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 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_autotrim_kick_cv
, NULL
, CV_DEFAULT
, NULL
);
700 cv_init(&vd
->vdev_trim_io_cv
, NULL
, CV_DEFAULT
, NULL
);
702 mutex_init(&vd
->vdev_rebuild_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
703 cv_init(&vd
->vdev_rebuild_cv
, NULL
, CV_DEFAULT
, NULL
);
705 for (int t
= 0; t
< DTL_TYPES
; t
++) {
706 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0,
710 txg_list_create(&vd
->vdev_ms_list
, spa
,
711 offsetof(struct metaslab
, ms_txg_node
));
712 txg_list_create(&vd
->vdev_dtl_list
, spa
,
713 offsetof(struct vdev
, vdev_dtl_node
));
714 vd
->vdev_stat
.vs_timestamp
= gethrtime();
722 * Allocate a new vdev. The 'alloctype' is used to control whether we are
723 * creating a new vdev or loading an existing one - the behavior is slightly
724 * different for each case.
727 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
732 uint64_t guid
= 0, islog
;
734 vdev_indirect_config_t
*vic
;
735 const char *tmp
= NULL
;
737 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
738 boolean_t top_level
= (parent
&& !parent
->vdev_parent
);
740 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
742 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
743 return (SET_ERROR(EINVAL
));
745 if ((ops
= vdev_getops(type
)) == NULL
)
746 return (SET_ERROR(EINVAL
));
749 * If this is a load, get the vdev guid from the nvlist.
750 * Otherwise, vdev_alloc_common() will generate one for us.
752 if (alloctype
== VDEV_ALLOC_LOAD
) {
755 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
757 return (SET_ERROR(EINVAL
));
759 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
760 return (SET_ERROR(EINVAL
));
761 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
762 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
763 return (SET_ERROR(EINVAL
));
764 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
765 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
766 return (SET_ERROR(EINVAL
));
767 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
768 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
769 return (SET_ERROR(EINVAL
));
773 * The first allocated vdev must be of type 'root'.
775 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
776 return (SET_ERROR(EINVAL
));
779 * Determine whether we're a log vdev.
782 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
783 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
784 return (SET_ERROR(ENOTSUP
));
786 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
787 return (SET_ERROR(ENOTSUP
));
789 if (top_level
&& alloctype
== VDEV_ALLOC_ADD
) {
793 * If creating a top-level vdev, check for allocation
796 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
798 alloc_bias
= vdev_derive_alloc_bias(bias
);
800 /* spa_vdev_add() expects feature to be enabled */
801 if (spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
802 !spa_feature_is_enabled(spa
,
803 SPA_FEATURE_ALLOCATION_CLASSES
)) {
804 return (SET_ERROR(ENOTSUP
));
808 /* spa_vdev_add() expects feature to be enabled */
809 if (ops
== &vdev_draid_ops
&&
810 spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
811 !spa_feature_is_enabled(spa
, SPA_FEATURE_DRAID
)) {
812 return (SET_ERROR(ENOTSUP
));
817 * Initialize the vdev specific data. This is done before calling
818 * vdev_alloc_common() since it may fail and this simplifies the
819 * error reporting and cleanup code paths.
822 if (ops
->vdev_op_init
!= NULL
) {
823 rc
= ops
->vdev_op_init(spa
, nv
, &tsd
);
829 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
831 vd
->vdev_islog
= islog
;
833 if (top_level
&& alloc_bias
!= VDEV_BIAS_NONE
)
834 vd
->vdev_alloc_bias
= alloc_bias
;
836 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &tmp
) == 0)
837 vd
->vdev_path
= spa_strdup(tmp
);
840 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
841 * fault on a vdev and want it to persist across imports (like with
844 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
845 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
846 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
847 vd
->vdev_faulted
= 1;
848 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
851 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &tmp
) == 0)
852 vd
->vdev_devid
= spa_strdup(tmp
);
853 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
, &tmp
) == 0)
854 vd
->vdev_physpath
= spa_strdup(tmp
);
856 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
858 vd
->vdev_enc_sysfs_path
= spa_strdup(tmp
);
860 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &tmp
) == 0)
861 vd
->vdev_fru
= spa_strdup(tmp
);
864 * Set the whole_disk property. If it's not specified, leave the value
867 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
868 &vd
->vdev_wholedisk
) != 0)
869 vd
->vdev_wholedisk
= -1ULL;
871 vic
= &vd
->vdev_indirect_config
;
873 ASSERT0(vic
->vic_mapping_object
);
874 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
875 &vic
->vic_mapping_object
);
876 ASSERT0(vic
->vic_births_object
);
877 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
878 &vic
->vic_births_object
);
879 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
880 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
881 &vic
->vic_prev_indirect_vdev
);
884 * Look for the 'not present' flag. This will only be set if the device
885 * was not present at the time of import.
887 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
888 &vd
->vdev_not_present
);
891 * Get the alignment requirement.
893 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
896 * Retrieve the vdev creation time.
898 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
902 * If we're a top-level vdev, try to load the allocation parameters.
905 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
906 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
908 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
910 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
912 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NONALLOCATING
,
914 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
916 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
919 ASSERT0(vd
->vdev_top_zap
);
922 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
923 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
924 alloctype
== VDEV_ALLOC_ADD
||
925 alloctype
== VDEV_ALLOC_SPLIT
||
926 alloctype
== VDEV_ALLOC_ROOTPOOL
);
927 /* Note: metaslab_group_create() is now deferred */
930 if (vd
->vdev_ops
->vdev_op_leaf
&&
931 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
932 (void) nvlist_lookup_uint64(nv
,
933 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
935 ASSERT0(vd
->vdev_leaf_zap
);
939 * If we're a leaf vdev, try to load the DTL object and other state.
942 if (vd
->vdev_ops
->vdev_op_leaf
&&
943 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
944 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
945 if (alloctype
== VDEV_ALLOC_LOAD
) {
946 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
947 &vd
->vdev_dtl_object
);
948 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
952 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
955 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
956 &spare
) == 0 && spare
)
960 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
963 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
964 &vd
->vdev_resilver_txg
);
966 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REBUILD_TXG
,
967 &vd
->vdev_rebuild_txg
);
969 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
970 vdev_defer_resilver(vd
);
973 * In general, when importing a pool we want to ignore the
974 * persistent fault state, as the diagnosis made on another
975 * system may not be valid in the current context. The only
976 * exception is if we forced a vdev to a persistently faulted
977 * state with 'zpool offline -f'. The persistent fault will
978 * remain across imports until cleared.
980 * Local vdevs will remain in the faulted state.
982 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
983 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
984 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
986 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
988 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
991 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
995 VDEV_AUX_ERR_EXCEEDED
;
996 if (nvlist_lookup_string(nv
,
997 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
998 strcmp(aux
, "external") == 0)
999 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
1001 vd
->vdev_faulted
= 0ULL;
1007 * Add ourselves to the parent's list of children.
1009 vdev_add_child(parent
, vd
);
1017 vdev_free(vdev_t
*vd
)
1019 spa_t
*spa
= vd
->vdev_spa
;
1021 ASSERT3P(vd
->vdev_initialize_thread
, ==, NULL
);
1022 ASSERT3P(vd
->vdev_trim_thread
, ==, NULL
);
1023 ASSERT3P(vd
->vdev_autotrim_thread
, ==, NULL
);
1024 ASSERT3P(vd
->vdev_rebuild_thread
, ==, NULL
);
1027 * Scan queues are normally destroyed at the end of a scan. If the
1028 * queue exists here, that implies the vdev is being removed while
1029 * the scan is still running.
1031 if (vd
->vdev_scan_io_queue
!= NULL
) {
1032 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
1033 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
1034 vd
->vdev_scan_io_queue
= NULL
;
1035 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
1039 * vdev_free() implies closing the vdev first. This is simpler than
1040 * trying to ensure complicated semantics for all callers.
1044 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
1045 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
1048 * Free all children.
1050 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1051 vdev_free(vd
->vdev_child
[c
]);
1053 ASSERT(vd
->vdev_child
== NULL
);
1054 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
1056 if (vd
->vdev_ops
->vdev_op_fini
!= NULL
)
1057 vd
->vdev_ops
->vdev_op_fini(vd
);
1060 * Discard allocation state.
1062 if (vd
->vdev_mg
!= NULL
) {
1063 vdev_metaslab_fini(vd
);
1064 metaslab_group_destroy(vd
->vdev_mg
);
1067 if (vd
->vdev_log_mg
!= NULL
) {
1068 ASSERT0(vd
->vdev_ms_count
);
1069 metaslab_group_destroy(vd
->vdev_log_mg
);
1070 vd
->vdev_log_mg
= NULL
;
1073 ASSERT0(vd
->vdev_stat
.vs_space
);
1074 ASSERT0(vd
->vdev_stat
.vs_dspace
);
1075 ASSERT0(vd
->vdev_stat
.vs_alloc
);
1078 * Remove this vdev from its parent's child list.
1080 vdev_remove_child(vd
->vdev_parent
, vd
);
1082 ASSERT(vd
->vdev_parent
== NULL
);
1083 ASSERT(!list_link_active(&vd
->vdev_leaf_node
));
1086 * Clean up vdev structure.
1088 vdev_queue_fini(vd
);
1089 vdev_cache_fini(vd
);
1092 spa_strfree(vd
->vdev_path
);
1094 spa_strfree(vd
->vdev_devid
);
1095 if (vd
->vdev_physpath
)
1096 spa_strfree(vd
->vdev_physpath
);
1098 if (vd
->vdev_enc_sysfs_path
)
1099 spa_strfree(vd
->vdev_enc_sysfs_path
);
1102 spa_strfree(vd
->vdev_fru
);
1104 if (vd
->vdev_isspare
)
1105 spa_spare_remove(vd
);
1106 if (vd
->vdev_isl2cache
)
1107 spa_l2cache_remove(vd
);
1109 txg_list_destroy(&vd
->vdev_ms_list
);
1110 txg_list_destroy(&vd
->vdev_dtl_list
);
1112 mutex_enter(&vd
->vdev_dtl_lock
);
1113 space_map_close(vd
->vdev_dtl_sm
);
1114 for (int t
= 0; t
< DTL_TYPES
; t
++) {
1115 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
1116 range_tree_destroy(vd
->vdev_dtl
[t
]);
1118 mutex_exit(&vd
->vdev_dtl_lock
);
1120 EQUIV(vd
->vdev_indirect_births
!= NULL
,
1121 vd
->vdev_indirect_mapping
!= NULL
);
1122 if (vd
->vdev_indirect_births
!= NULL
) {
1123 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
1124 vdev_indirect_births_close(vd
->vdev_indirect_births
);
1127 if (vd
->vdev_obsolete_sm
!= NULL
) {
1128 ASSERT(vd
->vdev_removing
||
1129 vd
->vdev_ops
== &vdev_indirect_ops
);
1130 space_map_close(vd
->vdev_obsolete_sm
);
1131 vd
->vdev_obsolete_sm
= NULL
;
1133 range_tree_destroy(vd
->vdev_obsolete_segments
);
1134 rw_destroy(&vd
->vdev_indirect_rwlock
);
1135 mutex_destroy(&vd
->vdev_obsolete_lock
);
1137 mutex_destroy(&vd
->vdev_dtl_lock
);
1138 mutex_destroy(&vd
->vdev_stat_lock
);
1139 mutex_destroy(&vd
->vdev_probe_lock
);
1140 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
1142 mutex_destroy(&vd
->vdev_initialize_lock
);
1143 mutex_destroy(&vd
->vdev_initialize_io_lock
);
1144 cv_destroy(&vd
->vdev_initialize_io_cv
);
1145 cv_destroy(&vd
->vdev_initialize_cv
);
1147 mutex_destroy(&vd
->vdev_trim_lock
);
1148 mutex_destroy(&vd
->vdev_autotrim_lock
);
1149 mutex_destroy(&vd
->vdev_trim_io_lock
);
1150 cv_destroy(&vd
->vdev_trim_cv
);
1151 cv_destroy(&vd
->vdev_autotrim_cv
);
1152 cv_destroy(&vd
->vdev_autotrim_kick_cv
);
1153 cv_destroy(&vd
->vdev_trim_io_cv
);
1155 mutex_destroy(&vd
->vdev_rebuild_lock
);
1156 cv_destroy(&vd
->vdev_rebuild_cv
);
1158 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
1159 zfs_ratelimit_fini(&vd
->vdev_deadman_rl
);
1160 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
1162 if (vd
== spa
->spa_root_vdev
)
1163 spa
->spa_root_vdev
= NULL
;
1165 kmem_free(vd
, sizeof (vdev_t
));
1169 * Transfer top-level vdev state from svd to tvd.
1172 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
1174 spa_t
*spa
= svd
->vdev_spa
;
1179 ASSERT(tvd
== tvd
->vdev_top
);
1181 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
1182 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
1183 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
1184 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
1185 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
1187 svd
->vdev_ms_array
= 0;
1188 svd
->vdev_ms_shift
= 0;
1189 svd
->vdev_ms_count
= 0;
1190 svd
->vdev_top_zap
= 0;
1193 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
1194 if (tvd
->vdev_log_mg
)
1195 ASSERT3P(tvd
->vdev_log_mg
, ==, svd
->vdev_log_mg
);
1196 tvd
->vdev_mg
= svd
->vdev_mg
;
1197 tvd
->vdev_log_mg
= svd
->vdev_log_mg
;
1198 tvd
->vdev_ms
= svd
->vdev_ms
;
1200 svd
->vdev_mg
= NULL
;
1201 svd
->vdev_log_mg
= NULL
;
1202 svd
->vdev_ms
= NULL
;
1204 if (tvd
->vdev_mg
!= NULL
)
1205 tvd
->vdev_mg
->mg_vd
= tvd
;
1206 if (tvd
->vdev_log_mg
!= NULL
)
1207 tvd
->vdev_log_mg
->mg_vd
= tvd
;
1209 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
1210 svd
->vdev_checkpoint_sm
= NULL
;
1212 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1213 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1215 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1216 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1217 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1219 svd
->vdev_stat
.vs_alloc
= 0;
1220 svd
->vdev_stat
.vs_space
= 0;
1221 svd
->vdev_stat
.vs_dspace
= 0;
1224 * State which may be set on a top-level vdev that's in the
1225 * process of being removed.
1227 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1228 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1229 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1230 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1231 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1232 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1233 ASSERT0(tvd
->vdev_noalloc
);
1234 ASSERT0(tvd
->vdev_removing
);
1235 ASSERT0(tvd
->vdev_rebuilding
);
1236 tvd
->vdev_noalloc
= svd
->vdev_noalloc
;
1237 tvd
->vdev_removing
= svd
->vdev_removing
;
1238 tvd
->vdev_rebuilding
= svd
->vdev_rebuilding
;
1239 tvd
->vdev_rebuild_config
= svd
->vdev_rebuild_config
;
1240 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1241 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1242 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1243 range_tree_swap(&svd
->vdev_obsolete_segments
,
1244 &tvd
->vdev_obsolete_segments
);
1245 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1246 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1247 svd
->vdev_indirect_config
.vic_births_object
= 0;
1248 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1249 svd
->vdev_indirect_mapping
= NULL
;
1250 svd
->vdev_indirect_births
= NULL
;
1251 svd
->vdev_obsolete_sm
= NULL
;
1252 svd
->vdev_noalloc
= 0;
1253 svd
->vdev_removing
= 0;
1254 svd
->vdev_rebuilding
= 0;
1256 for (t
= 0; t
< TXG_SIZE
; t
++) {
1257 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1258 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1259 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1260 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1261 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1262 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1265 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1266 vdev_config_clean(svd
);
1267 vdev_config_dirty(tvd
);
1270 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1271 vdev_state_clean(svd
);
1272 vdev_state_dirty(tvd
);
1275 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1276 svd
->vdev_deflate_ratio
= 0;
1278 tvd
->vdev_islog
= svd
->vdev_islog
;
1279 svd
->vdev_islog
= 0;
1281 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1285 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1292 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1293 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1297 * Add a mirror/replacing vdev above an existing vdev. There is no need to
1298 * call .vdev_op_init() since mirror/replacing vdevs do not have private state.
1301 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1303 spa_t
*spa
= cvd
->vdev_spa
;
1304 vdev_t
*pvd
= cvd
->vdev_parent
;
1307 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1309 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1311 mvd
->vdev_asize
= cvd
->vdev_asize
;
1312 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1313 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1314 mvd
->vdev_psize
= cvd
->vdev_psize
;
1315 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1316 mvd
->vdev_logical_ashift
= cvd
->vdev_logical_ashift
;
1317 mvd
->vdev_physical_ashift
= cvd
->vdev_physical_ashift
;
1318 mvd
->vdev_state
= cvd
->vdev_state
;
1319 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1321 vdev_remove_child(pvd
, cvd
);
1322 vdev_add_child(pvd
, mvd
);
1323 cvd
->vdev_id
= mvd
->vdev_children
;
1324 vdev_add_child(mvd
, cvd
);
1325 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1327 if (mvd
== mvd
->vdev_top
)
1328 vdev_top_transfer(cvd
, mvd
);
1334 * Remove a 1-way mirror/replacing vdev from the tree.
1337 vdev_remove_parent(vdev_t
*cvd
)
1339 vdev_t
*mvd
= cvd
->vdev_parent
;
1340 vdev_t
*pvd
= mvd
->vdev_parent
;
1342 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1344 ASSERT(mvd
->vdev_children
== 1);
1345 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1346 mvd
->vdev_ops
== &vdev_replacing_ops
||
1347 mvd
->vdev_ops
== &vdev_spare_ops
);
1348 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1349 cvd
->vdev_logical_ashift
= mvd
->vdev_logical_ashift
;
1350 cvd
->vdev_physical_ashift
= mvd
->vdev_physical_ashift
;
1351 vdev_remove_child(mvd
, cvd
);
1352 vdev_remove_child(pvd
, mvd
);
1355 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1356 * Otherwise, we could have detached an offline device, and when we
1357 * go to import the pool we'll think we have two top-level vdevs,
1358 * instead of a different version of the same top-level vdev.
1360 if (mvd
->vdev_top
== mvd
) {
1361 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1362 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1363 cvd
->vdev_guid
+= guid_delta
;
1364 cvd
->vdev_guid_sum
+= guid_delta
;
1367 * If pool not set for autoexpand, we need to also preserve
1368 * mvd's asize to prevent automatic expansion of cvd.
1369 * Otherwise if we are adjusting the mirror by attaching and
1370 * detaching children of non-uniform sizes, the mirror could
1371 * autoexpand, unexpectedly requiring larger devices to
1372 * re-establish the mirror.
