4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2018 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.
32 #include <sys/zfs_context.h>
33 #include <sys/fm/fs/zfs.h>
35 #include <sys/spa_impl.h>
36 #include <sys/bpobj.h>
38 #include <sys/dmu_tx.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/vdev_impl.h>
41 #include <sys/uberblock_impl.h>
42 #include <sys/metaslab.h>
43 #include <sys/metaslab_impl.h>
44 #include <sys/space_map.h>
45 #include <sys/space_reftree.h>
48 #include <sys/fs/zfs.h>
51 #include <sys/dsl_scan.h>
54 #include <sys/zfs_ratelimit.h>
56 /* target number of metaslabs per top-level vdev */
57 int vdev_max_ms_count
= 200;
59 /* minimum number of metaslabs per top-level vdev */
60 int vdev_min_ms_count
= 16;
62 /* practical upper limit of total metaslabs per top-level vdev */
63 int vdev_ms_count_limit
= 1ULL << 17;
65 /* lower limit for metaslab size (512M) */
66 int vdev_default_ms_shift
= 29;
68 /* upper limit for metaslab size (256G) */
69 int vdev_max_ms_shift
= 38;
71 int vdev_validate_skip
= B_FALSE
;
74 * Since the DTL space map of a vdev is not expected to have a lot of
75 * entries, we default its block size to 4K.
77 int vdev_dtl_sm_blksz
= (1 << 12);
80 * Rate limit slow IO (delay) events to this many per second.
82 unsigned int zfs_slow_io_events_per_second
= 20;
85 * Rate limit checksum events after this many checksum errors per second.
87 unsigned int zfs_checksum_events_per_second
= 20;
90 * Ignore errors during scrub/resilver. Allows to work around resilver
91 * upon import when there are pool errors.
93 int zfs_scan_ignore_errors
= 0;
96 * vdev-wide space maps that have lots of entries written to them at
97 * the end of each transaction can benefit from a higher I/O bandwidth
98 * (e.g. vdev_obsolete_sm), thus we default their block size to 128K.
100 int vdev_standard_sm_blksz
= (1 << 17);
104 vdev_dbgmsg(vdev_t
*vd
, const char *fmt
, ...)
110 (void) vsnprintf(buf
, sizeof (buf
), fmt
, adx
);
113 if (vd
->vdev_path
!= NULL
) {
114 zfs_dbgmsg("%s vdev '%s': %s", vd
->vdev_ops
->vdev_op_type
,
117 zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
118 vd
->vdev_ops
->vdev_op_type
,
119 (u_longlong_t
)vd
->vdev_id
,
120 (u_longlong_t
)vd
->vdev_guid
, buf
);
125 vdev_dbgmsg_print_tree(vdev_t
*vd
, int indent
)
129 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
) {
130 zfs_dbgmsg("%*svdev %u: %s", indent
, "", vd
->vdev_id
,
131 vd
->vdev_ops
->vdev_op_type
);
135 switch (vd
->vdev_state
) {
136 case VDEV_STATE_UNKNOWN
:
137 (void) snprintf(state
, sizeof (state
), "unknown");
139 case VDEV_STATE_CLOSED
:
140 (void) snprintf(state
, sizeof (state
), "closed");
142 case VDEV_STATE_OFFLINE
:
143 (void) snprintf(state
, sizeof (state
), "offline");
145 case VDEV_STATE_REMOVED
:
146 (void) snprintf(state
, sizeof (state
), "removed");
148 case VDEV_STATE_CANT_OPEN
:
149 (void) snprintf(state
, sizeof (state
), "can't open");
151 case VDEV_STATE_FAULTED
:
152 (void) snprintf(state
, sizeof (state
), "faulted");
154 case VDEV_STATE_DEGRADED
:
155 (void) snprintf(state
, sizeof (state
), "degraded");
157 case VDEV_STATE_HEALTHY
:
158 (void) snprintf(state
, sizeof (state
), "healthy");
161 (void) snprintf(state
, sizeof (state
), "<state %u>",
162 (uint_t
)vd
->vdev_state
);
165 zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent
,
166 "", (int)vd
->vdev_id
, vd
->vdev_ops
->vdev_op_type
,
167 vd
->vdev_islog
? " (log)" : "",
168 (u_longlong_t
)vd
->vdev_guid
,
169 vd
->vdev_path
? vd
->vdev_path
: "N/A", state
);
171 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
172 vdev_dbgmsg_print_tree(vd
->vdev_child
[i
], indent
+ 2);
176 * Virtual device management.
179 static vdev_ops_t
*vdev_ops_table
[] = {
194 * Given a vdev type, return the appropriate ops vector.
197 vdev_getops(const char *type
)
199 vdev_ops_t
*ops
, **opspp
;
201 for (opspp
= vdev_ops_table
; (ops
= *opspp
) != NULL
; opspp
++)
202 if (strcmp(ops
->vdev_op_type
, type
) == 0)
209 * Derive the enumerated alloction bias from string input.
210 * String origin is either the per-vdev zap or zpool(1M).
212 static vdev_alloc_bias_t
213 vdev_derive_alloc_bias(const char *bias
)
215 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
217 if (strcmp(bias
, VDEV_ALLOC_BIAS_LOG
) == 0)
218 alloc_bias
= VDEV_BIAS_LOG
;
219 else if (strcmp(bias
, VDEV_ALLOC_BIAS_SPECIAL
) == 0)
220 alloc_bias
= VDEV_BIAS_SPECIAL
;
221 else if (strcmp(bias
, VDEV_ALLOC_BIAS_DEDUP
) == 0)
222 alloc_bias
= VDEV_BIAS_DEDUP
;
228 * Default asize function: return the MAX of psize with the asize of
229 * all children. This is what's used by anything other than RAID-Z.
232 vdev_default_asize(vdev_t
*vd
, uint64_t psize
)
234 uint64_t asize
= P2ROUNDUP(psize
, 1ULL << vd
->vdev_top
->vdev_ashift
);
237 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
238 csize
= vdev_psize_to_asize(vd
->vdev_child
[c
], psize
);
239 asize
= MAX(asize
, csize
);
246 * Get the minimum allocatable size. We define the allocatable size as
247 * the vdev's asize rounded to the nearest metaslab. This allows us to
248 * replace or attach devices which don't have the same physical size but
249 * can still satisfy the same number of allocations.
252 vdev_get_min_asize(vdev_t
*vd
)
254 vdev_t
*pvd
= vd
->vdev_parent
;
257 * If our parent is NULL (inactive spare or cache) or is the root,
258 * just return our own asize.
261 return (vd
->vdev_asize
);
264 * The top-level vdev just returns the allocatable size rounded
265 * to the nearest metaslab.
267 if (vd
== vd
->vdev_top
)
268 return (P2ALIGN(vd
->vdev_asize
, 1ULL << vd
->vdev_ms_shift
));
271 * The allocatable space for a raidz vdev is N * sizeof(smallest child),
272 * so each child must provide at least 1/Nth of its asize.
274 if (pvd
->vdev_ops
== &vdev_raidz_ops
)
275 return ((pvd
->vdev_min_asize
+ pvd
->vdev_children
- 1) /
278 return (pvd
->vdev_min_asize
);
282 vdev_set_min_asize(vdev_t
*vd
)
284 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
286 for (int c
= 0; c
< vd
->vdev_children
; c
++)
287 vdev_set_min_asize(vd
->vdev_child
[c
]);
291 vdev_lookup_top(spa_t
*spa
, uint64_t vdev
)
293 vdev_t
*rvd
= spa
->spa_root_vdev
;
295 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
297 if (vdev
< rvd
->vdev_children
) {
298 ASSERT(rvd
->vdev_child
[vdev
] != NULL
);
299 return (rvd
->vdev_child
[vdev
]);
306 vdev_lookup_by_guid(vdev_t
*vd
, uint64_t guid
)
310 if (vd
->vdev_guid
== guid
)
313 for (int c
= 0; c
< vd
->vdev_children
; c
++)
314 if ((mvd
= vdev_lookup_by_guid(vd
->vdev_child
[c
], guid
)) !=
322 vdev_count_leaves_impl(vdev_t
*vd
)
326 if (vd
->vdev_ops
->vdev_op_leaf
)
329 for (int c
= 0; c
< vd
->vdev_children
; c
++)
330 n
+= vdev_count_leaves_impl(vd
->vdev_child
[c
]);
336 vdev_count_leaves(spa_t
*spa
)
340 spa_config_enter(spa
, SCL_VDEV
, FTAG
, RW_READER
);
341 rc
= vdev_count_leaves_impl(spa
->spa_root_vdev
);
342 spa_config_exit(spa
, SCL_VDEV
, FTAG
);
348 vdev_add_child(vdev_t
*pvd
, vdev_t
*cvd
)
350 size_t oldsize
, newsize
;
351 uint64_t id
= cvd
->vdev_id
;
354 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
355 ASSERT(cvd
->vdev_parent
== NULL
);
357 cvd
->vdev_parent
= pvd
;
362 ASSERT(id
>= pvd
->vdev_children
|| pvd
->vdev_child
[id
] == NULL
);
364 oldsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
365 pvd
->vdev_children
= MAX(pvd
->vdev_children
, id
+ 1);
366 newsize
= pvd
->vdev_children
* sizeof (vdev_t
*);
368 newchild
= kmem_alloc(newsize
, KM_SLEEP
);
369 if (pvd
->vdev_child
!= NULL
) {
370 bcopy(pvd
->vdev_child
, newchild
, oldsize
);
371 kmem_free(pvd
->vdev_child
, oldsize
);
374 pvd
->vdev_child
= newchild
;
375 pvd
->vdev_child
[id
] = cvd
;
377 cvd
->vdev_top
= (pvd
->vdev_top
? pvd
->vdev_top
: cvd
);
378 ASSERT(cvd
->vdev_top
->vdev_parent
->vdev_parent
== NULL
);
381 * Walk up all ancestors to update guid sum.
383 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
384 pvd
->vdev_guid_sum
+= cvd
->vdev_guid_sum
;
388 vdev_remove_child(vdev_t
*pvd
, vdev_t
*cvd
)
391 uint_t id
= cvd
->vdev_id
;
393 ASSERT(cvd
->vdev_parent
== pvd
);
398 ASSERT(id
< pvd
->vdev_children
);
399 ASSERT(pvd
->vdev_child
[id
] == cvd
);
401 pvd
->vdev_child
[id
] = NULL
;
402 cvd
->vdev_parent
= NULL
;
404 for (c
= 0; c
< pvd
->vdev_children
; c
++)
405 if (pvd
->vdev_child
[c
])
408 if (c
== pvd
->vdev_children
) {
409 kmem_free(pvd
->vdev_child
, c
* sizeof (vdev_t
*));
410 pvd
->vdev_child
= NULL
;
411 pvd
->vdev_children
= 0;
415 * Walk up all ancestors to update guid sum.
417 for (; pvd
!= NULL
; pvd
= pvd
->vdev_parent
)
418 pvd
->vdev_guid_sum
-= cvd
->vdev_guid_sum
;
422 * Remove any holes in the child array.
