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
;
734 const char *tmp
= NULL
;
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
, &tmp
) == 0)
836 vd
->vdev_path
= spa_strdup(tmp
);
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
, &tmp
) == 0)
851 vd
->vdev_devid
= spa_strdup(tmp
);
852 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
, &tmp
) == 0)
853 vd
->vdev_physpath
= spa_strdup(tmp
);
855 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
857 vd
->vdev_enc_sysfs_path
= spa_strdup(tmp
);
859 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &tmp
) == 0)
860 vd
->vdev_fru
= spa_strdup(tmp
);
863 * Set the whole_disk property. If it's not specified, leave the value
866 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
867 &vd
->vdev_wholedisk
) != 0)
868 vd
->vdev_wholedisk
= -1ULL;
870 vic
= &vd
->vdev_indirect_config
;
872 ASSERT0(vic
->vic_mapping_object
);
873 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
874 &vic
->vic_mapping_object
);
875 ASSERT0(vic
->vic_births_object
);
876 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
877 &vic
->vic_births_object
);
878 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
879 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
880 &vic
->vic_prev_indirect_vdev
);
883 * Look for the 'not present' flag. This will only be set if the device
884 * was not present at the time of import.
886 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
887 &vd
->vdev_not_present
);
890 * Get the alignment requirement.
892 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
895 * Retrieve the vdev creation time.
897 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
901 * If we're a top-level vdev, try to load the allocation parameters.
904 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
905 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
907 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
909 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
911 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NONALLOCATING
,
913 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
915 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
918 ASSERT0(vd
->vdev_top_zap
);
921 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
922 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
923 alloctype
== VDEV_ALLOC_ADD
||
924 alloctype
== VDEV_ALLOC_SPLIT
||
925 alloctype
== VDEV_ALLOC_ROOTPOOL
);
926 /* Note: metaslab_group_create() is now deferred */
929 if (vd
->vdev_ops
->vdev_op_leaf
&&
930 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
931 (void) nvlist_lookup_uint64(nv
,
932 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
934 ASSERT0(vd
->vdev_leaf_zap
);
938 * If we're a leaf vdev, try to load the DTL object and other state.
941 if (vd
->vdev_ops
->vdev_op_leaf
&&
942 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
943 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
944 if (alloctype
== VDEV_ALLOC_LOAD
) {
945 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
946 &vd
->vdev_dtl_object
);
947 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
951 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
954 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
955 &spare
) == 0 && spare
)
959 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
962 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
963 &vd
->vdev_resilver_txg
);
965 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REBUILD_TXG
,
966 &vd
->vdev_rebuild_txg
);
968 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
969 vdev_defer_resilver(vd
);
972 * In general, when importing a pool we want to ignore the
973 * persistent fault state, as the diagnosis made on another
974 * system may not be valid in the current context. The only
975 * exception is if we forced a vdev to a persistently faulted
976 * state with 'zpool offline -f'. The persistent fault will
977 * remain across imports until cleared.
979 * Local vdevs will remain in the faulted state.
981 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
982 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
983 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
985 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
987 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
990 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
994 VDEV_AUX_ERR_EXCEEDED
;
995 if (nvlist_lookup_string(nv
,
996 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
997 strcmp(aux
, "external") == 0)
998 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
1000 vd
->vdev_faulted
= 0ULL;
1006 * Add ourselves to the parent's list of children.
1008 vdev_add_child(parent
, vd
);
1016 vdev_free(vdev_t
*vd
)
1018 spa_t
*spa
= vd
->vdev_spa
;
1020 ASSERT3P(vd
->vdev_initialize_thread
, ==, NULL
);
1021 ASSERT3P(vd
->vdev_trim_thread
, ==, NULL
);
1022 ASSERT3P(vd
->vdev_autotrim_thread
, ==, NULL
);
1023 ASSERT3P(vd
->vdev_rebuild_thread
, ==, NULL
);
1026 * Scan queues are normally destroyed at the end of a scan. If the
1027 * queue exists here, that implies the vdev is being removed while
1028 * the scan is still running.
1030 if (vd
->vdev_scan_io_queue
!= NULL
) {
1031 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
1032 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
1033 vd
->vdev_scan_io_queue
= NULL
;
1034 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
1038 * vdev_free() implies closing the vdev first. This is simpler than
1039 * trying to ensure complicated semantics for all callers.
1043 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
1044 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
1047 * Free all children.
1049 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1050 vdev_free(vd
->vdev_child
[c
]);
1052 ASSERT(vd
->vdev_child
== NULL
);
1053 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
1055 if (vd
->vdev_ops
->vdev_op_fini
!= NULL
)
1056 vd
->vdev_ops
->vdev_op_fini(vd
);
1059 * Discard allocation state.
1061 if (vd
->vdev_mg
!= NULL
) {
1062 vdev_metaslab_fini(vd
);
1063 metaslab_group_destroy(vd
->vdev_mg
);
1066 if (vd
->vdev_log_mg
!= NULL
) {
1067 ASSERT0(vd
->vdev_ms_count
);
1068 metaslab_group_destroy(vd
->vdev_log_mg
);
1069 vd
->vdev_log_mg
= NULL
;
1072 ASSERT0(vd
->vdev_stat
.vs_space
);
1073 ASSERT0(vd
->vdev_stat
.vs_dspace
);
1074 ASSERT0(vd
->vdev_stat
.vs_alloc
);
1077 * Remove this vdev from its parent's child list.
1079 vdev_remove_child(vd
->vdev_parent
, vd
);
1081 ASSERT(vd
->vdev_parent
== NULL
);
1082 ASSERT(!list_link_active(&vd
->vdev_leaf_node
));
1085 * Clean up vdev structure.
1087 vdev_queue_fini(vd
);
1088 vdev_cache_fini(vd
);
1091 spa_strfree(vd
->vdev_path
);
1093 spa_strfree(vd
->vdev_devid
);
1094 if (vd
->vdev_physpath
)
1095 spa_strfree(vd
->vdev_physpath
);
1097 if (vd
->vdev_enc_sysfs_path
)
1098 spa_strfree(vd
->vdev_enc_sysfs_path
);
1101 spa_strfree(vd
->vdev_fru
);
1103 if (vd
->vdev_isspare
)
1104 spa_spare_remove(vd
);
1105 if (vd
->vdev_isl2cache
)
1106 spa_l2cache_remove(vd
);
1108 txg_list_destroy(&vd
->vdev_ms_list
);
1109 txg_list_destroy(&vd
->vdev_dtl_list
);
1111 mutex_enter(&vd
->vdev_dtl_lock
);
1112 space_map_close(vd
->vdev_dtl_sm
);
1113 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1114 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
1115 range_tree_destroy(vd
->vdev_dtl
[t
]);
1117 mutex_exit(&vd
->vdev_dtl_lock
);
1119 EQUIV(vd
->vdev_indirect_births
!= NULL
,
1120 vd
->vdev_indirect_mapping
!= NULL
);
1121 if (vd
->vdev_indirect_births
!= NULL
) {
1122 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
1123 vdev_indirect_births_close(vd
->vdev_indirect_births
);
1126 if (vd
->vdev_obsolete_sm
!= NULL
) {
1127 ASSERT(vd
->vdev_removing
||
1128 vd
->vdev_ops
== &vdev_indirect_ops
);
1129 space_map_close(vd
->vdev_obsolete_sm
);
1130 vd
->vdev_obsolete_sm
= NULL
;
1132 range_tree_destroy(vd
->vdev_obsolete_segments
);
1133 rw_destroy(&vd
->vdev_indirect_rwlock
);
1134 mutex_destroy(&vd
->vdev_obsolete_lock
);
1136 mutex_destroy(&vd
->vdev_dtl_lock
);
1137 mutex_destroy(&vd
->vdev_stat_lock
);
1138 mutex_destroy(&vd
->vdev_probe_lock
);
1139 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
1141 mutex_destroy(&vd
->vdev_initialize_lock
);
1142 mutex_destroy(&vd
->vdev_initialize_io_lock
);
1143 cv_destroy(&vd
->vdev_initialize_io_cv
);
1144 cv_destroy(&vd
->vdev_initialize_cv
);
1146 mutex_destroy(&vd
->vdev_trim_lock
);
1147 mutex_destroy(&vd
->vdev_autotrim_lock
);
1148 mutex_destroy(&vd
->vdev_trim_io_lock
);
1149 cv_destroy(&vd
->vdev_trim_cv
);
1150 cv_destroy(&vd
->vdev_autotrim_cv
);
1151 cv_destroy(&vd
->vdev_trim_io_cv
);
1153 mutex_destroy(&vd
->vdev_rebuild_lock
);
1154 cv_destroy(&vd
->vdev_rebuild_cv
);
1156 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
1157 zfs_ratelimit_fini(&vd
->vdev_deadman_rl
);
1158 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
1160 if (vd
== spa
->spa_root_vdev
)
1161 spa
->spa_root_vdev
= NULL
;
1163 kmem_free(vd
, sizeof (vdev_t
));
1167 * Transfer top-level vdev state from svd to tvd.
1170 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
1172 spa_t
*spa
= svd
->vdev_spa
;
1177 ASSERT(tvd
== tvd
->vdev_top
);
1179 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
1180 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
1181 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
1182 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
1183 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
1185 svd
->vdev_ms_array
= 0;
1186 svd
->vdev_ms_shift
= 0;
1187 svd
->vdev_ms_count
= 0;
1188 svd
->vdev_top_zap
= 0;
1191 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
1192 if (tvd
->vdev_log_mg
)
1193 ASSERT3P(tvd
->vdev_log_mg
, ==, svd
->vdev_log_mg
);
1194 tvd
->vdev_mg
= svd
->vdev_mg
;
1195 tvd
->vdev_log_mg
= svd
->vdev_log_mg
;
1196 tvd
->vdev_ms
= svd
->vdev_ms
;
1198 svd
->vdev_mg
= NULL
;
1199 svd
->vdev_log_mg
= NULL
;
1200 svd
->vdev_ms
= NULL
;
1202 if (tvd
->vdev_mg
!= NULL
)
1203 tvd
->vdev_mg
->mg_vd
= tvd
;
1204 if (tvd
->vdev_log_mg
!= NULL
)
1205 tvd
->vdev_log_mg
->mg_vd
= tvd
;
1207 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
1208 svd
->vdev_checkpoint_sm
= NULL
;
1210 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1211 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1213 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1214 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1215 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1217 svd
->vdev_stat
.vs_alloc
= 0;
1218 svd
->vdev_stat
.vs_space
= 0;
1219 svd
->vdev_stat
.vs_dspace
= 0;
1222 * State which may be set on a top-level vdev that's in the
1223 * process of being removed.
1225 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1226 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1227 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1228 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1229 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1230 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1231 ASSERT0(tvd
->vdev_noalloc
);
1232 ASSERT0(tvd
->vdev_removing
);
1233 ASSERT0(tvd
->vdev_rebuilding
);
1234 tvd
->vdev_noalloc
= svd
->vdev_noalloc
;
1235 tvd
->vdev_removing
= svd
->vdev_removing
;
1236 tvd
->vdev_rebuilding
= svd
->vdev_rebuilding
;
1237 tvd
->vdev_rebuild_config
= svd
->vdev_rebuild_config
;
1238 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1239 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1240 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1241 range_tree_swap(&svd
->vdev_obsolete_segments
,
1242 &tvd
->vdev_obsolete_segments
);
1243 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1244 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1245 svd
->vdev_indirect_config
.vic_births_object
= 0;
1246 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1247 svd
->vdev_indirect_mapping
= NULL
;
1248 svd
->vdev_indirect_births
= NULL
;
1249 svd
->vdev_obsolete_sm
= NULL
;
1250 svd
->vdev_noalloc
= 0;
1251 svd
->vdev_removing
= 0;
1252 svd
->vdev_rebuilding
= 0;
1254 for (t
= 0; t
< TXG_SIZE
; t
++) {
1255 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1256 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1257 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1258 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1259 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1260 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1263 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1264 vdev_config_clean(svd
);
1265 vdev_config_dirty(tvd
);
1268 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1269 vdev_state_clean(svd
);
1270 vdev_state_dirty(tvd
);
1273 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1274 svd
->vdev_deflate_ratio
= 0;
1276 tvd
->vdev_islog
= svd
->vdev_islog
;
1277 svd
->vdev_islog
= 0;
1279 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1283 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1290 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1291 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1295 * Add a mirror/replacing vdev above an existing vdev. There is no need to
1296 * call .vdev_op_init() since mirror/replacing vdevs do not have private state.
1299 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1301 spa_t
*spa
= cvd
->vdev_spa
;
1302 vdev_t
*pvd
= cvd
->vdev_parent
;
1305 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1307 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1309 mvd
->vdev_asize
= cvd
->vdev_asize
;
1310 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1311 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1312 mvd
->vdev_psize
= cvd
->vdev_psize
;
1313 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1314 mvd
->vdev_logical_ashift
= cvd
->vdev_logical_ashift
;
1315 mvd
->vdev_physical_ashift
= cvd
->vdev_physical_ashift
;
1316 mvd
->vdev_state
= cvd
->vdev_state
;
1317 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1319 vdev_remove_child(pvd
, cvd
);
1320 vdev_add_child(pvd
, mvd
);
1321 cvd
->vdev_id
= mvd
->vdev_children
;
1322 vdev_add_child(mvd
, cvd
);
1323 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1325 if (mvd
== mvd
->vdev_top
)
1326 vdev_top_transfer(cvd
, mvd
);
1332 * Remove a 1-way mirror/replacing vdev from the tree.
1335 vdev_remove_parent(vdev_t
*cvd
)
1337 vdev_t
*mvd
= cvd
->vdev_parent
;
1338 vdev_t
*pvd
= mvd
->vdev_parent
;
1340 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1342 ASSERT(mvd
->vdev_children
== 1);
1343 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1344 mvd
->vdev_ops
== &vdev_replacing_ops
||
1345 mvd
->vdev_ops
== &vdev_spare_ops
);
1346 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1347 cvd
->vdev_logical_ashift
= mvd
->vdev_logical_ashift
;
1348 cvd
->vdev_physical_ashift
= mvd
->vdev_physical_ashift
;
1349 vdev_remove_child(mvd
, cvd
);
1350 vdev_remove_child(pvd
, mvd
);
1353 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1354 * Otherwise, we could have detached an offline device, and when we
1355 * go to import the pool we'll think we have two top-level vdevs,
1356 * instead of a different version of the same top-level vdev.
1358 if (mvd
->vdev_top
== mvd
) {
1359 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1360 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1361 cvd
->vdev_guid
+= guid_delta
;
1362 cvd
->vdev_guid_sum
+= guid_delta
;
1365 * If pool not set for autoexpand, we need to also preserve
1366 * mvd's asize to prevent automatic expansion of cvd.
1367 * Otherwise if we are adjusting the mirror by attaching and
1368 * detaching children of non-uniform sizes, the mirror could
1369 * autoexpand, unexpectedly requiring larger devices to
1370 * re-establish the mirror.
1372 if (!cvd
->vdev_spa
->spa_autoexpand
)
1373 cvd
->vdev_asize
= mvd
->vdev_asize
;
1375 cvd
->vdev_id
= mvd
->vdev_id
;
1376 vdev_add_child(pvd
, cvd
);
1377 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1379 if (cvd
== cvd
->vdev_top
)
1380 vdev_top_transfer(mvd
, cvd
);
1382 ASSERT(mvd
->vdev_children
== 0);
1387 vdev_metaslab_group_create(vdev_t
*vd
)
1389 spa_t
*spa
= vd
->vdev_spa
;
1392 * metaslab_group_create was delayed until allocation bias was available
1394 if (vd
->vdev_mg
== NULL
) {
1395 metaslab_class_t
*mc
;
1397 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1398 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1400 ASSERT3U(vd
->vdev_islog
, ==,
1401 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1403 switch (vd
->vdev_alloc_bias
) {
1405 mc
= spa_log_class(spa
);
1407 case VDEV_BIAS_SPECIAL
:
1408 mc
= spa_special_class(spa
);
1410 case VDEV_BIAS_DEDUP
:
1411 mc
= spa_dedup_class(spa
);
1414 mc
= spa_normal_class(spa
);
1417 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1418 spa
->spa_alloc_count
);
1420 if (!vd
->vdev_islog
) {
1421 vd
->vdev_log_mg
= metaslab_group_create(
1422 spa_embedded_log_class(spa
), vd
, 1);
1426 * The spa ashift min/max only apply for the normal metaslab
1427 * class. Class destination is late binding so ashift boundary
1428 * setting had to wait until now.
1430 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1431 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1432 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1433 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1434 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1435 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1437 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
1438 if (min_alloc
< spa
->spa_min_alloc
)
1439 spa
->spa_min_alloc
= min_alloc
;
1445 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1447 spa_t
*spa
= vd
->vdev_spa
;
1448 uint64_t oldc
= vd
->vdev_ms_count
;
1449 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1452 boolean_t expanding
= (oldc
!= 0);
1454 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1457 * This vdev is not being allocated from yet or is a hole.
1459 if (vd
->vdev_ms_shift
== 0)
1462 ASSERT(!vd
->vdev_ishole
);
1464 ASSERT(oldc
<= newc
);
1466 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1469 memcpy(mspp
, vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1470 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1474 vd
->vdev_ms_count
= newc
;
1476 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1477 uint64_t object
= 0;
1479 * vdev_ms_array may be 0 if we are creating the "fake"
1480 * metaslabs for an indirect vdev for zdb's leak detection.
1481 * See zdb_leak_init().