1374 if (!cvd
->vdev_spa
->spa_autoexpand
)
1375 cvd
->vdev_asize
= mvd
->vdev_asize
;
1377 cvd
->vdev_id
= mvd
->vdev_id
;
1378 vdev_add_child(pvd
, cvd
);
1379 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1381 if (cvd
== cvd
->vdev_top
)
1382 vdev_top_transfer(mvd
, cvd
);
1384 ASSERT(mvd
->vdev_children
== 0);
1389 vdev_metaslab_group_create(vdev_t
*vd
)
1391 spa_t
*spa
= vd
->vdev_spa
;
1394 * metaslab_group_create was delayed until allocation bias was available
1396 if (vd
->vdev_mg
== NULL
) {
1397 metaslab_class_t
*mc
;
1399 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1400 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1402 ASSERT3U(vd
->vdev_islog
, ==,
1403 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1405 switch (vd
->vdev_alloc_bias
) {
1407 mc
= spa_log_class(spa
);
1409 case VDEV_BIAS_SPECIAL
:
1410 mc
= spa_special_class(spa
);
1412 case VDEV_BIAS_DEDUP
:
1413 mc
= spa_dedup_class(spa
);
1416 mc
= spa_normal_class(spa
);
1419 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1420 spa
->spa_alloc_count
);
1422 if (!vd
->vdev_islog
) {
1423 vd
->vdev_log_mg
= metaslab_group_create(
1424 spa_embedded_log_class(spa
), vd
, 1);
1428 * The spa ashift min/max only apply for the normal metaslab
1429 * class. Class destination is late binding so ashift boundary
1430 * setting had to wait until now.
1432 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1433 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1434 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1435 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1436 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1437 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1439 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
1440 if (min_alloc
< spa
->spa_min_alloc
)
1441 spa
->spa_min_alloc
= min_alloc
;
1447 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1449 spa_t
*spa
= vd
->vdev_spa
;
1450 uint64_t oldc
= vd
->vdev_ms_count
;
1451 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1454 boolean_t expanding
= (oldc
!= 0);
1456 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1459 * This vdev is not being allocated from yet or is a hole.
1461 if (vd
->vdev_ms_shift
== 0)
1464 ASSERT(!vd
->vdev_ishole
);
1466 ASSERT(oldc
<= newc
);
1468 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1471 memcpy(mspp
, vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1472 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1476 vd
->vdev_ms_count
= newc
;
1478 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1479 uint64_t object
= 0;
1481 * vdev_ms_array may be 0 if we are creating the "fake"
1482 * metaslabs for an indirect vdev for zdb's leak detection.
1483 * See zdb_leak_init().
1485 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1486 error
= dmu_read(spa
->spa_meta_objset
,
1488 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1491 vdev_dbgmsg(vd
, "unable to read the metaslab "
1492 "array [error=%d]", error
);
1497 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1500 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1507 * Find the emptiest metaslab on the vdev and mark it for use for
1508 * embedded slog by moving it from the regular to the log metaslab
1511 if (vd
->vdev_mg
->mg_class
== spa_normal_class(spa
) &&
1512 vd
->vdev_ms_count
> zfs_embedded_slog_min_ms
&&
1513 avl_is_empty(&vd
->vdev_log_mg
->mg_metaslab_tree
)) {
1514 uint64_t slog_msid
= 0;
1515 uint64_t smallest
= UINT64_MAX
;
1518 * Note, we only search the new metaslabs, because the old
1519 * (pre-existing) ones may be active (e.g. have non-empty
1520 * range_tree's), and we don't move them to the new
1523 for (uint64_t m
= oldc
; m
< newc
; m
++) {
1525 space_map_allocated(vd
->vdev_ms
[m
]->ms_sm
);
1526 if (alloc
< smallest
) {
1531 metaslab_t
*slog_ms
= vd
->vdev_ms
[slog_msid
];
1533 * The metaslab was marked as dirty at the end of
1534 * metaslab_init(). Remove it from the dirty list so that we
1535 * can uninitialize and reinitialize it to the new class.
1538 (void) txg_list_remove_this(&vd
->vdev_ms_list
,
1541 uint64_t sm_obj
= space_map_object(slog_ms
->ms_sm
);
1542 metaslab_fini(slog_ms
);
1543 VERIFY0(metaslab_init(vd
->vdev_log_mg
, slog_msid
, sm_obj
, txg
,
1544 &vd
->vdev_ms
[slog_msid
]));
1548 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1551 * If the vdev is marked as non-allocating then don't
1552 * activate the metaslabs since we want to ensure that
1553 * no allocations are performed on this device.
1555 if (vd
->vdev_noalloc
) {
1556 /* track non-allocating vdev space */
1557 spa
->spa_nonallocating_dspace
+= spa_deflate(spa
) ?
1558 vd
->vdev_stat
.vs_dspace
: vd
->vdev_stat
.vs_space
;
1559 } else if (!expanding
) {
1560 metaslab_group_activate(vd
->vdev_mg
);
1561 if (vd
->vdev_log_mg
!= NULL
)
1562 metaslab_group_activate(vd
->vdev_log_mg
);
1566 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1572 vdev_metaslab_fini(vdev_t
*vd
)
1574 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1575 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1576 SPA_FEATURE_POOL_CHECKPOINT
));
1577 space_map_close(vd
->vdev_checkpoint_sm
);
1579 * Even though we close the space map, we need to set its
1580 * pointer to NULL. The reason is that vdev_metaslab_fini()
1581 * may be called multiple times for certain operations
1582 * (i.e. when destroying a pool) so we need to ensure that
1583 * this clause never executes twice. This logic is similar
1584 * to the one used for the vdev_ms clause below.
1586 vd
->vdev_checkpoint_sm
= NULL
;
1589 if (vd
->vdev_ms
!= NULL
) {
1590 metaslab_group_t
*mg
= vd
->vdev_mg
;
1592 metaslab_group_passivate(mg
);
1593 if (vd
->vdev_log_mg
!= NULL
) {
1594 ASSERT(!vd
->vdev_islog
);
1595 metaslab_group_passivate(vd
->vdev_log_mg
);
1598 uint64_t count
= vd
->vdev_ms_count
;
1599 for (uint64_t m
= 0; m
< count
; m
++) {
1600 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1604 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1606 vd
->vdev_ms_count
= 0;
1608 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++) {
1609 ASSERT0(mg
->mg_histogram
[i
]);
1610 if (vd
->vdev_log_mg
!= NULL
)
1611 ASSERT0(vd
->vdev_log_mg
->mg_histogram
[i
]);
1614 ASSERT0(vd
->vdev_ms_count
);
1615 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1618 typedef struct vdev_probe_stats
{
1619 boolean_t vps_readable
;
1620 boolean_t vps_writeable
;
1622 } vdev_probe_stats_t
;
1625 vdev_probe_done(zio_t
*zio
)
1627 spa_t
*spa
= zio
->io_spa
;
1628 vdev_t
*vd
= zio
->io_vd
;
1629 vdev_probe_stats_t
*vps
= zio
->io_private
;
1631 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1633 if (zio
->io_type
== ZIO_TYPE_READ
) {
1634 if (zio
->io_error
== 0)
1635 vps
->vps_readable
= 1;
1636 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1637 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1638 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1639 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1640 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1642 abd_free(zio
->io_abd
);
1644 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1645 if (zio
->io_error
== 0)
1646 vps
->vps_writeable
= 1;
1647 abd_free(zio
->io_abd
);
1648 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1652 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1653 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1655 if (vdev_readable(vd
) &&
1656 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1659 ASSERT(zio
->io_error
!= 0);
1660 vdev_dbgmsg(vd
, "failed probe");
1661 (void) zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1662 spa
, vd
, NULL
, NULL
, 0);
1663 zio
->io_error
= SET_ERROR(ENXIO
);
1666 mutex_enter(&vd
->vdev_probe_lock
);
1667 ASSERT(vd
->vdev_probe_zio
== zio
);
1668 vd
->vdev_probe_zio
= NULL
;
1669 mutex_exit(&vd
->vdev_probe_lock
);
1672 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1673 if (!vdev_accessible(vd
, pio
))
1674 pio
->io_error
= SET_ERROR(ENXIO
);
1676 kmem_free(vps
, sizeof (*vps
));
1681 * Determine whether this device is accessible.
1683 * Read and write to several known locations: the pad regions of each
1684 * vdev label but the first, which we leave alone in case it contains
1688 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1690 spa_t
*spa
= vd
->vdev_spa
;
1691 vdev_probe_stats_t
*vps
= NULL
;
1694 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1697 * Don't probe the probe.
1699 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1703 * To prevent 'probe storms' when a device fails, we create
1704 * just one probe i/o at a time. All zios that want to probe
1705 * this vdev will become parents of the probe io.
1707 mutex_enter(&vd
->vdev_probe_lock
);
1709 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1710 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1712 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1713 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1716 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1718 * vdev_cant_read and vdev_cant_write can only
1719 * transition from TRUE to FALSE when we have the
1720 * SCL_ZIO lock as writer; otherwise they can only
1721 * transition from FALSE to TRUE. This ensures that
1722 * any zio looking at these values can assume that
1723 * failures persist for the life of the I/O. That's
1724 * important because when a device has intermittent
1725 * connectivity problems, we want to ensure that
1726 * they're ascribed to the device (ENXIO) and not
1729 * Since we hold SCL_ZIO as writer here, clear both
1730 * values so the probe can reevaluate from first
1733 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1734 vd
->vdev_cant_read
= B_FALSE
;
1735 vd
->vdev_cant_write
= B_FALSE
;
1738 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1739 vdev_probe_done
, vps
,
1740 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1743 * We can't change the vdev state in this context, so we
1744 * kick off an async task to do it on our behalf.
1747 vd
->vdev_probe_wanted
= B_TRUE
;
1748 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1753 zio_add_child(zio
, pio
);
1755 mutex_exit(&vd
->vdev_probe_lock
);
1758 ASSERT(zio
!= NULL
);
1762 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1763 zio_nowait(zio_read_phys(pio
, vd
,
1764 vdev_label_offset(vd
->vdev_psize
, l
,
1765 offsetof(vdev_label_t
, vl_be
)), VDEV_PAD_SIZE
,
1766 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1767 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1768 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1779 vdev_load_child(void *arg
)
1783 vd
->vdev_load_error
= vdev_load(vd
);
1787 vdev_open_child(void *arg
)
1791 vd
->vdev_open_thread
= curthread
;
1792 vd
->vdev_open_error
= vdev_open(vd
);
1793 vd
->vdev_open_thread
= NULL
;
1797 vdev_uses_zvols(vdev_t
*vd
)
1800 if (zvol_is_zvol(vd
->vdev_path
))
1804 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1805 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1812 * Returns B_TRUE if the passed child should be opened.
1815 vdev_default_open_children_func(vdev_t
*vd
)
1822 * Open the requested child vdevs. If any of the leaf vdevs are using
1823 * a ZFS volume then do the opens in a single thread. This avoids a
1824 * deadlock when the current thread is holding the spa_namespace_lock.
1827 vdev_open_children_impl(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1829 int children
= vd
->vdev_children
;
1831 taskq_t
*tq
= taskq_create("vdev_open", children
, minclsyspri
,
1832 children
, children
, TASKQ_PREPOPULATE
);
1833 vd
->vdev_nonrot
= B_TRUE
;
1835 for (int c
= 0; c
< children
; c
++) {
1836 vdev_t
*cvd
= vd
->vdev_child
[c
];
1838 if (open_func(cvd
) == B_FALSE
)
1841 if (tq
== NULL
|| vdev_uses_zvols(vd
)) {
1842 cvd
->vdev_open_error
= vdev_open(cvd
);
1844 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1845 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
1848 vd
->vdev_nonrot
&= cvd
->vdev_nonrot
;
1858 * Open all child vdevs.
1861 vdev_open_children(vdev_t
*vd
)
1863 vdev_open_children_impl(vd
, vdev_default_open_children_func
);
1867 * Conditionally open a subset of child vdevs.
1870 vdev_open_children_subset(vdev_t
*vd
, vdev_open_children_func_t
*open_func
)
1872 vdev_open_children_impl(vd
, open_func
);
1876 * Compute the raidz-deflation ratio. Note, we hard-code
1877 * in 128k (1 << 17) because it is the "typical" blocksize.
1878 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1879 * otherwise it would inconsistently account for existing bp's.
1882 vdev_set_deflate_ratio(vdev_t
*vd
)
1884 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1885 vd
->vdev_deflate_ratio
= (1 << 17) /
1886 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1891 * Choose the best of two ashifts, preferring one between logical ashift
1892 * (absolute minimum) and administrator defined maximum, otherwise take
1893 * the biggest of the two.
1896 vdev_best_ashift(uint64_t logical
, uint64_t a
, uint64_t b
)
1898 if (a
> logical
&& a
<= zfs_vdev_max_auto_ashift
) {
1899 if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1903 } else if (b
<= logical
|| b
> zfs_vdev_max_auto_ashift
)
1909 * Maximize performance by inflating the configured ashift for top level
1910 * vdevs to be as close to the physical ashift as possible while maintaining
1911 * administrator defined limits and ensuring it doesn't go below the
1915 vdev_ashift_optimize(vdev_t
*vd
)
1917 ASSERT(vd
== vd
->vdev_top
);
1919 if (vd
->vdev_ashift
< vd
->vdev_physical_ashift
&&
1920 vd
->vdev_physical_ashift
<= zfs_vdev_max_auto_ashift
) {
1921 vd
->vdev_ashift
= MIN(
1922 MAX(zfs_vdev_max_auto_ashift
, vd
->vdev_ashift
),
1923 MAX(zfs_vdev_min_auto_ashift
,
1924 vd
->vdev_physical_ashift
));
1927 * If the logical and physical ashifts are the same, then
1928 * we ensure that the top-level vdev's ashift is not smaller
1929 * than our minimum ashift value. For the unusual case
1930 * where logical ashift > physical ashift, we can't cap
1931 * the calculated ashift based on max ashift as that
1932 * would cause failures.
1933 * We still check if we need to increase it to match
1936 vd
->vdev_ashift
= MAX(zfs_vdev_min_auto_ashift
,
1942 * Prepare a virtual device for access.
1945 vdev_open(vdev_t
*vd
)
1947 spa_t
*spa
= vd
->vdev_spa
;
1950 uint64_t max_osize
= 0;
1951 uint64_t asize
, max_asize
, psize
;
1952 uint64_t logical_ashift
= 0;
1953 uint64_t physical_ashift
= 0;
1955 ASSERT(vd
->vdev_open_thread
== curthread
||
1956 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1957 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1958 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1959 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1961 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1962 vd
->vdev_cant_read
= B_FALSE
;
1963 vd
->vdev_cant_write
= B_FALSE
;
1964 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1967 * If this vdev is not removed, check its fault status. If it's
1968 * faulted, bail out of the open.
1970 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1971 ASSERT(vd
->vdev_children
== 0);
1972 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1973 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1974 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1975 vd
->vdev_label_aux
);
1976 return (SET_ERROR(ENXIO
));
1977 } else if (vd
->vdev_offline
) {
1978 ASSERT(vd
->vdev_children
== 0);
1979 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1980 return (SET_ERROR(ENXIO
));
1983 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
,
1984 &logical_ashift
, &physical_ashift
);
1986 /* Keep the device in removed state if unplugged */
1987 if (error
== ENOENT
&& vd
->vdev_removed
) {
1988 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_REMOVED
,
1994 * Physical volume size should never be larger than its max size, unless
1995 * the disk has shrunk while we were reading it or the device is buggy
1996 * or damaged: either way it's not safe for use, bail out of the open.
1998 if (osize
> max_osize
) {
1999 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2000 VDEV_AUX_OPEN_FAILED
);
2001 return (SET_ERROR(ENXIO
));
2005 * Reset the vdev_reopening flag so that we actually close
2006 * the vdev on error.
2008 vd
->vdev_reopening
= B_FALSE
;
2009 if (zio_injection_enabled
&& error
== 0)
2010 error
= zio_handle_device_injection(vd
, NULL
, SET_ERROR(ENXIO
));
2013 if (vd
->vdev_removed
&&
2014 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
2015 vd
->vdev_removed
= B_FALSE
;
2017 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
2018 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
2019 vd
->vdev_stat
.vs_aux
);
2021 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2022 vd
->vdev_stat
.vs_aux
);
2027 vd
->vdev_removed
= B_FALSE
;
2030 * Recheck the faulted flag now that we have confirmed that
2031 * the vdev is accessible. If we're faulted, bail.
2033 if (vd
->vdev_faulted
) {
2034 ASSERT(vd
->vdev_children
== 0);
2035 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
2036 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
2037 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2038 vd
->vdev_label_aux
);
2039 return (SET_ERROR(ENXIO
));
2042 if (vd
->vdev_degraded
) {
2043 ASSERT(vd
->vdev_children
== 0);
2044 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2045 VDEV_AUX_ERR_EXCEEDED
);
2047 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
2051 * For hole or missing vdevs we just return success.
2053 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
2056 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2057 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
2058 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
2064 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
2065 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
2067 if (vd
->vdev_children
== 0) {
2068 if (osize
< SPA_MINDEVSIZE
) {
2069 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2070 VDEV_AUX_TOO_SMALL
);
2071 return (SET_ERROR(EOVERFLOW
));
2074 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
2075 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
2076 VDEV_LABEL_END_SIZE
);
2078 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
2079 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
2080 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2081 VDEV_AUX_TOO_SMALL
);
2082 return (SET_ERROR(EOVERFLOW
));
2086 max_asize
= max_osize
;
2090 * If the vdev was expanded, record this so that we can re-create the
2091 * uberblock rings in labels {2,3}, during the next sync.
2093 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
2094 vd
->vdev_copy_uberblocks
= B_TRUE
;
2096 vd
->vdev_psize
= psize
;
2099 * Make sure the allocatable size hasn't shrunk too much.
2101 if (asize
< vd
->vdev_min_asize
) {
2102 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2103 VDEV_AUX_BAD_LABEL
);
2104 return (SET_ERROR(EINVAL
));
2108 * We can always set the logical/physical ashift members since
2109 * their values are only used to calculate the vdev_ashift when
2110 * the device is first added to the config. These values should
2111 * not be used for anything else since they may change whenever
2112 * the device is reopened and we don't store them in the label.
2114 vd
->vdev_physical_ashift
=
2115 MAX(physical_ashift
, vd
->vdev_physical_ashift
);
2116 vd
->vdev_logical_ashift
= MAX(logical_ashift
,
2117 vd
->vdev_logical_ashift
);
2119 if (vd
->vdev_asize
== 0) {
2121 * This is the first-ever open, so use the computed values.
2122 * For compatibility, a different ashift can be requested.
2124 vd
->vdev_asize
= asize
;
2125 vd
->vdev_max_asize
= max_asize
;
2128 * If the vdev_ashift was not overridden at creation time,
2129 * then set it the logical ashift and optimize the ashift.
2131 if (vd
->vdev_ashift
== 0) {
2132 vd
->vdev_ashift
= vd
->vdev_logical_ashift
;
2134 if (vd
->vdev_logical_ashift
> ASHIFT_MAX
) {
2135 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2136 VDEV_AUX_ASHIFT_TOO_BIG
);
2137 return (SET_ERROR(EDOM
));
2140 if (vd
->vdev_top
== vd
) {
2141 vdev_ashift_optimize(vd
);
2144 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
2145 vd
->vdev_ashift
> ASHIFT_MAX
)) {
2146 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2147 VDEV_AUX_BAD_ASHIFT
);
2148 return (SET_ERROR(EDOM
));
2152 * Make sure the alignment required hasn't increased.