425 vdev_compact_children(vdev_t
*pvd
)
427 vdev_t
**newchild
, *cvd
;
428 int oldc
= pvd
->vdev_children
;
431 ASSERT(spa_config_held(pvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
436 for (int c
= newc
= 0; c
< oldc
; c
++)
437 if (pvd
->vdev_child
[c
])
441 newchild
= kmem_zalloc(newc
* sizeof (vdev_t
*), KM_SLEEP
);
443 for (int c
= newc
= 0; c
< oldc
; c
++) {
444 if ((cvd
= pvd
->vdev_child
[c
]) != NULL
) {
445 newchild
[newc
] = cvd
;
446 cvd
->vdev_id
= newc
++;
453 kmem_free(pvd
->vdev_child
, oldc
* sizeof (vdev_t
*));
454 pvd
->vdev_child
= newchild
;
455 pvd
->vdev_children
= newc
;
459 * Allocate and minimally initialize a vdev_t.
462 vdev_alloc_common(spa_t
*spa
, uint_t id
, uint64_t guid
, vdev_ops_t
*ops
)
465 vdev_indirect_config_t
*vic
;
467 vd
= kmem_zalloc(sizeof (vdev_t
), KM_SLEEP
);
468 vic
= &vd
->vdev_indirect_config
;
470 if (spa
->spa_root_vdev
== NULL
) {
471 ASSERT(ops
== &vdev_root_ops
);
472 spa
->spa_root_vdev
= vd
;
473 spa
->spa_load_guid
= spa_generate_guid(NULL
);
476 if (guid
== 0 && ops
!= &vdev_hole_ops
) {
477 if (spa
->spa_root_vdev
== vd
) {
479 * The root vdev's guid will also be the pool guid,
480 * which must be unique among all pools.
482 guid
= spa_generate_guid(NULL
);
485 * Any other vdev's guid must be unique within the pool.
487 guid
= spa_generate_guid(spa
);
489 ASSERT(!spa_guid_exists(spa_guid(spa
), guid
));
494 vd
->vdev_guid
= guid
;
495 vd
->vdev_guid_sum
= guid
;
497 vd
->vdev_state
= VDEV_STATE_CLOSED
;
498 vd
->vdev_ishole
= (ops
== &vdev_hole_ops
);
499 vic
->vic_prev_indirect_vdev
= UINT64_MAX
;
501 rw_init(&vd
->vdev_indirect_rwlock
, NULL
, RW_DEFAULT
, NULL
);
502 mutex_init(&vd
->vdev_obsolete_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
503 vd
->vdev_obsolete_segments
= range_tree_create(NULL
, NULL
);
506 * Initialize rate limit structs for events. We rate limit ZIO delay
507 * and checksum events so that we don't overwhelm ZED with thousands
508 * of events when a disk is acting up.
510 zfs_ratelimit_init(&vd
->vdev_delay_rl
, &zfs_slow_io_events_per_second
,
512 zfs_ratelimit_init(&vd
->vdev_checksum_rl
,
513 &zfs_checksum_events_per_second
, 1);
515 list_link_init(&vd
->vdev_config_dirty_node
);
516 list_link_init(&vd
->vdev_state_dirty_node
);
517 mutex_init(&vd
->vdev_dtl_lock
, NULL
, MUTEX_NOLOCKDEP
, NULL
);
518 mutex_init(&vd
->vdev_stat_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
519 mutex_init(&vd
->vdev_probe_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
520 mutex_init(&vd
->vdev_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
521 mutex_init(&vd
->vdev_scan_io_queue_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
523 for (int t
= 0; t
< DTL_TYPES
; t
++) {
524 vd
->vdev_dtl
[t
] = range_tree_create(NULL
, NULL
);
526 txg_list_create(&vd
->vdev_ms_list
, spa
,
527 offsetof(struct metaslab
, ms_txg_node
));
528 txg_list_create(&vd
->vdev_dtl_list
, spa
,
529 offsetof(struct vdev
, vdev_dtl_node
));
530 vd
->vdev_stat
.vs_timestamp
= gethrtime();
538 * Allocate a new vdev. The 'alloctype' is used to control whether we are
539 * creating a new vdev or loading an existing one - the behavior is slightly
540 * different for each case.
543 vdev_alloc(spa_t
*spa
, vdev_t
**vdp
, nvlist_t
*nv
, vdev_t
*parent
, uint_t id
,
548 uint64_t guid
= 0, islog
, nparity
;
550 vdev_indirect_config_t
*vic
;
553 vdev_alloc_bias_t alloc_bias
= VDEV_BIAS_NONE
;
554 boolean_t top_level
= (parent
&& !parent
->vdev_parent
);
556 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
558 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_TYPE
, &type
) != 0)
559 return (SET_ERROR(EINVAL
));
561 if ((ops
= vdev_getops(type
)) == NULL
)
562 return (SET_ERROR(EINVAL
));
565 * If this is a load, get the vdev guid from the nvlist.
566 * Otherwise, vdev_alloc_common() will generate one for us.
568 if (alloctype
== VDEV_ALLOC_LOAD
) {
571 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ID
, &label_id
) ||
573 return (SET_ERROR(EINVAL
));
575 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
576 return (SET_ERROR(EINVAL
));
577 } else if (alloctype
== VDEV_ALLOC_SPARE
) {
578 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
579 return (SET_ERROR(EINVAL
));
580 } else if (alloctype
== VDEV_ALLOC_L2CACHE
) {
581 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
582 return (SET_ERROR(EINVAL
));
583 } else if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
584 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_GUID
, &guid
) != 0)
585 return (SET_ERROR(EINVAL
));
589 * The first allocated vdev must be of type 'root'.
591 if (ops
!= &vdev_root_ops
&& spa
->spa_root_vdev
== NULL
)
592 return (SET_ERROR(EINVAL
));
595 * Determine whether we're a log vdev.
598 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_LOG
, &islog
);
599 if (islog
&& spa_version(spa
) < SPA_VERSION_SLOGS
)
600 return (SET_ERROR(ENOTSUP
));
602 if (ops
== &vdev_hole_ops
&& spa_version(spa
) < SPA_VERSION_HOLES
)
603 return (SET_ERROR(ENOTSUP
));
606 * Set the nparity property for RAID-Z vdevs.
609 if (ops
== &vdev_raidz_ops
) {
610 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NPARITY
,
612 if (nparity
== 0 || nparity
> VDEV_RAIDZ_MAXPARITY
)
613 return (SET_ERROR(EINVAL
));
615 * Previous versions could only support 1 or 2 parity
619 spa_version(spa
) < SPA_VERSION_RAIDZ2
)
620 return (SET_ERROR(ENOTSUP
));
622 spa_version(spa
) < SPA_VERSION_RAIDZ3
)
623 return (SET_ERROR(ENOTSUP
));
626 * We require the parity to be specified for SPAs that
627 * support multiple parity levels.
629 if (spa_version(spa
) >= SPA_VERSION_RAIDZ2
)
630 return (SET_ERROR(EINVAL
));
632 * Otherwise, we default to 1 parity device for RAID-Z.
639 ASSERT(nparity
!= -1ULL);
642 * If creating a top-level vdev, check for allocation classes input
644 if (top_level
&& alloctype
== VDEV_ALLOC_ADD
) {
647 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_ALLOCATION_BIAS
,
649 alloc_bias
= vdev_derive_alloc_bias(bias
);
651 /* spa_vdev_add() expects feature to be enabled */
652 if (spa
->spa_load_state
!= SPA_LOAD_CREATE
&&
653 !spa_feature_is_enabled(spa
,
654 SPA_FEATURE_ALLOCATION_CLASSES
)) {
655 return (SET_ERROR(ENOTSUP
));
660 vd
= vdev_alloc_common(spa
, id
, guid
, ops
);
661 vic
= &vd
->vdev_indirect_config
;
663 vd
->vdev_islog
= islog
;
664 vd
->vdev_nparity
= nparity
;
665 if (top_level
&& alloc_bias
!= VDEV_BIAS_NONE
)
666 vd
->vdev_alloc_bias
= alloc_bias
;
668 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PATH
, &vd
->vdev_path
) == 0)
669 vd
->vdev_path
= spa_strdup(vd
->vdev_path
);
672 * ZPOOL_CONFIG_AUX_STATE = "external" means we previously forced a
673 * fault on a vdev and want it to persist across imports (like with
676 rc
= nvlist_lookup_string(nv
, ZPOOL_CONFIG_AUX_STATE
, &tmp
);
677 if (rc
== 0 && tmp
!= NULL
&& strcmp(tmp
, "external") == 0) {
678 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
679 vd
->vdev_faulted
= 1;
680 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
683 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_DEVID
, &vd
->vdev_devid
) == 0)
684 vd
->vdev_devid
= spa_strdup(vd
->vdev_devid
);
685 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_PHYS_PATH
,
686 &vd
->vdev_physpath
) == 0)
687 vd
->vdev_physpath
= spa_strdup(vd
->vdev_physpath
);
689 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_VDEV_ENC_SYSFS_PATH
,
690 &vd
->vdev_enc_sysfs_path
) == 0)
691 vd
->vdev_enc_sysfs_path
= spa_strdup(vd
->vdev_enc_sysfs_path
);
693 if (nvlist_lookup_string(nv
, ZPOOL_CONFIG_FRU
, &vd
->vdev_fru
) == 0)
694 vd
->vdev_fru
= spa_strdup(vd
->vdev_fru
);
697 * Set the whole_disk property. If it's not specified, leave the value
700 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_WHOLE_DISK
,
701 &vd
->vdev_wholedisk
) != 0)
702 vd
->vdev_wholedisk
= -1ULL;
704 ASSERT0(vic
->vic_mapping_object
);
705 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_OBJECT
,
706 &vic
->vic_mapping_object
);
707 ASSERT0(vic
->vic_births_object
);
708 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_INDIRECT_BIRTHS
,
709 &vic
->vic_births_object
);
710 ASSERT3U(vic
->vic_prev_indirect_vdev
, ==, UINT64_MAX
);
711 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_PREV_INDIRECT_VDEV
,
712 &vic
->vic_prev_indirect_vdev
);
715 * Look for the 'not present' flag. This will only be set if the device
716 * was not present at the time of import.
718 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_NOT_PRESENT
,
719 &vd
->vdev_not_present
);
722 * Get the alignment requirement.
724 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASHIFT
, &vd
->vdev_ashift
);
727 * Retrieve the vdev creation time.
729 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_CREATE_TXG
,
733 * If we're a top-level vdev, try to load the allocation parameters.
736 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
737 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_ARRAY
,
739 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_METASLAB_SHIFT
,
741 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_ASIZE
,
743 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVING
,
745 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_VDEV_TOP_ZAP
,
748 ASSERT0(vd
->vdev_top_zap
);
751 if (top_level
&& alloctype
!= VDEV_ALLOC_ATTACH
) {
752 ASSERT(alloctype
== VDEV_ALLOC_LOAD
||
753 alloctype
== VDEV_ALLOC_ADD
||
754 alloctype
== VDEV_ALLOC_SPLIT
||
755 alloctype
== VDEV_ALLOC_ROOTPOOL
);
756 /* Note: metaslab_group_create() is now deferred */
759 if (vd
->vdev_ops
->vdev_op_leaf
&&
760 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_SPLIT
)) {
761 (void) nvlist_lookup_uint64(nv
,
762 ZPOOL_CONFIG_VDEV_LEAF_ZAP
, &vd
->vdev_leaf_zap
);
764 ASSERT0(vd
->vdev_leaf_zap
);
768 * If we're a leaf vdev, try to load the DTL object and other state.
