&vd
->vdev_dtl_lock
);
3378 * We only destroy the leaf ZAP for detached leaves or for
3379 * removed log devices. Removed data devices handle leaf ZAP
3380 * cleanup later, once cancellation is no longer possible.
3382 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
3383 vd
->vdev_top
->vdev_islog
)) {
3384 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
3385 vd
->vdev_leaf_zap
= 0;
3392 if (vd
->vdev_dtl_sm
== NULL
) {
3393 uint64_t new_object
;
3395 new_object
= space_map_alloc(mos
, zfs_vdev_dtl_sm_blksz
, tx
);
3396 VERIFY3U(new_object
, !=, 0);
3398 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
3400 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3403 rtsync
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3405 mutex_enter(&vd
->vdev_dtl_lock
);
3406 range_tree_walk(rt
, range_tree_add
, rtsync
);
3407 mutex_exit(&vd
->vdev_dtl_lock
);
3409 space_map_truncate(vd
->vdev_dtl_sm
, zfs_vdev_dtl_sm_blksz
, tx
);
3410 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
3411 range_tree_vacate(rtsync
, NULL
, NULL
);
3413 range_tree_destroy(rtsync
);
3416 * If the object for the space map has changed then dirty
3417 * the top level so that we update the config.
3419 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
3420 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
3421 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
3422 (u_longlong_t
)object
,
3423 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
3424 vdev_config_dirty(vd
->vdev_top
);
3431 * Determine whether the specified vdev can be offlined/detached/removed
3432 * without losing data.
3435 vdev_dtl_required(vdev_t
*vd
)
3437 spa_t
*spa
= vd
->vdev_spa
;
3438 vdev_t
*tvd
= vd
->vdev_top
;
3439 uint8_t cant_read
= vd
->vdev_cant_read
;
3442 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3444 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
3448 * Temporarily mark the device as unreadable, and then determine
3449 * whether this results in any DTL outages in the top-level vdev.
3450 * If not, we can safely offline/detach/remove the device.
3452 vd
->vdev_cant_read
= B_TRUE
;
3453 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3454 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
3455 vd
->vdev_cant_read
= cant_read
;
3456 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3458 if (!required
&& zio_injection_enabled
) {
3459 required
= !!zio_handle_device_injection(vd
, NULL
,
3467 * Determine if resilver is needed, and if so the txg range.
3470 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
3472 boolean_t needed
= B_FALSE
;
3473 uint64_t thismin
= UINT64_MAX
;
3474 uint64_t thismax
= 0;
3476 if (vd
->vdev_children
== 0) {
3477 mutex_enter(&vd
->vdev_dtl_lock
);
3478 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3479 vdev_writeable(vd
)) {
3481 thismin
= vdev_dtl_min(vd
);
3482 thismax
= vdev_dtl_max(vd
);
3485 mutex_exit(&vd
->vdev_dtl_lock
);
3487 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3488 vdev_t
*cvd
= vd
->vdev_child
[c
];
3489 uint64_t cmin
, cmax
;
3491 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
3492 thismin
= MIN(thismin
, cmin
);
3493 thismax
= MAX(thismax
, cmax
);
3499 if (needed
&& minp
) {
3507 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3508 * will contain either the checkpoint spacemap object or zero if none exists.
3509 * All other errors are returned to the caller.
3512 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
3514 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
3516 if (vd
->vdev_top_zap
== 0) {
3521 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
3522 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
3523 if (error
== ENOENT
) {
3532 vdev_load(vdev_t
*vd
)
3534 int children
= vd
->vdev_children
;
3539 * It's only worthwhile to use the taskq for the root vdev, because the
3540 * slow part is metaslab_init, and that only happens for top-level
3543 if (vd
->vdev_ops
== &vdev_root_ops
&& vd
->vdev_children
> 0) {
3544 tq
= taskq_create("vdev_load", children
, minclsyspri
,
3545 children
, children
, TASKQ_PREPOPULATE
);
3549 * Recursively load all children.
3551 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3552 vdev_t
*cvd
= vd
->vdev_child
[c
];
3554 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
3555 cvd
->vdev_load_error
= vdev_load(cvd
);
3557 VERIFY(taskq_dispatch(tq
, vdev_load_child
,
3558 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
3567 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3568 int error
= vd
->vdev_child
[c
]->vdev_load_error
;
3574 vdev_set_deflate_ratio(vd
);
3577 * On spa_load path, grab the allocation bias from our zap
3579 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3580 spa_t
*spa
= vd
->vdev_spa
;
3583 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3584 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
3587 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
3588 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
3589 } else if (error
!= ENOENT
) {
3590 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3591 VDEV_AUX_CORRUPT_DATA
);
3592 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(top_zap=%llu) "
3593 "failed [error=%d]",
3594 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3599 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3600 spa_t
*spa
= vd
->vdev_spa
;
3603 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3604 vdev_prop_to_name(VDEV_PROP_FAILFAST
), sizeof (failfast
),
3607 vd
->vdev_failfast
= failfast
& 1;
3608 } else if (error
== ENOENT
) {
3609 vd
->vdev_failfast
= vdev_prop_default_numeric(
3610 VDEV_PROP_FAILFAST
);
3613 "vdev_load: zap_lookup(top_zap=%llu) "
3614 "failed [error=%d]",
3615 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3620 * Load any rebuild state from the top-level vdev zap.
3622 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3623 error
= vdev_rebuild_load(vd
);
3624 if (error
&& error
!= ENOTSUP
) {
3625 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3626 VDEV_AUX_CORRUPT_DATA
);
3627 vdev_dbgmsg(vd
, "vdev_load: vdev_rebuild_load "
3628 "failed [error=%d]", error
);
3633 if (vd
->vdev_top_zap
!= 0 || vd
->vdev_leaf_zap
!= 0) {
3636 if (vd
->vdev_top_zap
!= 0)
3637 zapobj
= vd
->vdev_top_zap
;
3639 zapobj
= vd
->vdev_leaf_zap
;
3641 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_N
,
3642 &vd
->vdev_checksum_n
);
3643 if (error
&& error
!= ENOENT
)
3644 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3645 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3647 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_T
,
3648 &vd
->vdev_checksum_t
);
3649 if (error
&& error
!= ENOENT
)
3650 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3651 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3653 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_N
,
3655 if (error
&& error
!= ENOENT
)
3656 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3657 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3659 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_T
,
3661 if (error
&& error
!= ENOENT
)
3662 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3663 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3667 * If this is a top-level vdev, initialize its metaslabs.
3669 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
3670 vdev_metaslab_group_create(vd
);
3672 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
3673 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3674 VDEV_AUX_CORRUPT_DATA
);
3675 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
3676 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
3677 (u_longlong_t
)vd
->vdev_asize
);
3678 return (SET_ERROR(ENXIO
));
3681 error
= vdev_metaslab_init(vd
, 0);
3683 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
3684 "[error=%d]", error
);
3685 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3686 VDEV_AUX_CORRUPT_DATA
);
3690 uint64_t checkpoint_sm_obj
;
3691 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
3692 if (error
== 0 && checkpoint_sm_obj
!= 0) {
3693 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3694 ASSERT(vd
->vdev_asize
!= 0);
3695 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
3697 error
= space_map_open(&vd
->vdev_checkpoint_sm
,
3698 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
3701 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3702 "failed for checkpoint spacemap (obj %llu) "
3704 (u_longlong_t
)checkpoint_sm_obj
, error
);
3707 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3710 * Since the checkpoint_sm contains free entries
3711 * exclusively we can use space_map_allocated() to
3712 * indicate the cumulative checkpointed space that
3715 vd
->vdev_stat
.vs_checkpoint_space
=
3716 -space_map_allocated(vd
->vdev_checkpoint_sm
);
3717 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3718 vd
->vdev_stat
.vs_checkpoint_space
;
3719 } else if (error
!= 0) {
3720 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3721 "checkpoint space map object from vdev ZAP "
3722 "[error=%d]", error
);
3728 * If this is a leaf vdev, load its DTL.
3730 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3731 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3732 VDEV_AUX_CORRUPT_DATA
);
3733 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3734 "[error=%d]", error
);
3738 uint64_t obsolete_sm_object
;
3739 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3740 if (error
== 0 && obsolete_sm_object
!= 0) {
3741 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3742 ASSERT(vd
->vdev_asize
!= 0);
3743 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3745 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3746 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3747 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3748 VDEV_AUX_CORRUPT_DATA
);
3749 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3750 "obsolete spacemap (obj %llu) [error=%d]",
3751 (u_longlong_t
)obsolete_sm_object
, error
);
3754 } else if (error
!= 0) {
3755 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve obsolete "
3756 "space map object from vdev ZAP [error=%d]", error
);
3764 * The special vdev case is used for hot spares and l2cache devices. Its
3765 * sole purpose it to set the vdev state for the associated vdev. To do this,
3766 * we make sure that we can open the underlying device, then try to read the
3767 * label, and make sure that the label is sane and that it hasn't been
3768 * repurposed to another pool.
3771 vdev_validate_aux(vdev_t
*vd
)
3774 uint64_t guid
, version
;
3777 if (!vdev_readable(vd
))
3780 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
3781 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3782 VDEV_AUX_CORRUPT_DATA
);
3786 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
3787 !SPA_VERSION_IS_SUPPORTED(version
) ||
3788 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
3789 guid
!= vd
->vdev_guid
||
3790 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
3791 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3792 VDEV_AUX_CORRUPT_DATA
);
3798 * We don't actually check the pool state here. If it's in fact in
3799 * use by another pool, we update this fact on the fly when requested.
3806 vdev_destroy_ms_flush_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3808 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3810 if (vd
->vdev_top_zap
== 0)
3813 uint64_t object
= 0;
3814 int err
= zap_lookup(mos
, vd
->vdev_top_zap
,
3815 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, sizeof (uint64_t), 1, &object
);
3820 VERIFY0(dmu_object_free(mos
, object
, tx
));
3821 VERIFY0(zap_remove(mos
, vd
->vdev_top_zap
,
3822 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, tx
));
3826 * Free the objects used to store this vdev's spacemaps, and the array
3827 * that points to them.