2154 if (vd
->vdev_ashift
> vd
->vdev_top
->vdev_ashift
&&
2155 vd
->vdev_ops
->vdev_op_leaf
) {
2156 (void) zfs_ereport_post(
2157 FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
2158 spa
, vd
, NULL
, NULL
, 0);
2159 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
2160 VDEV_AUX_BAD_LABEL
);
2161 return (SET_ERROR(EDOM
));
2163 vd
->vdev_max_asize
= max_asize
;
2167 * If all children are healthy we update asize if either:
2168 * The asize has increased, due to a device expansion caused by dynamic
2169 * LUN growth or vdev replacement, and automatic expansion is enabled;
2170 * making the additional space available.
2172 * The asize has decreased, due to a device shrink usually caused by a
2173 * vdev replace with a smaller device. This ensures that calculations
2174 * based of max_asize and asize e.g. esize are always valid. It's safe
2175 * to do this as we've already validated that asize is greater than
2178 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
2179 ((asize
> vd
->vdev_asize
&&
2180 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
2181 (asize
< vd
->vdev_asize
)))
2182 vd
->vdev_asize
= asize
;
2184 vdev_set_min_asize(vd
);
2187 * Ensure we can issue some IO before declaring the
2188 * vdev open for business.
2190 if (vd
->vdev_ops
->vdev_op_leaf
&&
2191 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
2192 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
2193 VDEV_AUX_ERR_EXCEEDED
);
2198 * Track the minimum allocation size.
2200 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
2201 vd
->vdev_islog
== 0 && vd
->vdev_aux
== NULL
) {
2202 uint64_t min_alloc
= vdev_get_min_alloc(vd
);
2203 if (min_alloc
< spa
->spa_min_alloc
)
2204 spa
->spa_min_alloc
= min_alloc
;
2208 * If this is a leaf vdev, assess whether a resilver is needed.
2209 * But don't do this if we are doing a reopen for a scrub, since
2210 * this would just restart the scrub we are already doing.
2212 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
)
2213 dsl_scan_assess_vdev(spa
->spa_dsl_pool
, vd
);
2219 vdev_validate_child(void *arg
)
2223 vd
->vdev_validate_thread
= curthread
;
2224 vd
->vdev_validate_error
= vdev_validate(vd
);
2225 vd
->vdev_validate_thread
= NULL
;
2229 * Called once the vdevs are all opened, this routine validates the label
2230 * contents. This needs to be done before vdev_load() so that we don't
2231 * inadvertently do repair I/Os to the wrong device.
2233 * This function will only return failure if one of the vdevs indicates that it
2234 * has since been destroyed or exported. This is only possible if
2235 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
2236 * will be updated but the function will return 0.
2239 vdev_validate(vdev_t
*vd
)
2241 spa_t
*spa
= vd
->vdev_spa
;
2244 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
2248 int children
= vd
->vdev_children
;
2250 if (vdev_validate_skip
)
2254 tq
= taskq_create("vdev_validate", children
, minclsyspri
,
2255 children
, children
, TASKQ_PREPOPULATE
);
2258 for (uint64_t c
= 0; c
< children
; c
++) {
2259 vdev_t
*cvd
= vd
->vdev_child
[c
];
2261 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
2262 vdev_validate_child(cvd
);
2264 VERIFY(taskq_dispatch(tq
, vdev_validate_child
, cvd
,
2265 TQ_SLEEP
) != TASKQID_INVALID
);
2272 for (int c
= 0; c
< children
; c
++) {
2273 int error
= vd
->vdev_child
[c
]->vdev_validate_error
;
2276 return (SET_ERROR(EBADF
));
2281 * If the device has already failed, or was marked offline, don't do
2282 * any further validation. Otherwise, label I/O will fail and we will
2283 * overwrite the previous state.
2285 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
2289 * If we are performing an extreme rewind, we allow for a label that
2290 * was modified at a point after the current txg.
2291 * If config lock is not held do not check for the txg. spa_sync could
2292 * be updating the vdev's label before updating spa_last_synced_txg.
2294 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
2295 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
2298 txg
= spa_last_synced_txg(spa
);
2300 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
2301 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2302 VDEV_AUX_BAD_LABEL
);
2303 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
2304 "txg %llu", (u_longlong_t
)txg
);
2309 * Determine if this vdev has been split off into another
2310 * pool. If so, then refuse to open it.
2312 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
2313 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
2314 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2315 VDEV_AUX_SPLIT_POOL
);
2317 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
2321 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
2322 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2323 VDEV_AUX_CORRUPT_DATA
);
2325 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2326 ZPOOL_CONFIG_POOL_GUID
);
2331 * If config is not trusted then ignore the spa guid check. This is
2332 * necessary because if the machine crashed during a re-guid the new
2333 * guid might have been written to all of the vdev labels, but not the
2334 * cached config. The check will be performed again once we have the
2335 * trusted config from the MOS.
2337 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
2338 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2339 VDEV_AUX_CORRUPT_DATA
);
2341 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
2342 "match config (%llu != %llu)", (u_longlong_t
)guid
,
2343 (u_longlong_t
)spa_guid(spa
));
2347 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
2348 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
2352 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
2353 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2354 VDEV_AUX_CORRUPT_DATA
);
2356 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2361 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
2363 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2364 VDEV_AUX_CORRUPT_DATA
);
2366 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2367 ZPOOL_CONFIG_TOP_GUID
);
2372 * If this vdev just became a top-level vdev because its sibling was
2373 * detached, it will have adopted the parent's vdev guid -- but the
2374 * label may or may not be on disk yet. Fortunately, either version
2375 * of the label will have the same top guid, so if we're a top-level
2376 * vdev, we can safely compare to that instead.
2377 * However, if the config comes from a cachefile that failed to update
2378 * after the detach, a top-level vdev will appear as a non top-level
2379 * vdev in the config. Also relax the constraints if we perform an
2382 * If we split this vdev off instead, then we also check the
2383 * original pool's guid. We don't want to consider the vdev
2384 * corrupt if it is partway through a split operation.
2386 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
2387 boolean_t mismatch
= B_FALSE
;
2388 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
2389 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
2392 if (vd
->vdev_guid
!= top_guid
&&
2393 vd
->vdev_top
->vdev_guid
!= guid
)
2398 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2399 VDEV_AUX_CORRUPT_DATA
);
2401 vdev_dbgmsg(vd
, "vdev_validate: config guid "
2402 "doesn't match label guid");
2403 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
2404 (u_longlong_t
)vd
->vdev_guid
,
2405 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
2406 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
2407 "aux_guid %llu", (u_longlong_t
)guid
,
2408 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
2413 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
2415 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2416 VDEV_AUX_CORRUPT_DATA
);
2418 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
2419 ZPOOL_CONFIG_POOL_STATE
);
2426 * If this is a verbatim import, no need to check the
2427 * state of the pool.
2429 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
2430 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
2431 state
!= POOL_STATE_ACTIVE
) {
2432 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
2433 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
2434 return (SET_ERROR(EBADF
));
2438 * If we were able to open and validate a vdev that was
2439 * previously marked permanently unavailable, clear that state
2442 if (vd
->vdev_not_present
)
2443 vd
->vdev_not_present
= 0;
2449 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2452 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2453 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2454 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2455 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2456 dvd
->vdev_path
, svd
->vdev_path
);
2457 spa_strfree(dvd
->vdev_path
);
2458 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2460 } else if (svd
->vdev_path
!= NULL
) {
2461 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2462 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2463 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2467 * Our enclosure sysfs path may have changed between imports
2469 old
= dvd
->vdev_enc_sysfs_path
;
2470 new = svd
->vdev_enc_sysfs_path
;
2471 if ((old
!= NULL
&& new == NULL
) ||
2472 (old
== NULL
&& new != NULL
) ||
2473 ((old
!= NULL
&& new != NULL
) && strcmp(new, old
) != 0)) {
2474 zfs_dbgmsg("vdev_copy_path: vdev %llu: vdev_enc_sysfs_path "
2475 "changed from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2478 if (dvd
->vdev_enc_sysfs_path
)
2479 spa_strfree(dvd
->vdev_enc_sysfs_path
);
2481 if (svd
->vdev_enc_sysfs_path
) {
2482 dvd
->vdev_enc_sysfs_path
= spa_strdup(
2483 svd
->vdev_enc_sysfs_path
);
2485 dvd
->vdev_enc_sysfs_path
= NULL
;
2491 * Recursively copy vdev paths from one vdev to another. Source and destination
2492 * vdev trees must have same geometry otherwise return error. Intended to copy
2493 * paths from userland config into MOS config.
2496 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2498 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2499 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2500 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2503 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2504 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2505 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2506 return (SET_ERROR(EINVAL
));
2509 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2510 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2511 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2512 (u_longlong_t
)dvd
->vdev_guid
);
2513 return (SET_ERROR(EINVAL
));
2516 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2517 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2518 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2519 (u_longlong_t
)dvd
->vdev_children
);
2520 return (SET_ERROR(EINVAL
));
2523 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2524 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2525 dvd
->vdev_child
[i
]);
2530 if (svd
->vdev_ops
->vdev_op_leaf
)
2531 vdev_copy_path_impl(svd
, dvd
);
2537 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2539 ASSERT(stvd
->vdev_top
== stvd
);
2540 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2542 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2543 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2546 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2550 * The idea here is that while a vdev can shift positions within
2551 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2552 * step outside of it.
2554 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2556 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2559 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2561 vdev_copy_path_impl(vd
, dvd
);
2565 * Recursively copy vdev paths from one root vdev to another. Source and
2566 * destination vdev trees may differ in geometry. For each destination leaf
2567 * vdev, search a vdev with the same guid and top vdev id in the source.
2568 * Intended to copy paths from userland config into MOS config.
2571 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2573 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2574 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2575 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2577 for (uint64_t i
= 0; i
< children
; i
++) {
2578 vdev_copy_path_search(srvd
->vdev_child
[i
],
2579 drvd
->vdev_child
[i
]);
2584 * Close a virtual device.
2587 vdev_close(vdev_t
*vd
)
2589 vdev_t
*pvd
= vd
->vdev_parent
;
2590 spa_t
*spa __maybe_unused
= vd
->vdev_spa
;
2593 ASSERT(vd
->vdev_open_thread
== curthread
||
2594 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2597 * If our parent is reopening, then we are as well, unless we are
2600 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2601 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2603 vd
->vdev_ops
->vdev_op_close(vd
);
2605 vdev_cache_purge(vd
);
2608 * We record the previous state before we close it, so that if we are
2609 * doing a reopen(), we don't generate FMA ereports if we notice that
2610 * it's still faulted.
2612 vd
->vdev_prevstate
= vd
->vdev_state
;
2614 if (vd
->vdev_offline
)
2615 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2617 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2618 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2622 vdev_hold(vdev_t
*vd
)
2624 spa_t
*spa
= vd
->vdev_spa
;
2626 ASSERT(spa_is_root(spa
));
2627 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2630 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2631 vdev_hold(vd
->vdev_child
[c
]);
2633 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_hold
!= NULL
)
2634 vd
->vdev_ops
->vdev_op_hold(vd
);
2638 vdev_rele(vdev_t
*vd
)
2640 ASSERT(spa_is_root(vd
->vdev_spa
));
2641 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2642 vdev_rele(vd
->vdev_child
[c
]);
2644 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_ops
->vdev_op_rele
!= NULL
)
2645 vd
->vdev_ops
->vdev_op_rele(vd
);
2649 * Reopen all interior vdevs and any unopened leaves. We don't actually
2650 * reopen leaf vdevs which had previously been opened as they might deadlock
2651 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2652 * If the leaf has never been opened then open it, as usual.
2655 vdev_reopen(vdev_t
*vd
)
2657 spa_t
*spa
= vd
->vdev_spa
;
2659 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2661 /* set the reopening flag unless we're taking the vdev offline */
2662 vd
->vdev_reopening
= !vd
->vdev_offline
;
2664 (void) vdev_open(vd
);
2667 * Call vdev_validate() here to make sure we have the same device.
2668 * Otherwise, a device with an invalid label could be successfully
2669 * opened in response to vdev_reopen().
2672 (void) vdev_validate_aux(vd
);
2673 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2674 vd
->vdev_aux
== &spa
->spa_l2cache
) {
2676 * In case the vdev is present we should evict all ARC
2677 * buffers and pointers to log blocks and reclaim their
2678 * space before restoring its contents to L2ARC.
2680 if (l2arc_vdev_present(vd
)) {
2681 l2arc_rebuild_vdev(vd
, B_TRUE
);
2683 l2arc_add_vdev(spa
, vd
);
2685 spa_async_request(spa
, SPA_ASYNC_L2CACHE_REBUILD
);
2686 spa_async_request(spa
, SPA_ASYNC_L2CACHE_TRIM
);
2689 (void) vdev_validate(vd
);
2693 * Reassess parent vdev's health.
2695 vdev_propagate_state(vd
);
2699 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2704 * Normally, partial opens (e.g. of a mirror) are allowed.
2705 * For a create, however, we want to fail the request if
2706 * there are any components we can't open.
2708 error
= vdev_open(vd
);
2710 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2712 return (error
? error
: SET_ERROR(ENXIO
));
2716 * Recursively load DTLs and initialize all labels.
2718 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2719 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2720 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2729 vdev_metaslab_set_size(vdev_t
*vd
)
2731 uint64_t asize
= vd
->vdev_asize
;
2732 uint64_t ms_count
= asize
>> zfs_vdev_default_ms_shift
;
2736 * There are two dimensions to the metaslab sizing calculation:
2737 * the size of the metaslab and the count of metaslabs per vdev.
2739 * The default values used below are a good balance between memory
2740 * usage (larger metaslab size means more memory needed for loaded
2741 * metaslabs; more metaslabs means more memory needed for the
2742 * metaslab_t structs), metaslab load time (larger metaslabs take
2743 * longer to load), and metaslab sync time (more metaslabs means
2744 * more time spent syncing all of them).
2746 * In general, we aim for zfs_vdev_default_ms_count (200) metaslabs.
2747 * The range of the dimensions are as follows:
2749 * 2^29 <= ms_size <= 2^34
2750 * 16 <= ms_count <= 131,072
2752 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2753 * at least 512MB (2^29) to minimize fragmentation effects when
2754 * testing with smaller devices. However, the count constraint
2755 * of at least 16 metaslabs will override this minimum size goal.
2757 * On the upper end of vdev sizes, we aim for a maximum metaslab
2758 * size of 16GB. However, we will cap the total count to 2^17
2759 * metaslabs to keep our memory footprint in check and let the
2760 * metaslab size grow from there if that limit is hit.
2762 * The net effect of applying above constrains is summarized below.
2764 * vdev size metaslab count
2765 * --------------|-----------------
2767 * 8GB - 100GB one per 512MB
2769 * 3TB - 2PB one per 16GB
2771 * --------------------------------
2773 * Finally, note that all of the above calculate the initial
2774 * number of metaslabs. Expanding a top-level vdev will result
2775 * in additional metaslabs being allocated making it possible
2776 * to exceed the zfs_vdev_ms_count_limit.
2779 if (ms_count
< zfs_vdev_min_ms_count
)
2780 ms_shift
= highbit64(asize
/ zfs_vdev_min_ms_count
);
2781 else if (ms_count
> zfs_vdev_default_ms_count
)
2782 ms_shift
= highbit64(asize
/ zfs_vdev_default_ms_count
);
2784 ms_shift
= zfs_vdev_default_ms_shift
;
2786 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2787 ms_shift
= SPA_MAXBLOCKSHIFT
;
2788 } else if (ms_shift
> zfs_vdev_max_ms_shift
) {
2789 ms_shift
= zfs_vdev_max_ms_shift
;
2790 /* cap the total count to constrain memory footprint */
2791 if ((asize
>> ms_shift
) > zfs_vdev_ms_count_limit
)
2792 ms_shift
= highbit64(asize
/ zfs_vdev_ms_count_limit
);
2795 vd
->vdev_ms_shift
= ms_shift
;
2796 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2800 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2802 ASSERT(vd
== vd
->vdev_top
);
2803 /* indirect vdevs don't have metaslabs or dtls */
2804 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2805 ASSERT(ISP2(flags
));
2806 ASSERT(spa_writeable(vd
->vdev_spa
));
2808 if (flags
& VDD_METASLAB
)
2809 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2811 if (flags
& VDD_DTL
)
2812 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2814 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2818 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2820 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2821 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2823 if (vd
->vdev_ops
->vdev_op_leaf
)
2824 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2830 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2831 * the vdev has less than perfect replication. There are four kinds of DTL:
2833 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2835 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2837 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2838 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2839 * txgs that was scrubbed.
2841 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2842 * persistent errors or just some device being offline.
2843 * Unlike the other three, the DTL_OUTAGE map is not generally
2844 * maintained; it's only computed when needed, typically to
2845 * determine whether a device can be detached.
2847 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2848 * either has the data or it doesn't.
2850 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2851 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2852 * if any child is less than fully replicated, then so is its parent.
2853 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2854 * comprising only those txgs which appear in 'maxfaults' or more children;
2855 * those are the txgs we don't have enough replication to read. For example,
2856 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2857 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2858 * two child DTL_MISSING maps.
2860 * It should be clear from the above that to compute the DTLs and outage maps
2861 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2862 * Therefore, that is all we keep on disk. When loading the pool, or after
2863 * a configuration change, we generate all other DTLs from first principles.
2866 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2868 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2870 ASSERT(t
< DTL_TYPES
);
2871 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2872 ASSERT(spa_writeable(vd
->vdev_spa
));
2874 mutex_enter(&vd
->vdev_dtl_lock
);
2875 if (!range_tree_contains(rt
, txg
, size
))
2876 range_tree_add(rt
, txg
, size
);
2877 mutex_exit(&vd
->vdev_dtl_lock
);
2881 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2883 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2884 boolean_t dirty
= B_FALSE
;
2886 ASSERT(t
< DTL_TYPES
);
2887 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2890 * While we are loading the pool, the DTLs have not been loaded yet.
2891 * This isn't a problem but it can result in devices being tried
2892 * which are known to not have the data. In which case, the import
2893 * is relying on the checksum to ensure that we get the right data.
2894 * Note that while importing we are only reading the MOS, which is
2895 * always checksummed.
2897 mutex_enter(&vd
->vdev_dtl_lock
);
2898 if (!range_tree_is_empty(rt
))
2899 dirty
= range_tree_contains(rt
, txg
, size
);
2900 mutex_exit(&vd
->vdev_dtl_lock
);
2906 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2908 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2911 mutex_enter(&vd
->vdev_dtl_lock
);
2912 empty
= range_tree_is_empty(rt
);
2913 mutex_exit(&vd
->vdev_dtl_lock
);
2919 * Check if the txg falls within the range which must be
2920 * resilvered. DVAs outside this range can always be skipped.