1483 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1484 error
= dmu_read(spa
->spa_meta_objset
,
1486 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1489 vdev_dbgmsg(vd
, "unable to read the metaslab "
1490 "array [error=%d]", error
);
1495 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1498 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1505 * Find the emptiest metaslab on the vdev and mark it for use for
1506 * embedded slog by moving it from the regular to the log metaslab
1509 if (vd
->vdev_mg
->mg_class
== spa_normal_class(spa
) &&
1510 vd
->vdev_ms_count
> zfs_embedded_slog_min_ms
&&
1511 avl_is_empty(&vd
->vdev_log_mg
->mg_metaslab_tree
)) {
1512 uint64_t slog_msid
= 0;
1513 uint64_t smallest
= UINT64_MAX
;
1516 * Note, we only search the new metaslabs, because the old
1517 * (pre-existing) ones may be active (e.g. have non-empty
1518 * range_tree's), and we don't move them to the new
1521 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1523 space_map_allocated(vd
->vdev_ms
[m
]->ms_sm
);
1524 if (alloc
< smallest
) {
1529 metaslab_t
*slog_ms
= vd
->vdev_ms
[slog_msid
];
1531 * The metaslab was marked as dirty at the end of
1532 * metaslab_init(). Remove it from the dirty list so that we
1533 * can uninitialize and reinitialize it to the new class.
1536 (void) txg_list_remove_this(&vd
->vdev_ms_list
,
1539 uint64_t sm_obj
= space_map_object(slog_ms
->ms_sm
);
1540 metaslab_fini(slog_ms
);
1541 VERIFY0(metaslab_init(vd
->vdev_log_mg
, slog_msid
, sm_obj
, txg
,
1542 &vd
->vdev_ms
[slog_msid
]));
1546 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1549 * If the vdev is marked as non-allocating then don't
1550 * activate the metaslabs since we want to ensure that
1551 * no allocations are performed on this device.
1553 if (vd
->vdev_noalloc
) {
1554 /* track non-allocating vdev space */
1555 spa
->spa_nonallocating_dspace
+= spa_deflate(spa
) ?
1556 vd
->vdev_stat
.vs_dspace
: vd
->vdev_stat
.vs_space
;
1557 } else if (!expanding
) {
1558 metaslab_group_activate(vd
->vdev_mg
);
1559 if (vd
->vdev_log_mg
!= NULL
)
1560 metaslab_group_activate(vd
->vdev_log_mg
);
1564 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1570 vdev_metaslab_fini(vdev_t
*vd
)
1572 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1573 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1574 SPA_FEATURE_POOL_CHECKPOINT
));
1575 space_map_close(vd
->vdev_checkpoint_sm
);
1577 * Even though we close the space map, we need to set its
1578 * pointer to NULL. The reason is that vdev_metaslab_fini()
1579 * may be called multiple times for certain operations
1580 * (i.e. when destroying a pool) so we need to ensure that
1581 * this clause never executes twice. This logic is similar
1582 * to the one used for the vdev_ms clause below.
1584 vd
->vdev_checkpoint_sm
= NULL
;
1587 if (vd
->vdev_ms
!= NULL
) {
1588 metaslab_group_t
*mg
= vd
->vdev_mg
;
1590 metaslab_group_passivate(mg
);
1591 if (vd
->vdev_log_mg
!= NULL
) {
1592 ASSERT(!vd
->vdev_islog
);
1593 metaslab_group_passivate(vd
->vdev_log_mg
);
1596 uint64_t count
= vd
->vdev_ms_count
;
1597 for (uint64_t m
= 0; m
< count
; m
++) {
1598 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1602 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1604 vd
->vdev_ms_count
= 0;
1606 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
1607 ASSERT0(mg
->mg_histogram
[i
]);
1608 if (vd
->vdev_log_mg
!= NULL
)
1609 ASSERT0(vd
->vdev_log_mg
->mg_histogram
[i
]);
1612 ASSERT0(vd
->vdev_ms_count
);
1613 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1616 typedef struct vdev_probe_stats
{
1617 boolean_t vps_readable
;
1618 boolean_t vps_writeable
;
1620 } vdev_probe_stats_t
;
1623 vdev_probe_done(zio_t
*zio
)
1625 spa_t
*spa
= zio
->io_spa
;
1626 vdev_t
*vd
= zio
->io_vd
;
1627 vdev_probe_stats_t
*vps
= zio
->io_private
;
1629 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1631 if (zio
->io_type
== ZIO_TYPE_READ
) {
1632 if (zio
->io_error
== 0)
1633 vps
->vps_readable
= 1;
1634 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1635 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1636 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1637 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1638 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1640 abd_free(zio
->io_abd
);
1642 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1643 if (zio
->io_error
== 0)
1644 vps
->vps_writeable
= 1;
1645 abd_free(zio
->io_abd
);
1646 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1650 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1651 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1653 if (vdev_readable(vd
) &&
1654 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1657 ASSERT(zio
->io_error
!= 0);
1658 vdev_dbgmsg(vd
, "failed probe");
1659 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1660 spa
, vd
, NULL
, NULL
, 0);
1661 zio
->io_error
= SET_ERROR(ENXIO
);
1664 mutex_enter(&vd
->vdev_probe_lock
);
1665 ASSERT(vd
->vdev_probe_zio
== zio
);
1666 vd
->vdev_probe_zio
= NULL
;
1667 mutex_exit(&vd
->vdev_probe_lock
);
1670 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1671 if (!vdev_accessible(vd
, pio
))
1672 pio
->io_error
= SET_ERROR(ENXIO
);
1674 kmem_free(vps
, sizeof (*vps
));
1679 * Determine whether this device is accessible.
1681 * Read and write to several known locations: the pad regions of each
1682 * vdev label but the first, which we leave alone in case it contains
1686 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1688 spa_t
*spa
= vd
->vdev_spa
;
1689 vdev_probe_stats_t
*vps
= NULL
;
1692 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1695 * Don't probe the probe.
1697 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1701 * To prevent 'probe storms' when a device fails, we create
1702 * just one probe i/o at a time. All zios that want to probe
1703 * this vdev will become parents of the probe io.
1705 mutex_enter(&vd
->vdev_probe_lock
);
1707 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1708 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1710 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1711 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1714 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1716 * vdev_cant_read and vdev_cant_write can only
1717 * transition from TRUE to FALSE when we have the
1718 * SCL_ZIO lock as writer; otherwise they can only
1719 * transition from FALSE to TRUE. This ensures that
1720 * any zio looking at these values can assume that
1721 * failures persist for the life of the I/O. That's
1722 * important because when a device has intermittent
1723 * connectivity problems, we want to ensure that
1724 * they're ascribed to the device (ENXIO) and not
1727 * Since we hold SCL_ZIO as writer here, clear both
1728 * values so the probe can reevaluate from first
1731 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1732 vd
->vdev_cant_read
= B_FALSE
;
1733 vd
->vdev_cant_write
= B_FALSE
;
1736 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1737 vdev_probe_done
, vps
,
1738 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1741 * We can't change the vdev state in this context, so we
1742 * kick off an async task to do it on our behalf.
1745 vd
->vdev_probe_wanted
= B_TRUE
;
1746 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1751 zio_add_child(zio
, pio
);
1753 mutex_exit(&vd
->vdev_probe_lock
);
1756 ASSERT(zio
!= NULL
);
1760 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1761 zio_nowait(zio_read_phys(pio
, vd
,
1762 vdev_label_offset(vd
->vdev_psize
, l
,
1763 offsetof(vdev_label_t
, vl_be
)), VDEV_PAD_SIZE
,
1764 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1765 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1766 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1777 vdev_load_child(void *arg
)
1781 vd
->vdev_load_error
= vdev_load(vd
);
1785 vdev_open_child(void *arg
)
1789 vd
->vdev_open_thread
= curthread
;
1790 vd
->vdev_open_error
= vdev_open(vd
);
1791 vd
->vdev_open_thread
= NULL
;
1795 vdev_uses_zvols(vdev_t
*vd
)
1798 if (zvol_is_zvol(vd
->vdev_path
))
1802 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1803 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1810 * Returns B_TRUE if the passed child should be opened.
1813 vdev_default_open_children_func(vdev_t
*vd
)
1820 * Open the requested child vdevs. If any of the leaf vdevs are using
1821 * a ZFS volume then do the opens in a single thread. This avoids a
1822 * deadlock when the current thread is holding the spa_namespace_lock.
1825 vdev_open_children_impl(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1827 int children
= vd
->vdev_children
;
1829 taskq_t
*tq
= taskq_create("vdev_open", children
, minclsyspri
,
1830 children
, children
, TASKQ_PREPOPULATE
);
1831 vd
->vdev_nonrot
= B_TRUE
;
1833 for (int c
= 0; c
< children
; c
++) {
1834 vdev_t
*cvd
= vd
->vdev_child
[c
];
1836 if (open_func(cvd
) == B_FALSE
)
1839 if (tq
== NULL
|| vdev_uses_zvols(vd
)) {
1840 cvd
->vdev_open_error
= vdev_open(cvd
);
1842 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1843 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
1846 vd
->vdev_nonrot
&= cvd
->vdev_nonrot
;
1856 * Open all child vdevs.
1859 vdev_open_children(vdev_t
*vd
)
1861 vdev_open_children_impl(vd
, vdev_default_open_children_func
);
1865 * Conditionally open a subset of child vdevs.
1868 vdev_open_children_subset(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1870 vdev_open_children_impl(vd
, open_func
);
1874 * Compute the raidz-deflation ratio. Note, we hard-code
1875 * in 128k (1 << 17) because it is the "typical" blocksize.
1876 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1877 * otherwise it would inconsistently account for existing bp's.
1880 vdev_set_deflate_ratio(vdev_t
*vd
)
1882 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1883 vd
->vdev_deflate_ratio
= (1 << 17) /
1884 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1889 * Choose the best of two ashifts, preferring one between logical ashift
1890 * (absolute minimum) and administrator defined maximum, otherwise take
1891 * the biggest of the two.
1894 vdev_best_ashift(uint64_t logical
, uint64_t a
, uint64_t b
)
1896 if (a
> logical
&& a
<= zfs_vdev_max_auto_ashift
) {
1897 if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1901 } else if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1907 * Maximize performance by inflating the configured ashift for top level
1908 * vdevs to be as close to the physical ashift as possible while maintaining
1909 * administrator defined limits and ensuring it doesn't go below the
1913 vdev_ashift_optimize(vdev_t
*vd
)
1915 ASSERT(vd
== vd
->vdev_top
);
1917 if (vd
->vdev_ashift
< vd
->vdev_physical_ashift
&&
1918 vd
->vdev_physical_ashift
<= zfs_vdev_max_auto_ashift
) {
1919 vd
->vdev_ashift
= MIN(
1920 MAX(zfs_vdev_max_auto_ashift
, vd
->vdev_ashift
),
1921 MAX(zfs_vdev_min_auto_ashift
,
1922 vd
->vdev_physical_ashift
));
1925 * If the logical and physical ashifts are the same, then
1926 * we ensure that the top-level vdev's ashift is not smaller
1927 * than our minimum ashift value. For the unusual case
1928 * where logical ashift > physical ashift, we can't cap
1929 * the calculated ashift based on max ashift as that
1930 * would cause failures.
1931 * We still check if we need to increase it to match
1934 vd
->vdev_ashift
= MAX(zfs_vdev_min_auto_ashift
,
1940 * Prepare a virtual device for access.
1943 vdev_open(vdev_t
*vd
)
1945 spa_t
*spa
= vd
->vdev_spa
;
1948 uint64_t max_osize
= 0;
1949 uint64_t asize
, max_asize
, psize
;
1950 uint64_t logical_ashift
= 0;
1951 uint64_t physical_ashift
= 0;
1953 ASSERT(vd
->vdev_open_thread
== curthread
||
1954 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1955 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1956 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1957 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1959 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1960 vd
->vdev_cant_read
= B_FALSE
;
1961 vd
->vdev_cant_write
= B_FALSE
;
1962 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1965 * If this vdev is not removed, check its fault status. If it's
1966 * faulted, bail out of the open.
1968 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1969 ASSERT(vd
->vdev_children
== 0);
1970 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1971 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1972 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1973 vd
->vdev_label_aux
);
1974 return (SET_ERROR(ENXIO
));
1975 } else if (vd
->vdev_offline
) {
1976 ASSERT(vd
->vdev_children
== 0);
1977 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1978 return (SET_ERROR(ENXIO
));
1981 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
,
1982 &logical_ashift
, &physical_ashift
);
1984 /* Keep the device in removed state if unplugged */
1985 if (error
== ENOENT
&& vd
->vdev_removed
) {
1986 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_REMOVED
,
1992 * Physical volume size should never be larger than its max size, unless
1993 * the disk has shrunk while we were reading it or the device is buggy
1994 * or damaged: either way it's not safe for use, bail out of the open.
1996 if (osize
> max_osize
) {
1997 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1998 VDEV_AUX_OPEN_FAILED
);
1999 return (SET_ERROR(ENXIO
));
2003 * Reset the vdev_reopening flag so that we actually close
2004 * the vdev on error.
2006 vd
->vdev_reopening
= B_FALSE
;
2007 if (zio_injection_enabled
&& error
== 0)
2008 error
= zio_handle_device_injection(vd
, NULL
, SET_ERROR(ENXIO
));
2011 if (vd
->vdev_removed
&&
2012 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
2013 vd
->vdev_removed
= B_FALSE
;
2015 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
2016 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
2017 vd
->vdev_stat
.vs_aux
);
2019 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2020 vd
->vdev_stat
.vs_aux
);
2025 vd
->vdev_removed
= B_FALSE
;
2028 * Recheck the faulted flag now that we have confirmed that
2029 * the vdev is accessible. If we're faulted, bail.
2031 if (vd
->vdev_faulted
) {
2032 ASSERT(vd
->vdev_children
== 0);
2033 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
2034 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
2035 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2036 vd
->vdev_label_aux
);
2037 return (SET_ERROR(ENXIO
));
2040 if (vd
->vdev_degraded
) {
2041 ASSERT(vd
->vdev_children
== 0);
2042 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2043 VDEV_AUX_ERR_EXCEEDED
);
2045 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
2049 * For hole or missing vdevs we just return success.
2051 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
2054 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2055 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
2056 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2062 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
2063 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
2065 if (vd
->vdev_children
== 0) {
2066 if (osize
< SPA_MINDEVSIZE
) {
2067 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2068 VDEV_AUX_TOO_SMALL
);
2069 return (SET_ERROR(EOVERFLOW
));
2072 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
2073 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
2074 VDEV_LABEL_END_SIZE
);
2076 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
2077 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
2078 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2079 VDEV_AUX_TOO_SMALL
);
2080 return (SET_ERROR(EOVERFLOW
));
2084 max_asize
= max_osize
;
2088 * If the vdev was expanded, record this so that we can re-create the
2089 * uberblock rings in labels {2,3}, during the next sync.
2091 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
2092 vd
->vdev_copy_uberblocks
= B_TRUE
;
2094 vd
->vdev_psize
= psize
;
2097 * Make sure the allocatable size hasn't shrunk too much.
2099 if (asize
< vd
->vdev_min_asize
) {
2100 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2101 VDEV_AUX_BAD_LABEL
);
2102 return (SET_ERROR(EINVAL
));
2106 * We can always set the logical/physical ashift members since
2107 * their values are only used to calculate the vdev_ashift when
2108 * the device is first added to the config. These values should
2109 * not be used for anything else since they may change whenever
2110 * the device is reopened and we don't store them in the label.
2112 vd
->vdev_physical_ashift
=
2113 MAX(physical_ashift
, vd
->vdev_physical_ashift
);
2114 vd
->vdev_logical_ashift
= MAX(logical_ashift
,
2115 vd
->vdev_logical_ashift
);
2117 if (vd
->vdev_asize
== 0) {
2119 * This is the first-ever open, so use the computed values.
2120 * For compatibility, a different ashift can be requested.
2122 vd
->vdev_asize
= asize
;
2123 vd
->vdev_max_asize
= max_asize
;
2126 * If the vdev_ashift was not overridden at creation time,
2127 * then set it the logical ashift and optimize the ashift.
2129 if (vd
->vdev_ashift
== 0) {
2130 vd
->vdev_ashift
= vd
->vdev_logical_ashift
;
2132 if (vd
->vdev_logical_ashift
> ASHIFT_MAX
) {
2133 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2134 VDEV_AUX_ASHIFT_TOO_BIG
);
2135 return (SET_ERROR(EDOM
));
2138 if (vd
->vdev_top
== vd
) {
2139 vdev_ashift_optimize(vd
);
2142 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
2143 vd
->vdev_ashift
> ASHIFT_MAX
)) {
2144 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2145 VDEV_AUX_BAD_ASHIFT
);
2146 return (SET_ERROR(EDOM
));
2150 * Make sure the alignment required hasn't increased.
2152 if (vd
->vdev_ashift
> vd
->vdev_top
->vdev_ashift
&&
2153 vd
->vdev_ops
->vdev_op_leaf
) {
2154 (void) zfs_ereport_post(
2155 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
2156 spa
, vd
, NULL
, NULL
, 0);
2157 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2158 VDEV_AUX_BAD_LABEL
);
2159 return (SET_ERROR(EDOM
));
2161 vd
->vdev_max_asize
= max_asize
;
2165 * If all children are healthy we update asize if either:
2166 * The asize has increased, due to a device expansion caused by dynamic
2167 * LUN growth or vdev replacement, and automatic expansion is enabled;
2168 * making the additional space available.
2170 * The asize has decreased, due to a device shrink usually caused by a
2171 * vdev replace with a smaller device. This ensures that calculations
2172 * based of max_asize and asize e.g. esize are always valid. It's safe
2173 * to do this as we've already validated that asize is greater than
2176 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
2177 ((asize
> vd
->vdev_asize
&&
2178 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
2179 (asize
< vd
->vdev_asize
)))
2180 vd
->vdev_asize
= asize
;
2182 vdev_set_min_asize(vd
);
2185 * Ensure we can issue some IO before declaring the
2186 * vdev open for business.
2188 if (vd
->vdev_ops
->vdev_op_leaf
&&
2189 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
2190 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2191 VDEV_AUX_ERR_EXCEEDED
);
2196 * Track the minimum allocation size.
2198 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
2199 vd
->vdev_islog
== 0 && vd
->vdev_aux
== NULL
) {
2200 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
2201 if (min_alloc
< spa
->spa_min_alloc
)
2202 spa
->spa_min_alloc
= min_alloc
;
2206 * If this is a leaf vdev, assess whether a resilver is needed.