3830 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3832 if (vd
->vdev_ms_array
== 0)
3835 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3836 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
3837 size_t array_bytes
= array_count
* sizeof (uint64_t);
3838 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
3839 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
3840 array_bytes
, smobj_array
, 0));
3842 for (uint64_t i
= 0; i
< array_count
; i
++) {
3843 uint64_t smobj
= smobj_array
[i
];
3847 space_map_free_obj(mos
, smobj
, tx
);
3850 kmem_free(smobj_array
, array_bytes
);
3851 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
3852 vdev_destroy_ms_flush_data(vd
, tx
);
3853 vd
->vdev_ms_array
= 0;
3857 vdev_remove_empty_log(vdev_t
*vd
, uint64_t txg
)
3859 spa_t
*spa
= vd
->vdev_spa
;
3861 ASSERT(vd
->vdev_islog
);
3862 ASSERT(vd
== vd
->vdev_top
);
3863 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
3865 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3867 vdev_destroy_spacemaps(vd
, tx
);
3868 if (vd
->vdev_top_zap
!= 0) {
3869 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3870 vd
->vdev_top_zap
= 0;
3877 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3880 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3882 ASSERT(vdev_is_concrete(vd
));
3884 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
3886 metaslab_sync_done(msp
, txg
);
3889 metaslab_sync_reassess(vd
->vdev_mg
);
3890 if (vd
->vdev_log_mg
!= NULL
)
3891 metaslab_sync_reassess(vd
->vdev_log_mg
);
3896 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3898 spa_t
*spa
= vd
->vdev_spa
;
3902 ASSERT3U(txg
, ==, spa
->spa_syncing_txg
);
3903 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3904 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3905 ASSERT(vd
->vdev_removing
||
3906 vd
->vdev_ops
== &vdev_indirect_ops
);
3908 vdev_indirect_sync_obsolete(vd
, tx
);
3911 * If the vdev is indirect, it can't have dirty
3912 * metaslabs or DTLs.
3914 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3915 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3916 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3922 ASSERT(vdev_is_concrete(vd
));
3924 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3925 !vd
->vdev_removing
) {
3926 ASSERT(vd
== vd
->vdev_top
);
3927 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3928 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3929 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3930 ASSERT(vd
->vdev_ms_array
!= 0);
3931 vdev_config_dirty(vd
);
3934 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3935 metaslab_sync(msp
, txg
);
3936 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3939 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3940 vdev_dtl_sync(lvd
, txg
);
3943 * If this is an empty log device being removed, destroy the
3944 * metadata associated with it.
3946 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3947 vdev_remove_empty_log(vd
, txg
);
3949 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3954 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3956 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3960 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3961 * not be opened, and no I/O is attempted.
3964 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3968 spa_vdev_state_enter(spa
, SCL_NONE
);
3970 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3971 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3973 if (!vd
->vdev_ops
->vdev_op_leaf
)
3974 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
3979 * If user did a 'zpool offline -f' then make the fault persist across
3982 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3984 * There are two kinds of forced faults: temporary and
3985 * persistent. Temporary faults go away at pool import, while
3986 * persistent faults stay set. Both types of faults can be
3987 * cleared with a zpool clear.
3989 * We tell if a vdev is persistently faulted by looking at the
3990 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3991 * import then it's a persistent fault. Otherwise, it's
3992 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3993 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3994 * tells vdev_config_generate() (which gets run later) to set
3995 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3997 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3998 vd
->vdev_tmpoffline
= B_FALSE
;
3999 aux
= VDEV_AUX_EXTERNAL
;
4001 vd
->vdev_tmpoffline
= B_TRUE
;
4005 * We don't directly use the aux state here, but if we do a
4006 * vdev_reopen(), we need this value to be present to remember why we
4009 vd
->vdev_label_aux
= aux
;
4012 * Faulted state takes precedence over degraded.
4014 vd
->vdev_delayed_close
= B_FALSE
;
4015 vd
->vdev_faulted
= 1ULL;
4016 vd
->vdev_degraded
= 0ULL;
4017 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
4020 * If this device has the only valid copy of the data, then
4021 * back off and simply mark the vdev as degraded instead.
4023 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
4024 vd
->vdev_degraded
= 1ULL;
4025 vd
->vdev_faulted
= 0ULL;
4028 * If we reopen the device and it's not dead, only then do we
4033 if (vdev_readable(vd
))
4034 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
4037 return (spa_vdev_state_exit(spa
, vd
, 0));
4041 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
4042 * user that something is wrong. The vdev continues to operate as normal as far
4043 * as I/O is concerned.
4046 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
4050 spa_vdev_state_enter(spa
, SCL_NONE
);
4052 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4053 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4055 if (!vd
->vdev_ops
->vdev_op_leaf
)
4056 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4059 * If the vdev is already faulted, then don't do anything.
4061 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
4062 return (spa_vdev_state_exit(spa
, NULL
, 0));
4064 vd
->vdev_degraded
= 1ULL;
4065 if (!vdev_is_dead(vd
))
4066 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
4069 return (spa_vdev_state_exit(spa
, vd
, 0));
4073 vdev_remove_wanted(spa_t
*spa
, uint64_t guid
)
4077 spa_vdev_state_enter(spa
, SCL_NONE
);
4079 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4080 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4083 * If the vdev is already removed, then don't do anything.
4085 if (vd
->vdev_removed
)
4086 return (spa_vdev_state_exit(spa
, NULL
, 0));
4088 vd
->vdev_remove_wanted
= B_TRUE
;
4089 spa_async_request(spa
, SPA_ASYNC_REMOVE
);
4091 return (spa_vdev_state_exit(spa
, vd
, 0));
4096 * Online the given vdev.
4098 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
4099 * spare device should be detached when the device finishes resilvering.
4100 * Second, the online should be treated like a 'test' online case, so no FMA
4101 * events are generated if the device fails to open.
4104 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
4106 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
4107 boolean_t wasoffline
;
4108 vdev_state_t oldstate
;
4110 spa_vdev_state_enter(spa
, SCL_NONE
);
4112 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4113 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4115 if (!vd
->vdev_ops
->vdev_op_leaf
)
4116 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4118 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
4119 oldstate
= vd
->vdev_state
;
4122 vd
->vdev_offline
= B_FALSE
;
4123 vd
->vdev_tmpoffline
= B_FALSE
;
4124 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
4125 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
4127 /* XXX - L2ARC 1.0 does not support expansion */
4128 if (!vd
->vdev_aux
) {
4129 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4130 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
4131 spa
->spa_autoexpand
);
4132 vd
->vdev_expansion_time
= gethrestime_sec();
4136 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
4138 if (!vd
->vdev_aux
) {
4139 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4140 pvd
->vdev_expanding
= B_FALSE
;
4144 *newstate
= vd
->vdev_state
;
4145 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
4146 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
4147 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4148 vd
->vdev_parent
->vdev_child
[0] == vd
)
4149 vd
->vdev_unspare
= B_TRUE
;
4151 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
4153 /* XXX - L2ARC 1.0 does not support expansion */
4155 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
4156 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4159 /* Restart initializing if necessary */
4160 mutex_enter(&vd
->vdev_initialize_lock
);
4161 if (vdev_writeable(vd
) &&
4162 vd
->vdev_initialize_thread
== NULL
&&
4163 vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) {
4164 (void) vdev_initialize(vd
);
4166 mutex_exit(&vd
->vdev_initialize_lock
);
4169 * Restart trimming if necessary. We do not restart trimming for cache
4170 * devices here. This is triggered by l2arc_rebuild_vdev()
4171 * asynchronously for the whole device or in l2arc_evict() as it evicts
4172 * space for upcoming writes.
4174 mutex_enter(&vd
->vdev_trim_lock
);
4175 if (vdev_writeable(vd
) && !vd
->vdev_isl2cache
&&
4176 vd
->vdev_trim_thread
== NULL
&&
4177 vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
) {
4178 (void) vdev_trim(vd
, vd
->vdev_trim_rate
, vd
->vdev_trim_partial
,
4179 vd
->vdev_trim_secure
);
4181 mutex_exit(&vd
->vdev_trim_lock
);
4184 (oldstate
< VDEV_STATE_DEGRADED
&&
4185 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
4186 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
4188 return (spa_vdev_state_exit(spa
, vd
, 0));
4192 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4196 uint64_t generation
;
4197 metaslab_group_t
*mg
;
4200 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4202 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4203 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4205 if (!vd
->vdev_ops
->vdev_op_leaf
)
4206 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4208 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
4209 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
4213 generation
= spa
->spa_config_generation
+ 1;
4216 * If the device isn't already offline, try to offline it.
4218 if (!vd
->vdev_offline
) {
4220 * If this device has the only valid copy of some data,
4221 * don't allow it to be offlined. Log devices are always
4224 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4225 vdev_dtl_required(vd
))
4226 return (spa_vdev_state_exit(spa
, NULL
,
4230 * If the top-level is a slog and it has had allocations
4231 * then proceed. We check that the vdev's metaslab group
4232 * is not NULL since it's possible that we may have just
4233 * added this vdev but not yet initialized its metaslabs.
4235 if (tvd
->vdev_islog
&& mg
!= NULL
) {
4237 * Prevent any future allocations.
4239 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
4240 metaslab_group_passivate(mg
);
4241 (void) spa_vdev_state_exit(spa
, vd
, 0);
4243 error
= spa_reset_logs(spa
);
4246 * If the log device was successfully reset but has
4247 * checkpointed data, do not offline it.
4250 tvd
->vdev_checkpoint_sm
!= NULL
) {
4251 ASSERT3U(space_map_allocated(
4252 tvd
->vdev_checkpoint_sm
), !=, 0);
4253 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
4256 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4259 * Check to see if the config has changed.