771 if (vd
->vdev_ops
->vdev_op_leaf
&&
772 (alloctype
== VDEV_ALLOC_LOAD
|| alloctype
== VDEV_ALLOC_L2CACHE
||
773 alloctype
== VDEV_ALLOC_ROOTPOOL
)) {
774 if (alloctype
== VDEV_ALLOC_LOAD
) {
775 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DTL
,
776 &vd
->vdev_dtl_object
);
777 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_UNSPARE
,
781 if (alloctype
== VDEV_ALLOC_ROOTPOOL
) {
784 if (nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_IS_SPARE
,
785 &spare
) == 0 && spare
)
789 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_OFFLINE
,
792 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_RESILVER_TXG
,
793 &vd
->vdev_resilver_txg
);
795 if (nvlist_exists(nv
, ZPOOL_CONFIG_RESILVER_DEFER
))
796 vdev_set_deferred_resilver(spa
, vd
);
799 * In general, when importing a pool we want to ignore the
800 * persistent fault state, as the diagnosis made on another
801 * system may not be valid in the current context. The only
802 * exception is if we forced a vdev to a persistently faulted
803 * state with 'zpool offline -f'. The persistent fault will
804 * remain across imports until cleared.
806 * Local vdevs will remain in the faulted state.
808 if (spa_load_state(spa
) == SPA_LOAD_OPEN
||
809 spa_load_state(spa
) == SPA_LOAD_IMPORT
) {
810 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_FAULTED
,
812 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_DEGRADED
,
814 (void) nvlist_lookup_uint64(nv
, ZPOOL_CONFIG_REMOVED
,
817 if (vd
->vdev_faulted
|| vd
->vdev_degraded
) {
821 VDEV_AUX_ERR_EXCEEDED
;
822 if (nvlist_lookup_string(nv
,
823 ZPOOL_CONFIG_AUX_STATE
, &aux
) == 0 &&
824 strcmp(aux
, "external") == 0)
825 vd
->vdev_label_aux
= VDEV_AUX_EXTERNAL
;
827 vd
->vdev_faulted
= 0ULL;
833 * Add ourselves to the parent's list of children.
835 vdev_add_child(parent
, vd
);
843 vdev_free(vdev_t
*vd
)
845 spa_t
*spa
= vd
->vdev_spa
;
848 * Scan queues are normally destroyed at the end of a scan. If the
849 * queue exists here, that implies the vdev is being removed while
850 * the scan is still running.
852 if (vd
->vdev_scan_io_queue
!= NULL
) {
853 mutex_enter(&vd
->vdev_scan_io_queue_lock
);
854 dsl_scan_io_queue_destroy(vd
->vdev_scan_io_queue
);
855 vd
->vdev_scan_io_queue
= NULL
;
856 mutex_exit(&vd
->vdev_scan_io_queue_lock
);
860 * vdev_free() implies closing the vdev first. This is simpler than
861 * trying to ensure complicated semantics for all callers.
865 ASSERT(!list_link_active(&vd
->vdev_config_dirty_node
));
866 ASSERT(!list_link_active(&vd
->vdev_state_dirty_node
));
871 for (int c
= 0; c
< vd
->vdev_children
; c
++)
872 vdev_free(vd
->vdev_child
[c
]);
874 ASSERT(vd
->vdev_child
== NULL
);
875 ASSERT(vd
->vdev_guid_sum
== vd
->vdev_guid
);
878 * Discard allocation state.
880 if (vd
->vdev_mg
!= NULL
) {
881 vdev_metaslab_fini(vd
);
882 metaslab_group_destroy(vd
->vdev_mg
);
885 ASSERT0(vd
->vdev_stat
.vs_space
);
886 ASSERT0(vd
->vdev_stat
.vs_dspace
);
887 ASSERT0(vd
->vdev_stat
.vs_alloc
);
890 * Remove this vdev from its parent's child list.
892 vdev_remove_child(vd
->vdev_parent
, vd
);
894 ASSERT(vd
->vdev_parent
== NULL
);
897 * Clean up vdev structure.
903 spa_strfree(vd
->vdev_path
);
905 spa_strfree(vd
->vdev_devid
);
906 if (vd
->vdev_physpath
)
907 spa_strfree(vd
->vdev_physpath
);
909 if (vd
->vdev_enc_sysfs_path
)
910 spa_strfree(vd
->vdev_enc_sysfs_path
);
913 spa_strfree(vd
->vdev_fru
);
915 if (vd
->vdev_isspare
)
916 spa_spare_remove(vd
);
917 if (vd
->vdev_isl2cache
)
918 spa_l2cache_remove(vd
);
920 txg_list_destroy(&vd
->vdev_ms_list
);
921 txg_list_destroy(&vd
->vdev_dtl_list
);
923 mutex_enter(&vd
->vdev_dtl_lock
);
924 space_map_close(vd
->vdev_dtl_sm
);
925 for (int t
= 0; t
< DTL_TYPES
; t
++) {
926 range_tree_vacate(vd
->vdev_dtl
[t
], NULL
, NULL
);
927 range_tree_destroy(vd
->vdev_dtl
[t
]);
929 mutex_exit(&vd
->vdev_dtl_lock
);
931 EQUIV(vd
->vdev_indirect_births
!= NULL
,
932 vd
->vdev_indirect_mapping
!= NULL
);
933 if (vd
->vdev_indirect_births
!= NULL
) {
934 vdev_indirect_mapping_close(vd
->vdev_indirect_mapping
);
935 vdev_indirect_births_close(vd
->vdev_indirect_births
);
938 if (vd
->vdev_obsolete_sm
!= NULL
) {
939 ASSERT(vd
->vdev_removing
||
940 vd
->vdev_ops
== &vdev_indirect_ops
);
941 space_map_close(vd
->vdev_obsolete_sm
);
942 vd
->vdev_obsolete_sm
= NULL
;
944 range_tree_destroy(vd
->vdev_obsolete_segments
);
945 rw_destroy(&vd
->vdev_indirect_rwlock
);
946 mutex_destroy(&vd
->vdev_obsolete_lock
);
948 mutex_destroy(&vd
->vdev_queue_lock
);
949 mutex_destroy(&vd
->vdev_dtl_lock
);
950 mutex_destroy(&vd
->vdev_stat_lock
);
951 mutex_destroy(&vd
->vdev_probe_lock
);
952 mutex_destroy(&vd
->vdev_scan_io_queue_lock
);
954 zfs_ratelimit_fini(&vd
->vdev_delay_rl
);
955 zfs_ratelimit_fini(&vd
->vdev_checksum_rl
);
957 if (vd
== spa
->spa_root_vdev
)
958 spa
->spa_root_vdev
= NULL
;
960 kmem_free(vd
, sizeof (vdev_t
));
964 * Transfer top-level vdev state from svd to tvd.
967 vdev_top_transfer(vdev_t
*svd
, vdev_t
*tvd
)
969 spa_t
*spa
= svd
->vdev_spa
;
974 ASSERT(tvd
== tvd
->vdev_top
);
976 tvd
->vdev_pending_fastwrite
= svd
->vdev_pending_fastwrite
;
977 tvd
->vdev_ms_array
= svd
->vdev_ms_array
;
978 tvd
->vdev_ms_shift
= svd
->vdev_ms_shift
;
979 tvd
->vdev_ms_count
= svd
->vdev_ms_count
;
980 tvd
->vdev_top_zap
= svd
->vdev_top_zap
;
982 svd
->vdev_ms_array
= 0;
983 svd
->vdev_ms_shift
= 0;
984 svd
->vdev_ms_count
= 0;
985 svd
->vdev_top_zap
= 0;
988 ASSERT3P(tvd
->vdev_mg
, ==, svd
->vdev_mg
);
989 tvd
->vdev_mg
= svd
->vdev_mg
;
990 tvd
->vdev_ms
= svd
->vdev_ms
;
995 if (tvd
->vdev_mg
!= NULL
)
996 tvd
->vdev_mg
->mg_vd
= tvd
;
998 tvd
->vdev_checkpoint_sm
= svd
->vdev_checkpoint_sm
;
999 svd
->vdev_checkpoint_sm
= NULL
;
1001 tvd
->vdev_alloc_bias
= svd
->vdev_alloc_bias
;
1002 svd
->vdev_alloc_bias
= VDEV_BIAS_NONE
;
1004 tvd
->vdev_stat
.vs_alloc
= svd
->vdev_stat
.vs_alloc
;
1005 tvd
->vdev_stat
.vs_space
= svd
->vdev_stat
.vs_space
;
1006 tvd
->vdev_stat
.vs_dspace
= svd
->vdev_stat
.vs_dspace
;
1008 svd
->vdev_stat
.vs_alloc
= 0;
1009 svd
->vdev_stat
.vs_space
= 0;
1010 svd
->vdev_stat
.vs_dspace
= 0;
1013 * State which may be set on a top-level vdev that's in the
1014 * process of being removed.
1016 ASSERT0(tvd
->vdev_indirect_config
.vic_births_object
);
1017 ASSERT0(tvd
->vdev_indirect_config
.vic_mapping_object
);
1018 ASSERT3U(tvd
->vdev_indirect_config
.vic_prev_indirect_vdev
, ==, -1ULL);
1019 ASSERT3P(tvd
->vdev_indirect_mapping
, ==, NULL
);
1020 ASSERT3P(tvd
->vdev_indirect_births
, ==, NULL
);
1021 ASSERT3P(tvd
->vdev_obsolete_sm
, ==, NULL
);
1022 ASSERT0(tvd
->vdev_removing
);
1023 tvd
->vdev_removing
= svd
->vdev_removing
;
1024 tvd
->vdev_indirect_config
= svd
->vdev_indirect_config
;
1025 tvd
->vdev_indirect_mapping
= svd
->vdev_indirect_mapping
;
1026 tvd
->vdev_indirect_births
= svd
->vdev_indirect_births
;
1027 range_tree_swap(&svd
->vdev_obsolete_segments
,
1028 &tvd
->vdev_obsolete_segments
);
1029 tvd
->vdev_obsolete_sm
= svd
->vdev_obsolete_sm
;
1030 svd
->vdev_indirect_config
.vic_mapping_object
= 0;
1031 svd
->vdev_indirect_config
.vic_births_object
= 0;
1032 svd
->vdev_indirect_config
.vic_prev_indirect_vdev
= -1ULL;
1033 svd
->vdev_indirect_mapping
= NULL
;
1034 svd
->vdev_indirect_births
= NULL
;
1035 svd
->vdev_obsolete_sm
= NULL
;
1036 svd
->vdev_removing
= 0;
1038 for (t
= 0; t
< TXG_SIZE
; t
++) {
1039 while ((msp
= txg_list_remove(&svd
->vdev_ms_list
, t
)) != NULL
)
1040 (void) txg_list_add(&tvd
->vdev_ms_list
, msp
, t
);
1041 while ((vd
= txg_list_remove(&svd
->vdev_dtl_list
, t
)) != NULL
)
1042 (void) txg_list_add(&tvd
->vdev_dtl_list
, vd
, t
);
1043 if (txg_list_remove_this(&spa
->spa_vdev_txg_list
, svd
, t
))
1044 (void) txg_list_add(&spa
->spa_vdev_txg_list
, tvd
, t
);
1047 if (list_link_active(&svd
->vdev_config_dirty_node
)) {
1048 vdev_config_clean(svd
);
1049 vdev_config_dirty(tvd
);
1052 if (list_link_active(&svd
->vdev_state_dirty_node
)) {
1053 vdev_state_clean(svd
);
1054 vdev_state_dirty(tvd
);
1057 tvd
->vdev_deflate_ratio
= svd
->vdev_deflate_ratio
;
1058 svd
->vdev_deflate_ratio
= 0;
1060 tvd
->vdev_islog
= svd
->vdev_islog
;
1061 svd
->vdev_islog
= 0;
1063 dsl_scan_io_queue_vdev_xfer(svd
, tvd
);
1067 vdev_top_update(vdev_t
*tvd
, vdev_t
*vd
)
1074 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1075 vdev_top_update(tvd
, vd
->vdev_child
[c
]);
1079 * Add a mirror/replacing vdev above an existing vdev.