2923 vdev_default_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2924 uint64_t phys_birth
)
2926 (void) dva
, (void) psize
;
2928 /* Set by sequential resilver. */
2929 if (phys_birth
== TXG_UNKNOWN
)
2932 return (vdev_dtl_contains(vd
, DTL_PARTIAL
, phys_birth
, 1));
2936 * Returns B_TRUE if the vdev determines the DVA needs to be resilvered.
2939 vdev_dtl_need_resilver(vdev_t
*vd
, const dva_t
*dva
, size_t psize
,
2940 uint64_t phys_birth
)
2942 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2944 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2945 vd
->vdev_ops
->vdev_op_leaf
)
2948 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, dva
, psize
,
2953 * Returns the lowest txg in the DTL range.
2956 vdev_dtl_min(vdev_t
*vd
)
2958 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2959 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2960 ASSERT0(vd
->vdev_children
);
2962 return (range_tree_min(vd
->vdev_dtl
[DTL_MISSING
]) - 1);
2966 * Returns the highest txg in the DTL.
2969 vdev_dtl_max(vdev_t
*vd
)
2971 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2972 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2973 ASSERT0(vd
->vdev_children
);
2975 return (range_tree_max(vd
->vdev_dtl
[DTL_MISSING
]));
2979 * Determine if a resilvering vdev should remove any DTL entries from
2980 * its range. If the vdev was resilvering for the entire duration of the
2981 * scan then it should excise that range from its DTLs. Otherwise, this
2982 * vdev is considered partially resilvered and should leave its DTL
2983 * entries intact. The comment in vdev_dtl_reassess() describes how we
2987 vdev_dtl_should_excise(vdev_t
*vd
, boolean_t rebuild_done
)
2989 ASSERT0(vd
->vdev_children
);
2991 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2994 if (vd
->vdev_resilver_deferred
)
2997 if (range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
3001 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3002 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
3004 /* Rebuild not initiated by attach */
3005 if (vd
->vdev_rebuild_txg
== 0)
3009 * When a rebuild completes without error then all missing data
3010 * up to the rebuild max txg has been reconstructed and the DTL
3011 * is eligible for excision.
3013 if (vrp
->vrp_rebuild_state
== VDEV_REBUILD_COMPLETE
&&
3014 vdev_dtl_max(vd
) <= vrp
->vrp_max_txg
) {
3015 ASSERT3U(vrp
->vrp_min_txg
, <=, vdev_dtl_min(vd
));
3016 ASSERT3U(vrp
->vrp_min_txg
, <, vd
->vdev_rebuild_txg
);
3017 ASSERT3U(vd
->vdev_rebuild_txg
, <=, vrp
->vrp_max_txg
);
3021 dsl_scan_t
*scn
= vd
->vdev_spa
->spa_dsl_pool
->dp_scan
;
3022 dsl_scan_phys_t
*scnp __maybe_unused
= &scn
->scn_phys
;
3024 /* Resilver not initiated by attach */
3025 if (vd
->vdev_resilver_txg
== 0)
3029 * When a resilver is initiated the scan will assign the
3030 * scn_max_txg value to the highest txg value that exists
3031 * in all DTLs. If this device's max DTL is not part of this
3032 * scan (i.e. it is not in the range (scn_min_txg, scn_max_txg]
3033 * then it is not eligible for excision.
3035 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
3036 ASSERT3U(scnp
->scn_min_txg
, <=, vdev_dtl_min(vd
));
3037 ASSERT3U(scnp
->scn_min_txg
, <, vd
->vdev_resilver_txg
);
3038 ASSERT3U(vd
->vdev_resilver_txg
, <=, scnp
->scn_max_txg
);
3047 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
3048 * write operations will be issued to the pool.
3051 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
,
3052 boolean_t scrub_done
, boolean_t rebuild_done
)
3054 spa_t
*spa
= vd
->vdev_spa
;
3058 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
3060 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3061 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
3062 scrub_txg
, scrub_done
, rebuild_done
);
3064 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
3067 if (vd
->vdev_ops
->vdev_op_leaf
) {
3068 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
3069 vdev_rebuild_t
*vr
= &vd
->vdev_top
->vdev_rebuild_config
;
3070 boolean_t check_excise
= B_FALSE
;
3071 boolean_t wasempty
= B_TRUE
;
3073 mutex_enter(&vd
->vdev_dtl_lock
);
3076 * If requested, pretend the scan or rebuild completed cleanly.
3078 if (zfs_scan_ignore_errors
) {
3080 scn
->scn_phys
.scn_errors
= 0;
3082 vr
->vr_rebuild_phys
.vrp_errors
= 0;
3085 if (scrub_txg
!= 0 &&
3086 !range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3088 zfs_dbgmsg("guid:%llu txg:%llu scrub:%llu started:%d "
3089 "dtl:%llu/%llu errors:%llu",
3090 (u_longlong_t
)vd
->vdev_guid
, (u_longlong_t
)txg
,
3091 (u_longlong_t
)scrub_txg
, spa
->spa_scrub_started
,
3092 (u_longlong_t
)vdev_dtl_min(vd
),
3093 (u_longlong_t
)vdev_dtl_max(vd
),
3094 (u_longlong_t
)(scn
? scn
->scn_phys
.scn_errors
: 0));
3098 * If we've completed a scrub/resilver or a rebuild cleanly
3099 * then determine if this vdev should remove any DTLs. We
3100 * only want to excise regions on vdevs that were available
3101 * during the entire duration of this scan.
3104 vr
!= NULL
&& vr
->vr_rebuild_phys
.vrp_errors
== 0) {
3105 check_excise
= B_TRUE
;
3107 if (spa
->spa_scrub_started
||
3108 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) {
3109 check_excise
= B_TRUE
;
3113 if (scrub_txg
&& check_excise
&&
3114 vdev_dtl_should_excise(vd
, rebuild_done
)) {
3116 * We completed a scrub, resilver or rebuild up to
3117 * scrub_txg. If we did it without rebooting, then
3118 * the scrub dtl will be valid, so excise the old
3119 * region and fold in the scrub dtl. Otherwise,
3120 * leave the dtl as-is if there was an error.
3122 * There's little trick here: to excise the beginning
3123 * of the DTL_MISSING map, we put it into a reference
3124 * tree and then add a segment with refcnt -1 that
3125 * covers the range [0, scrub_txg). This means
3126 * that each txg in that range has refcnt -1 or 0.
3127 * We then add DTL_SCRUB with a refcnt of 2, so that
3128 * entries in the range [0, scrub_txg) will have a
3129 * positive refcnt -- either 1 or 2. We then convert
3130 * the reference tree into the new DTL_MISSING map.
3132 space_reftree_create(&reftree
);
3133 space_reftree_add_map(&reftree
,
3134 vd
->vdev_dtl
[DTL_MISSING
], 1);
3135 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
3136 space_reftree_add_map(&reftree
,
3137 vd
->vdev_dtl
[DTL_SCRUB
], 2);
3138 space_reftree_generate_map(&reftree
,
3139 vd
->vdev_dtl
[DTL_MISSING
], 1);
3140 space_reftree_destroy(&reftree
);
3142 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
3143 zfs_dbgmsg("update DTL_MISSING:%llu/%llu",
3144 (u_longlong_t
)vdev_dtl_min(vd
),
3145 (u_longlong_t
)vdev_dtl_max(vd
));
3146 } else if (!wasempty
) {
3147 zfs_dbgmsg("DTL_MISSING is now empty");
3150 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
3151 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3152 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
3154 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
3155 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
3156 if (!vdev_readable(vd
))
3157 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
3159 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
3160 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
3163 * If the vdev was resilvering or rebuilding and no longer
3164 * has any DTLs then reset the appropriate flag and dirty
3165 * the top level so that we persist the change.
3168 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3169 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
3170 if (vd
->vdev_rebuild_txg
!= 0) {
3171 vd
->vdev_rebuild_txg
= 0;
3172 vdev_config_dirty(vd
->vdev_top
);
3173 } else if (vd
->vdev_resilver_txg
!= 0) {
3174 vd
->vdev_resilver_txg
= 0;
3175 vdev_config_dirty(vd
->vdev_top
);
3179 mutex_exit(&vd
->vdev_dtl_lock
);
3182 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
3186 mutex_enter(&vd
->vdev_dtl_lock
);
3187 for (int t
= 0; t
< DTL_TYPES
; t
++) {
3188 /* account for child's outage in parent's missing map */
3189 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
3191 continue; /* leaf vdevs only */
3192 if (t
== DTL_PARTIAL
)
3193 minref
= 1; /* i.e. non-zero */
3194 else if (vdev_get_nparity(vd
) != 0)
3195 minref
= vdev_get_nparity(vd
) + 1; /* RAID-Z, dRAID */
3197 minref
= vd
->vdev_children
; /* any kind of mirror */
3198 space_reftree_create(&reftree
);
3199 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3200 vdev_t
*cvd
= vd
->vdev_child
[c
];
3201 mutex_enter(&cvd
->vdev_dtl_lock
);
3202 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
3203 mutex_exit(&cvd
->vdev_dtl_lock
);
3205 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
3206 space_reftree_destroy(&reftree
);
3208 mutex_exit(&vd
->vdev_dtl_lock
);
3212 * Iterate over all the vdevs except spare, and post kobj events
3215 vdev_post_kobj_evt(vdev_t
*vd
)
3217 if (vd
->vdev_ops
->vdev_op_kobj_evt_post
&&
3218 vd
->vdev_kobj_flag
== B_FALSE
) {
3219 vd
->vdev_kobj_flag
= B_TRUE
;
3220 vd
->vdev_ops
->vdev_op_kobj_evt_post(vd
);
3223 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3224 vdev_post_kobj_evt(vd
->vdev_child
[c
]);
3228 * Iterate over all the vdevs except spare, and clear kobj events
3231 vdev_clear_kobj_evt(vdev_t
*vd
)
3233 vd
->vdev_kobj_flag
= B_FALSE
;
3235 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3236 vdev_clear_kobj_evt(vd
->vdev_child
[c
]);
3240 vdev_dtl_load(vdev_t
*vd
)
3242 spa_t
*spa
= vd
->vdev_spa
;
3243 objset_t
*mos
= spa
->spa_meta_objset
;
3247 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
3248 ASSERT(vdev_is_concrete(vd
));
3251 * If the dtl cannot be sync'd there is no need to open it.
3253 if (spa
->spa_mode
== SPA_MODE_READ
&& !spa
->spa_read_spacemaps
)
3256 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
3257 vd
->vdev_dtl_object
, 0, -1ULL, 0);
3260 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3262 rt
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3263 error
= space_map_load(vd
->vdev_dtl_sm
, rt
, SM_ALLOC
);
3265 mutex_enter(&vd
->vdev_dtl_lock
);
3266 range_tree_walk(rt
, range_tree_add
,
3267 vd
->vdev_dtl
[DTL_MISSING
]);
3268 mutex_exit(&vd
->vdev_dtl_lock
);
3271 range_tree_vacate(rt
, NULL
, NULL
);
3272 range_tree_destroy(rt
);
3277 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3278 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
3287 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3289 spa_t
*spa
= vd
->vdev_spa
;
3290 objset_t
*mos
= spa
->spa_meta_objset
;
3291 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
3294 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
3297 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
3298 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
3299 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
3301 ASSERT(string
!= NULL
);
3302 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
3303 1, strlen(string
) + 1, string
, tx
));
3305 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
3306 spa_activate_allocation_classes(spa
, tx
);
3311 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
3313 spa_t
*spa
= vd
->vdev_spa
;
3315 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
3316 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3321 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
3323 spa_t
*spa
= vd
->vdev_spa
;
3324 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
3325 DMU_OT_NONE
, 0, tx
);
3328 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
3335 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3337 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
3338 vd
->vdev_ops
!= &vdev_missing_ops
&&
3339 vd
->vdev_ops
!= &vdev_root_ops
&&
3340 !vd
->vdev_top
->vdev_removing
) {
3341 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
3342 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
3344 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
3345 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
3346 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
3347 vdev_zap_allocation_data(vd
, tx
);
3351 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
3352 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
3357 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
3359 spa_t
*spa
= vd
->vdev_spa
;
3360 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
3361 objset_t
*mos
= spa
->spa_meta_objset
;
3362 range_tree_t
*rtsync
;
3364 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
3366 ASSERT(vdev_is_concrete(vd
));
3367 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
3369 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3371 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
3372 mutex_enter(&vd
->vdev_dtl_lock
);
3373 space_map_free(vd
->vdev_dtl_sm
, tx
);
3374 space_map_close(vd
->vdev_dtl_sm
);
3375 vd
->vdev_dtl_sm
= NULL
;
3376 mutex_exit(&vd
->vdev_dtl_lock
);
3379 * We only destroy the leaf ZAP for detached leaves or for
3380 * removed log devices. Removed data devices handle leaf ZAP
3381 * cleanup later, once cancellation is no longer possible.
3383 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
3384 vd
->vdev_top
->vdev_islog
)) {
3385 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
3386 vd
->vdev_leaf_zap
= 0;
3393 if (vd
->vdev_dtl_sm
== NULL
) {
3394 uint64_t new_object
;
3396 new_object
= space_map_alloc(mos
, zfs_vdev_dtl_sm_blksz
, tx
);
3397 VERIFY3U(new_object
, !=, 0);
3399 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
3401 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
3404 rtsync
= range_tree_create(NULL
, RANGE_SEG64
, NULL
, 0, 0);
3406 mutex_enter(&vd
->vdev_dtl_lock
);
3407 range_tree_walk(rt
, range_tree_add
, rtsync
);
3408 mutex_exit(&vd
->vdev_dtl_lock
);
3410 space_map_truncate(vd
->vdev_dtl_sm
, zfs_vdev_dtl_sm_blksz
, tx
);
3411 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
3412 range_tree_vacate(rtsync
, NULL
, NULL
);
3414 range_tree_destroy(rtsync
);
3417 * If the object for the space map has changed then dirty
3418 * the top level so that we update the config.
3420 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
3421 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
3422 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
3423 (u_longlong_t
)object
,
3424 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
3425 vdev_config_dirty(vd
->vdev_top
);
3432 * Determine whether the specified vdev can be offlined/detached/removed
3433 * without losing data.
3436 vdev_dtl_required(vdev_t
*vd
)
3438 spa_t
*spa
= vd
->vdev_spa
;
3439 vdev_t
*tvd
= vd
->vdev_top
;
3440 uint8_t cant_read
= vd
->vdev_cant_read
;
3443 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3445 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
3449 * Temporarily mark the device as unreadable, and then determine
3450 * whether this results in any DTL outages in the top-level vdev.
3451 * If not, we can safely offline/detach/remove the device.
3453 vd
->vdev_cant_read
= B_TRUE
;
3454 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3455 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
3456 vd
->vdev_cant_read
= cant_read
;
3457 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
, B_FALSE
);
3459 if (!required
&& zio_injection_enabled
) {
3460 required
= !!zio_handle_device_injection(vd
, NULL
,
3468 * Determine if resilver is needed, and if so the txg range.
3471 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
3473 boolean_t needed
= B_FALSE
;
3474 uint64_t thismin
= UINT64_MAX
;
3475 uint64_t thismax
= 0;
3477 if (vd
->vdev_children
== 0) {
3478 mutex_enter(&vd
->vdev_dtl_lock
);
3479 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
3480 vdev_writeable(vd
)) {
3482 thismin
= vdev_dtl_min(vd
);
3483 thismax
= vdev_dtl_max(vd
);
3486 mutex_exit(&vd
->vdev_dtl_lock
);
3488 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3489 vdev_t
*cvd
= vd
->vdev_child
[c
];
3490 uint64_t cmin
, cmax
;
3492 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
3493 thismin
= MIN(thismin
, cmin
);
3494 thismax
= MAX(thismax
, cmax
);
3500 if (needed
&& minp
) {
3508 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
3509 * will contain either the checkpoint spacemap object or zero if none exists.
3510 * All other errors are returned to the caller.
3513 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
3515 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
3517 if (vd
->vdev_top_zap
== 0) {
3522 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
3523 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
3524 if (error
== ENOENT
) {
3533 vdev_load(vdev_t
*vd
)
3535 int children
= vd
->vdev_children
;
3540 * It's only worthwhile to use the taskq for the root vdev, because the
3541 * slow part is metaslab_init, and that only happens for top-level
3544 if (vd
->vdev_ops
== &vdev_root_ops
&& vd
->vdev_children
> 0) {
3545 tq
= taskq_create("vdev_load", children
, minclsyspri
,
3546 children
, children
, TASKQ_PREPOPULATE
);
3550 * Recursively load all children.
3552 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3553 vdev_t
*cvd
= vd
->vdev_child
[c
];
3555 if (tq
== NULL
|| vdev_uses_zvols(cvd
)) {
3556 cvd
->vdev_load_error
= vdev_load(cvd
);
3558 VERIFY(taskq_dispatch(tq
, vdev_load_child
,
3559 cvd
, TQ_SLEEP
) != TASKQID_INVALID
);
3568 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3569 int error
= vd
->vdev_child
[c
]->vdev_load_error
;
3575 vdev_set_deflate_ratio(vd
);
3578 * On spa_load path, grab the allocation bias from our zap
3580 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3581 spa_t
*spa
= vd
->vdev_spa
;
3584 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3585 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
3588 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
3589 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
3590 } else if (error
!= ENOENT
) {
3591 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3592 VDEV_AUX_CORRUPT_DATA
);
3593 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(top_zap=%llu) "
3594 "failed [error=%d]",
3595 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3600 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3601 spa_t
*spa
= vd
->vdev_spa
;
3604 error
= zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
3605 vdev_prop_to_name(VDEV_PROP_FAILFAST
), sizeof (failfast
),
3608 vd
->vdev_failfast
= failfast
& 1;
3609 } else if (error
== ENOENT
) {
3610 vd
->vdev_failfast
= vdev_prop_default_numeric(
3611 VDEV_PROP_FAILFAST
);
3614 "vdev_load: zap_lookup(top_zap=%llu) "
3615 "failed [error=%d]",
3616 (u_longlong_t
)vd
->vdev_top_zap
, error
);
3621 * Load any rebuild state from the top-level vdev zap.
3623 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
3624 error
= vdev_rebuild_load(vd
);
3625 if (error
&& error
!= ENOTSUP
) {
3626 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3627 VDEV_AUX_CORRUPT_DATA
);
3628 vdev_dbgmsg(vd
, "vdev_load: vdev_rebuild_load "
3629 "failed [error=%d]", error
);
3634 if (vd
->vdev_top_zap
!= 0 || vd
->vdev_leaf_zap
!= 0) {
3637 if (vd
->vdev_top_zap
!= 0)
3638 zapobj
= vd
->vdev_top_zap
;
3640 zapobj
= vd
->vdev_leaf_zap
;
3642 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_N
,
3643 &vd
->vdev_checksum_n
);
3644 if (error
&& error
!= ENOENT
)
3645 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3646 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3648 error
= vdev_prop_get_int(vd
, VDEV_PROP_CHECKSUM_T
,
3649 &vd
->vdev_checksum_t
);
3650 if (error
&& error
!= ENOENT
)
3651 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3652 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3654 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_N
,
3656 if (error
&& error
!= ENOENT
)
3657 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3658 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3660 error
= vdev_prop_get_int(vd
, VDEV_PROP_IO_T
,
3662 if (error
&& error
!= ENOENT
)
3663 vdev_dbgmsg(vd
, "vdev_load: zap_lookup(zap=%llu) "
3664 "failed [error=%d]", (u_longlong_t
)zapobj
, error
);
3668 * If this is a top-level vdev, initialize its metaslabs.