4261 if (error
|| generation
!= spa
->spa_config_generation
) {
4262 metaslab_group_activate(mg
);
4264 return (spa_vdev_state_exit(spa
,
4266 (void) spa_vdev_state_exit(spa
, vd
, 0);
4269 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
4273 * Offline this device and reopen its top-level vdev.
4274 * If the top-level vdev is a log device then just offline
4275 * it. Otherwise, if this action results in the top-level
4276 * vdev becoming unusable, undo it and fail the request.
4278 vd
->vdev_offline
= B_TRUE
;
4281 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4282 vdev_is_dead(tvd
)) {
4283 vd
->vdev_offline
= B_FALSE
;
4285 return (spa_vdev_state_exit(spa
, NULL
,
4290 * Add the device back into the metaslab rotor so that
4291 * once we online the device it's open for business.
4293 if (tvd
->vdev_islog
&& mg
!= NULL
)
4294 metaslab_group_activate(mg
);
4297 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
4299 return (spa_vdev_state_exit(spa
, vd
, 0));
4303 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4307 mutex_enter(&spa
->spa_vdev_top_lock
);
4308 error
= vdev_offline_locked(spa
, guid
, flags
);
4309 mutex_exit(&spa
->spa_vdev_top_lock
);
4315 * Clear the error counts associated with this vdev. Unlike vdev_online() and
4316 * vdev_offline(), we assume the spa config is locked. We also clear all
4317 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
4320 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
4322 vdev_t
*rvd
= spa
->spa_root_vdev
;
4324 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
4329 vd
->vdev_stat
.vs_read_errors
= 0;
4330 vd
->vdev_stat
.vs_write_errors
= 0;
4331 vd
->vdev_stat
.vs_checksum_errors
= 0;
4332 vd
->vdev_stat
.vs_slow_ios
= 0;
4334 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4335 vdev_clear(spa
, vd
->vdev_child
[c
]);
4338 * It makes no sense to "clear" an indirect or removed vdev.
4340 if (!vdev_is_concrete(vd
) || vd
->vdev_removed
)
4344 * If we're in the FAULTED state or have experienced failed I/O, then
4345 * clear the persistent state and attempt to reopen the device. We
4346 * also mark the vdev config dirty, so that the new faulted state is
4347 * written out to disk.
4349 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
4350 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
4352 * When reopening in response to a clear event, it may be due to
4353 * a fmadm repair request. In this case, if the device is
4354 * still broken, we want to still post the ereport again.
4356 vd
->vdev_forcefault
= B_TRUE
;
4358 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
4359 vd
->vdev_cant_read
= B_FALSE
;
4360 vd
->vdev_cant_write
= B_FALSE
;
4361 vd
->vdev_stat
.vs_aux
= 0;
4363 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
4365 vd
->vdev_forcefault
= B_FALSE
;
4367 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
4368 vdev_state_dirty(vd
->vdev_top
);
4370 /* If a resilver isn't required, check if vdevs can be culled */
4371 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
) &&
4372 !dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4373 !dsl_scan_resilver_scheduled(spa
->spa_dsl_pool
))
4374 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
4376 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
4380 * When clearing a FMA-diagnosed fault, we always want to
4381 * unspare the device, as we assume that the original spare was
4382 * done in response to the FMA fault.
4384 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
4385 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4386 vd
->vdev_parent
->vdev_child
[0] == vd
)
4387 vd
->vdev_unspare
= B_TRUE
;
4389 /* Clear recent error events cache (i.e. duplicate events tracking) */
4390 zfs_ereport_clear(spa
, vd
);
4394 vdev_is_dead(vdev_t
*vd
)
4397 * Holes and missing devices are always considered "dead".
4398 * This simplifies the code since we don't have to check for
4399 * these types of devices in the various code paths.
4400 * Instead we rely on the fact that we skip over dead devices
4401 * before issuing I/O to them.
4403 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
4404 vd
->vdev_ops
== &vdev_hole_ops
||
4405 vd
->vdev_ops
== &vdev_missing_ops
);
4409 vdev_readable(vdev_t
*vd
)
4411 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
4415 vdev_writeable(vdev_t
*vd
)
4417 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
4418 vdev_is_concrete(vd
));
4422 vdev_allocatable(vdev_t
*vd
)
4424 uint64_t state
= vd
->vdev_state
;
4427 * We currently allow allocations from vdevs which may be in the
4428 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
4429 * fails to reopen then we'll catch it later when we're holding
4430 * the proper locks. Note that we have to get the vdev state
4431 * in a local variable because although it changes atomically,
4432 * we're asking two separate questions about it.
4434 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
4435 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
4436 vd
->vdev_mg
->mg_initialized
);
4440 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
4442 ASSERT(zio
->io_vd
== vd
);
4444 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
4447 if (zio
->io_type
== ZIO_TYPE_READ
)
4448 return (!vd
->vdev_cant_read
);
4450 if (zio
->io_type
== ZIO_TYPE_WRITE
)
4451 return (!vd
->vdev_cant_write
);
4457 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
4460 * Exclude the dRAID spare when aggregating to avoid double counting
4461 * the ops and bytes. These IOs are counted by the physical leaves.
4463 if (cvd
->vdev_ops
== &vdev_draid_spare_ops
)
4466 for (int t
= 0; t
< VS_ZIO_TYPES
; t
++) {
4467 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
4468 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
4471 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
4475 * Get extended stats
4478 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
4483 for (t
= 0; t
< ZIO_TYPES
; t
++) {
4484 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
4485 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
4487 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
4488 vsx
->vsx_total_histo
[t
][b
] +=
4489 cvsx
->vsx_total_histo
[t
][b
];
4493 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
4494 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
4495 vsx
->vsx_queue_histo
[t
][b
] +=
4496 cvsx
->vsx_queue_histo
[t
][b
];
4498 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
4499 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
4501 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
4502 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
4504 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
4505 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
4511 vdev_is_spacemap_addressable(vdev_t
*vd
)
4513 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_SPACEMAP_V2
))
4517 * If double-word space map entries are not enabled we assume
4518 * 47 bits of the space map entry are dedicated to the entry's
4519 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4520 * to calculate the maximum address that can be described by a
4521 * space map entry for the given device.
4523 uint64_t shift
= vd
->vdev_ashift
+ SM_OFFSET_BITS
;
4525 if (shift
>= 63) /* detect potential overflow */
4528 return (vd
->vdev_asize
< (1ULL << shift
));
4532 * Get statistics for the given vdev.
4535 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4539 * If we're getting stats on the root vdev, aggregate the I/O counts
4540 * over all top-level vdevs (i.e. the direct children of the root).
4542 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4544 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
4545 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
4548 memset(vsx
, 0, sizeof (*vsx
));
4550 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4551 vdev_t
*cvd
= vd
->vdev_child
[c
];
4552 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
4553 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
4555 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
4557 vdev_get_child_stat(cvd
, vs
, cvs
);
4559 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
4563 * We're a leaf. Just copy our ZIO active queue stats in. The
4564 * other leaf stats are updated in vdev_stat_update().
4569 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
4571 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
4572 vsx
->vsx_active_queue
[t
] =
4573 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
4574 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
4575 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
4581 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4583 vdev_t
*tvd
= vd
->vdev_top
;
4584 mutex_enter(&vd
->vdev_stat_lock
);
4586 memcpy(vs
, &vd
->vdev_stat
, sizeof (*vs
));
4587 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
4588 vs
->vs_state
= vd
->vdev_state
;
4589 vs
->vs_rsize
= vdev_get_min_asize(vd
);
4591 if (vd
->vdev_ops
->vdev_op_leaf
) {
4592 vs
->vs_pspace
= vd
->vdev_psize
;
4593 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
4594 VDEV_LABEL_END_SIZE
;
4596 * Report initializing progress. Since we don't
4597 * have the initializing locks held, this is only
4598 * an estimate (although a fairly accurate one).
4600 vs
->vs_initialize_bytes_done
=
4601 vd
->vdev_initialize_bytes_done
;
4602 vs
->vs_initialize_bytes_est
=
4603 vd
->vdev_initialize_bytes_est
;
4604 vs
->vs_initialize_state
= vd
->vdev_initialize_state
;
4605 vs
->vs_initialize_action_time
=
4606 vd
->vdev_initialize_action_time
;
4609 * Report manual TRIM progress. Since we don't have
4610 * the manual TRIM locks held, this is only an
4611 * estimate (although fairly accurate one).
4613 vs
->vs_trim_notsup
= !vd
->vdev_has_trim
;
4614 vs
->vs_trim_bytes_done
= vd
->vdev_trim_bytes_done
;
4615 vs
->vs_trim_bytes_est
= vd
->vdev_trim_bytes_est
;
4616 vs
->vs_trim_state
= vd
->vdev_trim_state
;
4617 vs
->vs_trim_action_time
= vd
->vdev_trim_action_time
;
4619 /* Set when there is a deferred resilver. */
4620 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
4624 * Report expandable space on top-level, non-auxiliary devices
4625 * only. The expandable space is reported in terms of metaslab
4626 * sized units since that determines how much space the pool
4629 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
4630 vs
->vs_esize
= P2ALIGN(
4631 vd
->vdev_max_asize
- vd
->vdev_asize
,
4632 1ULL << tvd
->vdev_ms_shift
);
4635 vs
->vs_configured_ashift
= vd
->vdev_top
!= NULL
4636 ? vd
->vdev_top
->vdev_ashift
: vd
->vdev_ashift
;
4637 vs
->vs_logical_ashift
= vd
->vdev_logical_ashift
;
4638 if (vd
->vdev_physical_ashift
<= ASHIFT_MAX
)
4639 vs
->vs_physical_ashift
= vd
->vdev_physical_ashift
;
4641 vs
->vs_physical_ashift
= 0;
4644 * Report fragmentation and rebuild progress for top-level,
4645 * non-auxiliary, concrete devices.
4647 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
4648 vdev_is_concrete(vd
)) {
4650 * The vdev fragmentation rating doesn't take into
4651 * account the embedded slog metaslab (vdev_log_mg).