1082 vdev_add_parent(vdev_t
*cvd
, vdev_ops_t
*ops
)
1084 spa_t
*spa
= cvd
->vdev_spa
;
1085 vdev_t
*pvd
= cvd
->vdev_parent
;
1088 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1090 mvd
= vdev_alloc_common(spa
, cvd
->vdev_id
, 0, ops
);
1092 mvd
->vdev_asize
= cvd
->vdev_asize
;
1093 mvd
->vdev_min_asize
= cvd
->vdev_min_asize
;
1094 mvd
->vdev_max_asize
= cvd
->vdev_max_asize
;
1095 mvd
->vdev_psize
= cvd
->vdev_psize
;
1096 mvd
->vdev_ashift
= cvd
->vdev_ashift
;
1097 mvd
->vdev_state
= cvd
->vdev_state
;
1098 mvd
->vdev_crtxg
= cvd
->vdev_crtxg
;
1100 vdev_remove_child(pvd
, cvd
);
1101 vdev_add_child(pvd
, mvd
);
1102 cvd
->vdev_id
= mvd
->vdev_children
;
1103 vdev_add_child(mvd
, cvd
);
1104 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1106 if (mvd
== mvd
->vdev_top
)
1107 vdev_top_transfer(cvd
, mvd
);
1113 * Remove a 1-way mirror/replacing vdev from the tree.
1116 vdev_remove_parent(vdev_t
*cvd
)
1118 vdev_t
*mvd
= cvd
->vdev_parent
;
1119 vdev_t
*pvd
= mvd
->vdev_parent
;
1121 ASSERT(spa_config_held(cvd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
1123 ASSERT(mvd
->vdev_children
== 1);
1124 ASSERT(mvd
->vdev_ops
== &vdev_mirror_ops
||
1125 mvd
->vdev_ops
== &vdev_replacing_ops
||
1126 mvd
->vdev_ops
== &vdev_spare_ops
);
1127 cvd
->vdev_ashift
= mvd
->vdev_ashift
;
1129 vdev_remove_child(mvd
, cvd
);
1130 vdev_remove_child(pvd
, mvd
);
1133 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
1134 * Otherwise, we could have detached an offline device, and when we
1135 * go to import the pool we'll think we have two top-level vdevs,
1136 * instead of a different version of the same top-level vdev.
1138 if (mvd
->vdev_top
== mvd
) {
1139 uint64_t guid_delta
= mvd
->vdev_guid
- cvd
->vdev_guid
;
1140 cvd
->vdev_orig_guid
= cvd
->vdev_guid
;
1141 cvd
->vdev_guid
+= guid_delta
;
1142 cvd
->vdev_guid_sum
+= guid_delta
;
1145 * If pool not set for autoexpand, we need to also preserve
1146 * mvd's asize to prevent automatic expansion of cvd.
1147 * Otherwise if we are adjusting the mirror by attaching and
1148 * detaching children of non-uniform sizes, the mirror could
1149 * autoexpand, unexpectedly requiring larger devices to
1150 * re-establish the mirror.
1152 if (!cvd
->vdev_spa
->spa_autoexpand
)
1153 cvd
->vdev_asize
= mvd
->vdev_asize
;
1155 cvd
->vdev_id
= mvd
->vdev_id
;
1156 vdev_add_child(pvd
, cvd
);
1157 vdev_top_update(cvd
->vdev_top
, cvd
->vdev_top
);
1159 if (cvd
== cvd
->vdev_top
)
1160 vdev_top_transfer(mvd
, cvd
);
1162 ASSERT(mvd
->vdev_children
== 0);
1167 vdev_metaslab_group_create(vdev_t
*vd
)
1169 spa_t
*spa
= vd
->vdev_spa
;
1172 * metaslab_group_create was delayed until allocation bias was available
1174 if (vd
->vdev_mg
== NULL
) {
1175 metaslab_class_t
*mc
;
1177 if (vd
->vdev_islog
&& vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
)
1178 vd
->vdev_alloc_bias
= VDEV_BIAS_LOG
;
1180 ASSERT3U(vd
->vdev_islog
, ==,
1181 (vd
->vdev_alloc_bias
== VDEV_BIAS_LOG
));
1183 switch (vd
->vdev_alloc_bias
) {
1185 mc
= spa_log_class(spa
);
1187 case VDEV_BIAS_SPECIAL
:
1188 mc
= spa_special_class(spa
);
1190 case VDEV_BIAS_DEDUP
:
1191 mc
= spa_dedup_class(spa
);
1194 mc
= spa_normal_class(spa
);
1197 vd
->vdev_mg
= metaslab_group_create(mc
, vd
,
1198 spa
->spa_alloc_count
);
1201 * The spa ashift values currently only reflect the
1202 * general vdev classes. Class destination is late
1203 * binding so ashift checking had to wait until now
1205 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1206 mc
== spa_normal_class(spa
) && vd
->vdev_aux
== NULL
) {
1207 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1208 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1209 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1210 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1216 vdev_metaslab_init(vdev_t
*vd
, uint64_t txg
)
1218 spa_t
*spa
= vd
->vdev_spa
;
1219 objset_t
*mos
= spa
->spa_meta_objset
;
1221 uint64_t oldc
= vd
->vdev_ms_count
;
1222 uint64_t newc
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
1225 boolean_t expanding
= (oldc
!= 0);
1227 ASSERT(txg
== 0 || spa_config_held(spa
, SCL_ALLOC
, RW_WRITER
));
1230 * This vdev is not being allocated from yet or is a hole.
1232 if (vd
->vdev_ms_shift
== 0)
1235 ASSERT(!vd
->vdev_ishole
);
1237 ASSERT(oldc
<= newc
);
1239 mspp
= vmem_zalloc(newc
* sizeof (*mspp
), KM_SLEEP
);
1242 bcopy(vd
->vdev_ms
, mspp
, oldc
* sizeof (*mspp
));
1243 vmem_free(vd
->vdev_ms
, oldc
* sizeof (*mspp
));
1247 vd
->vdev_ms_count
= newc
;
1248 for (m
= oldc
; m
< newc
; m
++) {
1249 uint64_t object
= 0;
1252 * vdev_ms_array may be 0 if we are creating the "fake"
1253 * metaslabs for an indirect vdev for zdb's leak detection.
1254 * See zdb_leak_init().
1256 if (txg
== 0 && vd
->vdev_ms_array
!= 0) {
1257 error
= dmu_read(mos
, vd
->vdev_ms_array
,
1258 m
* sizeof (uint64_t), sizeof (uint64_t), &object
,
1261 vdev_dbgmsg(vd
, "unable to read the metaslab "
1262 "array [error=%d]", error
);
1269 * To accomodate zdb_leak_init() fake indirect
1270 * metaslabs, we allocate a metaslab group for
1271 * indirect vdevs which normally don't have one.
1273 if (vd
->vdev_mg
== NULL
) {
1274 ASSERT0(vdev_is_concrete(vd
));
1275 vdev_metaslab_group_create(vd
);
1278 error
= metaslab_init(vd
->vdev_mg
, m
, object
, txg
,
1281 vdev_dbgmsg(vd
, "metaslab_init failed [error=%d]",
1288 spa_config_enter(spa
, SCL_ALLOC
, FTAG
, RW_WRITER
);
1291 * If the vdev is being removed we don't activate
1292 * the metaslabs since we want to ensure that no new
1293 * allocations are performed on this device.
1295 if (!expanding
&& !vd
->vdev_removing
) {
1296 metaslab_group_activate(vd
->vdev_mg
);
1300 spa_config_exit(spa
, SCL_ALLOC
, FTAG
);
1306 vdev_metaslab_fini(vdev_t
*vd
)
1308 if (vd
->vdev_checkpoint_sm
!= NULL
) {
1309 ASSERT(spa_feature_is_active(vd
->vdev_spa
,
1310 SPA_FEATURE_POOL_CHECKPOINT
));
1311 space_map_close(vd
->vdev_checkpoint_sm
);
1313 * Even though we close the space map, we need to set its
1314 * pointer to NULL. The reason is that vdev_metaslab_fini()
1315 * may be called multiple times for certain operations
1316 * (i.e. when destroying a pool) so we need to ensure that
1317 * this clause never executes twice. This logic is similar
1318 * to the one used for the vdev_ms clause below.