2207 * But don't do this if we are doing a reopen for a scrub, since
2208 * this would just restart the scrub we are already doing.
2210 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
)
2211 dsl_scan_assess_vdev(spa
->spa_dsl_pool
, vd
);
2217 vdev_validate_child(void *arg
)
2221 vd
->vdev_validate_thread
= curthread
;
2222 vd
->vdev_validate_error
= vdev_validate(vd
);
2223 vd
->vdev_validate_thread
= NULL
;
2227 * Called once the vdevs are all opened, this routine validates the label
2228 * contents. This needs to be done before vdev_load() so that we don't
2229 * inadvertently do repair I/Os to the wrong device.
2231 * This function will only return failure if one of the vdevs indicates that it
2232 * has since been destroyed or exported. This is only possible if
2233 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
2234 * will be updated but the function will return 0.
2237 vdev_validate(vdev_t
*vd
)
2239 spa_t
*spa
= vd
->vdev_spa
;
2242 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
2246 int children
= vd
->vdev_children
;
2248 if (vdev_validate_skip
)
2252 tq
= taskq_create("vdev_validate", children
, minclsyspri
,
2253 children
, children
, TASKQ_PREPOPULATE
);
2256 for (uint64_t c
= 0; c
< children
; c
++) {
2257 vdev_t
*cvd
= vd
->vdev_child
[c
];
2259 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
2260 vdev_validate_child(cvd
);
2262 VERIFY(taskq_dispatch(tq
, vdev_validate_child
, cvd
,
2263 TQ_SLEEP
) != TASKQID_INVALID
);
2270 for (int c
= 0; c
< children
; c
++) {
2271 int error
= vd
->vdev_child
[c
]->vdev_validate_error
;
2274 return (SET_ERROR(EBADF
));
2279 * If the device has already failed, or was marked offline, don't do
2280 * any further validation. Otherwise, label I/O will fail and we will
2281 * overwrite the previous state.
2283 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
2287 * If we are performing an extreme rewind, we allow for a label that
2288 * was modified at a point after the current txg.
2289 * If config lock is not held do not check for the txg. spa_sync could
2290 * be updating the vdev's label before updating spa_last_synced_txg.
2292 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
2293 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
2296 txg
= spa_last_synced_txg(spa
);
2298 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
2299 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2300 VDEV_AUX_BAD_LABEL
);
2301 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
2302 "txg %llu", (u_longlong_t
)txg
);
2307 * Determine if this vdev has been split off into another
2308 * pool. If so, then refuse to open it.
2310 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
2311 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
2312 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2313 VDEV_AUX_SPLIT_POOL
);
2315 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
2319 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
2320 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2321 VDEV_AUX_CORRUPT_DATA
);
2323 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2324 ZPOOL_CONFIG_POOL_GUID
);
2329 * If config is not trusted then ignore the spa guid check. This is
2330 * necessary because if the machine crashed during a re-guid the new
2331 * guid might have been written to all of the vdev labels, but not the
2332 * cached config. The check will be performed again once we have the
2333 * trusted config from the MOS.
2335 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
2336 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2337 VDEV_AUX_CORRUPT_DATA
);
2339 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
2340 "match config (%llu != %llu)", (u_longlong_t
)guid
,
2341 (u_longlong_t
)spa_guid(spa
));
2345 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
2346 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
2350 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
2351 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2352 VDEV_AUX_CORRUPT_DATA
);
2354 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2359 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
2361 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2362 VDEV_AUX_CORRUPT_DATA
);
2364 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2365 ZPOOL_CONFIG_TOP_GUID
);
2370 * If this vdev just became a top-level vdev because its sibling was
2371 * detached, it will have adopted the parent's vdev guid -- but the
2372 * label may or may not be on disk yet. Fortunately, either version
2373 * of the label will have the same top guid, so if we're a top-level
2374 * vdev, we can safely compare to that instead.
2375 * However, if the config comes from a cachefile that failed to update
2376 * after the detach, a top-level vdev will appear as a non top-level
2377 * vdev in the config. Also relax the constraints if we perform an
2380 * If we split this vdev off instead, then we also check the
2381 * original pool's guid. We don't want to consider the vdev
2382 * corrupt if it is partway through a split operation.
2384 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
2385 boolean_t mismatch
= B_FALSE
;
2386 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
2387 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
2390 if (vd
->vdev_guid
!= top_guid
&&
2391 vd
->vdev_top
->vdev_guid
!= guid
)
2396 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2397 VDEV_AUX_CORRUPT_DATA
);
2399 vdev_dbgmsg(vd
, "vdev_validate: config guid "
2400 "doesn't match label guid");
2401 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
2402 (u_longlong_t
)vd
->vdev_guid
,
2403 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
2404 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
2405 "aux_guid %llu", (u_longlong_t
)guid
,
2406 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
2411 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
2413 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2414 VDEV_AUX_CORRUPT_DATA
);
2416 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2417 ZPOOL_CONFIG_POOL_STATE
);
2424 * If this is a verbatim import, no need to check the
2425 * state of the pool.
2427 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
2428 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
2429 state
!= POOL_STATE_ACTIVE
) {
2430 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
2431 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
2432 return (SET_ERROR(EBADF
));
2436 * If we were able to open and validate a vdev that was
2437 * previously marked permanently unavailable, clear that state
2440 if (vd
->vdev_not_present
)
2441 vd
->vdev_not_present
= 0;
2447 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2450 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2451 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2452 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2453 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2454 dvd
->vdev_path
, svd
->vdev_path
);
2455 spa_strfree(dvd
->vdev_path
);
2456 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2458 } else if (svd
->vdev_path
!= NULL
) {
2459 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2460 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2461 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2465 * Our enclosure sysfs path may have changed between imports
2467 old
= dvd
->vdev_enc_sysfs_path
;
2468 new = svd
->vdev_enc_sysfs_path
;
2469 if ((old
!= NULL
&& new == NULL
) ||
2470 (old
== NULL
&& new != NULL
) ||
2471 ((old
!= NULL
&& new != NULL
) && strcmp(new, old
) != 0)) {
2472 zfs_dbgmsg("vdev_copy_path: vdev %llu: vdev_enc_sysfs_path "
2473 "changed from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2476 if (dvd
->vdev_enc_sysfs_path
)
2477 spa_strfree(dvd
->vdev_enc_sysfs_path
);
2479 if (svd
->vdev_enc_sysfs_path
) {
2480 dvd
->vdev_enc_sysfs_path
= spa_strdup(
2481 svd
->vdev_enc_sysfs_path
);
2483 dvd
->vdev_enc_sysfs_path
= NULL
;
2489 * Recursively copy vdev paths from one vdev to another. Source and destination
2490 * vdev trees must have same geometry otherwise return error. Intended to copy
2491 * paths from userland config into MOS config.
2494 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2496 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2497 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2498 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2501 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2502 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2503 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2504 return (SET_ERROR(EINVAL
));
2507 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2508 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2509 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2510 (u_longlong_t
)dvd
->vdev_guid
);
2511 return (SET_ERROR(EINVAL
));
2514 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2515 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2516 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2517 (u_longlong_t
)dvd
->vdev_children
);
2518 return (SET_ERROR(EINVAL
));
2521 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2522 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2523 dvd
->vdev_child
[i
]);
2528 if (svd
->vdev_ops
->vdev_op_leaf
)
2529 vdev_copy_path_impl(svd
, dvd
);
2535 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2537 ASSERT(stvd
->vdev_top
== stvd
);
2538 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2540 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2541 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2544 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2548 * The idea here is that while a vdev can shift positions within
2549 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2550 * step outside of it.
2552 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2554 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2557 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2559 vdev_copy_path_impl(vd
, dvd
);
2563 * Recursively copy vdev paths from one root vdev to another. Source and
2564 * destination vdev trees may differ in geometry. For each destination leaf
2565 * vdev, search a vdev with the same guid and top vdev id in the source.
2566 * Intended to copy paths from userland config into MOS config.
2569 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2571 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2572 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2573 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2575 for (uint64_t i
= 0; i
< children
; i
++) {
2576 vdev_copy_path_search(srvd
->vdev_child
[i
],
2577 drvd
->vdev_child
[i
]);
2582 * Close a virtual device.
2585 vdev_close(vdev_t
*vd
)
2587 vdev_t
*pvd
= vd
->vdev_parent
;
2588 spa_t
*spa __maybe_unused
= vd
->vdev_spa
;
2591 ASSERT(vd
->vdev_open_thread
== curthread
||
2592 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2595 * If our parent is reopening, then we are as well, unless we are
2598 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2599 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2601 vd
->vdev_ops
->vdev_op_close(vd
);
2603 vdev_cache_purge(vd
);
2606 * We record the previous state before we close it, so that if we are
2607 * doing a reopen(), we don't generate FMA ereports if we notice that
2608 * it's still faulted.
2610 vd
->vdev_prevstate
= vd
->vdev_state
;
2612 if (vd
->vdev_offline
)
2613 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2615 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2616 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2620 vdev_hold(vdev_t
*vd
)
2622 spa_t
*spa
= vd
->vdev_spa
;
2624 ASSERT(spa_is_root(spa
));
2625 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2628 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2629 vdev_hold(vd
->vdev_child
[c
]);
2631 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_hold
!= NULL
)
2632 vd
->vdev_ops
->vdev_op_hold(vd
);
2636 vdev_rele(vdev_t
*vd
)
2638 ASSERT(spa_is_root(vd
->vdev_spa
));
2639 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2640 vdev_rele(vd
->vdev_child
[c
]);
2642 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_rele
!= NULL
)
2643 vd
->vdev_ops
->vdev_op_rele(vd
);
2647 * Reopen all interior vdevs and any unopened leaves. We don't actually
2648 * reopen leaf vdevs which had previously been opened as they might deadlock
2649 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2650 * If the leaf has never been opened then open it, as usual.
2653 vdev_reopen(vdev_t
*vd
)
2655 spa_t
*spa
= vd
->vdev_spa
;
2657 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2659 /* set the reopening flag unless we're taking the vdev offline */
2660 vd
->vdev_reopening
= !vd
->vdev_offline
;
2662 (void) vdev_open(vd
);
2665 * Call vdev_validate() here to make sure we have the same device.
2666 * Otherwise, a device with an invalid label could be successfully
2667 * opened in response to vdev_reopen().
2670 (void) vdev_validate_aux(vd
);
2671 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2672 vd
->vdev_aux
== &spa
->spa_l2cache
) {
2674 * In case the vdev is present we should evict all ARC
2675 * buffers and pointers to log blocks and reclaim their
2676 * space before restoring its contents to L2ARC.
2678 if (l2arc_vdev_present(vd
)) {
2679 l2arc_rebuild_vdev(vd
, B_TRUE
);
2681 l2arc_add_vdev(spa
, vd
);
2683 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
2684 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
2687 (void) vdev_validate(vd
);
2691 * Reassess parent vdev's health.
2693 vdev_propagate_state(vd
);
2697 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2702 * Normally, partial opens (e.g. of a mirror) are allowed.
2703 * For a create, however, we want to fail the request if
2704 * there are any components we can't open.
2706 error
= vdev_open(vd
);
2708 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2710 return (error
? error
: SET_ERROR(ENXIO
));
2714 * Recursively load DTLs and initialize all labels.
2716 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2717 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2718 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2727 vdev_metaslab_set_size(vdev_t
*vd
)
2729 uint64_t asize
= vd
->vdev_asize
;
2730 uint64_t ms_count
= asize
>> zfs_vdev_default_ms_shift
;
2734 * There are two dimensions to the metaslab sizing calculation:
2735 * the size of the metaslab and the count of metaslabs per vdev.
2737 * The default values used below are a good balance between memory
2738 * usage (larger metaslab size means more memory needed for loaded
2739 * metaslabs; more metaslabs means more memory needed for the
2740 * metaslab_t structs), metaslab load time (larger metaslabs take
2741 * longer to load), and metaslab sync time (more metaslabs means
2742 * more time spent syncing all of them).
2744 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2745 * The range of the dimensions are as follows:
2747 * 2^29 <= ms_size <= 2^34
2748 * 16 <= ms_count <= 131,072
2750 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2751 * at least 512MB (2^29) to minimize fragmentation effects when
2752 * testing with smaller devices. However, the count constraint
2753 * of at least 16 metaslabs will override this minimum size goal.
2755 * On the upper end of vdev sizes, we aim for a maximum metaslab
2756 * size of 16GB. However, we will cap the total count to 2^17
2757 * metaslabs to keep our memory footprint in check and let the
2758 * metaslab size grow from there if that limit is hit.
2760 * The net effect of applying above constrains is summarized below.
2762 * vdev size metaslab count
2763 * --------------|-----------------
2765 * 8GB - 100GB one per 512MB
2767 * 3TB - 2PB one per 16GB
2769 * --------------------------------
2771 * Finally, note that all of the above calculate the initial
2772 * number of metaslabs. Expanding a top-level vdev will result
2773 * in additional metaslabs being allocated making it possible
2774 * to exceed the zfs_vdev_ms_count_limit.
2777 if (ms_count
< zfs_vdev_min_ms_count
)
2778 ms_shift
= highbit64(asize
/ zfs_vdev_min_ms_count
);
2779 else if (ms_count
> zfs_vdev_default_ms_count
)
2780 ms_shift
= highbit64(asize
/ zfs_vdev_default_ms_count
);
2782 ms_shift
= zfs_vdev_default_ms_shift
;
2784 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2785 ms_shift
= SPA_MAXBLOCKSHIFT
;
2786 } else if (ms_shift
> zfs_vdev_max_ms_shift
) {
2787 ms_shift
= zfs_vdev_max_ms_shift
;
2788 /* cap the total count to constrain memory footprint */
2789 if ((asize
>> ms_shift
) > zfs_vdev_ms_count_limit
)
2790 ms_shift
= highbit64(asize
/ zfs_vdev_ms_count_limit
);
2793 vd
->vdev_ms_shift
= ms_shift
;
2794 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2798 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2800 ASSERT(vd
== vd
->vdev_top
);
2801 /* indirect vdevs don't have metaslabs or dtls */
2802 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2803 ASSERT(ISP2(flags
));
2804 ASSERT(spa_writeable(vd
->vdev_spa
));
2806 if (flags
& VDD_METASLAB
)
2807 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2809 if (flags
& VDD_DTL
)
2810 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2812 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2816 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2818 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2819 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2821 if (vd
->vdev_ops
->vdev_op_leaf
)
2822 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2828 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2829 * the vdev has less than perfect replication. There are four kinds of DTL:
2831 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2833 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2835 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2836 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2837 * txgs that was scrubbed.
2839 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2840 * persistent errors or just some device being offline.
2841 * Unlike the other three, the DTL_OUTAGE map is not generally
2842 * maintained; it's only computed when needed, typically to
2843 * determine whether a device can be detached.
2845 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2846 * either has the data or it doesn't.
2848 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2849 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2850 * if any child is less than fully replicated, then so is its parent.
2851 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2852 * comprising only those txgs which appear in 'maxfaults' or more children;
2853 * those are the txgs we don't have enough replication to read. For example,
2854 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2855 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2856 * two child DTL_MISSING maps.
2858 * It should be clear from the above that to compute the DTLs and outage maps
2859 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2860 * Therefore, that is all we keep on disk. When loading the pool, or after
2861 * a configuration change, we generate all other DTLs from first principles.
2864 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2866 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2868 ASSERT(t
< DTL_TYPES
);
2869 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2870 ASSERT(spa_writeable(vd
->vdev_spa
));
2872 mutex_enter(&vd
->vdev_dtl_lock
);
2873 if (!range_tree_contains(rt
, txg
, size
))
2874 range_tree_add(rt
, txg
, size
);
2875 mutex_exit(&vd
->vdev_dtl_lock
);
2879 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2881 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2882 boolean_t dirty
= B_FALSE
;
2884 ASSERT(t
< DTL_TYPES
);
2885 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2888 * While we are loading the pool, the DTLs have not been loaded yet.
2889 * This isn't a problem but it can result in devices being tried
2890 * which are known to not have the data. In which case, the import
2891 * is relying on the checksum to ensure that we get the right data.
2892 * Note that while importing we are only reading the MOS, which is
2893 * always checksummed.
2895 mutex_enter(&vd
->vdev_dtl_lock
);
2896 if (!range_tree_is_empty(rt
))
2897 dirty
= range_tree_contains(rt
, txg
, size
);
2898 mutex_exit(&vd
->vdev_dtl_lock
);
2904 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2906 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2909 mutex_enter(&vd
->vdev_dtl_lock
);
2910 empty
= range_tree_is_empty(rt
);
2911 mutex_exit(&vd
->vdev_dtl_lock
);
2917 * Check if the txg falls within the range which must be
2918 * resilvered. DVAs outside this range can always be skipped.
2921 vdev_default_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2922 uint64_t phys_birth
)
2924 (void) dva
, (void) psize
;
2926 /* Set by sequential resilver. */
2927 if (phys_birth
== TXG_UNKNOWN
)
2930 return (vdev_dtl_contains(vd
, DTL_PARTIAL
, phys_birth
, 1));
2934 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered.
2937 vdev_dtl_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2938 uint64_t phys_birth
)
2940 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2942 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2943 vd
->vdev_ops
->vdev_op_leaf
)
2946 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, dva
, psize
,
2951 * Returns the lowest txg in the DTL range.
2954 vdev_dtl_min(vdev_t
*vd
)
2956 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2957 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2958 ASSERT0(vd
->vdev_children
);
2960 return (range_tree_min(vd
->vdev_dtl
[DTL_MISSING
]) - 1);
2964 * Returns the highest txg in the DTL.