4652 * Since it's only one metaslab, it would have a tiny
4653 * impact on the overall fragmentation.
4655 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
4656 vd
->vdev_mg
->mg_fragmentation
: 0;
4658 vs
->vs_noalloc
= MAX(vd
->vdev_noalloc
,
4659 tvd
? tvd
->vdev_noalloc
: 0);
4662 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
4663 mutex_exit(&vd
->vdev_stat_lock
);
4667 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
4669 return (vdev_get_stats_ex(vd
, vs
, NULL
));
4673 vdev_clear_stats(vdev_t
*vd
)
4675 mutex_enter(&vd
->vdev_stat_lock
);
4676 vd
->vdev_stat
.vs_space
= 0;
4677 vd
->vdev_stat
.vs_dspace
= 0;
4678 vd
->vdev_stat
.vs_alloc
= 0;
4679 mutex_exit(&vd
->vdev_stat_lock
);
4683 vdev_scan_stat_init(vdev_t
*vd
)
4685 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4687 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4688 vdev_scan_stat_init(vd
->vdev_child
[c
]);
4690 mutex_enter(&vd
->vdev_stat_lock
);
4691 vs
->vs_scan_processed
= 0;
4692 mutex_exit(&vd
->vdev_stat_lock
);
4696 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
4698 spa_t
*spa
= zio
->io_spa
;
4699 vdev_t
*rvd
= spa
->spa_root_vdev
;
4700 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
4702 uint64_t txg
= zio
->io_txg
;
4703 vdev_stat_t
*vs
= vd
? &vd
->vdev_stat
: NULL
;
4704 vdev_stat_ex_t
*vsx
= vd
? &vd
->vdev_stat_ex
: NULL
;
4705 zio_type_t type
= zio
->io_type
;
4706 int flags
= zio
->io_flags
;
4709 * If this i/o is a gang leader, it didn't do any actual work.
4711 if (zio
->io_gang_tree
)
4714 if (zio
->io_error
== 0) {
4716 * If this is a root i/o, don't count it -- we've already
4717 * counted the top-level vdevs, and vdev_get_stats() will
4718 * aggregate them when asked. This reduces contention on
4719 * the root vdev_stat_lock and implicitly handles blocks
4720 * that compress away to holes, for which there is no i/o.
4721 * (Holes never create vdev children, so all the counters
4722 * remain zero, which is what we want.)
4724 * Note: this only applies to successful i/o (io_error == 0)
4725 * because unlike i/o counts, errors are not additive.
4726 * When reading a ditto block, for example, failure of
4727 * one top-level vdev does not imply a root-level error.
4732 ASSERT(vd
== zio
->io_vd
);
4734 if (flags
& ZIO_FLAG_IO_BYPASS
)
4737 mutex_enter(&vd
->vdev_stat_lock
);
4739 if (flags
& ZIO_FLAG_IO_REPAIR
) {
4741 * Repair is the result of a resilver issued by the
4742 * scan thread (spa_sync).
4744 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4745 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
4746 dsl_scan_phys_t
*scn_phys
= &scn
->scn_phys
;
4747 uint64_t *processed
= &scn_phys
->scn_processed
;
4749 if (vd
->vdev_ops
->vdev_op_leaf
)
4750 atomic_add_64(processed
, psize
);
4751 vs
->vs_scan_processed
+= psize
;
4755 * Repair is the result of a rebuild issued by the
4756 * rebuild thread (vdev_rebuild_thread). To avoid
4757 * double counting repaired bytes the virtual dRAID
4758 * spare vdev is excluded from the processed bytes.
4760 if (zio
->io_priority
== ZIO_PRIORITY_REBUILD
) {
4761 vdev_t
*tvd
= vd
->vdev_top
;
4762 vdev_rebuild_t
*vr
= &tvd
->vdev_rebuild_config
;
4763 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
4764 uint64_t *rebuilt
= &vrp
->vrp_bytes_rebuilt
;
4766 if (vd
->vdev_ops
->vdev_op_leaf
&&
4767 vd
->vdev_ops
!= &vdev_draid_spare_ops
) {
4768 atomic_add_64(rebuilt
, psize
);
4770 vs
->vs_rebuild_processed
+= psize
;
4773 if (flags
& ZIO_FLAG_SELF_HEAL
)
4774 vs
->vs_self_healed
+= psize
;
4778 * The bytes/ops/histograms are recorded at the leaf level and
4779 * aggregated into the higher level vdevs in vdev_get_stats().
4781 if (vd
->vdev_ops
->vdev_op_leaf
&&
4782 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
4783 zio_type_t vs_type
= type
;
4784 zio_priority_t priority
= zio
->io_priority
;
4787 * TRIM ops and bytes are reported to user space as
4788 * ZIO_TYPE_IOCTL. This is done to preserve the
4789 * vdev_stat_t structure layout for user space.
4791 if (type
== ZIO_TYPE_TRIM
)
4792 vs_type
= ZIO_TYPE_IOCTL
;
4795 * Solely for the purposes of 'zpool iostat -lqrw'
4796 * reporting use the priority to categorize the IO.
4797 * Only the following are reported to user space:
4799 * ZIO_PRIORITY_SYNC_READ,
4800 * ZIO_PRIORITY_SYNC_WRITE,
4801 * ZIO_PRIORITY_ASYNC_READ,
4802 * ZIO_PRIORITY_ASYNC_WRITE,
4803 * ZIO_PRIORITY_SCRUB,
4804 * ZIO_PRIORITY_TRIM,
4805 * ZIO_PRIORITY_REBUILD.
4807 if (priority
== ZIO_PRIORITY_INITIALIZING
) {
4808 ASSERT3U(type
, ==, ZIO_TYPE_WRITE
);
4809 priority
= ZIO_PRIORITY_ASYNC_WRITE
;
4810 } else if (priority
== ZIO_PRIORITY_REMOVAL
) {
4811 priority
= ((type
== ZIO_TYPE_WRITE
) ?
4812 ZIO_PRIORITY_ASYNC_WRITE
:
4813 ZIO_PRIORITY_ASYNC_READ
);
4816 vs
->vs_ops
[vs_type
]++;
4817 vs
->vs_bytes
[vs_type
] += psize
;
4819 if (flags
& ZIO_FLAG_DELEGATED
) {
4820 vsx
->vsx_agg_histo
[priority
]
4821 [RQ_HISTO(zio
->io_size
)]++;
4823 vsx
->vsx_ind_histo
[priority
]
4824 [RQ_HISTO(zio
->io_size
)]++;
4827 if (zio
->io_delta
&& zio
->io_delay
) {
4828 vsx
->vsx_queue_histo
[priority
]
4829 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
4830 vsx
->vsx_disk_histo
[type
]
4831 [L_HISTO(zio
->io_delay
)]++;
4832 vsx
->vsx_total_histo
[type
]
4833 [L_HISTO(zio
->io_delta
)]++;
4837 mutex_exit(&vd
->vdev_stat_lock
);
4841 if (flags
& ZIO_FLAG_SPECULATIVE
)
4845 * If this is an I/O error that is going to be retried, then ignore the
4846 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4847 * hard errors, when in reality they can happen for any number of
4848 * innocuous reasons (bus resets, MPxIO link failure, etc).
4850 if (zio
->io_error
== EIO
&&
4851 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4855 * Intent logs writes won't propagate their error to the root
4856 * I/O so don't mark these types of failures as pool-level
4859 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4862 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4863 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4864 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4865 spa
->spa_claiming
)) {
4867 * This is either a normal write (not a repair), or it's
4868 * a repair induced by the scrub thread, or it's a repair
4869 * made by zil_claim() during spa_load() in the first txg.
4870 * In the normal case, we commit the DTL change in the same
4871 * txg as the block was born. In the scrub-induced repair
4872 * case, we know that scrubs run in first-pass syncing context,
4873 * so we commit the DTL change in spa_syncing_txg(spa).
4874 * In the zil_claim() case, we commit in spa_first_txg(spa).
4876 * We currently do not make DTL entries for failed spontaneous
4877 * self-healing writes triggered by normal (non-scrubbing)
4878 * reads, because we have no transactional context in which to
4879 * do so -- and it's not clear that it'd be desirable anyway.
4881 if (vd
->vdev_ops
->vdev_op_leaf
) {
4882 uint64_t commit_txg
= txg
;
4883 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4884 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4885 ASSERT(spa_sync_pass(spa
) == 1);
4886 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4887 commit_txg
= spa_syncing_txg(spa
);
4888 } else if (spa
->spa_claiming
) {
4889 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4890 commit_txg
= spa_first_txg(spa
);
4892 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4893 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4895 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4896 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4897 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4900 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4905 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4907 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4908 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4910 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4914 * Update the in-core space usage stats for this vdev, its metaslab class,
4915 * and the root vdev.
4918 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4919 int64_t space_delta
)
4922 int64_t dspace_delta
;
4923 spa_t
*spa
= vd
->vdev_spa
;
4924 vdev_t
*rvd
= spa
->spa_root_vdev
;
4926 ASSERT(vd
== vd
->vdev_top
);
4929 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4930 * factor. We must calculate this here and not at the root vdev
4931 * because the root vdev's psize-to-asize is simply the max of its
4932 * children's, thus not accurate enough for us.
4934 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4936 mutex_enter(&vd
->vdev_stat_lock
);
4937 /* ensure we won't underflow */
4938 if (alloc_delta
< 0) {
4939 ASSERT3U(vd
->vdev_stat
.vs_alloc
, >=, -alloc_delta
);
4942 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4943 vd
->vdev_stat
.vs_space
+= space_delta
;
4944 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4945 mutex_exit(&vd
->vdev_stat_lock
);
4947 /* every class but log contributes to root space stats */
4948 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4949 ASSERT(!vd
->vdev_isl2cache
);
4950 mutex_enter(&rvd
->vdev_stat_lock
);
4951 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4952 rvd
->vdev_stat
.vs_space
+= space_delta
;
4953 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4954 mutex_exit(&rvd
->vdev_stat_lock
);
4956 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4960 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4961 * so that it will be written out next time the vdev configuration is synced.