1320 vd
->vdev_checkpoint_sm
= NULL
;
1323 if (vd
->vdev_ms
!= NULL
) {
1324 uint64_t count
= vd
->vdev_ms_count
;
1326 metaslab_group_passivate(vd
->vdev_mg
);
1327 for (uint64_t m
= 0; m
< count
; m
++) {
1328 metaslab_t
*msp
= vd
->vdev_ms
[m
];
1333 vmem_free(vd
->vdev_ms
, count
* sizeof (metaslab_t
*));
1336 vd
->vdev_ms_count
= 0;
1338 ASSERT0(vd
->vdev_ms_count
);
1339 ASSERT3U(vd
->vdev_pending_fastwrite
, ==, 0);
1342 typedef struct vdev_probe_stats
{
1343 boolean_t vps_readable
;
1344 boolean_t vps_writeable
;
1346 } vdev_probe_stats_t
;
1349 vdev_probe_done(zio_t
*zio
)
1351 spa_t
*spa
= zio
->io_spa
;
1352 vdev_t
*vd
= zio
->io_vd
;
1353 vdev_probe_stats_t
*vps
= zio
->io_private
;
1355 ASSERT(vd
->vdev_probe_zio
!= NULL
);
1357 if (zio
->io_type
== ZIO_TYPE_READ
) {
1358 if (zio
->io_error
== 0)
1359 vps
->vps_readable
= 1;
1360 if (zio
->io_error
== 0 && spa_writeable(spa
)) {
1361 zio_nowait(zio_write_phys(vd
->vdev_probe_zio
, vd
,
1362 zio
->io_offset
, zio
->io_size
, zio
->io_abd
,
1363 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1364 ZIO_PRIORITY_SYNC_WRITE
, vps
->vps_flags
, B_TRUE
));
1366 abd_free(zio
->io_abd
);
1368 } else if (zio
->io_type
== ZIO_TYPE_WRITE
) {
1369 if (zio
->io_error
== 0)
1370 vps
->vps_writeable
= 1;
1371 abd_free(zio
->io_abd
);
1372 } else if (zio
->io_type
== ZIO_TYPE_NULL
) {
1376 vd
->vdev_cant_read
|= !vps
->vps_readable
;
1377 vd
->vdev_cant_write
|= !vps
->vps_writeable
;
1379 if (vdev_readable(vd
) &&
1380 (vdev_writeable(vd
) || !spa_writeable(spa
))) {
1383 ASSERT(zio
->io_error
!= 0);
1384 vdev_dbgmsg(vd
, "failed probe");
1385 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE
,
1386 spa
, vd
, NULL
, NULL
, 0, 0);
1387 zio
->io_error
= SET_ERROR(ENXIO
);
1390 mutex_enter(&vd
->vdev_probe_lock
);
1391 ASSERT(vd
->vdev_probe_zio
== zio
);
1392 vd
->vdev_probe_zio
= NULL
;
1393 mutex_exit(&vd
->vdev_probe_lock
);
1396 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
)
1397 if (!vdev_accessible(vd
, pio
))
1398 pio
->io_error
= SET_ERROR(ENXIO
);
1400 kmem_free(vps
, sizeof (*vps
));
1405 * Determine whether this device is accessible.
1407 * Read and write to several known locations: the pad regions of each
1408 * vdev label but the first, which we leave alone in case it contains
1412 vdev_probe(vdev_t
*vd
, zio_t
*zio
)
1414 spa_t
*spa
= vd
->vdev_spa
;
1415 vdev_probe_stats_t
*vps
= NULL
;
1418 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
1421 * Don't probe the probe.
1423 if (zio
&& (zio
->io_flags
& ZIO_FLAG_PROBE
))
1427 * To prevent 'probe storms' when a device fails, we create
1428 * just one probe i/o at a time. All zios that want to probe
1429 * this vdev will become parents of the probe io.
1431 mutex_enter(&vd
->vdev_probe_lock
);
1433 if ((pio
= vd
->vdev_probe_zio
) == NULL
) {
1434 vps
= kmem_zalloc(sizeof (*vps
), KM_SLEEP
);
1436 vps
->vps_flags
= ZIO_FLAG_CANFAIL
| ZIO_FLAG_PROBE
|
1437 ZIO_FLAG_DONT_CACHE
| ZIO_FLAG_DONT_AGGREGATE
|
1440 if (spa_config_held(spa
, SCL_ZIO
, RW_WRITER
)) {
1442 * vdev_cant_read and vdev_cant_write can only
1443 * transition from TRUE to FALSE when we have the
1444 * SCL_ZIO lock as writer; otherwise they can only
1445 * transition from FALSE to TRUE. This ensures that
1446 * any zio looking at these values can assume that
1447 * failures persist for the life of the I/O. That's
1448 * important because when a device has intermittent
1449 * connectivity problems, we want to ensure that
1450 * they're ascribed to the device (ENXIO) and not
1453 * Since we hold SCL_ZIO as writer here, clear both
1454 * values so the probe can reevaluate from first
1457 vps
->vps_flags
|= ZIO_FLAG_CONFIG_WRITER
;
1458 vd
->vdev_cant_read
= B_FALSE
;
1459 vd
->vdev_cant_write
= B_FALSE
;
1462 vd
->vdev_probe_zio
= pio
= zio_null(NULL
, spa
, vd
,
1463 vdev_probe_done
, vps
,
1464 vps
->vps_flags
| ZIO_FLAG_DONT_PROPAGATE
);
1467 * We can't change the vdev state in this context, so we
1468 * kick off an async task to do it on our behalf.
1471 vd
->vdev_probe_wanted
= B_TRUE
;
1472 spa_async_request(spa
, SPA_ASYNC_PROBE
);
1477 zio_add_child(zio
, pio
);
1479 mutex_exit(&vd
->vdev_probe_lock
);
1482 ASSERT(zio
!= NULL
);
1486 for (int l
= 1; l
< VDEV_LABELS
; l
++) {
1487 zio_nowait(zio_read_phys(pio
, vd
,
1488 vdev_label_offset(vd
->vdev_psize
, l
,
1489 offsetof(vdev_label_t
, vl_pad2
)), VDEV_PAD_SIZE
,
1490 abd_alloc_for_io(VDEV_PAD_SIZE
, B_TRUE
),
1491 ZIO_CHECKSUM_OFF
, vdev_probe_done
, vps
,
1492 ZIO_PRIORITY_SYNC_READ
, vps
->vps_flags
, B_TRUE
));
1503 vdev_open_child(void *arg
)
1507 vd
->vdev_open_thread
= curthread
;
1508 vd
->vdev_open_error
= vdev_open(vd
);
1509 vd
->vdev_open_thread
= NULL
;
1513 vdev_uses_zvols(vdev_t
*vd
)
1516 if (zvol_is_zvol(vd
->vdev_path
))
1520 for (int c
= 0; c
< vd
->vdev_children
; c
++)
1521 if (vdev_uses_zvols(vd
->vdev_child
[c
]))
1528 vdev_open_children(vdev_t
*vd
)
1531 int children
= vd
->vdev_children
;
1534 * in order to handle pools on top of zvols, do the opens
1535 * in a single thread so that the same thread holds the
1536 * spa_namespace_lock
1538 if (vdev_uses_zvols(vd
)) {
1540 for (int c
= 0; c
< children
; c
++)
1541 vd
->vdev_child
[c
]->vdev_open_error
=
1542 vdev_open(vd
->vdev_child
[c
]);
1544 tq
= taskq_create("vdev_open", children
, minclsyspri
,
1545 children
, children
, TASKQ_PREPOPULATE
);
1549 for (int c
= 0; c
< children
; c
++)
1550 VERIFY(taskq_dispatch(tq
, vdev_open_child
,
1551 vd
->vdev_child
[c
], TQ_SLEEP
) != TASKQID_INVALID
);
1556 vd
->vdev_nonrot
= B_TRUE
;
1558 for (int c
= 0; c
< children
; c
++)
1559 vd
->vdev_nonrot
&= vd
->vdev_child
[c
]->vdev_nonrot
;
1563 * Compute the raidz-deflation ratio. Note, we hard-code
1564 * in 128k (1 << 17) because it is the "typical" blocksize.
1565 * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change,
1566 * otherwise it would inconsistently account for existing bp's.
1569 vdev_set_deflate_ratio(vdev_t
*vd
)
1571 if (vd
== vd
->vdev_top
&& !vd
->vdev_ishole
&& vd
->vdev_ashift
!= 0) {
1572 vd
->vdev_deflate_ratio
= (1 << 17) /
1573 (vdev_psize_to_asize(vd
, 1 << 17) >> SPA_MINBLOCKSHIFT
);
1578 * Prepare a virtual device for access.
1581 vdev_open(vdev_t
*vd
)
1583 spa_t
*spa
= vd
->vdev_spa
;
1586 uint64_t max_osize
= 0;
1587 uint64_t asize
, max_asize
, psize
;
1588 uint64_t ashift
= 0;
1590 ASSERT(vd
->vdev_open_thread
== curthread
||
1591 spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
1592 ASSERT(vd
->vdev_state
== VDEV_STATE_CLOSED
||
1593 vd
->vdev_state
== VDEV_STATE_CANT_OPEN
||
1594 vd
->vdev_state
== VDEV_STATE_OFFLINE
);
1596 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
1597 vd
->vdev_cant_read
= B_FALSE
;
1598 vd
->vdev_cant_write
= B_FALSE
;
1599 vd
->vdev_min_asize
= vdev_get_min_asize(vd
);
1602 * If this vdev is not removed, check its fault status. If it's
1603 * faulted, bail out of the open.
1605 if (!vd
->vdev_removed
&& vd
->vdev_faulted
) {
1606 ASSERT(vd
->vdev_children
== 0);
1607 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1608 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1609 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1610 vd
->vdev_label_aux
);
1611 return (SET_ERROR(ENXIO
));
1612 } else if (vd
->vdev_offline
) {
1613 ASSERT(vd
->vdev_children
== 0);
1614 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
, VDEV_AUX_NONE
);
1615 return (SET_ERROR(ENXIO
));
1618 error
= vd
->vdev_ops
->vdev_op_open(vd
, &osize
, &max_osize
, &ashift
);
1621 * Reset the vdev_reopening flag so that we actually close
1622 * the vdev on error.
1624 vd
->vdev_reopening
= B_FALSE
;
1625 if (zio_injection_enabled
&& error
== 0)
1626 error
= zio_handle_device_injection(vd
, NULL
, ENXIO
);
1629 if (vd
->vdev_removed
&&
1630 vd
->vdev_stat
.vs_aux
!= VDEV_AUX_OPEN_FAILED
)
1631 vd
->vdev_removed
= B_FALSE
;
1633 if (vd
->vdev_stat
.vs_aux
== VDEV_AUX_CHILDREN_OFFLINE
) {
1634 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_OFFLINE
,
1635 vd
->vdev_stat
.vs_aux
);
1637 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1638 vd
->vdev_stat
.vs_aux
);
1643 vd
->vdev_removed
= B_FALSE
;
1646 * Recheck the faulted flag now that we have confirmed that
1647 * the vdev is accessible. If we're faulted, bail.
1649 if (vd
->vdev_faulted
) {
1650 ASSERT(vd
->vdev_children
== 0);
1651 ASSERT(vd
->vdev_label_aux
== VDEV_AUX_ERR_EXCEEDED
||
1652 vd
->vdev_label_aux
== VDEV_AUX_EXTERNAL
);
1653 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1654 vd
->vdev_label_aux
);
1655 return (SET_ERROR(ENXIO
));
1658 if (vd
->vdev_degraded
) {
1659 ASSERT(vd
->vdev_children
== 0);
1660 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1661 VDEV_AUX_ERR_EXCEEDED
);
1663 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_HEALTHY
, 0);
1667 * For hole or missing vdevs we just return success.
1669 if (vd
->vdev_ishole
|| vd
->vdev_ops
== &vdev_missing_ops
)
1672 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
1673 if (vd
->vdev_child
[c
]->vdev_state
!= VDEV_STATE_HEALTHY
) {
1674 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_DEGRADED
,
1680 osize
= P2ALIGN(osize
, (uint64_t)sizeof (vdev_label_t
));
1681 max_osize
= P2ALIGN(max_osize
, (uint64_t)sizeof (vdev_label_t
));
1683 if (vd
->vdev_children
== 0) {
1684 if (osize
< SPA_MINDEVSIZE
) {
1685 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1686 VDEV_AUX_TOO_SMALL
);
1687 return (SET_ERROR(EOVERFLOW
));
1690 asize
= osize
- (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
);
1691 max_asize
= max_osize
- (VDEV_LABEL_START_SIZE
+
1692 VDEV_LABEL_END_SIZE
);
1694 if (vd
->vdev_parent
!= NULL
&& osize
< SPA_MINDEVSIZE
-
1695 (VDEV_LABEL_START_SIZE
+ VDEV_LABEL_END_SIZE
)) {
1696 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1697 VDEV_AUX_TOO_SMALL
);
1698 return (SET_ERROR(EOVERFLOW
));
1702 max_asize
= max_osize
;
1706 * If the vdev was expanded, record this so that we can re-create the
1707 * uberblock rings in labels {2,3}, during the next sync.