2967 vdev_dtl_max(vdev_t
*vd
)
2969 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2970 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2971 ASSERT0(vd
->vdev_children
);
2973 return (range_tree_max(vd
->vdev_dtl
[DTL_MISSING
]));
2977 * Determine if a resilvering vdev should remove any DTL entries from
2978 * its range. If the vdev was resilvering for the entire duration of the
2979 * scan then it should excise that range from its DTLs. Otherwise, this
2980 * vdev is considered partially resilvered and should leave its DTL
2981 * entries intact. The comment in vdev_dtl_reassess() describes how we
2985 vdev_dtl_should_excise(vdev_t
*vd
, boolean_t rebuild_done
)
2987 ASSERT0(vd
->vdev_children
);
2989 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2992 if (vd
->vdev_resilver_deferred
)
2995 if (range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
2999 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3000 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
3002 /* Rebuild not initiated by attach */
3003 if (vd
->vdev_rebuild_txg
== 0)
3007 * When a rebuild completes without error then all missing data
3008 * up to the rebuild max txg has been reconstructed and the DTL
3009 * is eligible for excision.
3011 if (vrp
->vrp_rebuild_state
== VDEV_REBUILD_COMPLETE
&&
3012 vdev_dtl_max(vd
) <= vrp
->vrp_max_txg
) {
3013 ASSERT3U(vrp
->vrp_min_txg
, <=, vdev_dtl_min(vd
));
3014 ASSERT3U(vrp
->vrp_min_txg
, <, vd
->vdev_rebuild_txg
);
3015 ASSERT3U(vd
->vdev_rebuild_txg
, <=, vrp
->vrp_max_txg
);
3019 dsl_scan_t
*scn
= vd
->vdev_spa
->spa_dsl_pool
->dp_scan
;
3020 dsl_scan_phys_t
*scnp __maybe_unused
= &scn
->scn_phys
;
3022 /* Resilver not initiated by attach */
3023 if (vd
->vdev_resilver_txg
== 0)
3027 * When a resilver is initiated the scan will assign the
3028 * scn_max_txg value to the highest txg value that exists
3029 * in all DTLs. If this device's max DTL is not part of this
3030 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
3031 * then it is not eligible for excision.
3033 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
3034 ASSERT3U(scnp
->scn_min_txg
, <=, vdev_dtl_min(vd
));
3035 ASSERT3U(scnp
->scn_min_txg
, <, vd
->vdev_resilver_txg
);
3036 ASSERT3U(vd
->vdev_resilver_txg
, <=, scnp
->scn_max_txg
);
3045 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
3046 * write operations will be issued to the pool.
3049 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
,
3050 boolean_t scrub_done
, boolean_t rebuild_done
)
3052 spa_t
*spa
= vd
->vdev_spa
;
3056 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
3058 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3059 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
3060 scrub_txg
, scrub_done
, rebuild_done
);
3062 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
3065 if (vd
->vdev_ops
->vdev_op_leaf
) {
3066 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
3067 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3068 boolean_t check_excise
= B_FALSE
;
3069 boolean_t wasempty
= B_TRUE
;
3071 mutex_enter(&vd
->vdev_dtl_lock
);
3074 * If requested, pretend the scan or rebuild completed cleanly.
3076 if (zfs_scan_ignore_errors
) {
3078 scn
->scn_phys
.scn_errors
= 0;
3080 vr
->vr_rebuild_phys
.vrp_errors
= 0;
3083 if (scrub_txg
!= 0 &&
3084 !range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3086 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
3087 "dtl:%llu/%llu errors:%llu",
3088 (u_longlong_t
)vd
->vdev_guid
, (u_longlong_t
)txg
,
3089 (u_longlong_t
)scrub_txg
, spa
->spa_scrub_started
,
3090 (u_longlong_t
)vdev_dtl_min(vd
),
3091 (u_longlong_t
)vdev_dtl_max(vd
),
3092 (u_longlong_t
)(scn
? scn
->scn_phys
.scn_errors
: 0));
3096 * If we've completed a scrub/resilver or a rebuild cleanly
3097 * then determine if this vdev should remove any DTLs. We
3098 * only want to excise regions on vdevs that were available
3099 * during the entire duration of this scan.
3102 vr
!= NULL
&& vr
->vr_rebuild_phys
.vrp_errors
== 0) {
3103 check_excise
= B_TRUE
;
3105 if (spa
->spa_scrub_started
||
3106 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) {
3107 check_excise
= B_TRUE
;
3111 if (scrub_txg
&& check_excise
&&
3112 vdev_dtl_should_excise(vd
, rebuild_done
)) {
3114 * We completed a scrub, resilver or rebuild up to
3115 * scrub_txg. If we did it without rebooting, then
3116 * the scrub dtl will be valid, so excise the old
3117 * region and fold in the scrub dtl. Otherwise,
3118 * leave the dtl as-is if there was an error.
3120 * There's little trick here: to excise the beginning
3121 * of the DTL_MISSING map, we put it into a reference
3122 * tree and then add a segment with refcnt -1 that
3123 * covers the range [0, scrub_txg). This means
3124 * that each txg in that range has refcnt -1 or 0.
3125 * We then add DTL_SCRUB with a refcnt of 2, so that
3126 * entries in the range [0, scrub_txg) will have a
3127 * positive refcnt -- either 1 or 2. We then convert
3128 * the reference tree into the new DTL_MISSING map.
3130 space_reftree_create(&reftree
);
3131 space_reftree_add_map(&reftree
,
3132 vd
->vdev_dtl
[DTL_MISSING
], 1);
3133 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
3134 space_reftree_add_map(&reftree
,
3135 vd
->vdev_dtl
[DTL_SCRUB
], 2);
3136 space_reftree_generate_map(&reftree
,
3137 vd
->vdev_dtl
[DTL_MISSING
], 1);
3138 space_reftree_destroy(&reftree
);
3140 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3141 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
3142 (u_longlong_t
)vdev_dtl_min(vd
),
3143 (u_longlong_t
)vdev_dtl_max(vd
));
3144 } else if (!wasempty
) {
3145 zfs_dbgmsg("DTL_MISSING is now empty");
3148 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
3149 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3150 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
3152 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
3153 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
3154 if (!vdev_readable(vd
))
3155 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
3157 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3158 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
3161 * If the vdev was resilvering or rebuilding and no longer
3162 * has any DTLs then reset the appropriate flag and dirty
3163 * the top level so that we persist the change.
3166 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3167 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
3168 if (vd
->vdev_rebuild_txg
!= 0) {
3169 vd
->vdev_rebuild_txg
= 0;
3170 vdev_config_dirty(vd
->vdev_top
);
3171 } else if (vd
->vdev_resilver_txg
!= 0) {
3172 vd
->vdev_resilver_txg
= 0;
3173 vdev_config_dirty(vd
->vdev_top
);
3177 mutex_exit(&vd
->vdev_dtl_lock
);
3180 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
3184 mutex_enter(&vd
->vdev_dtl_lock
);
3185 for (int t
= 0; t
< DTL_TYPES
; t
++) {
3186 /* account for child's outage in parent's missing map */
3187 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
3189 continue; /* leaf vdevs only */
3190 if (t
== DTL_PARTIAL
)
3191 minref
= 1; /* i.e. non-zero */
3192 else if (vdev_get_nparity(vd
) != 0)
3193 minref
= vdev_get_nparity(vd
) + 1; /* RAID-Z, dRAID */
3195 minref
= vd
->vdev_children
; /* any kind of mirror */
3196 space_reftree_create(&reftree
);
3197 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3198 vdev_t
*cvd
= vd
->vdev_child
[c
];
3199 mutex_enter(&cvd
->vdev_dtl_lock
);
3200 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
3201 mutex_exit(&cvd
->vdev_dtl_lock
);
3203 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
3204 space_reftree_destroy(&reftree
);
3206 mutex_exit(&vd
->vdev_dtl_lock
);
3210 * Iterate over all the vdevs except spare, and post kobj events
3213 vdev_post_kobj_evt(vdev_t
*vd
)
3215 if (vd
->vdev_ops
->vdev_op_kobj_evt_post
&&
3216 vd
->vdev_kobj_flag
== B_FALSE
) {
3217 vd
->vdev_kobj_flag
= B_TRUE
;
3218 vd
->vdev_ops
->vdev_op_kobj_evt_post(vd
);
3221 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3222 vdev_post_kobj_evt(vd
->vdev_child
[c
]);
3226 * Iterate over all the vdevs except spare, and clear kobj events
3229 vdev_clear_kobj_evt(vdev_t
*vd
)
3231 vd
->vdev_kobj_flag
= B_FALSE
;
3233 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3234 vdev_clear_kobj_evt(vd
->vdev_child
[c
]);
3238 vdev_dtl_load(vdev_t
*vd
)
3240 spa_t
*spa
= vd
->vdev_spa
;
3241 objset_t
*mos
= spa
->spa_meta_objset
;
3245 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
3246 ASSERT(vdev_is_concrete(vd
));
3249 * If the dtl cannot be sync'd there is no need to open it.
3251 if (spa
->spa_mode
== SPA_MODE_READ
&& !spa
->spa_read_spacemaps
)
3254 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
3255 vd
->vdev_dtl_object
, 0, -1ULL, 0);
3258 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3260 rt
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3261 error
= space_map_load(vd
->vdev_dtl_sm
, rt
, SM_ALLOC
);
3263 mutex_enter(&vd
->vdev_dtl_lock
);
3264 range_tree_walk(rt
, range_tree_add
,
3265 vd
->vdev_dtl
[DTL_MISSING
]);
3266 mutex_exit(&vd
->vdev_dtl_lock
);
3269 range_tree_vacate(rt
, NULL
, NULL
);
3270 range_tree_destroy(rt
);
3275 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3276 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
3285 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3287 spa_t
*spa
= vd
->vdev_spa
;
3288 objset_t
*mos
= spa
->spa_meta_objset
;
3289 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
3292 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
3295 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
3296 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
3297 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
3299 ASSERT(string
!= NULL
);
3300 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
3301 1, strlen(string
) + 1, string
, tx
));
3303 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
3304 spa_activate_allocation_classes(spa
, tx
);
3309 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
3311 spa_t
*spa
= vd
->vdev_spa
;
3313 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
3314 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3319 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
3321 spa_t
*spa
= vd
->vdev_spa
;
3322 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
3323 DMU_OT_NONE
, 0, tx
);
3326 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3333 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3335 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
3336 vd
->vdev_ops
!= &vdev_missing_ops
&&
3337 vd
->vdev_ops
!= &vdev_root_ops
&&
3338 !vd
->vdev_top
->vdev_removing
) {
3339 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
3340 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
3342 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
3343 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
3344 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
3345 vdev_zap_allocation_data(vd
, tx
);
3349 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3350 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
3355 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
3357 spa_t
*spa
= vd
->vdev_spa
;
3358 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
3359 objset_t
*mos
= spa
->spa_meta_objset
;
3360 range_tree_t
*rtsync
;
3362 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
3364 ASSERT(vdev_is_concrete(vd
));
3365 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3367 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3369 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
3370 mutex_enter(&vd
->vdev_dtl_lock
);
3371 space_map_free(vd
->vdev_dtl_sm
, tx
);
3372 space_map_close(vd
->vdev_dtl_sm
);
3373 vd
->vdev_dtl_sm
= NULL
;
3374 mutex_exit(&vd
->vdev_dtl_lock
);
3377 * We only destroy the leaf ZAP for detached leaves or for
3378 * removed log devices. Removed data devices handle leaf ZAP
3379 * cleanup later, once cancellation is no longer possible.
3381 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
3382 vd
->vdev_top
->vdev_islog
)) {
3383 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
3384 vd
->vdev_leaf_zap
= 0;
3391 if (vd
->vdev_dtl_sm
== NULL
) {
3392 uint64_t new_object
;
3394 new_object
= space_map_alloc(mos
, zfs_vdev_dtl_sm_blksz
, tx
);
3395 VERIFY3U(new_object
, !=, 0);
3397 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
3399 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3402 rtsync
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3404 mutex_enter(&vd
->vdev_dtl_lock
);
3405 range_tree_walk(rt
, range_tree_add
, rtsync
);
3406 mutex_exit(&vd
->vdev_dtl_lock
);
3408 space_map_truncate(vd
->vdev_dtl_sm
, zfs_vdev_dtl_sm_blksz
, tx
);
3409 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
3410 range_tree_vacate(rtsync
, NULL
, NULL
);
3412 range_tree_destroy(rtsync
);
3415 * If the object for the space map has changed then dirty
3416 * the top level so that we update the config.
3418 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
3419 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
3420 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
3421 (u_longlong_t
)object
,
3422 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
3423 vdev_config_dirty(vd
->vdev_top
);
3430 * Determine whether the specified vdev can be offlined/detached/removed
3431 * without losing data.
3434 vdev_dtl_required(vdev_t
*vd
)
3436 spa_t
*spa
= vd
->vdev_spa
;
3437 vdev_t
*tvd
= vd
->vdev_top
;
3438 uint8_t cant_read
= vd
->vdev_cant_read
;
3441 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3443 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
3447 * Temporarily mark the device as unreadable, and then determine
3448 * whether this results in any DTL outages in the top-level vdev.
3449 * If not, we can safely offline/detach/remove the device.
3451 vd
->vdev_cant_read
= B_TRUE
;
3452 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3453 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
3454 vd
->vdev_cant_read
= cant_read
;
3455 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3457 if (!required
&& zio_injection_enabled
) {
3458 required
= !!zio_handle_device_injection(vd
, NULL
,
3466 * Determine if resilver is needed, and if so the txg range.
3469 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
3471 boolean_t needed
= B_FALSE
;
3472 uint64_t thismin
= UINT64_MAX
;
3473 uint64_t thismax
= 0;
3475 if (vd
->vdev_children
== 0) {
3476 mutex_enter(&vd
->vdev_dtl_lock
);
3477 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3478 vdev_writeable(vd
)) {
3480 thismin
= vdev_dtl_min(vd
);
3481 thismax
= vdev_dtl_max(vd
);
3484 mutex_exit(&vd
->vdev_dtl_lock
);
3486 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3487 vdev_t
*cvd
= vd
->vdev_child
[c
];
3488 uint64_t cmin
, cmax
;
3490 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
3491 thismin
= MIN(thismin
, cmin
);
3492 thismax
= MAX(thismax
, cmax
);
3498 if (needed
&& minp
) {
3506 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3507 * will contain either the checkpoint spacemap object or zero if none exists.
3508 * All other errors are returned to the caller.
3511 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
3513 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
3515 if (vd
->vdev_top_zap
== 0) {
3520 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
3521 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
3522 if (error
== ENOENT
) {
3531 vdev_load(vdev_t
*vd
)
3533 int children
= vd
->vdev_children
;
3538 * It's only worthwhile to use the taskq for the root vdev, because the
3539 * slow part is metaslab_init, and that only happens for top-level
3542 if (vd
->vdev_ops
== &vdev_root_ops
&& vd
->vdev_children
> 0) {
3543 tq
= taskq_create("vdev_load", children
, minclsyspri
,
3544 children
, children
, TASKQ_PREPOPULATE
);
3548 * Recursively load all children.
3550 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3551 vdev_t
*cvd
= vd
->vdev_child
[c
];
3553 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
3554 cvd
->vdev_load_error
= vdev_load(cvd
);
3556 VERIFY(taskq_dispatch(tq
, vdev_load_child
,
3557 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
3566 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3567 int error
= vd
->vdev_child
[c
]->vdev_load_error
;
3573 vdev_set_deflate_ratio(vd
);
3576 * On spa_load path, grab the allocation bias from our zap
3578 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3579 spa_t
*spa
= vd
->vdev_spa
;
3582 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3583 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
3586 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
3587 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
3588 } else if (error
!= ENOENT
) {
3589 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3590 VDEV_AUX_CORRUPT_DATA
);
3591 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(top_zap=%llu) "
3592 "failed [error=%d]",
3593 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3598 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3599 spa_t
*spa
= vd
->vdev_spa
;
3602 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3603 vdev_prop_to_name(VDEV_PROP_FAILFAST
), sizeof (failfast
),
3606 vd
->vdev_failfast
= failfast
& 1;
3607 } else if (error
== ENOENT
) {
3608 vd
->vdev_failfast
= vdev_prop_default_numeric(
3609 VDEV_PROP_FAILFAST
);
3612 "vdev_load: zap_lookup(top_zap=%llu) "
3613 "failed [error=%d]",
3614 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3619 * Load any rebuild state from the top-level vdev zap.
3621 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3622 error
= vdev_rebuild_load(vd
);
3623 if (error
&& error
!= ENOTSUP
) {
3624 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3625 VDEV_AUX_CORRUPT_DATA
);
3626 vdev_dbgmsg(vd
, "vdev_load: vdev_rebuild_load "
3627 "failed [error=%d]", error
);
3632 if (vd
->vdev_top_zap
!= 0 || vd
->vdev_leaf_zap
!= 0) {
3635 if (vd
->vdev_top_zap
!= 0)
3636 zapobj
= vd
->vdev_top_zap
;
3638 zapobj
= vd
->vdev_leaf_zap
;
3640 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_N
,
3641 &vd
->vdev_checksum_n
);
3642 if (error
&& error
!= ENOENT
)
3643 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3644 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3646 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_T
,
3647 &vd
->vdev_checksum_t
);
3648 if (error
&& error
!= ENOENT
)
3649 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3650 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3652 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_N
,
3654 if (error
&& error
!= ENOENT
)
3655 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3656 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3658 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_T
,
3660 if (error
&& error
!= ENOENT
)
3661 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3662 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3666 * If this is a top-level vdev, initialize its metaslabs.