4962 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4965 vdev_config_dirty(vdev_t
*vd
)
4967 spa_t
*spa
= vd
->vdev_spa
;
4968 vdev_t
*rvd
= spa
->spa_root_vdev
;
4971 ASSERT(spa_writeable(spa
));
4974 * If this is an aux vdev (as with l2cache and spare devices), then we
4975 * update the vdev config manually and set the sync flag.
4977 if (vd
->vdev_aux
!= NULL
) {
4978 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
4982 for (c
= 0; c
< sav
->sav_count
; c
++) {
4983 if (sav
->sav_vdevs
[c
] == vd
)
4987 if (c
== sav
->sav_count
) {
4989 * We're being removed. There's nothing more to do.
4991 ASSERT(sav
->sav_sync
== B_TRUE
);
4995 sav
->sav_sync
= B_TRUE
;
4997 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
4998 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
4999 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
5000 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
5006 * Setting the nvlist in the middle if the array is a little
5007 * sketchy, but it will work.
5009 nvlist_free(aux
[c
]);
5010 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
5016 * The dirty list is protected by the SCL_CONFIG lock. The caller
5017 * must either hold SCL_CONFIG as writer, or must be the sync thread
5018 * (which holds SCL_CONFIG as reader). There's only one sync thread,
5019 * so this is sufficient to ensure mutual exclusion.
5021 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5022 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5023 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5026 for (c
= 0; c
< rvd
->vdev_children
; c
++)
5027 vdev_config_dirty(rvd
->vdev_child
[c
]);
5029 ASSERT(vd
== vd
->vdev_top
);
5031 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
5032 vdev_is_concrete(vd
)) {
5033 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
5039 vdev_config_clean(vdev_t
*vd
)
5041 spa_t
*spa
= vd
->vdev_spa
;
5043 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5044 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5045 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5047 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
5048 list_remove(&spa
->spa_config_dirty_list
, vd
);
5052 * Mark a top-level vdev's state as dirty, so that the next pass of
5053 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
5054 * the state changes from larger config changes because they require
5055 * much less locking, and are often needed for administrative actions.
5058 vdev_state_dirty(vdev_t
*vd
)
5060 spa_t
*spa
= vd
->vdev_spa
;
5062 ASSERT(spa_writeable(spa
));
5063 ASSERT(vd
== vd
->vdev_top
);
5066 * The state list is protected by the SCL_STATE lock. The caller
5067 * must either hold SCL_STATE as writer, or must be the sync thread
5068 * (which holds SCL_STATE as reader). There's only one sync thread,
5069 * so this is sufficient to ensure mutual exclusion.
5071 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5072 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5073 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5075 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
5076 vdev_is_concrete(vd
))
5077 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
5081 vdev_state_clean(vdev_t
*vd
)
5083 spa_t
*spa
= vd
->vdev_spa
;
5085 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5086 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5087 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5089 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
5090 list_remove(&spa
->spa_state_dirty_list
, vd
);
5094 * Propagate vdev state up from children to parent.
5097 vdev_propagate_state(vdev_t
*vd
)
5099 spa_t
*spa
= vd
->vdev_spa
;
5100 vdev_t
*rvd
= spa
->spa_root_vdev
;
5101 int degraded
= 0, faulted
= 0;
5105 if (vd
->vdev_children
> 0) {
5106 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5107 child
= vd
->vdev_child
[c
];
5110 * Don't factor holes or indirect vdevs into the
5113 if (!vdev_is_concrete(child
))
5116 if (!vdev_readable(child
) ||
5117 (!vdev_writeable(child
) && spa_writeable(spa
))) {
5119 * Root special: if there is a top-level log
5120 * device, treat the root vdev as if it were
5123 if (child
->vdev_islog
&& vd
== rvd
)
5127 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
5131 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
5135 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
5138 * Root special: if there is a top-level vdev that cannot be
5139 * opened due to corrupted metadata, then propagate the root
5140 * vdev's aux state as 'corrupt' rather than 'insufficient
5143 if (corrupted
&& vd
== rvd
&&
5144 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
5145 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
5146 VDEV_AUX_CORRUPT_DATA
);
5149 if (vd
->vdev_parent
)
5150 vdev_propagate_state(vd
->vdev_parent
);
5154 * Set a vdev's state. If this is during an open, we don't update the parent
5155 * state, because we're in the process of opening children depth-first.
5156 * Otherwise, we propagate the change to the parent.
5158 * If this routine places a device in a faulted state, an appropriate ereport is
5162 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
5164 uint64_t save_state
;
5165 spa_t
*spa
= vd
->vdev_spa
;
5167 if (state
== vd
->vdev_state
) {
5169 * Since vdev_offline() code path is already in an offline
5170 * state we can miss a statechange event to OFFLINE. Check
5171 * the previous state to catch this condition.
5173 if (vd
->vdev_ops
->vdev_op_leaf
&&
5174 (state
== VDEV_STATE_OFFLINE
) &&
5175 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
5176 /* post an offline state change */
5177 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
5179 vd
->vdev_stat
.vs_aux
= aux
;
5183 save_state
= vd
->vdev_state
;
5185 vd
->vdev_state
= state
;
5186 vd
->vdev_stat
.vs_aux
= aux
;
5189 * If we are setting the vdev state to anything but an open state, then
5190 * always close the underlying device unless the device has requested
5191 * a delayed close (i.e. we're about to remove or fault the device).
5192 * Otherwise, we keep accessible but invalid devices open forever.
5193 * We don't call vdev_close() itself, because that implies some extra
5194 * checks (offline, etc) that we don't want here. This is limited to
5195 * leaf devices, because otherwise closing the device will affect other
5198 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
5199 vd
->vdev_ops
->vdev_op_leaf
)
5200 vd
->vdev_ops
->vdev_op_close(vd
);
5202 if (vd
->vdev_removed
&&
5203 state
== VDEV_STATE_CANT_OPEN
&&
5204 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
5206 * If the previous state is set to VDEV_STATE_REMOVED, then this
5207 * device was previously marked removed and someone attempted to
5208 * reopen it. If this failed due to a nonexistent device, then
5209 * keep the device in the REMOVED state. We also let this be if
5210 * it is one of our special test online cases, which is only
5211 * attempting to online the device and shouldn't generate an FMA
5214 vd
->vdev_state
= VDEV_STATE_REMOVED
;
5215 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
5216 } else if (state
== VDEV_STATE_REMOVED
) {
5217 vd
->vdev_removed
= B_TRUE
;
5218 } else if (state
== VDEV_STATE_CANT_OPEN
) {
5220 * If we fail to open a vdev during an import or recovery, we
5221 * mark it as "not available", which signifies that it was
5222 * never there to begin with. Failure to open such a device
5223 * is not considered an error.
5225 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
5226 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
5227 vd
->vdev_ops
->vdev_op_leaf
)
5228 vd
->vdev_not_present
= 1;
5231 * Post the appropriate ereport. If the 'prevstate' field is
5232 * set to something other than VDEV_STATE_UNKNOWN, it indicates
5233 * that this is part of a vdev_reopen(). In this case, we don't
5234 * want to post the ereport if the device was already in the
5235 * CANT_OPEN state beforehand.
5237 * If the 'checkremove' flag is set, then this is an attempt to
5238 * online the device in response to an insertion event. If we
5239 * hit this case, then we have detected an insertion event for a
5240 * faulted or offline device that wasn't in the removed state.
5241 * In this scenario, we don't post an ereport because we are
5242 * about to replace the device, or attempt an online with
5243 * vdev_forcefault, which will generate the fault for us.
5245 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
5246 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
5247 vd
!= spa
->spa_root_vdev
) {
5251 case VDEV_AUX_OPEN_FAILED
:
5252 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
5254 case VDEV_AUX_CORRUPT_DATA
:
5255 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
5257 case VDEV_AUX_NO_REPLICAS
:
5258 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
5260 case VDEV_AUX_BAD_GUID_SUM
:
5261 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
5263 case VDEV_AUX_TOO_SMALL
:
5264 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
5266 case VDEV_AUX_BAD_LABEL
:
5267 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
5269 case VDEV_AUX_BAD_ASHIFT
:
5270 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
5273 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
5276 (void) zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
5280 /* Erase any notion of persistent removed state */
5281 vd
->vdev_removed
= B_FALSE
;
5283 vd
->vdev_removed
= B_FALSE
;
5287 * Notify ZED of any significant state-change on a leaf vdev.
5290 if (vd
->vdev_ops
->vdev_op_leaf
) {
5291 /* preserve original state from a vdev_reopen() */
5292 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
5293 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
5294 (save_state
<= VDEV_STATE_CLOSED
))
5295 save_state
= vd
->vdev_prevstate
;
5297 /* filter out state change due to initial vdev_open */
5298 if (save_state
> VDEV_STATE_CLOSED
)
5299 zfs_post_state_change(spa
, vd
, save_state
);
5302 if (!isopen
&& vd
->vdev_parent
)
5303 vdev_propagate_state(vd
->vdev_parent
);
5307 vdev_children_are_offline(vdev_t
*vd
)
5309 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
5311 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
5312 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
5320 * Check the vdev configuration to ensure that it's capable of supporting
5321 * a root pool. We do not support partial configuration.
5324 vdev_is_bootable(vdev_t
*vd
)
5326 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5327 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
5329 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
5333 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5334 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
5341 vdev_is_concrete(vdev_t
*vd
)
5343 vdev_ops_t
*ops
= vd
->vdev_ops
;
5344 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
5345 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
5353 * Determine if a log device has valid content. If the vdev was
5354 * removed or faulted in the MOS config then we know that
5355 * the content on the log device has already been written to the pool.
5358 vdev_log_state_valid(vdev_t
*vd
)
5360 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
5364 for (int c
= 0; c
< vd
->vdev_children
; c
++)
5365 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
5372 * Expand a vdev if possible.