1709 if ((psize
> vd
->vdev_psize
) && (vd
->vdev_psize
!= 0))
1710 vd
->vdev_copy_uberblocks
= B_TRUE
;
1712 vd
->vdev_psize
= psize
;
1715 * Make sure the allocatable size hasn't shrunk too much.
1717 if (asize
< vd
->vdev_min_asize
) {
1718 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1719 VDEV_AUX_BAD_LABEL
);
1720 return (SET_ERROR(EINVAL
));
1723 if (vd
->vdev_asize
== 0) {
1725 * This is the first-ever open, so use the computed values.
1726 * For compatibility, a different ashift can be requested.
1728 vd
->vdev_asize
= asize
;
1729 vd
->vdev_max_asize
= max_asize
;
1730 if (vd
->vdev_ashift
== 0) {
1731 vd
->vdev_ashift
= ashift
; /* use detected value */
1733 if (vd
->vdev_ashift
!= 0 && (vd
->vdev_ashift
< ASHIFT_MIN
||
1734 vd
->vdev_ashift
> ASHIFT_MAX
)) {
1735 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1736 VDEV_AUX_BAD_ASHIFT
);
1737 return (SET_ERROR(EDOM
));
1741 * Detect if the alignment requirement has increased.
1742 * We don't want to make the pool unavailable, just
1743 * post an event instead.
1745 if (ashift
> vd
->vdev_top
->vdev_ashift
&&
1746 vd
->vdev_ops
->vdev_op_leaf
) {
1747 zfs_ereport_post(FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
,
1748 spa
, vd
, NULL
, NULL
, 0, 0);
1751 vd
->vdev_max_asize
= max_asize
;
1755 * If all children are healthy we update asize if either:
1756 * The asize has increased, due to a device expansion caused by dynamic
1757 * LUN growth or vdev replacement, and automatic expansion is enabled;
1758 * making the additional space available.
1760 * The asize has decreased, due to a device shrink usually caused by a
1761 * vdev replace with a smaller device. This ensures that calculations
1762 * based of max_asize and asize e.g. esize are always valid. It's safe
1763 * to do this as we've already validated that asize is greater than
1766 if (vd
->vdev_state
== VDEV_STATE_HEALTHY
&&
1767 ((asize
> vd
->vdev_asize
&&
1768 (vd
->vdev_expanding
|| spa
->spa_autoexpand
)) ||
1769 (asize
< vd
->vdev_asize
)))
1770 vd
->vdev_asize
= asize
;
1772 vdev_set_min_asize(vd
);
1775 * Ensure we can issue some IO before declaring the
1776 * vdev open for business.
1778 if (vd
->vdev_ops
->vdev_op_leaf
&&
1779 (error
= zio_wait(vdev_probe(vd
, NULL
))) != 0) {
1780 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_FAULTED
,
1781 VDEV_AUX_ERR_EXCEEDED
);
1786 * Track the min and max ashift values for normal data devices.
1788 * DJB - TBD these should perhaps be tracked per allocation class
1789 * (e.g. spa_min_ashift is used to round up post compression buffers)
1791 if (vd
->vdev_top
== vd
&& vd
->vdev_ashift
!= 0 &&
1792 vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
&&
1793 vd
->vdev_aux
== NULL
) {
1794 if (vd
->vdev_ashift
> spa
->spa_max_ashift
)
1795 spa
->spa_max_ashift
= vd
->vdev_ashift
;
1796 if (vd
->vdev_ashift
< spa
->spa_min_ashift
)
1797 spa
->spa_min_ashift
= vd
->vdev_ashift
;
1801 * If a leaf vdev has a DTL, and seems healthy, then kick off a
1802 * resilver. But don't do this if we are doing a reopen for a scrub,
1803 * since this would just restart the scrub we are already doing.
1805 if (vd
->vdev_ops
->vdev_op_leaf
&& !spa
->spa_scrub_reopen
&&
1806 vdev_resilver_needed(vd
, NULL
, NULL
)) {
1807 if (dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
1808 spa_feature_is_enabled(spa
, SPA_FEATURE_RESILVER_DEFER
))
1809 vdev_set_deferred_resilver(spa
, vd
);
1811 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
1818 * Called once the vdevs are all opened, this routine validates the label
1819 * contents. This needs to be done before vdev_load() so that we don't
1820 * inadvertently do repair I/Os to the wrong device.
1822 * This function will only return failure if one of the vdevs indicates that it
1823 * has since been destroyed or exported. This is only possible if
1824 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state
1825 * will be updated but the function will return 0.
1828 vdev_validate(vdev_t
*vd
)
1830 spa_t
*spa
= vd
->vdev_spa
;
1832 uint64_t guid
= 0, aux_guid
= 0, top_guid
;
1837 if (vdev_validate_skip
)
1840 for (uint64_t c
= 0; c
< vd
->vdev_children
; c
++)
1841 if (vdev_validate(vd
->vdev_child
[c
]) != 0)
1842 return (SET_ERROR(EBADF
));
1845 * If the device has already failed, or was marked offline, don't do
1846 * any further validation. Otherwise, label I/O will fail and we will
1847 * overwrite the previous state.
1849 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_readable(vd
))
1853 * If we are performing an extreme rewind, we allow for a label that
1854 * was modified at a point after the current txg.
1855 * If config lock is not held do not check for the txg. spa_sync could
1856 * be updating the vdev's label before updating spa_last_synced_txg.
1858 if (spa
->spa_extreme_rewind
|| spa_last_synced_txg(spa
) == 0 ||
1859 spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) != SCL_CONFIG
)
1862 txg
= spa_last_synced_txg(spa
);
1864 if ((label
= vdev_label_read_config(vd
, txg
)) == NULL
) {
1865 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
1866 VDEV_AUX_BAD_LABEL
);
1867 vdev_dbgmsg(vd
, "vdev_validate: failed reading config for "
1868 "txg %llu", (u_longlong_t
)txg
);
1873 * Determine if this vdev has been split off into another
1874 * pool. If so, then refuse to open it.
1876 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_SPLIT_GUID
,
1877 &aux_guid
) == 0 && aux_guid
== spa_guid(spa
)) {
1878 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1879 VDEV_AUX_SPLIT_POOL
);
1881 vdev_dbgmsg(vd
, "vdev_validate: vdev split into other pool");
1885 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_GUID
, &guid
) != 0) {
1886 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1887 VDEV_AUX_CORRUPT_DATA
);
1889 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1890 ZPOOL_CONFIG_POOL_GUID
);
1895 * If config is not trusted then ignore the spa guid check. This is
1896 * necessary because if the machine crashed during a re-guid the new
1897 * guid might have been written to all of the vdev labels, but not the
1898 * cached config. The check will be performed again once we have the
1899 * trusted config from the MOS.
1901 if (spa
->spa_trust_config
&& guid
!= spa_guid(spa
)) {
1902 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1903 VDEV_AUX_CORRUPT_DATA
);
1905 vdev_dbgmsg(vd
, "vdev_validate: vdev label pool_guid doesn't "
1906 "match config (%llu != %llu)", (u_longlong_t
)guid
,
1907 (u_longlong_t
)spa_guid(spa
));
1911 if (nvlist_lookup_nvlist(label
, ZPOOL_CONFIG_VDEV_TREE
, &nvl
)
1912 != 0 || nvlist_lookup_uint64(nvl
, ZPOOL_CONFIG_ORIG_GUID
,
1916 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0) {
1917 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1918 VDEV_AUX_CORRUPT_DATA
);
1920 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1925 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_TOP_GUID
, &top_guid
)
1927 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1928 VDEV_AUX_CORRUPT_DATA
);
1930 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1931 ZPOOL_CONFIG_TOP_GUID
);
1936 * If this vdev just became a top-level vdev because its sibling was
1937 * detached, it will have adopted the parent's vdev guid -- but the
1938 * label may or may not be on disk yet. Fortunately, either version
1939 * of the label will have the same top guid, so if we're a top-level
1940 * vdev, we can safely compare to that instead.
1941 * However, if the config comes from a cachefile that failed to update
1942 * after the detach, a top-level vdev will appear as a non top-level
1943 * vdev in the config. Also relax the constraints if we perform an
1946 * If we split this vdev off instead, then we also check the
1947 * original pool's guid. We don't want to consider the vdev
1948 * corrupt if it is partway through a split operation.
1950 if (vd
->vdev_guid
!= guid
&& vd
->vdev_guid
!= aux_guid
) {
1951 boolean_t mismatch
= B_FALSE
;
1952 if (spa
->spa_trust_config
&& !spa
->spa_extreme_rewind
) {
1953 if (vd
!= vd
->vdev_top
|| vd
->vdev_guid
!= top_guid
)
1956 if (vd
->vdev_guid
!= top_guid
&&
1957 vd
->vdev_top
->vdev_guid
!= guid
)
1962 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1963 VDEV_AUX_CORRUPT_DATA
);
1965 vdev_dbgmsg(vd
, "vdev_validate: config guid "
1966 "doesn't match label guid");
1967 vdev_dbgmsg(vd
, "CONFIG: guid %llu, top_guid %llu",
1968 (u_longlong_t
)vd
->vdev_guid
,
1969 (u_longlong_t
)vd
->vdev_top
->vdev_guid
);
1970 vdev_dbgmsg(vd
, "LABEL: guid %llu, top_guid %llu, "
1971 "aux_guid %llu", (u_longlong_t
)guid
,
1972 (u_longlong_t
)top_guid
, (u_longlong_t
)aux_guid
);
1977 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
,
1979 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
1980 VDEV_AUX_CORRUPT_DATA
);
1982 vdev_dbgmsg(vd
, "vdev_validate: '%s' missing from label",
1983 ZPOOL_CONFIG_POOL_STATE
);
1990 * If this is a verbatim import, no need to check the
1991 * state of the pool.