3668 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
3669 vdev_metaslab_group_create(vd
);
3671 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
3672 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3673 VDEV_AUX_CORRUPT_DATA
);
3674 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
3675 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
3676 (u_longlong_t
)vd
->vdev_asize
);
3677 return (SET_ERROR(ENXIO
));
3680 error
= vdev_metaslab_init(vd
, 0);
3682 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
3683 "[error=%d]", error
);
3684 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3685 VDEV_AUX_CORRUPT_DATA
);
3689 uint64_t checkpoint_sm_obj
;
3690 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
3691 if (error
== 0 && checkpoint_sm_obj
!= 0) {
3692 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3693 ASSERT(vd
->vdev_asize
!= 0);
3694 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
3696 error
= space_map_open(&vd
->vdev_checkpoint_sm
,
3697 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
3700 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3701 "failed for checkpoint spacemap (obj %llu) "
3703 (u_longlong_t
)checkpoint_sm_obj
, error
);
3706 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3709 * Since the checkpoint_sm contains free entries
3710 * exclusively we can use space_map_allocated() to
3711 * indicate the cumulative checkpointed space that
3714 vd
->vdev_stat
.vs_checkpoint_space
=
3715 -space_map_allocated(vd
->vdev_checkpoint_sm
);
3716 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3717 vd
->vdev_stat
.vs_checkpoint_space
;
3718 } else if (error
!= 0) {
3719 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3720 "checkpoint space map object from vdev ZAP "
3721 "[error=%d]", error
);
3727 * If this is a leaf vdev, load its DTL.
3729 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3730 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3731 VDEV_AUX_CORRUPT_DATA
);
3732 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3733 "[error=%d]", error
);
3737 uint64_t obsolete_sm_object
;
3738 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3739 if (error
== 0 && obsolete_sm_object
!= 0) {
3740 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3741 ASSERT(vd
->vdev_asize
!= 0);
3742 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3744 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3745 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3746 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3747 VDEV_AUX_CORRUPT_DATA
);
3748 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3749 "obsolete spacemap (obj %llu) [error=%d]",
3750 (u_longlong_t
)obsolete_sm_object
, error
);
3753 } else if (error
!= 0) {
3754 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve obsolete "
3755 "space map object from vdev ZAP [error=%d]", error
);
3763 * The special vdev case is used for hot spares and l2cache devices. Its
3764 * sole purpose it to set the vdev state for the associated vdev. To do this,
3765 * we make sure that we can open the underlying device, then try to read the
3766 * label, and make sure that the label is sane and that it hasn't been
3767 * repurposed to another pool.
3770 vdev_validate_aux(vdev_t
*vd
)
3773 uint64_t guid
, version
;
3776 if (!vdev_readable(vd
))
3779 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
3780 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3781 VDEV_AUX_CORRUPT_DATA
);
3785 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
3786 !SPA_VERSION_IS_SUPPORTED(version
) ||
3787 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
3788 guid
!= vd
->vdev_guid
||
3789 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
3790 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3791 VDEV_AUX_CORRUPT_DATA
);
3797 * We don't actually check the pool state here. If it's in fact in
3798 * use by another pool, we update this fact on the fly when requested.
3805 vdev_destroy_ms_flush_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3807 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3809 if (vd
->vdev_top_zap
== 0)
3812 uint64_t object
= 0;
3813 int err
= zap_lookup(mos
, vd
->vdev_top_zap
,
3814 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, sizeof (uint64_t), 1, &object
);
3819 VERIFY0(dmu_object_free(mos
, object
, tx
));
3820 VERIFY0(zap_remove(mos
, vd
->vdev_top_zap
,
3821 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, tx
));
3825 * Free the objects used to store this vdev's spacemaps, and the array
3826 * that points to them.
3829 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3831 if (vd
->vdev_ms_array
== 0)
3834 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3835 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
3836 size_t array_bytes
= array_count
* sizeof (uint64_t);
3837 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
3838 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
3839 array_bytes
, smobj_array
, 0));
3841 for (uint64_t i
= 0; i
< array_count
; i
++) {
3842 uint64_t smobj
= smobj_array
[i
];
3846 space_map_free_obj(mos
, smobj
, tx
);
3849 kmem_free(smobj_array
, array_bytes
);
3850 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
3851 vdev_destroy_ms_flush_data(vd
, tx
);
3852 vd
->vdev_ms_array
= 0;
3856 vdev_remove_empty_log(vdev_t
*vd
, uint64_t txg
)
3858 spa_t
*spa
= vd
->vdev_spa
;
3860 ASSERT(vd
->vdev_islog
);
3861 ASSERT(vd
== vd
->vdev_top
);
3862 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
3864 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3866 vdev_destroy_spacemaps(vd
, tx
);
3867 if (vd
->vdev_top_zap
!= 0) {
3868 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3869 vd
->vdev_top_zap
= 0;
3876 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3879 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3881 ASSERT(vdev_is_concrete(vd
));
3883 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
3885 metaslab_sync_done(msp
, txg
);
3888 metaslab_sync_reassess(vd
->vdev_mg
);
3889 if (vd
->vdev_log_mg
!= NULL
)
3890 metaslab_sync_reassess(vd
->vdev_log_mg
);
3895 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3897 spa_t
*spa
= vd
->vdev_spa
;
3901 ASSERT3U(txg
, ==, spa
->spa_syncing_txg
);
3902 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3903 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3904 ASSERT(vd
->vdev_removing
||
3905 vd
->vdev_ops
== &vdev_indirect_ops
);
3907 vdev_indirect_sync_obsolete(vd
, tx
);
3910 * If the vdev is indirect, it can't have dirty
3911 * metaslabs or DTLs.
3913 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3914 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3915 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3921 ASSERT(vdev_is_concrete(vd
));
3923 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3924 !vd
->vdev_removing
) {
3925 ASSERT(vd
== vd
->vdev_top
);
3926 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3927 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3928 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3929 ASSERT(vd
->vdev_ms_array
!= 0);
3930 vdev_config_dirty(vd
);
3933 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3934 metaslab_sync(msp
, txg
);
3935 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3938 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3939 vdev_dtl_sync(lvd
, txg
);
3942 * If this is an empty log device being removed, destroy the
3943 * metadata associated with it.
3945 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3946 vdev_remove_empty_log(vd
, txg
);
3948 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3953 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3955 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3959 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3960 * not be opened, and no I/O is attempted.
3963 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3967 spa_vdev_state_enter(spa
, SCL_NONE
);
3969 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3970 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3972 if (!vd
->vdev_ops
->vdev_op_leaf
)
3973 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
3978 * If user did a 'zpool offline -f' then make the fault persist across
3981 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3983 * There are two kinds of forced faults: temporary and
3984 * persistent. Temporary faults go away at pool import, while
3985 * persistent faults stay set. Both types of faults can be
3986 * cleared with a zpool clear.
3988 * We tell if a vdev is persistently faulted by looking at the
3989 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3990 * import then it's a persistent fault. Otherwise, it's
3991 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3992 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3993 * tells vdev_config_generate() (which gets run later) to set
3994 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3996 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3997 vd
->vdev_tmpoffline
= B_FALSE
;
3998 aux
= VDEV_AUX_EXTERNAL
;
4000 vd
->vdev_tmpoffline
= B_TRUE
;
4004 * We don't directly use the aux state here, but if we do a
4005 * vdev_reopen(), we need this value to be present to remember why we
4008 vd
->vdev_label_aux
= aux
;
4011 * Faulted state takes precedence over degraded.
4013 vd
->vdev_delayed_close
= B_FALSE
;
4014 vd
->vdev_faulted
= 1ULL;
4015 vd
->vdev_degraded
= 0ULL;
4016 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
4019 * If this device has the only valid copy of the data, then
4020 * back off and simply mark the vdev as degraded instead.
4022 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
4023 vd
->vdev_degraded
= 1ULL;
4024 vd
->vdev_faulted
= 0ULL;
4027 * If we reopen the device and it's not dead, only then do we
4032 if (vdev_readable(vd
))
4033 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
4036 return (spa_vdev_state_exit(spa
, vd
, 0));
4040 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
4041 * user that something is wrong. The vdev continues to operate as normal as far
4042 * as I/O is concerned.
4045 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
4049 spa_vdev_state_enter(spa
, SCL_NONE
);
4051 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4052 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4054 if (!vd
->vdev_ops
->vdev_op_leaf
)
4055 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4058 * If the vdev is already faulted, then don't do anything.
4060 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
4061 return (spa_vdev_state_exit(spa
, NULL
, 0));
4063 vd
->vdev_degraded
= 1ULL;
4064 if (!vdev_is_dead(vd
))
4065 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
4068 return (spa_vdev_state_exit(spa
, vd
, 0));
4072 vdev_remove_wanted(spa_t
*spa
, uint64_t guid
)
4076 spa_vdev_state_enter(spa
, SCL_NONE
);
4078 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4079 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4082 * If the vdev is already removed, then don't do anything.
4084 if (vd
->vdev_removed
)
4085 return (spa_vdev_state_exit(spa
, NULL
, 0));
4087 vd
->vdev_remove_wanted
= B_TRUE
;
4088 spa_async_request(spa
, SPA_ASYNC_REMOVE
);
4090 return (spa_vdev_state_exit(spa
, vd
, 0));
4095 * Online the given vdev.
4097 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
4098 * spare device should be detached when the device finishes resilvering.
4099 * Second, the online should be treated like a 'test' online case, so no FMA
4100 * events are generated if the device fails to open.
4103 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
4105 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
4106 boolean_t wasoffline
;
4107 vdev_state_t oldstate
;
4109 spa_vdev_state_enter(spa
, SCL_NONE
);
4111 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4112 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4114 if (!vd
->vdev_ops
->vdev_op_leaf
)
4115 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4117 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
4118 oldstate
= vd
->vdev_state
;
4121 vd
->vdev_offline
= B_FALSE
;
4122 vd
->vdev_tmpoffline
= B_FALSE
;
4123 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
4124 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
4126 /* XXX - L2ARC 1.0 does not support expansion */
4127 if (!vd
->vdev_aux
) {
4128 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4129 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
4130 spa
->spa_autoexpand
);
4131 vd
->vdev_expansion_time
= gethrestime_sec();
4135 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
4137 if (!vd
->vdev_aux
) {
4138 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4139 pvd
->vdev_expanding
= B_FALSE
;
4143 *newstate
= vd
->vdev_state
;
4144 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
4145 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
4146 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4147 vd
->vdev_parent
->vdev_child
[0] == vd
)
4148 vd
->vdev_unspare
= B_TRUE
;
4150 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
4152 /* XXX - L2ARC 1.0 does not support expansion */
4154 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
4155 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4158 /* Restart initializing if necessary */
4159 mutex_enter(&vd
->vdev_initialize_lock
);
4160 if (vdev_writeable(vd
) &&
4161 vd
->vdev_initialize_thread
== NULL
&&
4162 vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) {
4163 (void) vdev_initialize(vd
);
4165 mutex_exit(&vd
->vdev_initialize_lock
);
4168 * Restart trimming if necessary. We do not restart trimming for cache
4169 * devices here. This is triggered by l2arc_rebuild_vdev()
4170 * asynchronously for the whole device or in l2arc_evict() as it evicts
4171 * space for upcoming writes.
4173 mutex_enter(&vd
->vdev_trim_lock
);
4174 if (vdev_writeable(vd
) && !vd
->vdev_isl2cache
&&
4175 vd
->vdev_trim_thread
== NULL
&&
4176 vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
) {
4177 (void) vdev_trim(vd
, vd
->vdev_trim_rate
, vd
->vdev_trim_partial
,
4178 vd
->vdev_trim_secure
);
4180 mutex_exit(&vd
->vdev_trim_lock
);
4183 (oldstate
< VDEV_STATE_DEGRADED
&&
4184 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
4185 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
4187 return (spa_vdev_state_exit(spa
, vd
, 0));
4191 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4195 uint64_t generation
;
4196 metaslab_group_t
*mg
;
4199 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4201 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4202 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4204 if (!vd
->vdev_ops
->vdev_op_leaf
)
4205 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4207 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
4208 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
4212 generation
= spa
->spa_config_generation
+ 1;
4215 * If the device isn't already offline, try to offline it.
4217 if (!vd
->vdev_offline
) {
4219 * If this device has the only valid copy of some data,
4220 * don't allow it to be offlined. Log devices are always
4223 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4224 vdev_dtl_required(vd
))
4225 return (spa_vdev_state_exit(spa
, NULL
,
4229 * If the top-level is a slog and it has had allocations
4230 * then proceed. We check that the vdev's metaslab group
4231 * is not NULL since it's possible that we may have just
4232 * added this vdev but not yet initialized its metaslabs.
4234 if (tvd
->vdev_islog
&& mg
!= NULL
) {
4236 * Prevent any future allocations.
4238 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
4239 metaslab_group_passivate(mg
);
4240 (void) spa_vdev_state_exit(spa
, vd
, 0);
4242 error
= spa_reset_logs(spa
);
4245 * If the log device was successfully reset but has
4246 * checkpointed data, do not offline it.
4249 tvd
->vdev_checkpoint_sm
!= NULL
) {
4250 ASSERT3U(space_map_allocated(
4251 tvd
->vdev_checkpoint_sm
), !=, 0);
4252 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
4255 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4258 * Check to see if the config has changed.
4260 if (error
|| generation
!= spa
->spa_config_generation
) {
4261 metaslab_group_activate(mg
);
4263 return (spa_vdev_state_exit(spa
,
4265 (void) spa_vdev_state_exit(spa
, vd
, 0);
4268 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
4272 * Offline this device and reopen its top-level vdev.
4273 * If the top-level vdev is a log device then just offline
4274 * it. Otherwise, if this action results in the top-level
4275 * vdev becoming unusable, undo it and fail the request.
4277 vd
->vdev_offline
= B_TRUE
;
4280 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4281 vdev_is_dead(tvd
)) {
4282 vd
->vdev_offline
= B_FALSE
;
4284 return (spa_vdev_state_exit(spa
, NULL
,
4289 * Add the device back into the metaslab rotor so that
4290 * once we online the device it's open for business.
4292 if (tvd
->vdev_islog
&& mg
!= NULL
)
4293 metaslab_group_activate(mg
);
4296 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
4298 return (spa_vdev_state_exit(spa
, vd
, 0));
4302 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4306 mutex_enter(&spa
->spa_vdev_top_lock
);
4307 error
= vdev_offline_locked(spa
, guid
, flags
);
4308 mutex_exit(&spa
->spa_vdev_top_lock
);
4314 * Clear the error counts associated with this vdev. Unlike vdev_online() and
4315 * vdev_offline(), we assume the spa config is locked. We also clear all
4316 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
4319 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
4321 vdev_t
*rvd
= spa
->spa_root_vdev
;
4323 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
4328 vd
->vdev_stat
.vs_read_errors
= 0;
4329 vd
->vdev_stat
.vs_write_errors
= 0;
4330 vd
->vdev_stat
.vs_checksum_errors
= 0;
4331 vd
->vdev_stat
.vs_slow_ios
= 0;
4333 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4334 vdev_clear(spa
, vd
->vdev_child
[c
]);
4337 * It makes no sense to "clear" an indirect or removed vdev.
4339 if (!vdev_is_concrete(vd
) || vd
->vdev_removed
)
4343 * If we're in the FAULTED state or have experienced failed I/O, then
4344 * clear the persistent state and attempt to reopen the device. We
4345 * also mark the vdev config dirty, so that the new faulted state is
4346 * written out to disk.
4348 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
4349 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
4351 * When reopening in response to a clear event, it may be due to
4352 * a fmadm repair request. In this case, if the device is
4353 * still broken, we want to still post the ereport again.
4355 vd
->vdev_forcefault
= B_TRUE
;
4357 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
4358 vd
->vdev_cant_read
= B_FALSE
;
4359 vd
->vdev_cant_write
= B_FALSE
;
4360 vd
->vdev_stat
.vs_aux
= 0;
4362 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
4364 vd
->vdev_forcefault
= B_FALSE
;
4366 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
4367 vdev_state_dirty(vd
->vdev_top
);
4369 /* If a resilver isn't required, check if vdevs can be culled */
4370 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
) &&
4371 !dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4372 !dsl_scan_resilver_scheduled(spa
->spa_dsl_pool
))
4373 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
4375 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
4379 * When clearing a FMA-diagnosed fault, we always want to
4380 * unspare the device, as we assume that the original spare was
4381 * done in response to the FMA fault.
4383 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
4384 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4385 vd
->vdev_parent
->vdev_child
[0] == vd
)
4386 vd
->vdev_unspare
= B_TRUE
;
4388 /* Clear recent error events cache (i.e. duplicate events tracking) */
4389 zfs_ereport_clear(spa
, vd
);
4393 vdev_is_dead(vdev_t
*vd
)
4396 * Holes and missing devices are always considered "dead".
4397 * This simplifies the code since we don't have to check for
4398 * these types of devices in the various code paths.
4399 * Instead we rely on the fact that we skip over dead devices
4400 * before issuing I/O to them.
4402 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
4403 vd
->vdev_ops
== &vdev_hole_ops
||
4404 vd
->vdev_ops
== &vdev_missing_ops
);
4408 vdev_readable(vdev_t
*vd
)
4410 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
4414 vdev_writeable(vdev_t
*vd
)
4416 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
4417 vdev_is_concrete(vd
));
4421 vdev_allocatable(vdev_t
*vd
)
4423 uint64_t state
= vd
->vdev_state
;
4426 * We currently allow allocations from vdevs which may be in the
4427 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
4428 * fails to reopen then we'll catch it later when we're holding
4429 * the proper locks. Note that we have to get the vdev state
4430 * in a local variable because although it changes atomically,
4431 * we're asking two separate questions about it.
4433 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
4434 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
4435 vd
->vdev_mg
->mg_initialized
);
4439 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
4441 ASSERT(zio
->io_vd
== vd
);
4443 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
4446 if (zio
->io_type
== ZIO_TYPE_READ
)
4447 return (!vd
->vdev_cant_read
);
4449 if (zio
->io_type
== ZIO_TYPE_WRITE
)
4450 return (!vd
->vdev_cant_write
);
4456 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
4459 * Exclude the dRAID spare when aggregating to avoid double counting
4460 * the ops and bytes. These IOs are counted by the physical leaves.