5375 vdev_expand(vdev_t
*vd
, uint64_t txg
)
5377 ASSERT(vd
->vdev_top
== vd
);
5378 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5379 ASSERT(vdev_is_concrete(vd
));
5381 vdev_set_deflate_ratio(vd
);
5383 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
5384 vdev_is_concrete(vd
)) {
5385 vdev_metaslab_group_create(vd
);
5386 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
5387 vdev_config_dirty(vd
);
5395 vdev_split(vdev_t
*vd
)
5397 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
5399 vdev_remove_child(pvd
, vd
);
5400 vdev_compact_children(pvd
);
5402 cvd
= pvd
->vdev_child
[0];
5403 if (pvd
->vdev_children
== 1) {
5404 vdev_remove_parent(cvd
);
5405 cvd
->vdev_splitting
= B_TRUE
;
5407 vdev_propagate_state(cvd
);
5411 vdev_deadman(vdev_t
*vd
, const char *tag
)
5413 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5414 vdev_t
*cvd
= vd
->vdev_child
[c
];
5416 vdev_deadman(cvd
, tag
);
5419 if (vd
->vdev_ops
->vdev_op_leaf
) {
5420 vdev_queue_t
*vq
= &vd
->vdev_queue
;
5422 mutex_enter(&vq
->vq_lock
);
5423 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
5424 spa_t
*spa
= vd
->vdev_spa
;
5428 zfs_dbgmsg("slow vdev: %s has %lu active IOs",
5429 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
5432 * Look at the head of all the pending queues,
5433 * if any I/O has been outstanding for longer than
5434 * the spa_deadman_synctime invoke the deadman logic.
5436 fio
= avl_first(&vq
->vq_active_tree
);
5437 delta
= gethrtime() - fio
->io_timestamp
;
5438 if (delta
> spa_deadman_synctime(spa
))
5439 zio_deadman(fio
, tag
);
5441 mutex_exit(&vq
->vq_lock
);
5446 vdev_defer_resilver(vdev_t
*vd
)
5448 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
5450 vd
->vdev_resilver_deferred
= B_TRUE
;
5451 vd
->vdev_spa
->spa_resilver_deferred
= B_TRUE
;
5455 * Clears the resilver deferred flag on all leaf devs under vd. Returns
5456 * B_TRUE if we have devices that need to be resilvered and are available to
5457 * accept resilver I/Os.
5460 vdev_clear_resilver_deferred(vdev_t
*vd
, dmu_tx_t
*tx
)
5462 boolean_t resilver_needed
= B_FALSE
;
5463 spa_t
*spa
= vd
->vdev_spa
;
5465 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5466 vdev_t
*cvd
= vd
->vdev_child
[c
];
5467 resilver_needed
|= vdev_clear_resilver_deferred(cvd
, tx
);
5470 if (vd
== spa
->spa_root_vdev
&&
5471 spa_feature_is_active(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
5472 spa_feature_decr(spa
, SPA_FEATURE_RESILVER_DEFER
, tx
);
5473 vdev_config_dirty(vd
);
5474 spa
->spa_resilver_deferred
= B_FALSE
;
5475 return (resilver_needed
);
5478 if (!vdev_is_concrete(vd
) || vd
->vdev_aux
||
5479 !vd
->vdev_ops
->vdev_op_leaf
)
5480 return (resilver_needed
);
5482 vd
->vdev_resilver_deferred
= B_FALSE
;
5484 return (!vdev_is_dead(vd
) && !vd
->vdev_offline
&&
5485 vdev_resilver_needed(vd
, NULL
, NULL
));
5489 vdev_xlate_is_empty(range_seg64_t
*rs
)
5491 return (rs
->rs_start
== rs
->rs_end
);
5495 * Translate a logical range to the first contiguous physical range for the
5496 * specified vdev_t. This function is initially called with a leaf vdev and
5497 * will walk each parent vdev until it reaches a top-level vdev. Once the
5498 * top-level is reached the physical range is initialized and the recursive
5499 * function begins to unwind. As it unwinds it calls the parent's vdev
5500 * specific translation function to do the real conversion.
5503 vdev_xlate(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5504 range_seg64_t
*physical_rs
, range_seg64_t
*remain_rs
)
5507 * Walk up the vdev tree
5509 if (vd
!= vd
->vdev_top
) {
5510 vdev_xlate(vd
->vdev_parent
, logical_rs
, physical_rs
,
5514 * We've reached the top-level vdev, initialize the physical
5515 * range to the logical range and set an empty remaining
5516 * range then start to unwind.
5518 physical_rs
->rs_start
= logical_rs
->rs_start
;
5519 physical_rs
->rs_end
= logical_rs
->rs_end
;
5521 remain_rs
->rs_start
= logical_rs
->rs_start
;
5522 remain_rs
->rs_end
= logical_rs
->rs_start
;
5527 vdev_t
*pvd
= vd
->vdev_parent
;
5528 ASSERT3P(pvd
, !=, NULL
);
5529 ASSERT3P(pvd
->vdev_ops
->vdev_op_xlate
, !=, NULL
);
5532 * As this recursive function unwinds, translate the logical
5533 * range into its physical and any remaining components by calling
5534 * the vdev specific translate function.
5536 range_seg64_t intermediate
= { 0 };
5537 pvd
->vdev_ops
->vdev_op_xlate(vd
, physical_rs
, &intermediate
, remain_rs
);
5539 physical_rs
->rs_start
= intermediate
.rs_start
;
5540 physical_rs
->rs_end
= intermediate
.rs_end
;
5544 vdev_xlate_walk(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5545 vdev_xlate_func_t
*func
, void *arg
)
5547 range_seg64_t iter_rs
= *logical_rs
;
5548 range_seg64_t physical_rs
;
5549 range_seg64_t remain_rs
;
5551 while (!vdev_xlate_is_empty(&iter_rs
)) {
5553 vdev_xlate(vd
, &iter_rs
, &physical_rs
, &remain_rs
);
5556 * With raidz and dRAID, it's possible that the logical range
5557 * does not live on this leaf vdev. Only when there is a non-
5558 * zero physical size call the provided function.
5560 if (!vdev_xlate_is_empty(&physical_rs
))
5561 func(arg
, &physical_rs
);
5563 iter_rs
= remain_rs
;
5568 vdev_name(vdev_t
*vd
, char *buf
, int buflen
)
5570 if (vd
->vdev_path
== NULL
) {
5571 if (strcmp(vd
->vdev_ops
->vdev_op_type
, "root") == 0) {
5572 strlcpy(buf
, vd
->vdev_spa
->spa_name
, buflen
);
5573 } else if (!vd
->vdev_ops
->vdev_op_leaf
) {
5574 snprintf(buf
, buflen
, "%s-%llu",
5575 vd
->vdev_ops
->vdev_op_type
,
5576 (u_longlong_t
)vd
->vdev_id
);
5579 strlcpy(buf
, vd
->vdev_path
, buflen
);
5585 * Look at the vdev tree and determine whether any devices are currently being
5589 vdev_replace_in_progress(vdev_t
*vdev
)
5591 ASSERT(spa_config_held(vdev
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
5593 if (vdev
->vdev_ops
== &vdev_replacing_ops
)
5597 * A 'spare' vdev indicates that we have a replace in progress, unless
5598 * it has exactly two children, and the second, the hot spare, has
5599 * finished being resilvered.
5601 if (vdev
->vdev_ops
== &vdev_spare_ops
&& (vdev
->vdev_children
> 2 ||
5602 !vdev_dtl_empty(vdev
->vdev_child
[1], DTL_MISSING
)))
5605 for (int i
= 0; i
< vdev
->vdev_children
; i
++) {
5606 if (vdev_replace_in_progress(vdev
->vdev_child
[i
]))
5614 * Add a (source=src, propname=propval) list to an nvlist.
5617 vdev_prop_add_list(nvlist_t
*nvl
, const char *propname
, char *strval
,
5618 uint64_t intval
, zprop_source_t src
)
5622 propval
= fnvlist_alloc();
5623 fnvlist_add_uint64(propval
, ZPROP_SOURCE
, src
);
5626 fnvlist_add_string(propval
, ZPROP_VALUE
, strval
);
5628 fnvlist_add_uint64(propval
, ZPROP_VALUE
, intval
);
5630 fnvlist_add_nvlist(nvl
, propname
, propval
);
5631 nvlist_free(propval
);
5635 vdev_props_set_sync(void *arg
, dmu_tx_t
*tx
)
5638 nvlist_t
*nvp
= arg
;
5639 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5640 objset_t
*mos
= spa
->spa_meta_objset
;
5641 nvpair_t
*elem
= NULL
;
5645 vdev_guid
= fnvlist_lookup_uint64(nvp
, ZPOOL_VDEV_PROPS_SET_VDEV
);
5646 nvprops
= fnvlist_lookup_nvlist(nvp
, ZPOOL_VDEV_PROPS_SET_PROPS
);
5647 vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
);
5649 /* this vdev could get removed while waiting for this sync task */
5653 mutex_enter(&spa
->spa_props_lock
);
5655 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5656 uint64_t intval
, objid
= 0;
5659 const char *propname
= nvpair_name(elem
);
5660 zprop_type_t proptype
;
5663 * Set vdev property values in the vdev props mos object.