1993 if (!(spa
->spa_import_flags
& ZFS_IMPORT_VERBATIM
) &&
1994 spa_load_state(spa
) == SPA_LOAD_OPEN
&&
1995 state
!= POOL_STATE_ACTIVE
) {
1996 vdev_dbgmsg(vd
, "vdev_validate: invalid pool state (%llu) "
1997 "for spa %s", (u_longlong_t
)state
, spa
->spa_name
);
1998 return (SET_ERROR(EBADF
));
2002 * If we were able to open and validate a vdev that was
2003 * previously marked permanently unavailable, clear that state
2006 if (vd
->vdev_not_present
)
2007 vd
->vdev_not_present
= 0;
2013 vdev_copy_path_impl(vdev_t
*svd
, vdev_t
*dvd
)
2015 if (svd
->vdev_path
!= NULL
&& dvd
->vdev_path
!= NULL
) {
2016 if (strcmp(svd
->vdev_path
, dvd
->vdev_path
) != 0) {
2017 zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
2018 "from '%s' to '%s'", (u_longlong_t
)dvd
->vdev_guid
,
2019 dvd
->vdev_path
, svd
->vdev_path
);
2020 spa_strfree(dvd
->vdev_path
);
2021 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2023 } else if (svd
->vdev_path
!= NULL
) {
2024 dvd
->vdev_path
= spa_strdup(svd
->vdev_path
);
2025 zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
2026 (u_longlong_t
)dvd
->vdev_guid
, dvd
->vdev_path
);
2031 * Recursively copy vdev paths from one vdev to another. Source and destination
2032 * vdev trees must have same geometry otherwise return error. Intended to copy
2033 * paths from userland config into MOS config.
2036 vdev_copy_path_strict(vdev_t
*svd
, vdev_t
*dvd
)
2038 if ((svd
->vdev_ops
== &vdev_missing_ops
) ||
2039 (svd
->vdev_ishole
&& dvd
->vdev_ishole
) ||
2040 (dvd
->vdev_ops
== &vdev_indirect_ops
))
2043 if (svd
->vdev_ops
!= dvd
->vdev_ops
) {
2044 vdev_dbgmsg(svd
, "vdev_copy_path: vdev type mismatch: %s != %s",
2045 svd
->vdev_ops
->vdev_op_type
, dvd
->vdev_ops
->vdev_op_type
);
2046 return (SET_ERROR(EINVAL
));
2049 if (svd
->vdev_guid
!= dvd
->vdev_guid
) {
2050 vdev_dbgmsg(svd
, "vdev_copy_path: guids mismatch (%llu != "
2051 "%llu)", (u_longlong_t
)svd
->vdev_guid
,
2052 (u_longlong_t
)dvd
->vdev_guid
);
2053 return (SET_ERROR(EINVAL
));
2056 if (svd
->vdev_children
!= dvd
->vdev_children
) {
2057 vdev_dbgmsg(svd
, "vdev_copy_path: children count mismatch: "
2058 "%llu != %llu", (u_longlong_t
)svd
->vdev_children
,
2059 (u_longlong_t
)dvd
->vdev_children
);
2060 return (SET_ERROR(EINVAL
));
2063 for (uint64_t i
= 0; i
< svd
->vdev_children
; i
++) {
2064 int error
= vdev_copy_path_strict(svd
->vdev_child
[i
],
2065 dvd
->vdev_child
[i
]);
2070 if (svd
->vdev_ops
->vdev_op_leaf
)
2071 vdev_copy_path_impl(svd
, dvd
);
2077 vdev_copy_path_search(vdev_t
*stvd
, vdev_t
*dvd
)
2079 ASSERT(stvd
->vdev_top
== stvd
);
2080 ASSERT3U(stvd
->vdev_id
, ==, dvd
->vdev_top
->vdev_id
);
2082 for (uint64_t i
= 0; i
< dvd
->vdev_children
; i
++) {
2083 vdev_copy_path_search(stvd
, dvd
->vdev_child
[i
]);
2086 if (!dvd
->vdev_ops
->vdev_op_leaf
|| !vdev_is_concrete(dvd
))
2090 * The idea here is that while a vdev can shift positions within
2091 * a top vdev (when replacing, attaching mirror, etc.) it cannot
2092 * step outside of it.
2094 vdev_t
*vd
= vdev_lookup_by_guid(stvd
, dvd
->vdev_guid
);
2096 if (vd
== NULL
|| vd
->vdev_ops
!= dvd
->vdev_ops
)
2099 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2101 vdev_copy_path_impl(vd
, dvd
);
2105 * Recursively copy vdev paths from one root vdev to another. Source and
2106 * destination vdev trees may differ in geometry. For each destination leaf
2107 * vdev, search a vdev with the same guid and top vdev id in the source.
2108 * Intended to copy paths from userland config into MOS config.
2111 vdev_copy_path_relaxed(vdev_t
*srvd
, vdev_t
*drvd
)
2113 uint64_t children
= MIN(srvd
->vdev_children
, drvd
->vdev_children
);
2114 ASSERT(srvd
->vdev_ops
== &vdev_root_ops
);
2115 ASSERT(drvd
->vdev_ops
== &vdev_root_ops
);
2117 for (uint64_t i
= 0; i
< children
; i
++) {
2118 vdev_copy_path_search(srvd
->vdev_child
[i
],
2119 drvd
->vdev_child
[i
]);
2124 * Close a virtual device.
2127 vdev_close(vdev_t
*vd
)
2129 vdev_t
*pvd
= vd
->vdev_parent
;
2130 ASSERTV(spa_t
*spa
= vd
->vdev_spa
);
2132 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2135 * If our parent is reopening, then we are as well, unless we are
2138 if (pvd
!= NULL
&& pvd
->vdev_reopening
)
2139 vd
->vdev_reopening
= (pvd
->vdev_reopening
&& !vd
->vdev_offline
);
2141 vd
->vdev_ops
->vdev_op_close(vd
);
2143 vdev_cache_purge(vd
);
2146 * We record the previous state before we close it, so that if we are
2147 * doing a reopen(), we don't generate FMA ereports if we notice that
2148 * it's still faulted.
2150 vd
->vdev_prevstate
= vd
->vdev_state
;
2152 if (vd
->vdev_offline
)
2153 vd
->vdev_state
= VDEV_STATE_OFFLINE
;
2155 vd
->vdev_state
= VDEV_STATE_CLOSED
;
2156 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
2160 vdev_hold(vdev_t
*vd
)
2162 spa_t
*spa
= vd
->vdev_spa
;
2164 ASSERT(spa_is_root(spa
));
2165 if (spa
->spa_state
== POOL_STATE_UNINITIALIZED
)
2168 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2169 vdev_hold(vd
->vdev_child
[c
]);
2171 if (vd
->vdev_ops
->vdev_op_leaf
)
2172 vd
->vdev_ops
->vdev_op_hold(vd
);
2176 vdev_rele(vdev_t
*vd
)
2178 ASSERT(spa_is_root(vd
->vdev_spa
));
2179 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2180 vdev_rele(vd
->vdev_child
[c
]);
2182 if (vd
->vdev_ops
->vdev_op_leaf
)
2183 vd
->vdev_ops
->vdev_op_rele(vd
);
2187 * Reopen all interior vdevs and any unopened leaves. We don't actually
2188 * reopen leaf vdevs which had previously been opened as they might deadlock
2189 * on the spa_config_lock. Instead we only obtain the leaf's physical size.
2190 * If the leaf has never been opened then open it, as usual.
2193 vdev_reopen(vdev_t
*vd
)
2195 spa_t
*spa
= vd
->vdev_spa
;
2197 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2199 /* set the reopening flag unless we're taking the vdev offline */
2200 vd
->vdev_reopening
= !vd
->vdev_offline
;
2202 (void) vdev_open(vd
);
2205 * Call vdev_validate() here to make sure we have the same device.
2206 * Otherwise, a device with an invalid label could be successfully
2207 * opened in response to vdev_reopen().
2210 (void) vdev_validate_aux(vd
);
2211 if (vdev_readable(vd
) && vdev_writeable(vd
) &&
2212 vd
->vdev_aux
== &spa
->spa_l2cache
&&
2213 !l2arc_vdev_present(vd
))
2214 l2arc_add_vdev(spa
, vd
);
2216 (void) vdev_validate(vd
);
2220 * Reassess parent vdev's health.
2222 vdev_propagate_state(vd
);
2226 vdev_create(vdev_t
*vd
, uint64_t txg
, boolean_t isreplacing
)
2231 * Normally, partial opens (e.g. of a mirror) are allowed.
2232 * For a create, however, we want to fail the request if
2233 * there are any components we can't open.
2235 error
= vdev_open(vd
);
2237 if (error
|| vd
->vdev_state
!= VDEV_STATE_HEALTHY
) {
2239 return (error
? error
: ENXIO
);
2243 * Recursively load DTLs and initialize all labels.
2245 if ((error
= vdev_dtl_load(vd
)) != 0 ||
2246 (error
= vdev_label_init(vd
, txg
, isreplacing
?
2247 VDEV_LABEL_REPLACE
: VDEV_LABEL_CREATE
)) != 0) {
2256 vdev_metaslab_set_size(vdev_t
*vd
)
2258 uint64_t asize
= vd
->vdev_asize
;
2259 uint64_t ms_count
= asize
>> vdev_default_ms_shift
;
2263 * There are two dimensions to the metaslab sizing calculation:
2264 * the size of the metaslab and the count of metaslabs per vdev.
2265 * In general, we aim for vdev_max_ms_count (200) metaslabs. The
2266 * range of the dimensions are as follows:
2268 * 2^29 <= ms_size <= 2^38
2269 * 16 <= ms_count <= 131,072
2271 * On the lower end of vdev sizes, we aim for metaslabs sizes of
2272 * at least 512MB (2^29) to minimize fragmentation effects when
2273 * testing with smaller devices. However, the count constraint
2274 * of at least 16 metaslabs will override this minimum size goal.
2276 * On the upper end of vdev sizes, we aim for a maximum metaslab
2277 * size of 256GB. However, we will cap the total count to 2^17
2278 * metaslabs to keep our memory footprint in check.
2280 * The net effect of applying above constrains is summarized below.
2282 * vdev size metaslab count
2283 * -------------|-----------------
2285 * 8GB - 100GB one per 512MB
2287 * 50TB - 32PB one per 256GB
2289 * -------------------------------
2292 if (ms_count
< vdev_min_ms_count
)
2293 ms_shift
= highbit64(asize
/ vdev_min_ms_count
);
2294 else if (ms_count
> vdev_max_ms_count
)
2295 ms_shift
= highbit64(asize
/ vdev_max_ms_count
);
2297 ms_shift
= vdev_default_ms_shift
;
2299 if (ms_shift
< SPA_MAXBLOCKSHIFT
) {
2300 ms_shift
= SPA_MAXBLOCKSHIFT
;
2301 } else if (ms_shift
> vdev_max_ms_shift
) {
2302 ms_shift
= vdev_max_ms_shift
;
2303 /* cap the total count to constrain memory footprint */
2304 if ((asize
>> ms_shift
) > vdev_ms_count_limit
)
2305 ms_shift
= highbit64(asize
/ vdev_ms_count_limit
);
2308 vd
->vdev_ms_shift
= ms_shift
;
2309 ASSERT3U(vd
->vdev_ms_shift
, >=, SPA_MAXBLOCKSHIFT
);
2313 vdev_dirty(vdev_t
*vd
, int flags
, void *arg
, uint64_t txg
)
2315 ASSERT(vd
== vd
->vdev_top
);
2316 /* indirect vdevs don't have metaslabs or dtls */
2317 ASSERT(vdev_is_concrete(vd
) || flags
== 0);
2318 ASSERT(ISP2(flags
));
2319 ASSERT(spa_writeable(vd
->vdev_spa
));
2321 if (flags
& VDD_METASLAB
)
2322 (void) txg_list_add(&vd
->vdev_ms_list
, arg
, txg
);
2324 if (flags
& VDD_DTL
)
2325 (void) txg_list_add(&vd
->vdev_dtl_list
, arg
, txg
);
2327 (void) txg_list_add(&vd
->vdev_spa
->spa_vdev_txg_list
, vd
, txg
);
2331 vdev_dirty_leaves(vdev_t
*vd
, int flags
, uint64_t txg
)
2333 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2334 vdev_dirty_leaves(vd
->vdev_child
[c
], flags
, txg
);
2336 if (vd
->vdev_ops
->vdev_op_leaf
)
2337 vdev_dirty(vd
->vdev_top
, flags
, vd
, txg
);
2343 * A vdev's DTL (dirty time log) is the set of transaction groups for which
2344 * the vdev has less than perfect replication. There are four kinds of DTL:
2346 * DTL_MISSING: txgs for which the vdev has no valid copies of the data
2348 * DTL_PARTIAL: txgs for which data is available, but not fully replicated
2350 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
2351 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
2352 * txgs that was scrubbed.