4462 if (cvd
->vdev_ops
== &vdev_draid_spare_ops
)
4465 for (int t
= 0; t
< VS_ZIO_TYPES
; t
++) {
4466 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
4467 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
4470 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
4474 * Get extended stats
4477 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
4482 for (t
= 0; t
< ZIO_TYPES
; t
++) {
4483 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
4484 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
4486 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
4487 vsx
->vsx_total_histo
[t
][b
] +=
4488 cvsx
->vsx_total_histo
[t
][b
];
4492 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
4493 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
4494 vsx
->vsx_queue_histo
[t
][b
] +=
4495 cvsx
->vsx_queue_histo
[t
][b
];
4497 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
4498 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
4500 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
4501 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
4503 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
4504 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
4510 vdev_is_spacemap_addressable(vdev_t
*vd
)
4512 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_SPACEMAP_V2
))
4516 * If double-word space map entries are not enabled we assume
4517 * 47 bits of the space map entry are dedicated to the entry's
4518 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4519 * to calculate the maximum address that can be described by a
4520 * space map entry for the given device.
4522 uint64_t shift
= vd
->vdev_ashift
+ SM_OFFSET_BITS
;
4524 if (shift
>= 63) /* detect potential overflow */
4527 return (vd
->vdev_asize
< (1ULL << shift
));
4531 * Get statistics for the given vdev.
4534 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4538 * If we're getting stats on the root vdev, aggregate the I/O counts
4539 * over all top-level vdevs (i.e. the direct children of the root).
4541 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4543 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
4544 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
4547 memset(vsx
, 0, sizeof (*vsx
));
4549 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4550 vdev_t
*cvd
= vd
->vdev_child
[c
];
4551 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
4552 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
4554 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
4556 vdev_get_child_stat(cvd
, vs
, cvs
);
4558 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
4562 * We're a leaf. Just copy our ZIO active queue stats in. The
4563 * other leaf stats are updated in vdev_stat_update().
4568 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
4570 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
4571 vsx
->vsx_active_queue
[t
] =
4572 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
4573 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
4574 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
4580 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4582 vdev_t
*tvd
= vd
->vdev_top
;
4583 mutex_enter(&vd
->vdev_stat_lock
);
4585 memcpy(vs
, &vd
->vdev_stat
, sizeof (*vs
));
4586 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
4587 vs
->vs_state
= vd
->vdev_state
;
4588 vs
->vs_rsize
= vdev_get_min_asize(vd
);
4590 if (vd
->vdev_ops
->vdev_op_leaf
) {
4591 vs
->vs_pspace
= vd
->vdev_psize
;
4592 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
4593 VDEV_LABEL_END_SIZE
;
4595 * Report initializing progress. Since we don't
4596 * have the initializing locks held, this is only
4597 * an estimate (although a fairly accurate one).
4599 vs
->vs_initialize_bytes_done
=
4600 vd
->vdev_initialize_bytes_done
;
4601 vs
->vs_initialize_bytes_est
=
4602 vd
->vdev_initialize_bytes_est
;
4603 vs
->vs_initialize_state
= vd
->vdev_initialize_state
;
4604 vs
->vs_initialize_action_time
=
4605 vd
->vdev_initialize_action_time
;
4608 * Report manual TRIM progress. Since we don't have
4609 * the manual TRIM locks held, this is only an
4610 * estimate (although fairly accurate one).
4612 vs
->vs_trim_notsup
= !vd
->vdev_has_trim
;
4613 vs
->vs_trim_bytes_done
= vd
->vdev_trim_bytes_done
;
4614 vs
->vs_trim_bytes_est
= vd
->vdev_trim_bytes_est
;
4615 vs
->vs_trim_state
= vd
->vdev_trim_state
;
4616 vs
->vs_trim_action_time
= vd
->vdev_trim_action_time
;
4618 /* Set when there is a deferred resilver. */
4619 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
4623 * Report expandable space on top-level, non-auxiliary devices
4624 * only. The expandable space is reported in terms of metaslab
4625 * sized units since that determines how much space the pool
4628 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
4629 vs
->vs_esize
= P2ALIGN(
4630 vd
->vdev_max_asize
- vd
->vdev_asize
,
4631 1ULL << tvd
->vdev_ms_shift
);
4634 vs
->vs_configured_ashift
= vd
->vdev_top
!= NULL
4635 ? vd
->vdev_top
->vdev_ashift
: vd
->vdev_ashift
;
4636 vs
->vs_logical_ashift
= vd
->vdev_logical_ashift
;
4637 if (vd
->vdev_physical_ashift
<= ASHIFT_MAX
)
4638 vs
->vs_physical_ashift
= vd
->vdev_physical_ashift
;
4640 vs
->vs_physical_ashift
= 0;
4643 * Report fragmentation and rebuild progress for top-level,
4644 * non-auxiliary, concrete devices.
4646 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
4647 vdev_is_concrete(vd
)) {
4649 * The vdev fragmentation rating doesn't take into
4650 * account the embedded slog metaslab (vdev_log_mg).
4651 * Since it's only one metaslab, it would have a tiny
4652 * impact on the overall fragmentation.
4654 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
4655 vd
->vdev_mg
->mg_fragmentation
: 0;
4657 vs
->vs_noalloc
= MAX(vd
->vdev_noalloc
,
4658 tvd
? tvd
->vdev_noalloc
: 0);
4661 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
4662 mutex_exit(&vd
->vdev_stat_lock
);
4666 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
4668 return (vdev_get_stats_ex(vd
, vs
, NULL
));
4672 vdev_clear_stats(vdev_t
*vd
)
4674 mutex_enter(&vd
->vdev_stat_lock
);
4675 vd
->vdev_stat
.vs_space
= 0;
4676 vd
->vdev_stat
.vs_dspace
= 0;
4677 vd
->vdev_stat
.vs_alloc
= 0;
4678 mutex_exit(&vd
->vdev_stat_lock
);
4682 vdev_scan_stat_init(vdev_t
*vd
)
4684 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4686 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4687 vdev_scan_stat_init(vd
->vdev_child
[c
]);
4689 mutex_enter(&vd
->vdev_stat_lock
);
4690 vs
->vs_scan_processed
= 0;
4691 mutex_exit(&vd
->vdev_stat_lock
);
4695 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
4697 spa_t
*spa
= zio
->io_spa
;
4698 vdev_t
*rvd
= spa
->spa_root_vdev
;
4699 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
4701 uint64_t txg
= zio
->io_txg
;
4702 /* Suppress ASAN false positive */
4703 #ifdef __SANITIZE_ADDRESS__
4704 vdev_stat_t
*vs
= vd
? &vd
->vdev_stat
: NULL
;
4705 vdev_stat_ex_t
*vsx
= vd
? &vd
->vdev_stat_ex
: NULL
;
4707 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4708 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
4710 zio_type_t type
= zio
->io_type
;
4711 int flags
= zio
->io_flags
;
4714 * If this i/o is a gang leader, it didn't do any actual work.
4716 if (zio
->io_gang_tree
)
4719 if (zio
->io_error
== 0) {
4721 * If this is a root i/o, don't count it -- we've already
4722 * counted the top-level vdevs, and vdev_get_stats() will
4723 * aggregate them when asked. This reduces contention on
4724 * the root vdev_stat_lock and implicitly handles blocks
4725 * that compress away to holes, for which there is no i/o.
4726 * (Holes never create vdev children, so all the counters
4727 * remain zero, which is what we want.)
4729 * Note: this only applies to successful i/o (io_error == 0)
4730 * because unlike i/o counts, errors are not additive.
4731 * When reading a ditto block, for example, failure of
4732 * one top-level vdev does not imply a root-level error.
4737 ASSERT(vd
== zio
->io_vd
);
4739 if (flags
& ZIO_FLAG_IO_BYPASS
)
4742 mutex_enter(&vd
->vdev_stat_lock
);
4744 if (flags
& ZIO_FLAG_IO_REPAIR
) {
4746 * Repair is the result of a resilver issued by the
4747 * scan thread (spa_sync).
4749 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4750 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
4751 dsl_scan_phys_t
*scn_phys
= &scn
->scn_phys
;
4752 uint64_t *processed
= &scn_phys
->scn_processed
;
4754 if (vd
->vdev_ops
->vdev_op_leaf
)
4755 atomic_add_64(processed
, psize
);
4756 vs
->vs_scan_processed
+= psize
;
4760 * Repair is the result of a rebuild issued by the
4761 * rebuild thread (vdev_rebuild_thread). To avoid
4762 * double counting repaired bytes the virtual dRAID
4763 * spare vdev is excluded from the processed bytes.
4765 if (zio
->io_priority
== ZIO_PRIORITY_REBUILD
) {
4766 vdev_t
*tvd
= vd
->vdev_top
;
4767 vdev_rebuild_t
*vr
= &tvd
->vdev_rebuild_config
;
4768 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
4769 uint64_t *rebuilt
= &vrp
->vrp_bytes_rebuilt
;
4771 if (vd
->vdev_ops
->vdev_op_leaf
&&
4772 vd
->vdev_ops
!= &vdev_draid_spare_ops
) {
4773 atomic_add_64(rebuilt
, psize
);
4775 vs
->vs_rebuild_processed
+= psize
;
4778 if (flags
& ZIO_FLAG_SELF_HEAL
)
4779 vs
->vs_self_healed
+= psize
;
4783 * The bytes/ops/histograms are recorded at the leaf level and
4784 * aggregated into the higher level vdevs in vdev_get_stats().
4786 if (vd
->vdev_ops
->vdev_op_leaf
&&
4787 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
4788 zio_type_t vs_type
= type
;
4789 zio_priority_t priority
= zio
->io_priority
;
4792 * TRIM ops and bytes are reported to user space as
4793 * ZIO_TYPE_IOCTL. This is done to preserve the
4794 * vdev_stat_t structure layout for user space.
4796 if (type
== ZIO_TYPE_TRIM
)
4797 vs_type
= ZIO_TYPE_IOCTL
;
4800 * Solely for the purposes of 'zpool iostat -lqrw'
4801 * reporting use the priority to categorize the IO.
4802 * Only the following are reported to user space:
4804 * ZIO_PRIORITY_SYNC_READ,
4805 * ZIO_PRIORITY_SYNC_WRITE,
4806 * ZIO_PRIORITY_ASYNC_READ,
4807 * ZIO_PRIORITY_ASYNC_WRITE,
4808 * ZIO_PRIORITY_SCRUB,
4809 * ZIO_PRIORITY_TRIM,
4810 * ZIO_PRIORITY_REBUILD.
4812 if (priority
== ZIO_PRIORITY_INITIALIZING
) {
4813 ASSERT3U(type
, ==, ZIO_TYPE_WRITE
);
4814 priority
= ZIO_PRIORITY_ASYNC_WRITE
;
4815 } else if (priority
== ZIO_PRIORITY_REMOVAL
) {
4816 priority
= ((type
== ZIO_TYPE_WRITE
) ?
4817 ZIO_PRIORITY_ASYNC_WRITE
:
4818 ZIO_PRIORITY_ASYNC_READ
);
4821 vs
->vs_ops
[vs_type
]++;
4822 vs
->vs_bytes
[vs_type
] += psize
;
4824 if (flags
& ZIO_FLAG_DELEGATED
) {
4825 vsx
->vsx_agg_histo
[priority
]
4826 [RQ_HISTO(zio
->io_size
)]++;
4828 vsx
->vsx_ind_histo
[priority
]
4829 [RQ_HISTO(zio
->io_size
)]++;
4832 if (zio
->io_delta
&& zio
->io_delay
) {
4833 vsx
->vsx_queue_histo
[priority
]
4834 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
4835 vsx
->vsx_disk_histo
[type
]
4836 [L_HISTO(zio
->io_delay
)]++;
4837 vsx
->vsx_total_histo
[type
]
4838 [L_HISTO(zio
->io_delta
)]++;
4842 mutex_exit(&vd
->vdev_stat_lock
);
4846 if (flags
& ZIO_FLAG_SPECULATIVE
)
4850 * If this is an I/O error that is going to be retried, then ignore the
4851 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4852 * hard errors, when in reality they can happen for any number of
4853 * innocuous reasons (bus resets, MPxIO link failure, etc).
4855 if (zio
->io_error
== EIO
&&
4856 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4860 * Intent logs writes won't propagate their error to the root
4861 * I/O so don't mark these types of failures as pool-level
4864 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4867 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4868 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4869 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4870 spa
->spa_claiming
)) {
4872 * This is either a normal write (not a repair), or it's
4873 * a repair induced by the scrub thread, or it's a repair
4874 * made by zil_claim() during spa_load() in the first txg.
4875 * In the normal case, we commit the DTL change in the same
4876 * txg as the block was born. In the scrub-induced repair
4877 * case, we know that scrubs run in first-pass syncing context,
4878 * so we commit the DTL change in spa_syncing_txg(spa).
4879 * In the zil_claim() case, we commit in spa_first_txg(spa).
4881 * We currently do not make DTL entries for failed spontaneous
4882 * self-healing writes triggered by normal (non-scrubbing)
4883 * reads, because we have no transactional context in which to
4884 * do so -- and it's not clear that it'd be desirable anyway.
4886 if (vd
->vdev_ops
->vdev_op_leaf
) {
4887 uint64_t commit_txg
= txg
;
4888 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4889 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4890 ASSERT(spa_sync_pass(spa
) == 1);
4891 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4892 commit_txg
= spa_syncing_txg(spa
);
4893 } else if (spa
->spa_claiming
) {
4894 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4895 commit_txg
= spa_first_txg(spa
);
4897 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4898 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4900 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4901 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4902 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4905 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4910 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4912 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4913 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4915 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4919 * Update the in-core space usage stats for this vdev, its metaslab class,
4920 * and the root vdev.
4923 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4924 int64_t space_delta
)
4927 int64_t dspace_delta
;
4928 spa_t
*spa
= vd
->vdev_spa
;
4929 vdev_t
*rvd
= spa
->spa_root_vdev
;
4931 ASSERT(vd
== vd
->vdev_top
);
4934 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4935 * factor. We must calculate this here and not at the root vdev
4936 * because the root vdev's psize-to-asize is simply the max of its
4937 * children's, thus not accurate enough for us.
4939 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4941 mutex_enter(&vd
->vdev_stat_lock
);
4942 /* ensure we won't underflow */
4943 if (alloc_delta
< 0) {
4944 ASSERT3U(vd
->vdev_stat
.vs_alloc
, >=, -alloc_delta
);
4947 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4948 vd
->vdev_stat
.vs_space
+= space_delta
;
4949 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4950 mutex_exit(&vd
->vdev_stat_lock
);
4952 /* every class but log contributes to root space stats */
4953 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4954 ASSERT(!vd
->vdev_isl2cache
);
4955 mutex_enter(&rvd
->vdev_stat_lock
);
4956 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4957 rvd
->vdev_stat
.vs_space
+= space_delta
;
4958 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4959 mutex_exit(&rvd
->vdev_stat_lock
);
4961 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4965 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4966 * so that it will be written out next time the vdev configuration is synced.
4967 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4970 vdev_config_dirty(vdev_t
*vd
)
4972 spa_t
*spa
= vd
->vdev_spa
;
4973 vdev_t
*rvd
= spa
->spa_root_vdev
;
4976 ASSERT(spa_writeable(spa
));
4979 * If this is an aux vdev (as with l2cache and spare devices), then we
4980 * update the vdev config manually and set the sync flag.
4982 if (vd
->vdev_aux
!= NULL
) {
4983 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
4987 for (c
= 0; c
< sav
->sav_count
; c
++) {
4988 if (sav
->sav_vdevs
[c
] == vd
)
4992 if (c
== sav
->sav_count
) {
4994 * We're being removed. There's nothing more to do.
4996 ASSERT(sav
->sav_sync
== B_TRUE
);
5000 sav
->sav_sync
= B_TRUE
;
5002 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
5003 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
5004 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
5005 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
5011 * Setting the nvlist in the middle if the array is a little
5012 * sketchy, but it will work.
5014 nvlist_free(aux
[c
]);
5015 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
5021 * The dirty list is protected by the SCL_CONFIG lock. The caller
5022 * must either hold SCL_CONFIG as writer, or must be the sync thread
5023 * (which holds SCL_CONFIG as reader). There's only one sync thread,
5024 * so this is sufficient to ensure mutual exclusion.
5026 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5027 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5028 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5031 for (c
= 0; c
< rvd
->vdev_children
; c
++)
5032 vdev_config_dirty(rvd
->vdev_child
[c
]);
5034 ASSERT(vd
== vd
->vdev_top
);
5036 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
5037 vdev_is_concrete(vd
)) {
5038 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
5044 vdev_config_clean(vdev_t
*vd
)
5046 spa_t
*spa
= vd
->vdev_spa
;
5048 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5049 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5050 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5052 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
5053 list_remove(&spa
->spa_config_dirty_list
, vd
);
5057 * Mark a top-level vdev's state as dirty, so that the next pass of
5058 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
5059 * the state changes from larger config changes because they require
5060 * much less locking, and are often needed for administrative actions.
5063 vdev_state_dirty(vdev_t
*vd
)
5065 spa_t
*spa
= vd
->vdev_spa
;
5067 ASSERT(spa_writeable(spa
));
5068 ASSERT(vd
== vd
->vdev_top
);
5071 * The state list is protected by the SCL_STATE lock. The caller
5072 * must either hold SCL_STATE as writer, or must be the sync thread
5073 * (which holds SCL_STATE as reader). There's only one sync thread,
5074 * so this is sufficient to ensure mutual exclusion.
5076 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5077 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5078 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5080 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
5081 vdev_is_concrete(vd
))
5082 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
5086 vdev_state_clean(vdev_t
*vd
)
5088 spa_t
*spa
= vd
->vdev_spa
;
5090 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5091 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5092 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5094 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
5095 list_remove(&spa
->spa_state_dirty_list
, vd
);
5099 * Propagate vdev state up from children to parent.