5665 if (vd
->vdev_top_zap
!= 0) {
5666 objid
= vd
->vdev_top_zap
;
5667 } else if (vd
->vdev_leaf_zap
!= 0) {
5668 objid
= vd
->vdev_leaf_zap
;
5670 panic("vdev not top or leaf");
5673 switch (prop
= vdev_name_to_prop(propname
)) {
5674 case VDEV_PROP_USERPROP
:
5675 if (vdev_prop_user(propname
)) {
5676 strval
= fnvpair_value_string(elem
);
5677 if (strlen(strval
) == 0) {
5678 /* remove the property if value == "" */
5679 (void) zap_remove(mos
, objid
, propname
,
5682 VERIFY0(zap_update(mos
, objid
, propname
,
5683 1, strlen(strval
) + 1, strval
, tx
));
5685 spa_history_log_internal(spa
, "vdev set", tx
,
5686 "vdev_guid=%llu: %s=%s",
5687 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5692 /* normalize the property name */
5693 propname
= vdev_prop_to_name(prop
);
5694 proptype
= vdev_prop_get_type(prop
);
5696 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5697 ASSERT(proptype
== PROP_TYPE_STRING
);
5698 strval
= fnvpair_value_string(elem
);
5699 VERIFY0(zap_update(mos
, objid
, propname
,
5700 1, strlen(strval
) + 1, strval
, tx
));
5701 spa_history_log_internal(spa
, "vdev set", tx
,
5702 "vdev_guid=%llu: %s=%s",
5703 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5705 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
5706 intval
= fnvpair_value_uint64(elem
);
5708 if (proptype
== PROP_TYPE_INDEX
) {
5710 VERIFY0(vdev_prop_index_to_string(
5711 prop
, intval
, &unused
));
5713 VERIFY0(zap_update(mos
, objid
, propname
,
5714 sizeof (uint64_t), 1, &intval
, tx
));
5715 spa_history_log_internal(spa
, "vdev set", tx
,
5716 "vdev_guid=%llu: %s=%lld",
5717 (u_longlong_t
)vdev_guid
,
5718 nvpair_name(elem
), (longlong_t
)intval
);
5720 panic("invalid vdev property type %u",
5727 mutex_exit(&spa
->spa_props_lock
);
5731 vdev_prop_set(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5733 spa_t
*spa
= vd
->vdev_spa
;
5734 nvpair_t
*elem
= NULL
;
5741 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_SET_VDEV
,
5743 return (SET_ERROR(EINVAL
));
5745 if (nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_SET_PROPS
,
5747 return (SET_ERROR(EINVAL
));
5749 if ((vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
)) == NULL
)
5750 return (SET_ERROR(EINVAL
));
5752 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5753 char *propname
= nvpair_name(elem
);
5754 vdev_prop_t prop
= vdev_name_to_prop(propname
);
5755 uint64_t intval
= 0;
5756 char *strval
= NULL
;
5758 if (prop
== VDEV_PROP_USERPROP
&& !vdev_prop_user(propname
)) {
5763 if (vdev_prop_readonly(prop
)) {
5768 /* Special Processing */
5770 case VDEV_PROP_PATH
:
5771 if (vd
->vdev_path
== NULL
) {
5775 if (nvpair_value_string(elem
, &strval
) != 0) {
5779 /* New path must start with /dev/ */
5780 if (strncmp(strval
, "/dev/", 5)) {
5784 error
= spa_vdev_setpath(spa
, vdev_guid
, strval
);
5786 case VDEV_PROP_ALLOCATING
:
5787 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5791 if (intval
!= vd
->vdev_noalloc
)
5794 error
= spa_vdev_noalloc(spa
, vdev_guid
);
5796 error
= spa_vdev_alloc(spa
, vdev_guid
);
5798 case VDEV_PROP_FAILFAST
:
5799 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5803 vd
->vdev_failfast
= intval
& 1;
5805 case VDEV_PROP_CHECKSUM_N
:
5806 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5810 vd
->vdev_checksum_n
= intval
;
5812 case VDEV_PROP_CHECKSUM_T
:
5813 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5817 vd
->vdev_checksum_t
= intval
;
5819 case VDEV_PROP_IO_N
:
5820 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5824 vd
->vdev_io_n
= intval
;
5826 case VDEV_PROP_IO_T
:
5827 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5831 vd
->vdev_io_t
= intval
;
5834 /* Most processing is done in vdev_props_set_sync */
5840 vdev_prop_add_list(outnvl
, propname
, strval
, intval
, 0);
5845 return (dsl_sync_task(spa
->spa_name
, NULL
, vdev_props_set_sync
,
5846 innvl
, 6, ZFS_SPACE_CHECK_EXTRA_RESERVED
));
5850 vdev_prop_get(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5852 spa_t
*spa
= vd
->vdev_spa
;
5853 objset_t
*mos
= spa
->spa_meta_objset
;
5857 nvpair_t
*elem
= NULL
;
5858 nvlist_t
*nvprops
= NULL
;
5859 uint64_t intval
= 0;
5860 char *strval
= NULL
;
5861 const char *propname
= NULL
;
5865 ASSERT(mos
!= NULL
);
5867 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_GET_VDEV
,
5869 return (SET_ERROR(EINVAL
));
5871 nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_GET_PROPS
, &nvprops
);
5873 if (vd
->vdev_top_zap
!= 0) {
5874 objid
= vd
->vdev_top_zap
;
5875 } else if (vd
->vdev_leaf_zap
!= 0) {
5876 objid
= vd
->vdev_leaf_zap
;
5878 return (SET_ERROR(EINVAL
));
5882 mutex_enter(&spa
->spa_props_lock
);
5884 if (nvprops
!= NULL
) {
5885 char namebuf
[64] = { 0 };
5887 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5890 propname
= nvpair_name(elem
);
5891 prop
= vdev_name_to_prop(propname
);
5892 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
5893 uint64_t integer_size
, num_integers
;
5896 /* Special Read-only Properties */
5897 case VDEV_PROP_NAME
:
5898 strval
= vdev_name(vd
, namebuf
,
5902 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
5905 case VDEV_PROP_CAPACITY
:
5907 intval
= (vd
->vdev_stat
.vs_dspace
== 0) ? 0 :
5908 (vd
->vdev_stat
.vs_alloc
* 100 /
5909 vd
->vdev_stat
.vs_dspace
);
5910 vdev_prop_add_list(outnvl
, propname
, NULL
,
5911 intval
, ZPROP_SRC_NONE
);
5913 case VDEV_PROP_STATE
:
5914 vdev_prop_add_list(outnvl
, propname
, NULL
,
5915 vd
->vdev_state
, ZPROP_SRC_NONE
);
5917 case VDEV_PROP_GUID
:
5918 vdev_prop_add_list(outnvl
, propname
, NULL
,
5919 vd
->vdev_guid
, ZPROP_SRC_NONE
);
5921 case VDEV_PROP_ASIZE
:
5922 vdev_prop_add_list(outnvl
, propname
, NULL
,
5923 vd
->vdev_asize
, ZPROP_SRC_NONE
);
5925 case VDEV_PROP_PSIZE
:
5926 vdev_prop_add_list(outnvl
, propname
, NULL
,
5927 vd
->vdev_psize
, ZPROP_SRC_NONE
);
5929 case VDEV_PROP_ASHIFT
:
5930 vdev_prop_add_list(outnvl
, propname
, NULL
,
5931 vd
->vdev_ashift
, ZPROP_SRC_NONE
);
5933 case VDEV_PROP_SIZE
:
5934 vdev_prop_add_list(outnvl
, propname
, NULL
,
5935 vd
->vdev_stat
.vs_dspace
, ZPROP_SRC_NONE
);
5937 case VDEV_PROP_FREE
:
5938 vdev_prop_add_list(outnvl
, propname
, NULL
,
5939 vd
->vdev_stat
.vs_dspace
-
5940 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
5942 case VDEV_PROP_ALLOCATED
:
5943 vdev_prop_add_list(outnvl
, propname
, NULL
,
5944 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
5946 case VDEV_PROP_EXPANDSZ
:
5947 vdev_prop_add_list(outnvl
, propname
, NULL
,
5948 vd
->vdev_stat
.vs_esize
, ZPROP_SRC_NONE
);
5950 case VDEV_PROP_FRAGMENTATION
:
5951 vdev_prop_add_list(outnvl
, propname
, NULL
,
5952 vd
->vdev_stat
.vs_fragmentation
,
5955 case VDEV_PROP_PARITY
:
5956 vdev_prop_add_list(outnvl
, propname
, NULL
,
5957 vdev_get_nparity(vd
), ZPROP_SRC_NONE
);
5959 case VDEV_PROP_PATH
:
5960 if (vd
->vdev_path
== NULL
)
5962 vdev_prop_add_list(outnvl
, propname
,
5963 vd
->vdev_path
, 0, ZPROP_SRC_NONE
);
5965 case VDEV_PROP_DEVID
:
5966 if (vd
->vdev_devid
== NULL
)
5968 vdev_prop_add_list(outnvl
, propname
,
5969 vd
->vdev_devid
, 0, ZPROP_SRC_NONE
);
5971 case VDEV_PROP_PHYS_PATH
:
5972 if (vd
->vdev_physpath
== NULL
)
5974 vdev_prop_add_list(outnvl
, propname
,
5975 vd
->vdev_physpath
, 0, ZPROP_SRC_NONE
);
5977 case VDEV_PROP_ENC_PATH
:
5978 if (vd
->vdev_enc_sysfs_path
== NULL
)
5980 vdev_prop_add_list(outnvl
, propname
,
5981 vd
->vdev_enc_sysfs_path
, 0, ZPROP_SRC_NONE
);
5984 if (vd
->vdev_fru
== NULL
)
5986 vdev_prop_add_list(outnvl
, propname
,
5987 vd
->vdev_fru
, 0, ZPROP_SRC_NONE
);
5989 case VDEV_PROP_PARENT
:
5990 if (vd
->vdev_parent
!= NULL
) {
5991 strval
= vdev_name(vd
->vdev_parent
,
5992 namebuf
, sizeof (namebuf
));
5993 vdev_prop_add_list(outnvl
, propname
,
5994 strval
, 0, ZPROP_SRC_NONE
);
5997 case VDEV_PROP_CHILDREN
:
5998 if (vd
->vdev_children
> 0)
5999 strval
= kmem_zalloc(ZAP_MAXVALUELEN
,
6001 for (uint64_t i