3670 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
3671 vdev_metaslab_group_create(vd
);
3673 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
3674 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3675 VDEV_AUX_CORRUPT_DATA
);
3676 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
3677 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
3678 (u_longlong_t
)vd
->vdev_asize
);
3679 return (SET_ERROR(ENXIO
));
3682 error
= vdev_metaslab_init(vd
, 0);
3684 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
3685 "[error=%d]", error
);
3686 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3687 VDEV_AUX_CORRUPT_DATA
);
3691 uint64_t checkpoint_sm_obj
;
3692 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
3693 if (error
== 0 && checkpoint_sm_obj
!= 0) {
3694 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3695 ASSERT(vd
->vdev_asize
!= 0);
3696 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
3698 error
= space_map_open(&vd
->vdev_checkpoint_sm
,
3699 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
3702 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3703 "failed for checkpoint spacemap (obj %llu) "
3705 (u_longlong_t
)checkpoint_sm_obj
, error
);
3708 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3711 * Since the checkpoint_sm contains free entries
3712 * exclusively we can use space_map_allocated() to
3713 * indicate the cumulative checkpointed space that
3716 vd
->vdev_stat
.vs_checkpoint_space
=
3717 -space_map_allocated(vd
->vdev_checkpoint_sm
);
3718 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3719 vd
->vdev_stat
.vs_checkpoint_space
;
3720 } else if (error
!= 0) {
3721 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3722 "checkpoint space map object from vdev ZAP "
3723 "[error=%d]", error
);
3729 * If this is a leaf vdev, load its DTL.
3731 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3732 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3733 VDEV_AUX_CORRUPT_DATA
);
3734 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3735 "[error=%d]", error
);
3739 uint64_t obsolete_sm_object
;
3740 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3741 if (error
== 0 && obsolete_sm_object
!= 0) {
3742 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3743 ASSERT(vd
->vdev_asize
!= 0);
3744 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3746 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3747 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3748 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3749 VDEV_AUX_CORRUPT_DATA
);
3750 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3751 "obsolete spacemap (obj %llu) [error=%d]",
3752 (u_longlong_t
)obsolete_sm_object
, error
);
3755 } else if (error
!= 0) {
3756 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve obsolete "
3757 "space map object from vdev ZAP [error=%d]", error
);
3765 * The special vdev case is used for hot spares and l2cache devices. Its
3766 * sole purpose it to set the vdev state for the associated vdev. To do this,
3767 * we make sure that we can open the underlying device, then try to read the
3768 * label, and make sure that the label is sane and that it hasn't been
3769 * repurposed to another pool.
3772 vdev_validate_aux(vdev_t
*vd
)
3775 uint64_t guid
, version
;
3778 if (!vdev_readable(vd
))
3781 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
3782 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3783 VDEV_AUX_CORRUPT_DATA
);
3787 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
3788 !SPA_VERSION_IS_SUPPORTED(version
) ||
3789 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
3790 guid
!= vd
->vdev_guid
||
3791 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
3792 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3793 VDEV_AUX_CORRUPT_DATA
);
3799 * We don't actually check the pool state here. If it's in fact in
3800 * use by another pool, we update this fact on the fly when requested.
3807 vdev_destroy_ms_flush_data(vdev_t
*vd
, dmu_tx_t
*tx
)
3809 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
3811 if (vd
->vdev_top_zap
== 0)
3814 uint64_t object
= 0;
3815 int err
= zap_lookup(mos
, vd
->vdev_top_zap
,
3816 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, sizeof (uint64_t), 1, &object
);
3821 VERIFY0(dmu_object_free(mos
, object
, tx
));
3822 VERIFY0(zap_remove(mos
, vd
->vdev_top_zap
,
3823 VDEV_TOP_ZAP_MS_UNFLUSHED_PHYS_TXGS
, tx
));
3827 * Free the objects used to store this vdev's spacemaps, and the array
3828 * that points to them.
3831 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3833 if (vd
->vdev_ms_array
== 0)
3836 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3837 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
3838 size_t array_bytes
= array_count
* sizeof (uint64_t);
3839 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
3840 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
3841 array_bytes
, smobj_array
, 0));
3843 for (uint64_t i
= 0; i
< array_count
; i
++) {
3844 uint64_t smobj
= smobj_array
[i
];
3848 space_map_free_obj(mos
, smobj
, tx
);
3851 kmem_free(smobj_array
, array_bytes
);
3852 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
3853 vdev_destroy_ms_flush_data(vd
, tx
);
3854 vd
->vdev_ms_array
= 0;
3858 vdev_remove_empty_log(vdev_t
*vd
, uint64_t txg
)
3860 spa_t
*spa
= vd
->vdev_spa
;
3862 ASSERT(vd
->vdev_islog
);
3863 ASSERT(vd
== vd
->vdev_top
);
3864 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
3866 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3868 vdev_destroy_spacemaps(vd
, tx
);
3869 if (vd
->vdev_top_zap
!= 0) {
3870 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3871 vd
->vdev_top_zap
= 0;
3878 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3881 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3883 ASSERT(vdev_is_concrete(vd
));
3885 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
)))
3887 metaslab_sync_done(msp
, txg
);
3890 metaslab_sync_reassess(vd
->vdev_mg
);
3891 if (vd
->vdev_log_mg
!= NULL
)
3892 metaslab_sync_reassess(vd
->vdev_log_mg
);
3897 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3899 spa_t
*spa
= vd
->vdev_spa
;
3903 ASSERT3U(txg
, ==, spa
->spa_syncing_txg
);
3904 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3905 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3906 ASSERT(vd
->vdev_removing
||
3907 vd
->vdev_ops
== &vdev_indirect_ops
);
3909 vdev_indirect_sync_obsolete(vd
, tx
);
3912 * If the vdev is indirect, it can't have dirty
3913 * metaslabs or DTLs.
3915 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3916 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3917 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3923 ASSERT(vdev_is_concrete(vd
));
3925 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3926 !vd
->vdev_removing
) {
3927 ASSERT(vd
== vd
->vdev_top
);
3928 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3929 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3930 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3931 ASSERT(vd
->vdev_ms_array
!= 0);
3932 vdev_config_dirty(vd
);
3935 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3936 metaslab_sync(msp
, txg
);
3937 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3940 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3941 vdev_dtl_sync(lvd
, txg
);
3944 * If this is an empty log device being removed, destroy the
3945 * metadata associated with it.
3947 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3948 vdev_remove_empty_log(vd
, txg
);
3950 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3955 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3957 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3961 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3962 * not be opened, and no I/O is attempted.
3965 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3969 spa_vdev_state_enter(spa
, SCL_NONE
);
3971 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3972 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
3974 if (!vd
->vdev_ops
->vdev_op_leaf
)
3975 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
3980 * If user did a 'zpool offline -f' then make the fault persist across
3983 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3985 * There are two kinds of forced faults: temporary and
3986 * persistent. Temporary faults go away at pool import, while
3987 * persistent faults stay set. Both types of faults can be
3988 * cleared with a zpool clear.
3990 * We tell if a vdev is persistently faulted by looking at the
3991 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3992 * import then it's a persistent fault. Otherwise, it's
3993 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3994 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3995 * tells vdev_config_generate() (which gets run later) to set
3996 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3998 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3999 vd
->vdev_tmpoffline
= B_FALSE
;
4000 aux
= VDEV_AUX_EXTERNAL
;
4002 vd
->vdev_tmpoffline
= B_TRUE
;
4006 * We don't directly use the aux state here, but if we do a
4007 * vdev_reopen(), we need this value to be present to remember why we
4010 vd
->vdev_label_aux
= aux
;
4013 * Faulted state takes precedence over degraded.
4015 vd
->vdev_delayed_close
= B_FALSE
;
4016 vd
->vdev_faulted
= 1ULL;
4017 vd
->vdev_degraded
= 0ULL;
4018 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
4021 * If this device has the only valid copy of the data, then
4022 * back off and simply mark the vdev as degraded instead.
4024 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
4025 vd
->vdev_degraded
= 1ULL;
4026 vd
->vdev_faulted
= 0ULL;
4029 * If we reopen the device and it's not dead, only then do we
4034 if (vdev_readable(vd
))
4035 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
4038 return (spa_vdev_state_exit(spa
, vd
, 0));
4042 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
4043 * user that something is wrong. The vdev continues to operate as normal as far
4044 * as I/O is concerned.
4047 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
4051 spa_vdev_state_enter(spa
, SCL_NONE
);
4053 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4054 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4056 if (!vd
->vdev_ops
->vdev_op_leaf
)
4057 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4060 * If the vdev is already faulted, then don't do anything.
4062 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
4063 return (spa_vdev_state_exit(spa
, NULL
, 0));
4065 vd
->vdev_degraded
= 1ULL;
4066 if (!vdev_is_dead(vd
))
4067 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
4070 return (spa_vdev_state_exit(spa
, vd
, 0));
4074 vdev_remove_wanted(spa_t
*spa
, uint64_t guid
)
4078 spa_vdev_state_enter(spa
, SCL_NONE
);
4080 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4081 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4084 * If the vdev is already removed, then don't do anything.
4086 if (vd
->vdev_removed
)
4087 return (spa_vdev_state_exit(spa
, NULL
, 0));
4089 vd
->vdev_remove_wanted
= B_TRUE
;
4090 spa_async_request(spa
, SPA_ASYNC_REMOVE
);
4092 return (spa_vdev_state_exit(spa
, vd
, 0));
4097 * Online the given vdev.
4099 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
4100 * spare device should be detached when the device finishes resilvering.
4101 * Second, the online should be treated like a 'test' online case, so no FMA
4102 * events are generated if the device fails to open.
4105 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
4107 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
4108 boolean_t wasoffline
;
4109 vdev_state_t oldstate
;
4111 spa_vdev_state_enter(spa
, SCL_NONE
);
4113 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4114 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4116 if (!vd
->vdev_ops
->vdev_op_leaf
)
4117 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4119 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
4120 oldstate
= vd
->vdev_state
;
4123 vd
->vdev_offline
= B_FALSE
;
4124 vd
->vdev_tmpoffline
= B_FALSE
;
4125 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
4126 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
4128 /* XXX - L2ARC 1.0 does not support expansion */
4129 if (!vd
->vdev_aux
) {
4130 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4131 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
4132 spa
->spa_autoexpand
);
4133 vd
->vdev_expansion_time
= gethrestime_sec();
4137 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
4139 if (!vd
->vdev_aux
) {
4140 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4141 pvd
->vdev_expanding
= B_FALSE
;
4145 *newstate
= vd
->vdev_state
;
4146 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
4147 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
4148 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4149 vd
->vdev_parent
->vdev_child
[0] == vd
)
4150 vd
->vdev_unspare
= B_TRUE
;
4152 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
4154 /* XXX - L2ARC 1.0 does not support expansion */
4156 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
4157 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
4160 /* Restart initializing if necessary */
4161 mutex_enter(&vd
->vdev_initialize_lock
);
4162 if (vdev_writeable(vd
) &&
4163 vd
->vdev_initialize_thread
== NULL
&&
4164 vd
->vdev_initialize_state
== VDEV_INITIALIZE_ACTIVE
) {
4165 (void) vdev_initialize(vd
);
4167 mutex_exit(&vd
->vdev_initialize_lock
);
4170 * Restart trimming if necessary. We do not restart trimming for cache
4171 * devices here. This is triggered by l2arc_rebuild_vdev()
4172 * asynchronously for the whole device or in l2arc_evict() as it evicts
4173 * space for upcoming writes.
4175 mutex_enter(&vd
->vdev_trim_lock
);
4176 if (vdev_writeable(vd
) && !vd
->vdev_isl2cache
&&
4177 vd
->vdev_trim_thread
== NULL
&&
4178 vd
->vdev_trim_state
== VDEV_TRIM_ACTIVE
) {
4179 (void) vdev_trim(vd
, vd
->vdev_trim_rate
, vd
->vdev_trim_partial
,
4180 vd
->vdev_trim_secure
);
4182 mutex_exit(&vd
->vdev_trim_lock
);
4185 (oldstate
< VDEV_STATE_DEGRADED
&&
4186 vd
->vdev_state
>= VDEV_STATE_DEGRADED
)) {
4187 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
4190 * Asynchronously detach spare vdev if resilver or
4191 * rebuild is not required
4193 if (vd
->vdev_unspare
&&
4194 !dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4195 !dsl_scan_resilver_scheduled(spa
->spa_dsl_pool
) &&
4196 !vdev_rebuild_active(tvd
))
4197 spa_async_request(spa
, SPA_ASYNC_DETACH_SPARE
);
4199 return (spa_vdev_state_exit(spa
, vd
, 0));
4203 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4207 uint64_t generation
;
4208 metaslab_group_t
*mg
;
4211 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4213 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
4214 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENODEV
)));
4216 if (!vd
->vdev_ops
->vdev_op_leaf
)
4217 return (spa_vdev_state_exit(spa
, NULL
, SET_ERROR(ENOTSUP
)));
4219 if (vd
->vdev_ops
== &vdev_draid_spare_ops
)
4220 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
4224 generation
= spa
->spa_config_generation
+ 1;
4227 * If the device isn't already offline, try to offline it.
4229 if (!vd
->vdev_offline
) {
4231 * If this device has the only valid copy of some data,
4232 * don't allow it to be offlined. Log devices are always
4235 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4236 vdev_dtl_required(vd
))
4237 return (spa_vdev_state_exit(spa
, NULL
,
4241 * If the top-level is a slog and it has had allocations
4242 * then proceed. We check that the vdev's metaslab group
4243 * is not NULL since it's possible that we may have just
4244 * added this vdev but not yet initialized its metaslabs.
4246 if (tvd
->vdev_islog
&& mg
!= NULL
) {
4248 * Prevent any future allocations.
4250 ASSERT3P(tvd
->vdev_log_mg
, ==, NULL
);
4251 metaslab_group_passivate(mg
);
4252 (void) spa_vdev_state_exit(spa
, vd
, 0);
4254 error
= spa_reset_logs(spa
);
4257 * If the log device was successfully reset but has
4258 * checkpointed data, do not offline it.
4261 tvd
->vdev_checkpoint_sm
!= NULL
) {
4262 ASSERT3U(space_map_allocated(
4263 tvd
->vdev_checkpoint_sm
), !=, 0);
4264 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
4267 spa_vdev_state_enter(spa
, SCL_ALLOC
);
4270 * Check to see if the config has changed.
4272 if (error
|| generation
!= spa
->spa_config_generation
) {
4273 metaslab_group_activate(mg
);
4275 return (spa_vdev_state_exit(spa
,
4277 (void) spa_vdev_state_exit(spa
, vd
, 0);
4280 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
4284 * Offline this device and reopen its top-level vdev.
4285 * If the top-level vdev is a log device then just offline
4286 * it. Otherwise, if this action results in the top-level
4287 * vdev becoming unusable, undo it and fail the request.
4289 vd
->vdev_offline
= B_TRUE
;
4292 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
4293 vdev_is_dead(tvd
)) {
4294 vd
->vdev_offline
= B_FALSE
;
4296 return (spa_vdev_state_exit(spa
, NULL
,
4301 * Add the device back into the metaslab rotor so that
4302 * once we online the device it's open for business.
4304 if (tvd
->vdev_islog
&& mg
!= NULL
)
4305 metaslab_group_activate(mg
);
4308 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
4310 return (spa_vdev_state_exit(spa
, vd
, 0));
4314 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
4318 mutex_enter(&spa
->spa_vdev_top_lock
);
4319 error
= vdev_offline_locked(spa
, guid
, flags
);
4320 mutex_exit(&spa
->spa_vdev_top_lock
);
4326 * Clear the error counts associated with this vdev. Unlike vdev_online() and
4327 * vdev_offline(), we assume the spa config is locked. We also clear all
4328 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
4331 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
4333 vdev_t
*rvd
= spa
->spa_root_vdev
;
4335 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
4340 vd
->vdev_stat
.vs_read_errors
= 0;
4341 vd
->vdev_stat
.vs_write_errors
= 0;
4342 vd
->vdev_stat
.vs_checksum_errors
= 0;
4343 vd
->vdev_stat
.vs_slow_ios
= 0;
4345 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4346 vdev_clear(spa
, vd
->vdev_child
[c
]);
4349 * It makes no sense to "clear" an indirect or removed vdev.
4351 if (!vdev_is_concrete(vd
) || vd
->vdev_removed
)
4355 * If we're in the FAULTED state or have experienced failed I/O, then
4356 * clear the persistent state and attempt to reopen the device. We
4357 * also mark the vdev config dirty, so that the new faulted state is
4358 * written out to disk.
4360 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
4361 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
4363 * When reopening in response to a clear event, it may be due to
4364 * a fmadm repair request. In this case, if the device is
4365 * still broken, we want to still post the ereport again.
4367 vd
->vdev_forcefault
= B_TRUE
;
4369 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
4370 vd
->vdev_cant_read
= B_FALSE
;
4371 vd
->vdev_cant_write
= B_FALSE
;
4372 vd
->vdev_stat
.vs_aux
= 0;
4374 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
4376 vd
->vdev_forcefault
= B_FALSE
;
4378 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
4379 vdev_state_dirty(vd
->vdev_top
);
4381 /* If a resilver isn't required, check if vdevs can be culled */
4382 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
) &&
4383 !dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
4384 !dsl_scan_resilver_scheduled(spa
->spa_dsl_pool
))
4385 spa_async_request(spa
, SPA_ASYNC_RESILVER_DONE
);
4387 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
4391 * When clearing a FMA-diagnosed fault, we always want to
4392 * unspare the device, as we assume that the original spare was
4393 * done in response to the FMA fault.
4395 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
4396 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
4397 vd
->vdev_parent
->vdev_child
[0] == vd
)
4398 vd
->vdev_unspare
= B_TRUE
;
4400 /* Clear recent error events cache (i.e. duplicate events tracking) */
4401 zfs_ereport_clear(spa
, vd
);
4405 vdev_is_dead(vdev_t
*vd
)
4408 * Holes and missing devices are always considered "dead".
4409 * This simplifies the code since we don't have to check for
4410 * these types of devices in the various code paths.
4411 * Instead we rely on the fact that we skip over dead devices
4412 * before issuing I/O to them.
4414 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
4415 vd
->vdev_ops
== &vdev_hole_ops
||
4416 vd
->vdev_ops
== &vdev_missing_ops
);
4420 vdev_readable(vdev_t
*vd
)
4422 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
4426 vdev_writeable(vdev_t
*vd
)
4428 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
4429 vdev_is_concrete(vd
));
4433 vdev_allocatable(vdev_t
*vd
)
4435 uint64_t state
= vd
->vdev_state
;
4438 * We currently allow allocations from vdevs which may be in the
4439 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
4440 * fails to reopen then we'll catch it later when we're holding
4441 * the proper locks. Note that we have to get the vdev state
4442 * in a local variable because although it changes atomically,
4443 * we're asking two separate questions about it.