5102 vdev_propagate_state(vdev_t
*vd
)
5104 spa_t
*spa
= vd
->vdev_spa
;
5105 vdev_t
*rvd
= spa
->spa_root_vdev
;
5106 int degraded
= 0, faulted
= 0;
5110 if (vd
->vdev_children
> 0) {
5111 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5112 child
= vd
->vdev_child
[c
];
5115 * Don't factor holes or indirect vdevs into the
5118 if (!vdev_is_concrete(child
))
5121 if (!vdev_readable(child
) ||
5122 (!vdev_writeable(child
) && spa_writeable(spa
))) {
5124 * Root special: if there is a top-level log
5125 * device, treat the root vdev as if it were
5128 if (child
->vdev_islog
&& vd
== rvd
)
5132 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
5136 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
5140 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
5143 * Root special: if there is a top-level vdev that cannot be
5144 * opened due to corrupted metadata, then propagate the root
5145 * vdev's aux state as 'corrupt' rather than 'insufficient
5148 if (corrupted
&& vd
== rvd
&&
5149 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
5150 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
5151 VDEV_AUX_CORRUPT_DATA
);
5154 if (vd
->vdev_parent
)
5155 vdev_propagate_state(vd
->vdev_parent
);
5159 * Set a vdev's state. If this is during an open, we don't update the parent
5160 * state, because we're in the process of opening children depth-first.
5161 * Otherwise, we propagate the change to the parent.
5163 * If this routine places a device in a faulted state, an appropriate ereport is
5167 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
5169 uint64_t save_state
;
5170 spa_t
*spa
= vd
->vdev_spa
;
5172 if (state
== vd
->vdev_state
) {
5174 * Since vdev_offline() code path is already in an offline
5175 * state we can miss a statechange event to OFFLINE. Check
5176 * the previous state to catch this condition.
5178 if (vd
->vdev_ops
->vdev_op_leaf
&&
5179 (state
== VDEV_STATE_OFFLINE
) &&
5180 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
5181 /* post an offline state change */
5182 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
5184 vd
->vdev_stat
.vs_aux
= aux
;
5188 save_state
= vd
->vdev_state
;
5190 vd
->vdev_state
= state
;
5191 vd
->vdev_stat
.vs_aux
= aux
;
5194 * If we are setting the vdev state to anything but an open state, then
5195 * always close the underlying device unless the device has requested
5196 * a delayed close (i.e. we're about to remove or fault the device).
5197 * Otherwise, we keep accessible but invalid devices open forever.
5198 * We don't call vdev_close() itself, because that implies some extra
5199 * checks (offline, etc) that we don't want here. This is limited to
5200 * leaf devices, because otherwise closing the device will affect other
5203 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
5204 vd
->vdev_ops
->vdev_op_leaf
)
5205 vd
->vdev_ops
->vdev_op_close(vd
);
5207 if (vd
->vdev_removed
&&
5208 state
== VDEV_STATE_CANT_OPEN
&&
5209 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
5211 * If the previous state is set to VDEV_STATE_REMOVED, then this
5212 * device was previously marked removed and someone attempted to
5213 * reopen it. If this failed due to a nonexistent device, then
5214 * keep the device in the REMOVED state. We also let this be if
5215 * it is one of our special test online cases, which is only
5216 * attempting to online the device and shouldn't generate an FMA
5219 vd
->vdev_state
= VDEV_STATE_REMOVED
;
5220 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
5221 } else if (state
== VDEV_STATE_REMOVED
) {
5222 vd
->vdev_removed
= B_TRUE
;
5223 } else if (state
== VDEV_STATE_CANT_OPEN
) {
5225 * If we fail to open a vdev during an import or recovery, we
5226 * mark it as "not available", which signifies that it was
5227 * never there to begin with. Failure to open such a device
5228 * is not considered an error.
5230 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
5231 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
5232 vd
->vdev_ops
->vdev_op_leaf
)
5233 vd
->vdev_not_present
= 1;
5236 * Post the appropriate ereport. If the 'prevstate' field is
5237 * set to something other than VDEV_STATE_UNKNOWN, it indicates
5238 * that this is part of a vdev_reopen(). In this case, we don't
5239 * want to post the ereport if the device was already in the
5240 * CANT_OPEN state beforehand.
5242 * If the 'checkremove' flag is set, then this is an attempt to
5243 * online the device in response to an insertion event. If we
5244 * hit this case, then we have detected an insertion event for a
5245 * faulted or offline device that wasn't in the removed state.
5246 * In this scenario, we don't post an ereport because we are
5247 * about to replace the device, or attempt an online with
5248 * vdev_forcefault, which will generate the fault for us.
5250 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
5251 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
5252 vd
!= spa
->spa_root_vdev
) {
5256 case VDEV_AUX_OPEN_FAILED
:
5257 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
5259 case VDEV_AUX_CORRUPT_DATA
:
5260 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
5262 case VDEV_AUX_NO_REPLICAS
:
5263 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
5265 case VDEV_AUX_BAD_GUID_SUM
:
5266 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
5268 case VDEV_AUX_TOO_SMALL
:
5269 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
5271 case VDEV_AUX_BAD_LABEL
:
5272 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
5274 case VDEV_AUX_BAD_ASHIFT
:
5275 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
5278 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
5281 (void) zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
5285 /* Erase any notion of persistent removed state */
5286 vd
->vdev_removed
= B_FALSE
;
5288 vd
->vdev_removed
= B_FALSE
;
5292 * Notify ZED of any significant state-change on a leaf vdev.
5295 if (vd
->vdev_ops
->vdev_op_leaf
) {
5296 /* preserve original state from a vdev_reopen() */
5297 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
5298 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
5299 (save_state
<= VDEV_STATE_CLOSED
))
5300 save_state
= vd
->vdev_prevstate
;
5302 /* filter out state change due to initial vdev_open */
5303 if (save_state
> VDEV_STATE_CLOSED
)
5304 zfs_post_state_change(spa
, vd
, save_state
);
5307 if (!isopen
&& vd
->vdev_parent
)
5308 vdev_propagate_state(vd
->vdev_parent
);
5312 vdev_children_are_offline(vdev_t
*vd
)
5314 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
5316 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
5317 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
5325 * Check the vdev configuration to ensure that it's capable of supporting
5326 * a root pool. We do not support partial configuration.
5329 vdev_is_bootable(vdev_t
*vd
)
5331 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5332 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
5334 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
5338 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5339 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
5346 vdev_is_concrete(vdev_t
*vd
)
5348 vdev_ops_t
*ops
= vd
->vdev_ops
;
5349 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
5350 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
5358 * Determine if a log device has valid content. If the vdev was
5359 * removed or faulted in the MOS config then we know that
5360 * the content on the log device has already been written to the pool.
5363 vdev_log_state_valid(vdev_t
*vd
)
5365 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
5369 for (int c
= 0; c
< vd
->vdev_children
; c
++)
5370 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
5377 * Expand a vdev if possible.
5380 vdev_expand(vdev_t
*vd
, uint64_t txg
)
5382 ASSERT(vd
->vdev_top
== vd
);
5383 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5384 ASSERT(vdev_is_concrete(vd
));
5386 vdev_set_deflate_ratio(vd
);
5388 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
5389 vdev_is_concrete(vd
)) {
5390 vdev_metaslab_group_create(vd
);
5391 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
5392 vdev_config_dirty(vd
);
5400 vdev_split(vdev_t
*vd
)
5402 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
5404 VERIFY3U(pvd
->vdev_children
, >, 1);
5406 vdev_remove_child(pvd
, vd
);
5407 vdev_compact_children(pvd
);
5409 ASSERT3P(pvd
->vdev_child
, !=, NULL
);
5411 cvd
= pvd
->vdev_child
[0];
5412 if (pvd
->vdev_children
== 1) {
5413 vdev_remove_parent(cvd
);
5414 cvd
->vdev_splitting
= B_TRUE
;
5416 vdev_propagate_state(cvd
);
5420 vdev_deadman(vdev_t
*vd
, const char *tag
)
5422 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5423 vdev_t
*cvd
= vd
->vdev_child
[c
];
5425 vdev_deadman(cvd
, tag
);
5428 if (vd
->vdev_ops
->vdev_op_leaf
) {
5429 vdev_queue_t
*vq
= &vd
->vdev_queue
;
5431 mutex_enter(&vq
->vq_lock
);
5432 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
5433 spa_t
*spa
= vd
->vdev_spa
;
5437 zfs_dbgmsg("slow vdev: %s has %lu active IOs",
5438 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
5441 * Look at the head of all the pending queues,
5442 * if any I/O has been outstanding for longer than
5443 * the spa_deadman_synctime invoke the deadman logic.
5445 fio
= avl_first(&vq
->vq_active_tree
);
5446 delta
= gethrtime() - fio
->io_timestamp
;
5447 if (delta
> spa_deadman_synctime(spa
))
5448 zio_deadman(fio
, tag
);
5450 mutex_exit(&vq
->vq_lock
);
5455 vdev_defer_resilver(vdev_t
*vd
)
5457 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
5459 vd
->vdev_resilver_deferred
= B_TRUE
;
5460 vd
->vdev_spa
->spa_resilver_deferred
= B_TRUE
;
5464 * Clears the resilver deferred flag on all leaf devs under vd. Returns
5465 * B_TRUE if we have devices that need to be resilvered and are available to
5466 * accept resilver I/Os.
5469 vdev_clear_resilver_deferred(vdev_t
*vd
, dmu_tx_t
*tx
)
5471 boolean_t resilver_needed
= B_FALSE
;
5472 spa_t
*spa
= vd
->vdev_spa
;
5474 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5475 vdev_t
*cvd
= vd
->vdev_child
[c
];
5476 resilver_needed
|= vdev_clear_resilver_deferred(cvd
, tx
);
5479 if (vd
== spa
->spa_root_vdev
&&
5480 spa_feature_is_active(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
5481 spa_feature_decr(spa
, SPA_FEATURE_RESILVER_DEFER
, tx
);
5482 vdev_config_dirty(vd
);
5483 spa
->spa_resilver_deferred
= B_FALSE
;
5484 return (resilver_needed
);
5487 if (!vdev_is_concrete(vd
) || vd
->vdev_aux
||
5488 !vd
->vdev_ops
->vdev_op_leaf
)
5489 return (resilver_needed
);
5491 vd
->vdev_resilver_deferred
= B_FALSE
;
5493 return (!vdev_is_dead(vd
) && !vd
->vdev_offline
&&
5494 vdev_resilver_needed(vd
, NULL
, NULL
));
5498 vdev_xlate_is_empty(range_seg64_t
*rs
)
5500 return (rs
->rs_start
== rs
->rs_end
);
5504 * Translate a logical range to the first contiguous physical range for the
5505 * specified vdev_t. This function is initially called with a leaf vdev and
5506 * will walk each parent vdev until it reaches a top-level vdev. Once the
5507 * top-level is reached the physical range is initialized and the recursive
5508 * function begins to unwind. As it unwinds it calls the parent's vdev
5509 * specific translation function to do the real conversion.
5512 vdev_xlate(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5513 range_seg64_t
*physical_rs
, range_seg64_t
*remain_rs
)
5516 * Walk up the vdev tree
5518 if (vd
!= vd
->vdev_top
) {
5519 vdev_xlate(vd
->vdev_parent
, logical_rs
, physical_rs
,
5523 * We've reached the top-level vdev, initialize the physical
5524 * range to the logical range and set an empty remaining
5525 * range then start to unwind.
5527 physical_rs
->rs_start
= logical_rs
->rs_start
;
5528 physical_rs
->rs_end
= logical_rs
->rs_end
;
5530 remain_rs
->rs_start
= logical_rs
->rs_start
;
5531 remain_rs
->rs_end
= logical_rs
->rs_start
;
5536 vdev_t
*pvd
= vd
->vdev_parent
;
5537 ASSERT3P(pvd
, !=, NULL
);
5538 ASSERT3P(pvd
->vdev_ops
->vdev_op_xlate
, !=, NULL
);
5541 * As this recursive function unwinds, translate the logical
5542 * range into its physical and any remaining components by calling
5543 * the vdev specific translate function.
5545 range_seg64_t intermediate
= { 0 };
5546 pvd
->vdev_ops
->vdev_op_xlate(vd
, physical_rs
, &intermediate
, remain_rs
);
5548 physical_rs
->rs_start
= intermediate
.rs_start
;
5549 physical_rs
->rs_end
= intermediate
.rs_end
;
5553 vdev_xlate_walk(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5554 vdev_xlate_func_t
*func
, void *arg
)
5556 range_seg64_t iter_rs
= *logical_rs
;
5557 range_seg64_t physical_rs
;
5558 range_seg64_t remain_rs
;
5560 while (!vdev_xlate_is_empty(&iter_rs
)) {
5562 vdev_xlate(vd
, &iter_rs
, &physical_rs
, &remain_rs
);
5565 * With raidz and dRAID, it's possible that the logical range
5566 * does not live on this leaf vdev. Only when there is a non-
5567 * zero physical size call the provided function.
5569 if (!vdev_xlate_is_empty(&physical_rs
))
5570 func(arg
, &physical_rs
);
5572 iter_rs
= remain_rs
;
5577 vdev_name(vdev_t
*vd
, char *buf
, int buflen
)
5579 if (vd
->vdev_path
== NULL
) {
5580 if (strcmp(vd
->vdev_ops
->vdev_op_type
, "root") == 0) {
5581 strlcpy(buf
, vd
->vdev_spa
->spa_name
, buflen
);
5582 } else if (!vd
->vdev_ops
->vdev_op_leaf
) {
5583 snprintf(buf
, buflen
, "%s-%llu",
5584 vd
->vdev_ops
->vdev_op_type
,
5585 (u_longlong_t
)vd
->vdev_id
);
5588 strlcpy(buf
, vd
->vdev_path
, buflen
);
5594 * Look at the vdev tree and determine whether any devices are currently being
5598 vdev_replace_in_progress(vdev_t
*vdev
)
5600 ASSERT(spa_config_held(vdev
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
5602 if (vdev
->vdev_ops
== &vdev_replacing_ops
)
5606 * A 'spare' vdev indicates that we have a replace in progress, unless
5607 * it has exactly two children, and the second, the hot spare, has
5608 * finished being resilvered.
5610 if (vdev
->vdev_ops
== &vdev_spare_ops
&& (vdev
->vdev_children
> 2 ||
5611 !vdev_dtl_empty(vdev
->vdev_child
[1], DTL_MISSING
)))
5614 for (int i
= 0; i
< vdev
->vdev_children
; i
++) {
5615 if (vdev_replace_in_progress(vdev
->vdev_child
[i
]))
5623 * Add a (source=src, propname=propval) list to an nvlist.
5626 vdev_prop_add_list(nvlist_t
*nvl
, const char *propname
, const char *strval
,
5627 uint64_t intval
, zprop_source_t src
)
5631 propval
= fnvlist_alloc();
5632 fnvlist_add_uint64(propval
, ZPROP_SOURCE
, src
);
5635 fnvlist_add_string(propval
, ZPROP_VALUE
, strval
);
5637 fnvlist_add_uint64(propval
, ZPROP_VALUE
, intval
);
5639 fnvlist_add_nvlist(nvl
, propname
, propval
);
5640 nvlist_free(propval
);
5644 vdev_props_set_sync(void *arg
, dmu_tx_t
*tx
)
5647 nvlist_t
*nvp
= arg
;
5648 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5649 objset_t
*mos
= spa
->spa_meta_objset
;
5650 nvpair_t
*elem
= NULL
;
5654 vdev_guid
= fnvlist_lookup_uint64(nvp
, ZPOOL_VDEV_PROPS_SET_VDEV
);
5655 nvprops
= fnvlist_lookup_nvlist(nvp
, ZPOOL_VDEV_PROPS_SET_PROPS
);
5656 vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
);
5658 /* this vdev could get removed while waiting for this sync task */
5662 mutex_enter(&spa
->spa_props_lock
);
5664 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5665 uint64_t intval
, objid
= 0;
5668 const char *propname
= nvpair_name(elem
);
5669 zprop_type_t proptype
;
5672 * Set vdev property values in the vdev props mos object.