= 0; i
< vd
->vdev_children
;
6005 vname
= vdev_name(vd
->vdev_child
[i
],
6006 namebuf
, sizeof (namebuf
));
6008 vname
= "(unknown)";
6009 if (strlen(strval
) > 0)
6010 strlcat(strval
, ",",
6012 strlcat(strval
, vname
, ZAP_MAXVALUELEN
);
6014 if (strval
!= NULL
) {
6015 vdev_prop_add_list(outnvl
, propname
,
6016 strval
, 0, ZPROP_SRC_NONE
);
6017 kmem_free(strval
, ZAP_MAXVALUELEN
);
6020 case VDEV_PROP_NUMCHILDREN
:
6021 vdev_prop_add_list(outnvl
, propname
, NULL
,
6022 vd
->vdev_children
, ZPROP_SRC_NONE
);
6024 case VDEV_PROP_READ_ERRORS
:
6025 vdev_prop_add_list(outnvl
, propname
, NULL
,
6026 vd
->vdev_stat
.vs_read_errors
,
6029 case VDEV_PROP_WRITE_ERRORS
:
6030 vdev_prop_add_list(outnvl
, propname
, NULL
,
6031 vd
->vdev_stat
.vs_write_errors
,
6034 case VDEV_PROP_CHECKSUM_ERRORS
:
6035 vdev_prop_add_list(outnvl
, propname
, NULL
,
6036 vd
->vdev_stat
.vs_checksum_errors
,
6039 case VDEV_PROP_INITIALIZE_ERRORS
:
6040 vdev_prop_add_list(outnvl
, propname
, NULL
,
6041 vd
->vdev_stat
.vs_initialize_errors
,
6044 case VDEV_PROP_OPS_NULL
:
6045 vdev_prop_add_list(outnvl
, propname
, NULL
,
6046 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_NULL
],
6049 case VDEV_PROP_OPS_READ
:
6050 vdev_prop_add_list(outnvl
, propname
, NULL
,
6051 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_READ
],
6054 case VDEV_PROP_OPS_WRITE
:
6055 vdev_prop_add_list(outnvl
, propname
, NULL
,
6056 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_WRITE
],
6059 case VDEV_PROP_OPS_FREE
:
6060 vdev_prop_add_list(outnvl
, propname
, NULL
,
6061 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_FREE
],
6064 case VDEV_PROP_OPS_CLAIM
:
6065 vdev_prop_add_list(outnvl
, propname
, NULL
,
6066 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_CLAIM
],
6069 case VDEV_PROP_OPS_TRIM
:
6071 * TRIM ops and bytes are reported to user
6072 * space as ZIO_TYPE_IOCTL. This is done to
6073 * preserve the vdev_stat_t structure layout
6076 vdev_prop_add_list(outnvl
, propname
, NULL
,
6077 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_IOCTL
],
6080 case VDEV_PROP_BYTES_NULL
:
6081 vdev_prop_add_list(outnvl
, propname
, NULL
,
6082 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_NULL
],
6085 case VDEV_PROP_BYTES_READ
:
6086 vdev_prop_add_list(outnvl
, propname
, NULL
,
6087 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_READ
],
6090 case VDEV_PROP_BYTES_WRITE
:
6091 vdev_prop_add_list(outnvl
, propname
, NULL
,
6092 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_WRITE
],
6095 case VDEV_PROP_BYTES_FREE
:
6096 vdev_prop_add_list(outnvl
, propname
, NULL
,
6097 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_FREE
],
6100 case VDEV_PROP_BYTES_CLAIM
:
6101 vdev_prop_add_list(outnvl
, propname
, NULL
,
6102 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_CLAIM
],
6105 case VDEV_PROP_BYTES_TRIM
:
6107 * TRIM ops and bytes are reported to user
6108 * space as ZIO_TYPE_IOCTL. This is done to
6109 * preserve the vdev_stat_t structure layout
6112 vdev_prop_add_list(outnvl
, propname
, NULL
,
6113 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_IOCTL
],
6116 case VDEV_PROP_REMOVING
:
6117 vdev_prop_add_list(outnvl
, propname
, NULL
,
6118 vd
->vdev_removing
, ZPROP_SRC_NONE
);
6120 /* Numeric Properites */
6121 case VDEV_PROP_ALLOCATING
:
6122 /* Leaf vdevs cannot have this property */
6123 if (vd
->vdev_mg
== NULL
&&
6124 vd
->vdev_top
!= NULL
) {
6125 src
= ZPROP_SRC_NONE
;
6126 intval
= ZPROP_BOOLEAN_NA
;
6128 err
= vdev_prop_get_int(vd
, prop
,
6130 if (err
&& err
!= ENOENT
)
6134 vdev_prop_default_numeric(prop
))
6135 src
= ZPROP_SRC_DEFAULT
;
6137 src
= ZPROP_SRC_LOCAL
;
6140 vdev_prop_add_list(outnvl
, propname
, NULL
,
6143 case VDEV_PROP_FAILFAST
:
6144 src
= ZPROP_SRC_LOCAL
;
6147 err
= zap_lookup(mos
, objid
, nvpair_name(elem
),
6148 sizeof (uint64_t), 1, &intval
);
6149 if (err
== ENOENT
) {
6150 intval
= vdev_prop_default_numeric(
6156 if (intval
== vdev_prop_default_numeric(prop
))
6157 src
= ZPROP_SRC_DEFAULT
;
6159 vdev_prop_add_list(outnvl
, propname
, strval
,
6162 case VDEV_PROP_CHECKSUM_N
:
6163 case VDEV_PROP_CHECKSUM_T
:
6164 case VDEV_PROP_IO_N
:
6165 case VDEV_PROP_IO_T
:
6166 err
= vdev_prop_get_int(vd
, prop
, &intval
);
6167 if (err
&& err
!= ENOENT
)
6170 if (intval
== vdev_prop_default_numeric(prop
))
6171 src
= ZPROP_SRC_DEFAULT
;
6173 src
= ZPROP_SRC_LOCAL
;
6175 vdev_prop_add_list(outnvl
, propname
, NULL
,
6178 /* Text Properties */
6179 case VDEV_PROP_COMMENT
:
6180 /* Exists in the ZAP below */
6182 case VDEV_PROP_USERPROP
:
6183 /* User Properites */
6184 src
= ZPROP_SRC_LOCAL
;
6186 err
= zap_length(mos
, objid
, nvpair_name(elem
),
6187 &integer_size
, &num_integers
);
6191 switch (integer_size
) {
6193 /* User properties cannot be integers */
6197 /* string property */
6198 strval
= kmem_alloc(num_integers
,
6200 err
= zap_lookup(mos
, objid
,
6201 nvpair_name(elem
), 1,
6202 num_integers
, strval
);
6208 vdev_prop_add_list(outnvl
, propname
,
6210 kmem_free(strval
, num_integers
);
6223 * Get all properties from the MOS vdev property object.
6227 for (zap_cursor_init(&zc
, mos
, objid
);
6228 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
6229 zap_cursor_advance(&zc
)) {
6232 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
6233 propname
= za
.za_name
;
6235 switch (za
.za_integer_length
) {
6237 /* We do not allow integer user properties */
6238 /* This is likely an internal value */
6241 /* string property */
6242 strval
= kmem_alloc(za
.za_num_integers
,
6244 err
= zap_lookup(mos
, objid
, za
.za_name
, 1,
6245 za
.za_num_integers
, strval
);
6247 kmem_free(strval
, za
.za_num_integers
);
6250 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
6252 kmem_free(strval
, za
.za_num_integers
);
6259 zap_cursor_fini(&zc
);
6262 mutex_exit(&spa
->spa_props_lock
);
6263 if (err
&& err
!= ENOENT
) {
6270 EXPORT_SYMBOL(vdev_fault
);
6271 EXPORT_SYMBOL(vdev_degrade
);
6272 EXPORT_SYMBOL(vdev_online
);
6273 EXPORT_SYMBOL(vdev_offline
);
6274 EXPORT_SYMBOL(vdev_clear
);
6276 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_count
, UINT
, ZMOD_RW
,
6277 "Target number of metaslabs per top-level vdev");
6279 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_shift
, UINT
, ZMOD_RW
,
6280 "Default limit for metaslab size");
6282 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, min_ms_count
, UINT
, ZMOD_RW
,
6283 "Minimum number of metaslabs per top-level vdev");
6285 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, ms_count_limit
, UINT
, ZMOD_RW
,
6286 "Practical upper limit of total metaslabs per top-level vdev");
6288 ZFS_MODULE_PARAM(zfs
, zfs_
, slow_io_events_per_second
, UINT
, ZMOD_RW
,
6289 "Rate limit slow IO (delay) events to this many per second");
6292 ZFS_MODULE_PARAM(zfs
, zfs_
, checksum_events_per_second
, UINT
, ZMOD_RW
,
6293 "Rate limit checksum events to this many checksum errors per second "
6294 "(do not set below ZED threshold).");
6297 ZFS_MODULE_PARAM(zfs
, zfs_
, scan_ignore_errors
, INT
, ZMOD_RW
,
6298 "Ignore errors during resilver/scrub");
6300 ZFS_MODULE_PARAM(zfs_vdev
, vdev_
, validate_skip
, INT
, ZMOD_RW
,
6301 "Bypass vdev_validate()");
6303 ZFS_MODULE_PARAM(zfs
, zfs_
, nocacheflush
, INT
, ZMOD_RW
,
6304 "Disable cache flushes");
6306 ZFS_MODULE_PARAM(zfs
, zfs_
, embedded_slog_min_ms
, UINT
, ZMOD_RW
,
6307 "Minimum number of metaslabs required to dedicate one for log blocks");
6310 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, min_auto_ashift
,
6311 param_set_min_auto_ashift
, param_get_uint
, ZMOD_RW
,
6312 "Minimum ashift used when creating new top-level vdevs");
6314 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, max_auto_ashift
,
6315 param_set_max_auto_ashift
, param_get_uint
, ZMOD_RW
,
6316 "Maximum ashift used when optimizing for logical -> physical sector "
6317 "size on new top-level vdevs");