4445 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
4446 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
4447 vd
->vdev_mg
->mg_initialized
);
4451 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
4453 ASSERT(zio
->io_vd
== vd
);
4455 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
4458 if (zio
->io_type
== ZIO_TYPE_READ
)
4459 return (!vd
->vdev_cant_read
);
4461 if (zio
->io_type
== ZIO_TYPE_WRITE
)
4462 return (!vd
->vdev_cant_write
);
4468 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
4471 * Exclude the dRAID spare when aggregating to avoid double counting
4472 * the ops and bytes. These IOs are counted by the physical leaves.
4474 if (cvd
->vdev_ops
== &vdev_draid_spare_ops
)
4477 for (int t
= 0; t
< VS_ZIO_TYPES
; t
++) {
4478 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
4479 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
4482 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
4486 * Get extended stats
4489 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
4494 for (t
= 0; t
< ZIO_TYPES
; t
++) {
4495 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
4496 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
4498 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
4499 vsx
->vsx_total_histo
[t
][b
] +=
4500 cvsx
->vsx_total_histo
[t
][b
];
4504 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
4505 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
4506 vsx
->vsx_queue_histo
[t
][b
] +=
4507 cvsx
->vsx_queue_histo
[t
][b
];
4509 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
4510 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
4512 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
4513 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
4515 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
4516 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
4522 vdev_is_spacemap_addressable(vdev_t
*vd
)
4524 if (spa_feature_is_active(vd
->vdev_spa
, SPA_FEATURE_SPACEMAP_V2
))
4528 * If double-word space map entries are not enabled we assume
4529 * 47 bits of the space map entry are dedicated to the entry's
4530 * offset (see SM_OFFSET_BITS in space_map.h). We then use that
4531 * to calculate the maximum address that can be described by a
4532 * space map entry for the given device.
4534 uint64_t shift
= vd
->vdev_ashift
+ SM_OFFSET_BITS
;
4536 if (shift
>= 63) /* detect potential overflow */
4539 return (vd
->vdev_asize
< (1ULL << shift
));
4543 * Get statistics for the given vdev.
4546 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4550 * If we're getting stats on the root vdev, aggregate the I/O counts
4551 * over all top-level vdevs (i.e. the direct children of the root).
4553 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4555 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
4556 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
4559 memset(vsx
, 0, sizeof (*vsx
));
4561 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4562 vdev_t
*cvd
= vd
->vdev_child
[c
];
4563 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
4564 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
4566 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
4568 vdev_get_child_stat(cvd
, vs
, cvs
);
4570 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
4574 * We're a leaf. Just copy our ZIO active queue stats in. The
4575 * other leaf stats are updated in vdev_stat_update().
4580 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
4582 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
4583 vsx
->vsx_active_queue
[t
] =
4584 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
4585 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
4586 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
4592 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
4594 vdev_t
*tvd
= vd
->vdev_top
;
4595 mutex_enter(&vd
->vdev_stat_lock
);
4597 memcpy(vs
, &vd
->vdev_stat
, sizeof (*vs
));
4598 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
4599 vs
->vs_state
= vd
->vdev_state
;
4600 vs
->vs_rsize
= vdev_get_min_asize(vd
);
4602 if (vd
->vdev_ops
->vdev_op_leaf
) {
4603 vs
->vs_pspace
= vd
->vdev_psize
;
4604 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
4605 VDEV_LABEL_END_SIZE
;
4607 * Report initializing progress. Since we don't
4608 * have the initializing locks held, this is only
4609 * an estimate (although a fairly accurate one).
4611 vs
->vs_initialize_bytes_done
=
4612 vd
->vdev_initialize_bytes_done
;
4613 vs
->vs_initialize_bytes_est
=
4614 vd
->vdev_initialize_bytes_est
;
4615 vs
->vs_initialize_state
= vd
->vdev_initialize_state
;
4616 vs
->vs_initialize_action_time
=
4617 vd
->vdev_initialize_action_time
;
4620 * Report manual TRIM progress. Since we don't have
4621 * the manual TRIM locks held, this is only an
4622 * estimate (although fairly accurate one).
4624 vs
->vs_trim_notsup
= !vd
->vdev_has_trim
;
4625 vs
->vs_trim_bytes_done
= vd
->vdev_trim_bytes_done
;
4626 vs
->vs_trim_bytes_est
= vd
->vdev_trim_bytes_est
;
4627 vs
->vs_trim_state
= vd
->vdev_trim_state
;
4628 vs
->vs_trim_action_time
= vd
->vdev_trim_action_time
;
4630 /* Set when there is a deferred resilver. */
4631 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
4635 * Report expandable space on top-level, non-auxiliary devices
4636 * only. The expandable space is reported in terms of metaslab
4637 * sized units since that determines how much space the pool
4640 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
4641 vs
->vs_esize
= P2ALIGN(
4642 vd
->vdev_max_asize
- vd
->vdev_asize
,
4643 1ULL << tvd
->vdev_ms_shift
);
4646 vs
->vs_configured_ashift
= vd
->vdev_top
!= NULL
4647 ? vd
->vdev_top
->vdev_ashift
: vd
->vdev_ashift
;
4648 vs
->vs_logical_ashift
= vd
->vdev_logical_ashift
;
4649 if (vd
->vdev_physical_ashift
<= ASHIFT_MAX
)
4650 vs
->vs_physical_ashift
= vd
->vdev_physical_ashift
;
4652 vs
->vs_physical_ashift
= 0;
4655 * Report fragmentation and rebuild progress for top-level,
4656 * non-auxiliary, concrete devices.
4658 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
4659 vdev_is_concrete(vd
)) {
4661 * The vdev fragmentation rating doesn't take into
4662 * account the embedded slog metaslab (vdev_log_mg).
4663 * Since it's only one metaslab, it would have a tiny
4664 * impact on the overall fragmentation.
4666 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
4667 vd
->vdev_mg
->mg_fragmentation
: 0;
4669 vs
->vs_noalloc
= MAX(vd
->vdev_noalloc
,
4670 tvd
? tvd
->vdev_noalloc
: 0);
4673 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
4674 mutex_exit(&vd
->vdev_stat_lock
);
4678 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
4680 return (vdev_get_stats_ex(vd
, vs
, NULL
));
4684 vdev_clear_stats(vdev_t
*vd
)
4686 mutex_enter(&vd
->vdev_stat_lock
);
4687 vd
->vdev_stat
.vs_space
= 0;
4688 vd
->vdev_stat
.vs_dspace
= 0;
4689 vd
->vdev_stat
.vs_alloc
= 0;
4690 mutex_exit(&vd
->vdev_stat_lock
);
4694 vdev_scan_stat_init(vdev_t
*vd
)
4696 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4698 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4699 vdev_scan_stat_init(vd
->vdev_child
[c
]);
4701 mutex_enter(&vd
->vdev_stat_lock
);
4702 vs
->vs_scan_processed
= 0;
4703 mutex_exit(&vd
->vdev_stat_lock
);
4707 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
4709 spa_t
*spa
= zio
->io_spa
;
4710 vdev_t
*rvd
= spa
->spa_root_vdev
;
4711 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
4713 uint64_t txg
= zio
->io_txg
;
4714 /* Suppress ASAN false positive */
4715 #ifdef __SANITIZE_ADDRESS__
4716 vdev_stat_t
*vs
= vd
? &vd
->vdev_stat
: NULL
;
4717 vdev_stat_ex_t
*vsx
= vd
? &vd
->vdev_stat_ex
: NULL
;
4719 vdev_stat_t
*vs
= &vd
->vdev_stat
;
4720 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
4722 zio_type_t type
= zio
->io_type
;
4723 int flags
= zio
->io_flags
;
4726 * If this i/o is a gang leader, it didn't do any actual work.
4728 if (zio
->io_gang_tree
)
4731 if (zio
->io_error
== 0) {
4733 * If this is a root i/o, don't count it -- we've already
4734 * counted the top-level vdevs, and vdev_get_stats() will
4735 * aggregate them when asked. This reduces contention on
4736 * the root vdev_stat_lock and implicitly handles blocks
4737 * that compress away to holes, for which there is no i/o.
4738 * (Holes never create vdev children, so all the counters
4739 * remain zero, which is what we want.)
4741 * Note: this only applies to successful i/o (io_error == 0)
4742 * because unlike i/o counts, errors are not additive.
4743 * When reading a ditto block, for example, failure of
4744 * one top-level vdev does not imply a root-level error.
4749 ASSERT(vd
== zio
->io_vd
);
4751 if (flags
& ZIO_FLAG_IO_BYPASS
)
4754 mutex_enter(&vd
->vdev_stat_lock
);
4756 if (flags
& ZIO_FLAG_IO_REPAIR
) {
4758 * Repair is the result of a resilver issued by the
4759 * scan thread (spa_sync).
4761 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4762 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
4763 dsl_scan_phys_t
*scn_phys
= &scn
->scn_phys
;
4764 uint64_t *processed
= &scn_phys
->scn_processed
;
4766 if (vd
->vdev_ops
->vdev_op_leaf
)
4767 atomic_add_64(processed
, psize
);
4768 vs
->vs_scan_processed
+= psize
;
4772 * Repair is the result of a rebuild issued by the
4773 * rebuild thread (vdev_rebuild_thread). To avoid
4774 * double counting repaired bytes the virtual dRAID
4775 * spare vdev is excluded from the processed bytes.
4777 if (zio
->io_priority
== ZIO_PRIORITY_REBUILD
) {
4778 vdev_t
*tvd
= vd
->vdev_top
;
4779 vdev_rebuild_t
*vr
= &tvd
->vdev_rebuild_config
;
4780 vdev_rebuild_phys_t
*vrp
= &vr
->vr_rebuild_phys
;
4781 uint64_t *rebuilt
= &vrp
->vrp_bytes_rebuilt
;
4783 if (vd
->vdev_ops
->vdev_op_leaf
&&
4784 vd
->vdev_ops
!= &vdev_draid_spare_ops
) {
4785 atomic_add_64(rebuilt
, psize
);
4787 vs
->vs_rebuild_processed
+= psize
;
4790 if (flags
& ZIO_FLAG_SELF_HEAL
)
4791 vs
->vs_self_healed
+= psize
;
4795 * The bytes/ops/histograms are recorded at the leaf level and
4796 * aggregated into the higher level vdevs in vdev_get_stats().
4798 if (vd
->vdev_ops
->vdev_op_leaf
&&
4799 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
4800 zio_type_t vs_type
= type
;
4801 zio_priority_t priority
= zio
->io_priority
;
4804 * TRIM ops and bytes are reported to user space as
4805 * ZIO_TYPE_IOCTL. This is done to preserve the
4806 * vdev_stat_t structure layout for user space.
4808 if (type
== ZIO_TYPE_TRIM
)
4809 vs_type
= ZIO_TYPE_IOCTL
;
4812 * Solely for the purposes of 'zpool iostat -lqrw'
4813 * reporting use the priority to categorize the IO.
4814 * Only the following are reported to user space:
4816 * ZIO_PRIORITY_SYNC_READ,
4817 * ZIO_PRIORITY_SYNC_WRITE,
4818 * ZIO_PRIORITY_ASYNC_READ,
4819 * ZIO_PRIORITY_ASYNC_WRITE,
4820 * ZIO_PRIORITY_SCRUB,
4821 * ZIO_PRIORITY_TRIM,
4822 * ZIO_PRIORITY_REBUILD.
4824 if (priority
== ZIO_PRIORITY_INITIALIZING
) {
4825 ASSERT3U(type
, ==, ZIO_TYPE_WRITE
);
4826 priority
= ZIO_PRIORITY_ASYNC_WRITE
;
4827 } else if (priority
== ZIO_PRIORITY_REMOVAL
) {
4828 priority
= ((type
== ZIO_TYPE_WRITE
) ?
4829 ZIO_PRIORITY_ASYNC_WRITE
:
4830 ZIO_PRIORITY_ASYNC_READ
);
4833 vs
->vs_ops
[vs_type
]++;
4834 vs
->vs_bytes
[vs_type
] += psize
;
4836 if (flags
& ZIO_FLAG_DELEGATED
) {
4837 vsx
->vsx_agg_histo
[priority
]
4838 [RQ_HISTO(zio
->io_size
)]++;
4840 vsx
->vsx_ind_histo
[priority
]
4841 [RQ_HISTO(zio
->io_size
)]++;
4844 if (zio
->io_delta
&& zio
->io_delay
) {
4845 vsx
->vsx_queue_histo
[priority
]
4846 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
4847 vsx
->vsx_disk_histo
[type
]
4848 [L_HISTO(zio
->io_delay
)]++;
4849 vsx
->vsx_total_histo
[type
]
4850 [L_HISTO(zio
->io_delta
)]++;
4854 mutex_exit(&vd
->vdev_stat_lock
);
4858 if (flags
& ZIO_FLAG_SPECULATIVE
)
4862 * If this is an I/O error that is going to be retried, then ignore the
4863 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
4864 * hard errors, when in reality they can happen for any number of
4865 * innocuous reasons (bus resets, MPxIO link failure, etc).
4867 if (zio
->io_error
== EIO
&&
4868 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4872 * Intent logs writes won't propagate their error to the root
4873 * I/O so don't mark these types of failures as pool-level
4876 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4879 if (type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4880 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4881 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4882 spa
->spa_claiming
)) {
4884 * This is either a normal write (not a repair), or it's
4885 * a repair induced by the scrub thread, or it's a repair
4886 * made by zil_claim() during spa_load() in the first txg.
4887 * In the normal case, we commit the DTL change in the same
4888 * txg as the block was born. In the scrub-induced repair
4889 * case, we know that scrubs run in first-pass syncing context,
4890 * so we commit the DTL change in spa_syncing_txg(spa).
4891 * In the zil_claim() case, we commit in spa_first_txg(spa).
4893 * We currently do not make DTL entries for failed spontaneous
4894 * self-healing writes triggered by normal (non-scrubbing)
4895 * reads, because we have no transactional context in which to
4896 * do so -- and it's not clear that it'd be desirable anyway.
4898 if (vd
->vdev_ops
->vdev_op_leaf
) {
4899 uint64_t commit_txg
= txg
;
4900 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4901 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4902 ASSERT(spa_sync_pass(spa
) == 1);
4903 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4904 commit_txg
= spa_syncing_txg(spa
);
4905 } else if (spa
->spa_claiming
) {
4906 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4907 commit_txg
= spa_first_txg(spa
);
4909 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4910 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4912 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4913 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4914 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4917 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4922 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4924 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4925 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4927 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4931 * Update the in-core space usage stats for this vdev, its metaslab class,
4932 * and the root vdev.
4935 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4936 int64_t space_delta
)
4939 int64_t dspace_delta
;
4940 spa_t
*spa
= vd
->vdev_spa
;
4941 vdev_t
*rvd
= spa
->spa_root_vdev
;
4943 ASSERT(vd
== vd
->vdev_top
);
4946 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4947 * factor. We must calculate this here and not at the root vdev
4948 * because the root vdev's psize-to-asize is simply the max of its
4949 * children's, thus not accurate enough for us.
4951 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4953 mutex_enter(&vd
->vdev_stat_lock
);
4954 /* ensure we won't underflow */
4955 if (alloc_delta
< 0) {
4956 ASSERT3U(vd
->vdev_stat
.vs_alloc
, >=, -alloc_delta
);
4959 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4960 vd
->vdev_stat
.vs_space
+= space_delta
;
4961 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4962 mutex_exit(&vd
->vdev_stat_lock
);
4964 /* every class but log contributes to root space stats */
4965 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4966 ASSERT(!vd
->vdev_isl2cache
);
4967 mutex_enter(&rvd
->vdev_stat_lock
);
4968 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4969 rvd
->vdev_stat
.vs_space
+= space_delta
;
4970 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4971 mutex_exit(&rvd
->vdev_stat_lock
);
4973 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4977 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4978 * so that it will be written out next time the vdev configuration is synced.
4979 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4982 vdev_config_dirty(vdev_t
*vd
)
4984 spa_t
*spa
= vd
->vdev_spa
;
4985 vdev_t
*rvd
= spa
->spa_root_vdev
;
4988 ASSERT(spa_writeable(spa
));
4991 * If this is an aux vdev (as with l2cache and spare devices), then we
4992 * update the vdev config manually and set the sync flag.
4994 if (vd
->vdev_aux
!= NULL
) {
4995 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
4999 for (c
= 0; c
< sav
->sav_count
; c
++) {
5000 if (sav
->sav_vdevs
[c
] == vd
)
5004 if (c
== sav
->sav_count
) {
5006 * We're being removed. There's nothing more to do.
5008 ASSERT(sav
->sav_sync
== B_TRUE
);
5012 sav
->sav_sync
= B_TRUE
;
5014 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
5015 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
5016 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
5017 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
5023 * Setting the nvlist in the middle if the array is a little
5024 * sketchy, but it will work.
5026 nvlist_free(aux
[c
]);
5027 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
5033 * The dirty list is protected by the SCL_CONFIG lock. The caller
5034 * must either hold SCL_CONFIG as writer, or must be the sync thread
5035 * (which holds SCL_CONFIG as reader). There's only one sync thread,
5036 * so this is sufficient to ensure mutual exclusion.
5038 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5039 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5040 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5043 for (c
= 0; c
< rvd
->vdev_children
; c
++)
5044 vdev_config_dirty(rvd
->vdev_child
[c
]);
5046 ASSERT(vd
== vd
->vdev_top
);
5048 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
5049 vdev_is_concrete(vd
)) {
5050 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
5056 vdev_config_clean(vdev_t
*vd
)
5058 spa_t
*spa
= vd
->vdev_spa
;
5060 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
5061 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5062 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
5064 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
5065 list_remove(&spa
->spa_config_dirty_list
, vd
);
5069 * Mark a top-level vdev's state as dirty, so that the next pass of
5070 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
5071 * the state changes from larger config changes because they require
5072 * much less locking, and are often needed for administrative actions.
5075 vdev_state_dirty(vdev_t
*vd
)
5077 spa_t
*spa
= vd
->vdev_spa
;
5079 ASSERT(spa_writeable(spa
));
5080 ASSERT(vd
== vd
->vdev_top
);
5083 * The state list is protected by the SCL_STATE lock. The caller
5084 * must either hold SCL_STATE as writer, or must be the sync thread
5085 * (which holds SCL_STATE as reader). There's only one sync thread,
5086 * so this is sufficient to ensure mutual exclusion.
5088 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5089 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5090 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5092 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
5093 vdev_is_concrete(vd
))
5094 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
5098 vdev_state_clean(vdev_t
*vd
)
5100 spa_t
*spa
= vd
->vdev_spa
;
5102 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
5103 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
5104 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
5106 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
5107 list_remove(&spa
->spa_state_dirty_list
, vd
);
5111 * Propagate vdev state up from children to parent.