5674 if (vd
->vdev_top_zap
!= 0) {
5675 objid
= vd
->vdev_top_zap
;
5676 } else if (vd
->vdev_leaf_zap
!= 0) {
5677 objid
= vd
->vdev_leaf_zap
;
5679 panic("vdev not top or leaf");
5682 switch (prop
= vdev_name_to_prop(propname
)) {
5683 case VDEV_PROP_USERPROP
:
5684 if (vdev_prop_user(propname
)) {
5685 strval
= fnvpair_value_string(elem
);
5686 if (strlen(strval
) == 0) {
5687 /* remove the property if value == "" */
5688 (void) zap_remove(mos
, objid
, propname
,
5691 VERIFY0(zap_update(mos
, objid
, propname
,
5692 1, strlen(strval
) + 1, strval
, tx
));
5694 spa_history_log_internal(spa
, "vdev set", tx
,
5695 "vdev_guid=%llu: %s=%s",
5696 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5701 /* normalize the property name */
5702 propname
= vdev_prop_to_name(prop
);
5703 proptype
= vdev_prop_get_type(prop
);
5705 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5706 ASSERT(proptype
== PROP_TYPE_STRING
);
5707 strval
= fnvpair_value_string(elem
);
5708 VERIFY0(zap_update(mos
, objid
, propname
,
5709 1, strlen(strval
) + 1, strval
, tx
));
5710 spa_history_log_internal(spa
, "vdev set", tx
,
5711 "vdev_guid=%llu: %s=%s",
5712 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5714 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
5715 intval
= fnvpair_value_uint64(elem
);
5717 if (proptype
== PROP_TYPE_INDEX
) {
5719 VERIFY0(vdev_prop_index_to_string(
5720 prop
, intval
, &unused
));
5722 VERIFY0(zap_update(mos
, objid
, propname
,
5723 sizeof (uint64_t), 1, &intval
, tx
));
5724 spa_history_log_internal(spa
, "vdev set", tx
,
5725 "vdev_guid=%llu: %s=%lld",
5726 (u_longlong_t
)vdev_guid
,
5727 nvpair_name(elem
), (longlong_t
)intval
);
5729 panic("invalid vdev property type %u",
5736 mutex_exit(&spa
->spa_props_lock
);
5740 vdev_prop_set(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5742 spa_t
*spa
= vd
->vdev_spa
;
5743 nvpair_t
*elem
= NULL
;
5750 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_SET_VDEV
,
5752 return (SET_ERROR(EINVAL
));
5754 if (nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_SET_PROPS
,
5756 return (SET_ERROR(EINVAL
));
5758 if ((vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
)) == NULL
)
5759 return (SET_ERROR(EINVAL
));
5761 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5762 const char *propname
= nvpair_name(elem
);
5763 vdev_prop_t prop
= vdev_name_to_prop(propname
);
5764 uint64_t intval
= 0;
5765 const char *strval
= NULL
;
5767 if (prop
== VDEV_PROP_USERPROP
&& !vdev_prop_user(propname
)) {
5772 if (vdev_prop_readonly(prop
)) {
5777 /* Special Processing */
5779 case VDEV_PROP_PATH
:
5780 if (vd
->vdev_path
== NULL
) {
5784 if (nvpair_value_string(elem
, &strval
) != 0) {
5788 /* New path must start with /dev/ */
5789 if (strncmp(strval
, "/dev/", 5)) {
5793 error
= spa_vdev_setpath(spa
, vdev_guid
, strval
);
5795 case VDEV_PROP_ALLOCATING
:
5796 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5800 if (intval
!= vd
->vdev_noalloc
)
5803 error
= spa_vdev_noalloc(spa
, vdev_guid
);
5805 error
= spa_vdev_alloc(spa
, vdev_guid
);
5807 case VDEV_PROP_FAILFAST
:
5808 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5812 vd
->vdev_failfast
= intval
& 1;
5814 case VDEV_PROP_CHECKSUM_N
:
5815 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5819 vd
->vdev_checksum_n
= intval
;
5821 case VDEV_PROP_CHECKSUM_T
:
5822 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5826 vd
->vdev_checksum_t
= intval
;
5828 case VDEV_PROP_IO_N
:
5829 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5833 vd
->vdev_io_n
= intval
;
5835 case VDEV_PROP_IO_T
:
5836 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5840 vd
->vdev_io_t
= intval
;
5843 /* Most processing is done in vdev_props_set_sync */
5849 vdev_prop_add_list(outnvl
, propname
, strval
, intval
, 0);
5854 return (dsl_sync_task(spa
->spa_name
, NULL
, vdev_props_set_sync
,
5855 innvl
, 6, ZFS_SPACE_CHECK_EXTRA_RESERVED
));
5859 vdev_prop_get(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5861 spa_t
*spa
= vd
->vdev_spa
;
5862 objset_t
*mos
= spa
->spa_meta_objset
;
5866 nvpair_t
*elem
= NULL
;
5867 nvlist_t
*nvprops
= NULL
;
5868 uint64_t intval
= 0;
5869 char *strval
= NULL
;
5870 const char *propname
= NULL
;
5874 ASSERT(mos
!= NULL
);
5876 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_GET_VDEV
,
5878 return (SET_ERROR(EINVAL
));
5880 nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_GET_PROPS
, &nvprops
);
5882 if (vd
->vdev_top_zap
!= 0) {
5883 objid
= vd
->vdev_top_zap
;
5884 } else if (vd
->vdev_leaf_zap
!= 0) {
5885 objid
= vd
->vdev_leaf_zap
;
5887 return (SET_ERROR(EINVAL
));
5891 mutex_enter(&spa
->spa_props_lock
);
5893 if (nvprops
!= NULL
) {
5894 char namebuf
[64] = { 0 };
5896 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5899 propname
= nvpair_name(elem
);
5900 prop
= vdev_name_to_prop(propname
);
5901 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
5902 uint64_t integer_size
, num_integers
;
5905 /* Special Read-only Properties */
5906 case VDEV_PROP_NAME
:
5907 strval
= vdev_name(vd
, namebuf
,
5911 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
5914 case VDEV_PROP_CAPACITY
:
5916 intval
= (vd
->vdev_stat
.vs_dspace
== 0) ? 0 :
5917 (vd
->vdev_stat
.vs_alloc
* 100 /
5918 vd
->vdev_stat
.vs_dspace
);
5919 vdev_prop_add_list(outnvl
, propname
, NULL
,
5920 intval
, ZPROP_SRC_NONE
);
5922 case VDEV_PROP_STATE
:
5923 vdev_prop_add_list(outnvl
, propname
, NULL
,
5924 vd
->vdev_state
, ZPROP_SRC_NONE
);
5926 case VDEV_PROP_GUID
:
5927 vdev_prop_add_list(outnvl
, propname
, NULL
,
5928 vd
->vdev_guid
, ZPROP_SRC_NONE
);
5930 case VDEV_PROP_ASIZE
:
5931 vdev_prop_add_list(outnvl
, propname
, NULL
,
5932 vd
->vdev_asize
, ZPROP_SRC_NONE
);
5934 case VDEV_PROP_PSIZE
:
5935 vdev_prop_add_list(outnvl
, propname
, NULL
,
5936 vd
->vdev_psize
, ZPROP_SRC_NONE
);
5938 case VDEV_PROP_ASHIFT
:
5939 vdev_prop_add_list(outnvl
, propname
, NULL
,
5940 vd
->vdev_ashift
, ZPROP_SRC_NONE
);
5942 case VDEV_PROP_SIZE
:
5943 vdev_prop_add_list(outnvl
, propname
, NULL
,
5944 vd
->vdev_stat
.vs_dspace
, ZPROP_SRC_NONE
);
5946 case VDEV_PROP_FREE
:
5947 vdev_prop_add_list(outnvl
, propname
, NULL
,
5948 vd
->vdev_stat
.vs_dspace
-
5949 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
5951 case VDEV_PROP_ALLOCATED
:
5952 vdev_prop_add_list(outnvl
, propname
, NULL
,
5953 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
5955 case VDEV_PROP_EXPANDSZ
:
5956 vdev_prop_add_list(outnvl
, propname
, NULL
,
5957 vd
->vdev_stat
.vs_esize
, ZPROP_SRC_NONE
);
5959 case VDEV_PROP_FRAGMENTATION
:
5960 vdev_prop_add_list(outnvl
, propname
, NULL
,
5961 vd
->vdev_stat
.vs_fragmentation
,
5964 case VDEV_PROP_PARITY
:
5965 vdev_prop_add_list(outnvl
, propname
, NULL
,
5966 vdev_get_nparity(vd
), ZPROP_SRC_NONE
);
5968 case VDEV_PROP_PATH
:
5969 if (vd
->vdev_path
== NULL
)
5971 vdev_prop_add_list(outnvl
, propname
,
5972 vd
->vdev_path
, 0, ZPROP_SRC_NONE
);
5974 case VDEV_PROP_DEVID
:
5975 if (vd
->vdev_devid
== NULL
)
5977 vdev_prop_add_list(outnvl
, propname
,
5978 vd
->vdev_devid
, 0, ZPROP_SRC_NONE
);
5980 case VDEV_PROP_PHYS_PATH
:
5981 if (vd
->vdev_physpath
== NULL
)
5983 vdev_prop_add_list(outnvl
, propname
,
5984 vd
->vdev_physpath
, 0, ZPROP_SRC_NONE
);
5986 case VDEV_PROP_ENC_PATH
:
5987 if (vd
->vdev_enc_sysfs_path
== NULL
)
5989 vdev_prop_add_list(outnvl
, propname
,
5990 vd
->vdev_enc_sysfs_path
, 0, ZPROP_SRC_NONE
);
5993 if (vd
->vdev_fru
== NULL
)
5995 vdev_prop_add_list(outnvl
, propname
,
5996 vd
->vdev_fru
, 0, ZPROP_SRC_NONE
);
5998 case VDEV_PROP_PARENT
:
5999 if (vd
->vdev_parent
!= NULL
) {
6000 strval
= vdev_name(vd
->vdev_parent
,
6001 namebuf
, sizeof (namebuf
));
6002 vdev_prop_add_list(outnvl
, propname
,
6003 strval
, 0, ZPROP_SRC_NONE
);
6006 case VDEV_PROP_CHILDREN
:
6007 if (vd
->vdev_children
> 0)
6008 strval
= kmem_zalloc(ZAP_MAXVALUELEN
,
6010 for (uint64_t i
= 0; i
< vd
->vdev_children
;
6014 vname
= vdev_name(vd
->vdev_child
[i
],
6015 namebuf
, sizeof (namebuf
));
6017 vname
= "(unknown)";
6018 if (strlen(strval
) > 0)
6019 strlcat(strval
, ",",
6021 strlcat(strval
, vname
, ZAP_MAXVALUELEN
);
6023 if (strval
!= NULL
) {
6024 vdev_prop_add_list(outnvl
, propname
,
6025 strval
, 0, ZPROP_SRC_NONE
);
6026 kmem_free(strval
, ZAP_MAXVALUELEN
);
6029 case VDEV_PROP_NUMCHILDREN
:
6030 vdev_prop_add_list(outnvl
, propname
, NULL
,
6031 vd
->vdev_children
, ZPROP_SRC_NONE
);
6033 case VDEV_PROP_READ_ERRORS
:
6034 vdev_prop_add_list(outnvl
, propname
, NULL
,
6035 vd
->vdev_stat
.vs_read_errors
,
6038 case VDEV_PROP_WRITE_ERRORS
:
6039 vdev_prop_add_list(outnvl
, propname
, NULL
,
6040 vd
->vdev_stat
.vs_write_errors
,
6043 case VDEV_PROP_CHECKSUM_ERRORS
:
6044 vdev_prop_add_list(outnvl
, propname
, NULL
,
6045 vd
->vdev_stat
.vs_checksum_errors
,
6048 case VDEV_PROP_INITIALIZE_ERRORS
:
6049 vdev_prop_add_list(outnvl
, propname
, NULL
,
6050 vd
->vdev_stat
.vs_initialize_errors
,
6053 case VDEV_PROP_OPS_NULL
:
6054 vdev_prop_add_list(outnvl
, propname
, NULL
,
6055 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_NULL
],
6058 case VDEV_PROP_OPS_READ
:
6059 vdev_prop_add_list(outnvl
, propname
, NULL
,
6060 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_READ
],
6063 case VDEV_PROP_OPS_WRITE
:
6064 vdev_prop_add_list(outnvl
, propname
, NULL
,
6065 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_WRITE
],
6068 case VDEV_PROP_OPS_FREE
:
6069 vdev_prop_add_list(outnvl
, propname
, NULL
,
6070 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_FREE
],
6073 case VDEV_PROP_OPS_CLAIM
:
6074 vdev_prop_add_list(outnvl
, propname
, NULL
,
6075 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_CLAIM
],
6078 case VDEV_PROP_OPS_TRIM
:
6080 * TRIM ops and bytes are reported to user
6081 * space as ZIO_TYPE_IOCTL. This is done to
6082 * preserve the vdev_stat_t structure layout
6085 vdev_prop_add_list(outnvl
, propname
, NULL
,
6086 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_IOCTL
],
6089 case VDEV_PROP_BYTES_NULL
:
6090 vdev_prop_add_list(outnvl
, propname
, NULL
,
6091 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_NULL
],
6094 case VDEV_PROP_BYTES_READ
:
6095 vdev_prop_add_list(outnvl
, propname
, NULL
,
6096 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_READ
],
6099 case VDEV_PROP_BYTES_WRITE
:
6100 vdev_prop_add_list(outnvl
, propname
, NULL
,
6101 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_WRITE
],
6104 case VDEV_PROP_BYTES_FREE
:
6105 vdev_prop_add_list(outnvl
, propname
, NULL
,
6106 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_FREE
],
6109 case VDEV_PROP_BYTES_CLAIM
:
6110 vdev_prop_add_list(outnvl
, propname
, NULL
,
6111 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_CLAIM
],
6114 case VDEV_PROP_BYTES_TRIM
:
6116 * TRIM ops and bytes are reported to user
6117 * space as ZIO_TYPE_IOCTL. This is done to
6118 * preserve the vdev_stat_t structure layout
6121 vdev_prop_add_list(outnvl
, propname
, NULL
,
6122 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_IOCTL
],
6125 case VDEV_PROP_REMOVING
:
6126 vdev_prop_add_list(outnvl
, propname
, NULL
,
6127 vd
->vdev_removing
, ZPROP_SRC_NONE
);
6129 /* Numeric Properites */
6130 case VDEV_PROP_ALLOCATING
:
6131 /* Leaf vdevs cannot have this property */
6132 if (vd
->vdev_mg
== NULL
&&
6133 vd
->vdev_top
!= NULL
) {
6134 src
= ZPROP_SRC_NONE
;
6135 intval
= ZPROP_BOOLEAN_NA
;
6137 err
= vdev_prop_get_int(vd
, prop
,
6139 if (err
&& err
!= ENOENT
)
6143 vdev_prop_default_numeric(prop
))
6144 src
= ZPROP_SRC_DEFAULT
;
6146 src
= ZPROP_SRC_LOCAL
;
6149 vdev_prop_add_list(outnvl
, propname
, NULL
,
6152 case VDEV_PROP_FAILFAST
:
6153 src
= ZPROP_SRC_LOCAL
;
6156 err
= zap_lookup(mos
, objid
, nvpair_name(elem
),
6157 sizeof (uint64_t), 1, &intval
);
6158 if (err
== ENOENT
) {
6159 intval
= vdev_prop_default_numeric(
6165 if (intval
== vdev_prop_default_numeric(prop
))
6166 src
= ZPROP_SRC_DEFAULT
;
6168 vdev_prop_add_list(outnvl
, propname
, strval
,
6171 case VDEV_PROP_CHECKSUM_N
:
6172 case VDEV_PROP_CHECKSUM_T
:
6173 case VDEV_PROP_IO_N
:
6174 case VDEV_PROP_IO_T
:
6175 err
= vdev_prop_get_int(vd
, prop
, &intval
);
6176 if (err
&& err
!= ENOENT
)
6179 if (intval
== vdev_prop_default_numeric(prop
))
6180 src
= ZPROP_SRC_DEFAULT
;
6182 src
= ZPROP_SRC_LOCAL
;
6184 vdev_prop_add_list(outnvl
, propname
, NULL
,
6187 /* Text Properties */
6188 case VDEV_PROP_COMMENT
:
6189 /* Exists in the ZAP below */
6191 case VDEV_PROP_USERPROP
:
6192 /* User Properites */
6193 src
= ZPROP_SRC_LOCAL
;
6195 err
= zap_length(mos
, objid
, nvpair_name(elem
),
6196 &integer_size
, &num_integers
);
6200 switch (integer_size
) {
6202 /* User properties cannot be integers */
6206 /* string property */
6207 strval
= kmem_alloc(num_integers
,
6209 err
= zap_lookup(mos
, objid
,
6210 nvpair_name(elem
), 1,
6211 num_integers
, strval
);
6217 vdev_prop_add_list(outnvl
, propname
,
6219 kmem_free(strval
, num_integers
);
6232 * Get all properties from the MOS vdev property object.
6236 for (zap_cursor_init(&zc
, mos
, objid
);
6237 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
6238 zap_cursor_advance(&zc
)) {
6241 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
6242 propname
= za
.za_name
;
6244 switch (za
.za_integer_length
) {
6246 /* We do not allow integer user properties */
6247 /* This is likely an internal value */
6250 /* string property */
6251 strval
= kmem_alloc(za
.za_num_integers
,
6253 err
= zap_lookup(mos
, objid
, za
.za_name
, 1,
6254 za
.za_num_integers
, strval
);
6256 kmem_free(strval
, za
.za_num_integers
);
6259 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
6261 kmem_free(strval
, za
.za_num_integers
);
6268 zap_cursor_fini(&zc
);
6271 mutex_exit(&spa
->spa_props_lock
);
6272 if (err
&& err
!= ENOENT
) {
6279 EXPORT_SYMBOL(vdev_fault
);
6280 EXPORT_SYMBOL(vdev_degrade
);
6281 EXPORT_SYMBOL(vdev_online
);
6282 EXPORT_SYMBOL(vdev_offline
);
6283 EXPORT_SYMBOL(vdev_clear
);
6285 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_count
, UINT
, ZMOD_RW
,
6286 "Target number of metaslabs per top-level vdev");
6288 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_shift
, UINT
, ZMOD_RW
,
6289 "Default limit for metaslab size");
6291 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, min_ms_count
, UINT
, ZMOD_RW
,
6292 "Minimum number of metaslabs per top-level vdev");
6294 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, ms_count_limit
, UINT
, ZMOD_RW
,
6295 "Practical upper limit of total metaslabs per top-level vdev");
6297 ZFS_MODULE_PARAM(zfs
, zfs_
, slow_io_events_per_second
, UINT
, ZMOD_RW
,
6298 "Rate limit slow IO (delay) events to this many per second");
6301 ZFS_MODULE_PARAM(zfs
, zfs_
, checksum_events_per_second
, UINT
, ZMOD_RW
,
6302 "Rate limit checksum events to this many checksum errors per second "
6303 "(do not set below ZED threshold).");
6306 ZFS_MODULE_PARAM(zfs
, zfs_
, scan_ignore_errors
, INT
, ZMOD_RW
,
6307 "Ignore errors during resilver/scrub");
6309 ZFS_MODULE_PARAM(zfs_vdev
, vdev_
, validate_skip
, INT
, ZMOD_RW
,
6310 "Bypass vdev_validate()");
6312 ZFS_MODULE_PARAM(zfs
, zfs_
, nocacheflush
, INT
, ZMOD_RW
,
6313 "Disable cache flushes");
6315 ZFS_MODULE_PARAM(zfs
, zfs_
, embedded_slog_min_ms
, UINT
, ZMOD_RW
,
6316 "Minimum number of metaslabs required to dedicate one for log blocks");
6319 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, min_auto_ashift
,
6320 param_set_min_auto_ashift
, param_get_uint
, ZMOD_RW
,
6321 "Minimum ashift used when creating new top-level vdevs");
6323 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, max_auto_ashift
,
6324 param_set_max_auto_ashift
, param_get_uint
, ZMOD_RW
,
6325 "Maximum ashift used when optimizing for logical -> physical sector "
6326 "size on new top-level vdevs");