5114 vdev_propagate_state(vdev_t
*vd
)
5116 spa_t
*spa
= vd
->vdev_spa
;
5117 vdev_t
*rvd
= spa
->spa_root_vdev
;
5118 int degraded
= 0, faulted
= 0;
5122 if (vd
->vdev_children
> 0) {
5123 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5124 child
= vd
->vdev_child
[c
];
5127 * Don't factor holes or indirect vdevs into the
5130 if (!vdev_is_concrete(child
))
5133 if (!vdev_readable(child
) ||
5134 (!vdev_writeable(child
) && spa_writeable(spa
))) {
5136 * Root special: if there is a top-level log
5137 * device, treat the root vdev as if it were
5140 if (child
->vdev_islog
&& vd
== rvd
)
5144 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
5148 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
5152 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
5155 * Root special: if there is a top-level vdev that cannot be
5156 * opened due to corrupted metadata, then propagate the root
5157 * vdev's aux state as 'corrupt' rather than 'insufficient
5160 if (corrupted
&& vd
== rvd
&&
5161 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
5162 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
5163 VDEV_AUX_CORRUPT_DATA
);
5166 if (vd
->vdev_parent
)
5167 vdev_propagate_state(vd
->vdev_parent
);
5171 * Set a vdev's state. If this is during an open, we don't update the parent
5172 * state, because we're in the process of opening children depth-first.
5173 * Otherwise, we propagate the change to the parent.
5175 * If this routine places a device in a faulted state, an appropriate ereport is
5179 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
5181 uint64_t save_state
;
5182 spa_t
*spa
= vd
->vdev_spa
;
5184 if (state
== vd
->vdev_state
) {
5186 * Since vdev_offline() code path is already in an offline
5187 * state we can miss a statechange event to OFFLINE. Check
5188 * the previous state to catch this condition.
5190 if (vd
->vdev_ops
->vdev_op_leaf
&&
5191 (state
== VDEV_STATE_OFFLINE
) &&
5192 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
5193 /* post an offline state change */
5194 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
5196 vd
->vdev_stat
.vs_aux
= aux
;
5200 save_state
= vd
->vdev_state
;
5202 vd
->vdev_state
= state
;
5203 vd
->vdev_stat
.vs_aux
= aux
;
5206 * If we are setting the vdev state to anything but an open state, then
5207 * always close the underlying device unless the device has requested
5208 * a delayed close (i.e. we're about to remove or fault the device).
5209 * Otherwise, we keep accessible but invalid devices open forever.
5210 * We don't call vdev_close() itself, because that implies some extra
5211 * checks (offline, etc) that we don't want here. This is limited to
5212 * leaf devices, because otherwise closing the device will affect other
5215 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
5216 vd
->vdev_ops
->vdev_op_leaf
)
5217 vd
->vdev_ops
->vdev_op_close(vd
);
5219 if (vd
->vdev_removed
&&
5220 state
== VDEV_STATE_CANT_OPEN
&&
5221 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
5223 * If the previous state is set to VDEV_STATE_REMOVED, then this
5224 * device was previously marked removed and someone attempted to
5225 * reopen it. If this failed due to a nonexistent device, then
5226 * keep the device in the REMOVED state. We also let this be if
5227 * it is one of our special test online cases, which is only
5228 * attempting to online the device and shouldn't generate an FMA
5231 vd
->vdev_state
= VDEV_STATE_REMOVED
;
5232 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
5233 } else if (state
== VDEV_STATE_REMOVED
) {
5234 vd
->vdev_removed
= B_TRUE
;
5235 } else if (state
== VDEV_STATE_CANT_OPEN
) {
5237 * If we fail to open a vdev during an import or recovery, we
5238 * mark it as "not available", which signifies that it was
5239 * never there to begin with. Failure to open such a device
5240 * is not considered an error.
5242 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
5243 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
5244 vd
->vdev_ops
->vdev_op_leaf
)
5245 vd
->vdev_not_present
= 1;
5248 * Post the appropriate ereport. If the 'prevstate' field is
5249 * set to something other than VDEV_STATE_UNKNOWN, it indicates
5250 * that this is part of a vdev_reopen(). In this case, we don't
5251 * want to post the ereport if the device was already in the
5252 * CANT_OPEN state beforehand.
5254 * If the 'checkremove' flag is set, then this is an attempt to
5255 * online the device in response to an insertion event. If we
5256 * hit this case, then we have detected an insertion event for a
5257 * faulted or offline device that wasn't in the removed state.
5258 * In this scenario, we don't post an ereport because we are
5259 * about to replace the device, or attempt an online with
5260 * vdev_forcefault, which will generate the fault for us.
5262 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
5263 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
5264 vd
!= spa
->spa_root_vdev
) {
5268 case VDEV_AUX_OPEN_FAILED
:
5269 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
5271 case VDEV_AUX_CORRUPT_DATA
:
5272 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
5274 case VDEV_AUX_NO_REPLICAS
:
5275 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
5277 case VDEV_AUX_BAD_GUID_SUM
:
5278 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
5280 case VDEV_AUX_TOO_SMALL
:
5281 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
5283 case VDEV_AUX_BAD_LABEL
:
5284 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
5286 case VDEV_AUX_BAD_ASHIFT
:
5287 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
5290 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
5293 (void) zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
5297 /* Erase any notion of persistent removed state */
5298 vd
->vdev_removed
= B_FALSE
;
5300 vd
->vdev_removed
= B_FALSE
;
5304 * Notify ZED of any significant state-change on a leaf vdev.
5307 if (vd
->vdev_ops
->vdev_op_leaf
) {
5308 /* preserve original state from a vdev_reopen() */
5309 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
5310 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
5311 (save_state
<= VDEV_STATE_CLOSED
))
5312 save_state
= vd
->vdev_prevstate
;
5314 /* filter out state change due to initial vdev_open */
5315 if (save_state
> VDEV_STATE_CLOSED
)
5316 zfs_post_state_change(spa
, vd
, save_state
);
5319 if (!isopen
&& vd
->vdev_parent
)
5320 vdev_propagate_state(vd
->vdev_parent
);
5324 vdev_children_are_offline(vdev_t
*vd
)
5326 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
5328 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
5329 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
5337 * Check the vdev configuration to ensure that it's capable of supporting
5338 * a root pool. We do not support partial configuration.
5341 vdev_is_bootable(vdev_t
*vd
)
5343 if (!vd
->vdev_ops
->vdev_op_leaf
) {
5344 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
5346 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0)
5350 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5351 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
5358 vdev_is_concrete(vdev_t
*vd
)
5360 vdev_ops_t
*ops
= vd
->vdev_ops
;
5361 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
5362 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
5370 * Determine if a log device has valid content. If the vdev was
5371 * removed or faulted in the MOS config then we know that
5372 * the content on the log device has already been written to the pool.
5375 vdev_log_state_valid(vdev_t
*vd
)
5377 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
5381 for (int c
= 0; c
< vd
->vdev_children
; c
++)
5382 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
5389 * Expand a vdev if possible.
5392 vdev_expand(vdev_t
*vd
, uint64_t txg
)
5394 ASSERT(vd
->vdev_top
== vd
);
5395 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
5396 ASSERT(vdev_is_concrete(vd
));
5398 vdev_set_deflate_ratio(vd
);
5400 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
5401 vdev_is_concrete(vd
)) {
5402 vdev_metaslab_group_create(vd
);
5403 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
5404 vdev_config_dirty(vd
);
5412 vdev_split(vdev_t
*vd
)
5414 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
5416 VERIFY3U(pvd
->vdev_children
, >, 1);
5418 vdev_remove_child(pvd
, vd
);
5419 vdev_compact_children(pvd
);
5421 ASSERT3P(pvd
->vdev_child
, !=, NULL
);
5423 cvd
= pvd
->vdev_child
[0];
5424 if (pvd
->vdev_children
== 1) {
5425 vdev_remove_parent(cvd
);
5426 cvd
->vdev_splitting
= B_TRUE
;
5428 vdev_propagate_state(cvd
);
5432 vdev_deadman(vdev_t
*vd
, const char *tag
)
5434 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5435 vdev_t
*cvd
= vd
->vdev_child
[c
];
5437 vdev_deadman(cvd
, tag
);
5440 if (vd
->vdev_ops
->vdev_op_leaf
) {
5441 vdev_queue_t
*vq
= &vd
->vdev_queue
;
5443 mutex_enter(&vq
->vq_lock
);
5444 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
5445 spa_t
*spa
= vd
->vdev_spa
;
5449 zfs_dbgmsg("slow vdev: %s has %lu active IOs",
5450 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
5453 * Look at the head of all the pending queues,
5454 * if any I/O has been outstanding for longer than
5455 * the spa_deadman_synctime invoke the deadman logic.
5457 fio
= avl_first(&vq
->vq_active_tree
);
5458 delta
= gethrtime() - fio
->io_timestamp
;
5459 if (delta
> spa_deadman_synctime(spa
))
5460 zio_deadman(fio
, tag
);
5462 mutex_exit(&vq
->vq_lock
);
5467 vdev_defer_resilver(vdev_t
*vd
)
5469 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
5471 vd
->vdev_resilver_deferred
= B_TRUE
;
5472 vd
->vdev_spa
->spa_resilver_deferred
= B_TRUE
;
5476 * Clears the resilver deferred flag on all leaf devs under vd. Returns
5477 * B_TRUE if we have devices that need to be resilvered and are available to
5478 * accept resilver I/Os.
5481 vdev_clear_resilver_deferred(vdev_t
*vd
, dmu_tx_t
*tx
)
5483 boolean_t resilver_needed
= B_FALSE
;
5484 spa_t
*spa
= vd
->vdev_spa
;
5486 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
5487 vdev_t
*cvd
= vd
->vdev_child
[c
];
5488 resilver_needed
|= vdev_clear_resilver_deferred(cvd
, tx
);
5491 if (vd
== spa
->spa_root_vdev
&&
5492 spa_feature_is_active(spa
, SPA_FEATURE_RESILVER_DEFER
)) {
5493 spa_feature_decr(spa
, SPA_FEATURE_RESILVER_DEFER
, tx
);
5494 vdev_config_dirty(vd
);
5495 spa
->spa_resilver_deferred
= B_FALSE
;
5496 return (resilver_needed
);
5499 if (!vdev_is_concrete(vd
) || vd
->vdev_aux
||
5500 !vd
->vdev_ops
->vdev_op_leaf
)
5501 return (resilver_needed
);
5503 vd
->vdev_resilver_deferred
= B_FALSE
;
5505 return (!vdev_is_dead(vd
) && !vd
->vdev_offline
&&
5506 vdev_resilver_needed(vd
, NULL
, NULL
));
5510 vdev_xlate_is_empty(range_seg64_t
*rs
)
5512 return (rs
->rs_start
== rs
->rs_end
);
5516 * Translate a logical range to the first contiguous physical range for the
5517 * specified vdev_t. This function is initially called with a leaf vdev and
5518 * will walk each parent vdev until it reaches a top-level vdev. Once the
5519 * top-level is reached the physical range is initialized and the recursive
5520 * function begins to unwind. As it unwinds it calls the parent's vdev
5521 * specific translation function to do the real conversion.
5524 vdev_xlate(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5525 range_seg64_t
*physical_rs
, range_seg64_t
*remain_rs
)
5528 * Walk up the vdev tree
5530 if (vd
!= vd
->vdev_top
) {
5531 vdev_xlate(vd
->vdev_parent
, logical_rs
, physical_rs
,
5535 * We've reached the top-level vdev, initialize the physical
5536 * range to the logical range and set an empty remaining
5537 * range then start to unwind.
5539 physical_rs
->rs_start
= logical_rs
->rs_start
;
5540 physical_rs
->rs_end
= logical_rs
->rs_end
;
5542 remain_rs
->rs_start
= logical_rs
->rs_start
;
5543 remain_rs
->rs_end
= logical_rs
->rs_start
;
5548 vdev_t
*pvd
= vd
->vdev_parent
;
5549 ASSERT3P(pvd
, !=, NULL
);
5550 ASSERT3P(pvd
->vdev_ops
->vdev_op_xlate
, !=, NULL
);
5553 * As this recursive function unwinds, translate the logical
5554 * range into its physical and any remaining components by calling
5555 * the vdev specific translate function.
5557 range_seg64_t intermediate
= { 0 };
5558 pvd
->vdev_ops
->vdev_op_xlate(vd
, physical_rs
, &intermediate
, remain_rs
);
5560 physical_rs
->rs_start
= intermediate
.rs_start
;
5561 physical_rs
->rs_end
= intermediate
.rs_end
;
5565 vdev_xlate_walk(vdev_t
*vd
, const range_seg64_t
*logical_rs
,
5566 vdev_xlate_func_t
*func
, void *arg
)
5568 range_seg64_t iter_rs
= *logical_rs
;
5569 range_seg64_t physical_rs
;
5570 range_seg64_t remain_rs
;
5572 while (!vdev_xlate_is_empty(&iter_rs
)) {
5574 vdev_xlate(vd
, &iter_rs
, &physical_rs
, &remain_rs
);
5577 * With raidz and dRAID, it's possible that the logical range
5578 * does not live on this leaf vdev. Only when there is a non-
5579 * zero physical size call the provided function.
5581 if (!vdev_xlate_is_empty(&physical_rs
))
5582 func(arg
, &physical_rs
);
5584 iter_rs
= remain_rs
;
5589 vdev_name(vdev_t
*vd
, char *buf
, int buflen
)
5591 if (vd
->vdev_path
== NULL
) {
5592 if (strcmp(vd
->vdev_ops
->vdev_op_type
, "root") == 0) {
5593 strlcpy(buf
, vd
->vdev_spa
->spa_name
, buflen
);
5594 } else if (!vd
->vdev_ops
->vdev_op_leaf
) {
5595 snprintf(buf
, buflen
, "%s-%llu",
5596 vd
->vdev_ops
->vdev_op_type
,
5597 (u_longlong_t
)vd
->vdev_id
);
5600 strlcpy(buf
, vd
->vdev_path
, buflen
);
5606 * Look at the vdev tree and determine whether any devices are currently being
5610 vdev_replace_in_progress(vdev_t
*vdev
)
5612 ASSERT(spa_config_held(vdev
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
5614 if (vdev
->vdev_ops
== &vdev_replacing_ops
)
5618 * A 'spare' vdev indicates that we have a replace in progress, unless
5619 * it has exactly two children, and the second, the hot spare, has
5620 * finished being resilvered.
5622 if (vdev
->vdev_ops
== &vdev_spare_ops
&& (vdev
->vdev_children
> 2 ||
5623 !vdev_dtl_empty(vdev
->vdev_child
[1], DTL_MISSING
)))
5626 for (int i
= 0; i
< vdev
->vdev_children
; i
++) {
5627 if (vdev_replace_in_progress(vdev
->vdev_child
[i
]))
5635 * Add a (source=src, propname=propval) list to an nvlist.
5638 vdev_prop_add_list(nvlist_t
*nvl
, const char *propname
, const char *strval
,
5639 uint64_t intval
, zprop_source_t src
)
5643 propval
= fnvlist_alloc();
5644 fnvlist_add_uint64(propval
, ZPROP_SOURCE
, src
);
5647 fnvlist_add_string(propval
, ZPROP_VALUE
, strval
);
5649 fnvlist_add_uint64(propval
, ZPROP_VALUE
, intval
);
5651 fnvlist_add_nvlist(nvl
, propname
, propval
);
5652 nvlist_free(propval
);
5656 vdev_props_set_sync(void *arg
, dmu_tx_t
*tx
)
5659 nvlist_t
*nvp
= arg
;
5660 spa_t
*spa
= dmu_tx_pool(tx
)->dp_spa
;
5661 objset_t
*mos
= spa
->spa_meta_objset
;
5662 nvpair_t
*elem
= NULL
;
5666 vdev_guid
= fnvlist_lookup_uint64(nvp
, ZPOOL_VDEV_PROPS_SET_VDEV
);
5667 nvprops
= fnvlist_lookup_nvlist(nvp
, ZPOOL_VDEV_PROPS_SET_PROPS
);
5668 vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
);
5670 /* this vdev could get removed while waiting for this sync task */
5674 mutex_enter(&spa
->spa_props_lock
);
5676 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5677 uint64_t intval
, objid
= 0;
5680 const char *propname
= nvpair_name(elem
);
5681 zprop_type_t proptype
;
5684 * Set vdev property values in the vdev props mos object.
5686 if (vd
->vdev_top_zap
!= 0) {
5687 objid
= vd
->vdev_top_zap
;
5688 } else if (vd
->vdev_leaf_zap
!= 0) {
5689 objid
= vd
->vdev_leaf_zap
;
5691 panic("vdev not top or leaf");
5694 switch (prop
= vdev_name_to_prop(propname
)) {
5695 case VDEV_PROP_USERPROP
:
5696 if (vdev_prop_user(propname
)) {
5697 strval
= fnvpair_value_string(elem
);
5698 if (strlen(strval
) == 0) {
5699 /* remove the property if value == "" */
5700 (void) zap_remove(mos
, objid
, propname
,
5703 VERIFY0(zap_update(mos
, objid
, propname
,
5704 1, strlen(strval
) + 1, strval
, tx
));
5706 spa_history_log_internal(spa
, "vdev set", tx
,
5707 "vdev_guid=%llu: %s=%s",
5708 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5713 /* normalize the property name */
5714 propname
= vdev_prop_to_name(prop
);
5715 proptype
= vdev_prop_get_type(prop
);
5717 if (nvpair_type(elem
) == DATA_TYPE_STRING
) {
5718 ASSERT(proptype
== PROP_TYPE_STRING
);
5719 strval
= fnvpair_value_string(elem
);
5720 VERIFY0(zap_update(mos
, objid
, propname
,
5721 1, strlen(strval
) + 1, strval
, tx
));
5722 spa_history_log_internal(spa
, "vdev set", tx
,
5723 "vdev_guid=%llu: %s=%s",
5724 (u_longlong_t
)vdev_guid
, nvpair_name(elem
),
5726 } else if (nvpair_type(elem
) == DATA_TYPE_UINT64
) {
5727 intval
= fnvpair_value_uint64(elem
);
5729 if (proptype
== PROP_TYPE_INDEX
) {
5731 VERIFY0(vdev_prop_index_to_string(
5732 prop
, intval
, &unused
));
5734 VERIFY0(zap_update(mos
, objid
, propname
,
5735 sizeof (uint64_t), 1, &intval
, tx
));
5736 spa_history_log_internal(spa
, "vdev set", tx
,
5737 "vdev_guid=%llu: %s=%lld",
5738 (u_longlong_t
)vdev_guid
,
5739 nvpair_name(elem
), (longlong_t
)intval
);
5741 panic("invalid vdev property type %u",
5748 mutex_exit(&spa
->spa_props_lock
);
5752 vdev_prop_set(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5754 spa_t
*spa
= vd
->vdev_spa
;
5755 nvpair_t
*elem
= NULL
;
5762 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_SET_VDEV
,
5764 return (SET_ERROR(EINVAL
));
5766 if (nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_SET_PROPS
,
5768 return (SET_ERROR(EINVAL
));
5770 if ((vd
= spa_lookup_by_guid(spa
, vdev_guid
, B_TRUE
)) == NULL
)
5771 return (SET_ERROR(EINVAL
));
5773 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5774 const char *propname
= nvpair_name(elem
);
5775 vdev_prop_t prop
= vdev_name_to_prop(propname
);
5776 uint64_t intval
= 0;
5777 const char *strval
= NULL
;
5779 if (prop
== VDEV_PROP_USERPROP
&& !vdev_prop_user(propname
)) {
5784 if (vdev_prop_readonly(prop
)) {
5789 /* Special Processing */
5791 case VDEV_PROP_PATH
:
5792 if (vd
->vdev_path
== NULL
) {
5796 if (nvpair_value_string(elem
, &strval
) != 0) {
5800 /* New path must start with /dev/ */
5801 if (strncmp(strval
, "/dev/", 5)) {
5805 error
= spa_vdev_setpath(spa
, vdev_guid
, strval
);
5807 case VDEV_PROP_ALLOCATING
:
5808 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5812 if (intval
!= vd
->vdev_noalloc
)
5815 error
= spa_vdev_noalloc(spa
, vdev_guid
);
5817 error
= spa_vdev_alloc(spa
, vdev_guid
);
5819 case VDEV_PROP_FAILFAST
:
5820 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5824 vd
->vdev_failfast
= intval
& 1;
5826 case VDEV_PROP_CHECKSUM_N
:
5827 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5831 vd
->vdev_checksum_n
= intval
;
5833 case VDEV_PROP_CHECKSUM_T
:
5834 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5838 vd
->vdev_checksum_t
= intval
;
5840 case VDEV_PROP_IO_N
:
5841 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5845 vd
->vdev_io_n
= intval
;
5847 case VDEV_PROP_IO_T
:
5848 if (nvpair_value_uint64(elem
, &intval
) != 0) {
5852 vd
->vdev_io_t
= intval
;
5855 /* Most processing is done in vdev_props_set_sync */
5861 vdev_prop_add_list(outnvl
, propname
, strval
, intval
, 0);
5866 return (dsl_sync_task(spa
->spa_name
, NULL
, vdev_props_set_sync
,
5867 innvl
, 6, ZFS_SPACE_CHECK_EXTRA_RESERVED
));
5871 vdev_prop_get(vdev_t
*vd
, nvlist_t
*innvl
, nvlist_t
*outnvl
)
5873 spa_t
*spa
= vd
->vdev_spa
;
5874 objset_t
*mos
= spa
->spa_meta_objset
;
5878 nvpair_t
*elem
= NULL
;
5879 nvlist_t
*nvprops
= NULL
;
5880 uint64_t intval
= 0;
5881 char *strval
= NULL
;
5882 const char *propname
= NULL
;
5886 ASSERT(mos
!= NULL
);
5888 if (nvlist_lookup_uint64(innvl
, ZPOOL_VDEV_PROPS_GET_VDEV
,
5890 return (SET_ERROR(EINVAL
));
5892 nvlist_lookup_nvlist(innvl
, ZPOOL_VDEV_PROPS_GET_PROPS
, &nvprops
);
5894 if (vd
->vdev_top_zap
!= 0) {
5895 objid
= vd
->vdev_top_zap
;
5896 } else if (vd
->vdev_leaf_zap
!= 0) {
5897 objid
= vd
->vdev_leaf_zap
;
5899 return (SET_ERROR(EINVAL
));
5903 mutex_enter(&spa
->spa_props_lock
);
5905 if (nvprops
!= NULL
) {
5906 char namebuf
[64] = { 0 };
5908 while ((elem
= nvlist_next_nvpair(nvprops
, elem
)) != NULL
) {
5911 propname
= nvpair_name(elem
);
5912 prop
= vdev_name_to_prop(propname
);
5913 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
5914 uint64_t integer_size
, num_integers
;
5917 /* Special Read-only Properties */
5918 case VDEV_PROP_NAME
:
5919 strval
= vdev_name(vd
, namebuf
,
5923 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
5926 case VDEV_PROP_CAPACITY
:
5928 intval
= (vd
->vdev_stat
.vs_dspace
== 0) ? 0 :
5929 (vd
->vdev_stat
.vs_alloc
* 100 /
5930 vd
->vdev_stat
.vs_dspace
);
5931 vdev_prop_add_list(outnvl
, propname
, NULL
,
5932 intval
, ZPROP_SRC_NONE
);
5934 case VDEV_PROP_STATE
:
5935 vdev_prop_add_list(outnvl
, propname
, NULL
,
5936 vd
->vdev_state
, ZPROP_SRC_NONE
);
5938 case VDEV_PROP_GUID
:
5939 vdev_prop_add_list(outnvl
, propname
, NULL
,
5940 vd
->vdev_guid
, ZPROP_SRC_NONE
);
5942 case VDEV_PROP_ASIZE
:
5943 vdev_prop_add_list(outnvl
, propname
, NULL
,
5944 vd
->vdev_asize
, ZPROP_SRC_NONE
);
5946 case VDEV_PROP_PSIZE
:
5947 vdev_prop_add_list(outnvl
, propname
, NULL
,
5948 vd
->vdev_psize
, ZPROP_SRC_NONE
);
5950 case VDEV_PROP_ASHIFT
:
5951 vdev_prop_add_list(outnvl
, propname
, NULL
,
5952 vd
->vdev_ashift
, ZPROP_SRC_NONE
);
5954 case VDEV_PROP_SIZE
:
5955 vdev_prop_add_list(outnvl
, propname
, NULL
,
5956 vd
->vdev_stat
.vs_dspace
, ZPROP_SRC_NONE
);
5958 case VDEV_PROP_FREE
:
5959 vdev_prop_add_list(outnvl
, propname
, NULL
,
5960 vd
->vdev_stat
.vs_dspace
-
5961 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
5963 case VDEV_PROP_ALLOCATED
:
5964 vdev_prop_add_list(outnvl
, propname
, NULL
,
5965 vd
->vdev_stat
.vs_alloc
, ZPROP_SRC_NONE
);
5967 case VDEV_PROP_EXPANDSZ
:
5968 vdev_prop_add_list(outnvl
, propname
, NULL
,
5969 vd
->vdev_stat
.vs_esize
, ZPROP_SRC_NONE
);
5971 case VDEV_PROP_FRAGMENTATION
:
5972 vdev_prop_add_list(outnvl
, propname
, NULL
,
5973 vd
->vdev_stat
.vs_fragmentation
,
5976 case VDEV_PROP_PARITY
:
5977 vdev_prop_add_list(outnvl
, propname
, NULL
,
5978 vdev_get_nparity(vd
), ZPROP_SRC_NONE
);
5980 case VDEV_PROP_PATH
:
5981 if (vd
->vdev_path
== NULL
)
5983 vdev_prop_add_list(outnvl
, propname
,
5984 vd
->vdev_path
, 0, ZPROP_SRC_NONE
);
5986 case VDEV_PROP_DEVID
:
5987 if (vd
->vdev_devid
== NULL
)
5989 vdev_prop_add_list(outnvl
, propname
,
5990 vd
->vdev_devid
, 0, ZPROP_SRC_NONE
);
5992 case VDEV_PROP_PHYS_PATH
:
5993 if (vd
->vdev_physpath
== NULL
)
5995 vdev_prop_add_list(outnvl
, propname
,
5996 vd
->vdev_physpath
, 0, ZPROP_SRC_NONE
);
5998 case VDEV_PROP_ENC_PATH
:
5999 if (vd
->vdev_enc_sysfs_path
== NULL
)
6001 vdev_prop_add_list(outnvl
, propname
,
6002 vd
->vdev_enc_sysfs_path
, 0, ZPROP_SRC_NONE
);
6005 if (vd
->vdev_fru
== NULL
)
6007 vdev_prop_add_list(outnvl
, propname
,
6008 vd
->vdev_fru
, 0, ZPROP_SRC_NONE
);
6010 case VDEV_PROP_PARENT
:
6011 if (vd
->vdev_parent
!= NULL
) {
6012 strval
= vdev_name(vd
->vdev_parent
,
6013 namebuf
, sizeof (namebuf
));
6014 vdev_prop_add_list(outnvl
, propname
,
6015 strval
, 0, ZPROP_SRC_NONE
);
6018 case VDEV_PROP_CHILDREN
:
6019 if (vd
->vdev_children
> 0)
6020 strval
= kmem_zalloc(ZAP_MAXVALUELEN
,
6022 for (uint64_t i
= 0; i
< vd
->vdev_children
;
6026 vname
= vdev_name(vd
->vdev_child
[i
],
6027 namebuf
, sizeof (namebuf
));
6029 vname
= "(unknown)";
6030 if (strlen(strval
) > 0)
6031 strlcat(strval
, ",",
6033 strlcat(strval
, vname
, ZAP_MAXVALUELEN
);
6035 if (strval
!= NULL
) {
6036 vdev_prop_add_list(outnvl
, propname
,
6037 strval
, 0, ZPROP_SRC_NONE
);
6038 kmem_free(strval
, ZAP_MAXVALUELEN
);
6041 case VDEV_PROP_NUMCHILDREN
:
6042 vdev_prop_add_list(outnvl
, propname
, NULL
,
6043 vd
->vdev_children
, ZPROP_SRC_NONE
);
6045 case VDEV_PROP_READ_ERRORS
:
6046 vdev_prop_add_list(outnvl
, propname
, NULL
,
6047 vd
->vdev_stat
.vs_read_errors
,
6050 case VDEV_PROP_WRITE_ERRORS
:
6051 vdev_prop_add_list(outnvl
, propname
, NULL
,
6052 vd
->vdev_stat
.vs_write_errors
,
6055 case VDEV_PROP_CHECKSUM_ERRORS
:
6056 vdev_prop_add_list(outnvl
, propname
, NULL
,
6057 vd
->vdev_stat
.vs_checksum_errors
,
6060 case VDEV_PROP_INITIALIZE_ERRORS
:
6061 vdev_prop_add_list(outnvl
, propname
, NULL
,
6062 vd
->vdev_stat
.vs_initialize_errors
,
6065 case VDEV_PROP_OPS_NULL
:
6066 vdev_prop_add_list(outnvl
, propname
, NULL
,
6067 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_NULL
],
6070 case VDEV_PROP_OPS_READ
:
6071 vdev_prop_add_list(outnvl
, propname
, NULL
,
6072 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_READ
],
6075 case VDEV_PROP_OPS_WRITE
:
6076 vdev_prop_add_list(outnvl
, propname
, NULL
,
6077 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_WRITE
],
6080 case VDEV_PROP_OPS_FREE
:
6081 vdev_prop_add_list(outnvl
, propname
, NULL
,
6082 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_FREE
],
6085 case VDEV_PROP_OPS_CLAIM
:
6086 vdev_prop_add_list(outnvl
, propname
, NULL
,
6087 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_CLAIM
],
6090 case VDEV_PROP_OPS_TRIM
:
6092 * TRIM ops and bytes are reported to user
6093 * space as ZIO_TYPE_IOCTL. This is done to
6094 * preserve the vdev_stat_t structure layout
6097 vdev_prop_add_list(outnvl
, propname
, NULL
,
6098 vd
->vdev_stat
.vs_ops
[ZIO_TYPE_IOCTL
],
6101 case VDEV_PROP_BYTES_NULL
:
6102 vdev_prop_add_list(outnvl
, propname
, NULL
,
6103 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_NULL
],
6106 case VDEV_PROP_BYTES_READ
:
6107 vdev_prop_add_list(outnvl
, propname
, NULL
,
6108 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_READ
],
6111 case VDEV_PROP_BYTES_WRITE
:
6112 vdev_prop_add_list(outnvl
, propname
, NULL
,
6113 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_WRITE
],
6116 case VDEV_PROP_BYTES_FREE
:
6117 vdev_prop_add_list(outnvl
, propname
, NULL
,
6118 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_FREE
],
6121 case VDEV_PROP_BYTES_CLAIM
:
6122 vdev_prop_add_list(outnvl
, propname
, NULL
,
6123 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_CLAIM
],
6126 case VDEV_PROP_BYTES_TRIM
:
6128 * TRIM ops and bytes are reported to user
6129 * space as ZIO_TYPE_IOCTL. This is done to
6130 * preserve the vdev_stat_t structure layout
6133 vdev_prop_add_list(outnvl
, propname
, NULL
,
6134 vd
->vdev_stat
.vs_bytes
[ZIO_TYPE_IOCTL
],
6137 case VDEV_PROP_REMOVING
:
6138 vdev_prop_add_list(outnvl
, propname
, NULL
,
6139 vd
->vdev_removing
, ZPROP_SRC_NONE
);
6141 /* Numeric Properites */
6142 case VDEV_PROP_ALLOCATING
:
6143 /* Leaf vdevs cannot have this property */
6144 if (vd
->vdev_mg
== NULL
&&
6145 vd
->vdev_top
!= NULL
) {
6146 src
= ZPROP_SRC_NONE
;
6147 intval
= ZPROP_BOOLEAN_NA
;
6149 err
= vdev_prop_get_int(vd
, prop
,
6151 if (err
&& err
!= ENOENT
)
6155 vdev_prop_default_numeric(prop
))
6156 src
= ZPROP_SRC_DEFAULT
;
6158 src
= ZPROP_SRC_LOCAL
;
6161 vdev_prop_add_list(outnvl
, propname
, NULL
,
6164 case VDEV_PROP_FAILFAST
:
6165 src
= ZPROP_SRC_LOCAL
;
6168 err
= zap_lookup(mos
, objid
, nvpair_name(elem
),
6169 sizeof (uint64_t), 1, &intval
);
6170 if (err
== ENOENT
) {
6171 intval
= vdev_prop_default_numeric(
6177 if (intval
== vdev_prop_default_numeric(prop
))
6178 src
= ZPROP_SRC_DEFAULT
;
6180 vdev_prop_add_list(outnvl
, propname
, strval
,
6183 case VDEV_PROP_CHECKSUM_N
:
6184 case VDEV_PROP_CHECKSUM_T
:
6185 case VDEV_PROP_IO_N
:
6186 case VDEV_PROP_IO_T
:
6187 err
= vdev_prop_get_int(vd
, prop
, &intval
);
6188 if (err
&& err
!= ENOENT
)
6191 if (intval
== vdev_prop_default_numeric(prop
))
6192 src
= ZPROP_SRC_DEFAULT
;
6194 src
= ZPROP_SRC_LOCAL
;
6196 vdev_prop_add_list(outnvl
, propname
, NULL
,
6199 /* Text Properties */
6200 case VDEV_PROP_COMMENT
:
6201 /* Exists in the ZAP below */
6203 case VDEV_PROP_USERPROP
:
6204 /* User Properites */
6205 src
= ZPROP_SRC_LOCAL
;
6207 err
= zap_length(mos
, objid
, nvpair_name(elem
),
6208 &integer_size
, &num_integers
);
6212 switch (integer_size
) {
6214 /* User properties cannot be integers */
6218 /* string property */
6219 strval
= kmem_alloc(num_integers
,
6221 err
= zap_lookup(mos
, objid
,
6222 nvpair_name(elem
), 1,
6223 num_integers
, strval
);
6229 vdev_prop_add_list(outnvl
, propname
,
6231 kmem_free(strval
, num_integers
);
6244 * Get all properties from the MOS vdev property object.
6248 for (zap_cursor_init(&zc
, mos
, objid
);
6249 (err
= zap_cursor_retrieve(&zc
, &za
)) == 0;
6250 zap_cursor_advance(&zc
)) {
6253 zprop_source_t src
= ZPROP_SRC_DEFAULT
;
6254 propname
= za
.za_name
;
6256 switch (za
.za_integer_length
) {
6258 /* We do not allow integer user properties */
6259 /* This is likely an internal value */
6262 /* string property */
6263 strval
= kmem_alloc(za
.za_num_integers
,
6265 err
= zap_lookup(mos
, objid
, za
.za_name
, 1,
6266 za
.za_num_integers
, strval
);
6268 kmem_free(strval
, za
.za_num_integers
);
6271 vdev_prop_add_list(outnvl
, propname
, strval
, 0,
6273 kmem_free(strval
, za
.za_num_integers
);
6280 zap_cursor_fini(&zc
);
6283 mutex_exit(&spa
->spa_props_lock
);
6284 if (err
&& err
!= ENOENT
) {
6291 EXPORT_SYMBOL(vdev_fault
);
6292 EXPORT_SYMBOL(vdev_degrade
);
6293 EXPORT_SYMBOL(vdev_online
);
6294 EXPORT_SYMBOL(vdev_offline
);
6295 EXPORT_SYMBOL(vdev_clear
);
6297 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_count
, UINT
, ZMOD_RW
,
6298 "Target number of metaslabs per top-level vdev");
6300 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, default_ms_shift
, UINT
, ZMOD_RW
,
6301 "Default lower limit for metaslab size");
6303 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, max_ms_shift
, UINT
, ZMOD_RW
,
6304 "Default upper limit for metaslab size");
6306 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, min_ms_count
, UINT
, ZMOD_RW
,
6307 "Minimum number of metaslabs per top-level vdev");
6309 ZFS_MODULE_PARAM(zfs_vdev
, zfs_vdev_
, ms_count_limit
, UINT
, ZMOD_RW
,
6310 "Practical upper limit of total metaslabs per top-level vdev");
6312 ZFS_MODULE_PARAM(zfs
, zfs_
, slow_io_events_per_second
, UINT
, ZMOD_RW
,
6313 "Rate limit slow IO (delay) events to this many per second");
6316 ZFS_MODULE_PARAM(zfs
, zfs_
, checksum_events_per_second
, UINT
, ZMOD_RW
,
6317 "Rate limit checksum events to this many checksum errors per second "
6318 "(do not set below ZED threshold).");
6321 ZFS_MODULE_PARAM(zfs
, zfs_
, scan_ignore_errors
, INT
, ZMOD_RW
,
6322 "Ignore errors during resilver/scrub");
6324 ZFS_MODULE_PARAM(zfs_vdev
, vdev_
, validate_skip
, INT
, ZMOD_RW
,
6325 "Bypass vdev_validate()");
6327 ZFS_MODULE_PARAM(zfs
, zfs_
, nocacheflush
, INT
, ZMOD_RW
,
6328 "Disable cache flushes");
6330 ZFS_MODULE_PARAM(zfs
, zfs_
, embedded_slog_min_ms
, UINT
, ZMOD_RW
,
6331 "Minimum number of metaslabs required to dedicate one for log blocks");
6334 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, min_auto_ashift
,
6335 param_set_min_auto_ashift
, param_get_uint
, ZMOD_RW
,
6336 "Minimum ashift used when creating new top-level vdevs");
6338 ZFS_MODULE_PARAM_CALL(zfs_vdev
, zfs_vdev_
, max_auto_ashift
,
6339 param_set_max_auto_ashift
, param_get_uint
, ZMOD_RW
,
6340 "Maximum ashift used when optimizing for logical -> physical sector "
6341 "size on new top-level vdevs");