2354 * DTL_OUTAGE: txgs which cannot currently be read, whether due to
2355 * persistent errors or just some device being offline.
2356 * Unlike the other three, the DTL_OUTAGE map is not generally
2357 * maintained; it's only computed when needed, typically to
2358 * determine whether a device can be detached.
2360 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
2361 * either has the data or it doesn't.
2363 * For interior vdevs such as mirror and RAID-Z the picture is more complex.
2364 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
2365 * if any child is less than fully replicated, then so is its parent.
2366 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
2367 * comprising only those txgs which appear in 'maxfaults' or more children;
2368 * those are the txgs we don't have enough replication to read. For example,
2369 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
2370 * thus, its DTL_MISSING consists of the set of txgs that appear in more than
2371 * two child DTL_MISSING maps.
2373 * It should be clear from the above that to compute the DTLs and outage maps
2374 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
2375 * Therefore, that is all we keep on disk. When loading the pool, or after
2376 * a configuration change, we generate all other DTLs from first principles.
2379 vdev_dtl_dirty(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2381 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2383 ASSERT(t
< DTL_TYPES
);
2384 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2385 ASSERT(spa_writeable(vd
->vdev_spa
));
2387 mutex_enter(&vd
->vdev_dtl_lock
);
2388 if (!range_tree_contains(rt
, txg
, size
))
2389 range_tree_add(rt
, txg
, size
);
2390 mutex_exit(&vd
->vdev_dtl_lock
);
2394 vdev_dtl_contains(vdev_t
*vd
, vdev_dtl_type_t t
, uint64_t txg
, uint64_t size
)
2396 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2397 boolean_t dirty
= B_FALSE
;
2399 ASSERT(t
< DTL_TYPES
);
2400 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2403 * While we are loading the pool, the DTLs have not been loaded yet.
2404 * Ignore the DTLs and try all devices. This avoids a recursive
2405 * mutex enter on the vdev_dtl_lock, and also makes us try hard
2406 * when loading the pool (relying on the checksum to ensure that
2407 * we get the right data -- note that we while loading, we are
2408 * only reading the MOS, which is always checksummed).
2410 if (vd
->vdev_spa
->spa_load_state
!= SPA_LOAD_NONE
)
2413 mutex_enter(&vd
->vdev_dtl_lock
);
2414 if (!range_tree_is_empty(rt
))
2415 dirty
= range_tree_contains(rt
, txg
, size
);
2416 mutex_exit(&vd
->vdev_dtl_lock
);
2422 vdev_dtl_empty(vdev_t
*vd
, vdev_dtl_type_t t
)
2424 range_tree_t
*rt
= vd
->vdev_dtl
[t
];
2427 mutex_enter(&vd
->vdev_dtl_lock
);
2428 empty
= range_tree_is_empty(rt
);
2429 mutex_exit(&vd
->vdev_dtl_lock
);
2435 * Returns B_TRUE if vdev determines offset needs to be resilvered.
2438 vdev_dtl_need_resilver(vdev_t
*vd
, uint64_t offset
, size_t psize
)
2440 ASSERT(vd
!= vd
->vdev_spa
->spa_root_vdev
);
2442 if (vd
->vdev_ops
->vdev_op_need_resilver
== NULL
||
2443 vd
->vdev_ops
->vdev_op_leaf
)
2446 return (vd
->vdev_ops
->vdev_op_need_resilver(vd
, offset
, psize
));
2450 * Returns the lowest txg in the DTL range.
2453 vdev_dtl_min(vdev_t
*vd
)
2457 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2458 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2459 ASSERT0(vd
->vdev_children
);
2461 rs
= avl_first(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2462 return (rs
->rs_start
- 1);
2466 * Returns the highest txg in the DTL.
2469 vdev_dtl_max(vdev_t
*vd
)
2473 ASSERT(MUTEX_HELD(&vd
->vdev_dtl_lock
));
2474 ASSERT3U(range_tree_space(vd
->vdev_dtl
[DTL_MISSING
]), !=, 0);
2475 ASSERT0(vd
->vdev_children
);
2477 rs
= avl_last(&vd
->vdev_dtl
[DTL_MISSING
]->rt_root
);
2478 return (rs
->rs_end
);
2482 * Determine if a resilvering vdev should remove any DTL entries from
2483 * its range. If the vdev was resilvering for the entire duration of the
2484 * scan then it should excise that range from its DTLs. Otherwise, this
2485 * vdev is considered partially resilvered and should leave its DTL
2486 * entries intact. The comment in vdev_dtl_reassess() describes how we
2490 vdev_dtl_should_excise(vdev_t
*vd
)
2492 spa_t
*spa
= vd
->vdev_spa
;
2493 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2495 ASSERT0(scn
->scn_phys
.scn_errors
);
2496 ASSERT0(vd
->vdev_children
);
2498 if (vd
->vdev_state
< VDEV_STATE_DEGRADED
)
2501 if (vd
->vdev_resilver_deferred
)
2504 if (vd
->vdev_resilver_txg
== 0 ||
2505 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]))
2509 * When a resilver is initiated the scan will assign the scn_max_txg
2510 * value to the highest txg value that exists in all DTLs. If this
2511 * device's max DTL is not part of this scan (i.e. it is not in
2512 * the range (scn_min_txg, scn_max_txg] then it is not eligible
2515 if (vdev_dtl_max(vd
) <= scn
->scn_phys
.scn_max_txg
) {
2516 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <=, vdev_dtl_min(vd
));
2517 ASSERT3U(scn
->scn_phys
.scn_min_txg
, <, vd
->vdev_resilver_txg
);
2518 ASSERT3U(vd
->vdev_resilver_txg
, <=, scn
->scn_phys
.scn_max_txg
);
2525 * Reassess DTLs after a config change or scrub completion. If txg == 0 no
2526 * write operations will be issued to the pool.
2529 vdev_dtl_reassess(vdev_t
*vd
, uint64_t txg
, uint64_t scrub_txg
, int scrub_done
)
2531 spa_t
*spa
= vd
->vdev_spa
;
2535 ASSERT(spa_config_held(spa
, SCL_ALL
, RW_READER
) != 0);
2537 for (int c
= 0; c
< vd
->vdev_children
; c
++)
2538 vdev_dtl_reassess(vd
->vdev_child
[c
], txg
,
2539 scrub_txg
, scrub_done
);
2541 if (vd
== spa
->spa_root_vdev
|| !vdev_is_concrete(vd
) || vd
->vdev_aux
)
2544 if (vd
->vdev_ops
->vdev_op_leaf
) {
2545 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
2547 mutex_enter(&vd
->vdev_dtl_lock
);
2550 * If requested, pretend the scan completed cleanly.
2552 if (zfs_scan_ignore_errors
&& scn
)
2553 scn
->scn_phys
.scn_errors
= 0;
2556 * If we've completed a scan cleanly then determine
2557 * if this vdev should remove any DTLs. We only want to
2558 * excise regions on vdevs that were available during
2559 * the entire duration of this scan.
2561 if (scrub_txg
!= 0 &&
2562 (spa
->spa_scrub_started
||
2563 (scn
!= NULL
&& scn
->scn_phys
.scn_errors
== 0)) &&
2564 vdev_dtl_should_excise(vd
)) {
2566 * We completed a scrub up to scrub_txg. If we
2567 * did it without rebooting, then the scrub dtl
2568 * will be valid, so excise the old region and
2569 * fold in the scrub dtl. Otherwise, leave the
2570 * dtl as-is if there was an error.
2572 * There's little trick here: to excise the beginning
2573 * of the DTL_MISSING map, we put it into a reference
2574 * tree and then add a segment with refcnt -1 that
2575 * covers the range [0, scrub_txg). This means
2576 * that each txg in that range has refcnt -1 or 0.
2577 * We then add DTL_SCRUB with a refcnt of 2, so that
2578 * entries in the range [0, scrub_txg) will have a
2579 * positive refcnt -- either 1 or 2. We then convert
2580 * the reference tree into the new DTL_MISSING map.
2582 space_reftree_create(&reftree
);
2583 space_reftree_add_map(&reftree
,
2584 vd
->vdev_dtl
[DTL_MISSING
], 1);
2585 space_reftree_add_seg(&reftree
, 0, scrub_txg
, -1);
2586 space_reftree_add_map(&reftree
,
2587 vd
->vdev_dtl
[DTL_SCRUB
], 2);
2588 space_reftree_generate_map(&reftree
,
2589 vd
->vdev_dtl
[DTL_MISSING
], 1);
2590 space_reftree_destroy(&reftree
);
2592 range_tree_vacate(vd
->vdev_dtl
[DTL_PARTIAL
], NULL
, NULL
);
2593 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2594 range_tree_add
, vd
->vdev_dtl
[DTL_PARTIAL
]);
2596 range_tree_vacate(vd
->vdev_dtl
[DTL_SCRUB
], NULL
, NULL
);
2597 range_tree_vacate(vd
->vdev_dtl
[DTL_OUTAGE
], NULL
, NULL
);
2598 if (!vdev_readable(vd
))
2599 range_tree_add(vd
->vdev_dtl
[DTL_OUTAGE
], 0, -1ULL);
2601 range_tree_walk(vd
->vdev_dtl
[DTL_MISSING
],
2602 range_tree_add
, vd
->vdev_dtl
[DTL_OUTAGE
]);
2605 * If the vdev was resilvering and no longer has any
2606 * DTLs then reset its resilvering flag and dirty
2607 * the top level so that we persist the change.
2609 if (txg
!= 0 && vd
->vdev_resilver_txg
!= 0 &&
2610 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2611 range_tree_is_empty(vd
->vdev_dtl
[DTL_OUTAGE
])) {
2612 vd
->vdev_resilver_txg
= 0;
2613 vdev_config_dirty(vd
->vdev_top
);
2616 mutex_exit(&vd
->vdev_dtl_lock
);
2619 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, txg
);
2623 mutex_enter(&vd
->vdev_dtl_lock
);
2624 for (int t
= 0; t
< DTL_TYPES
; t
++) {
2625 /* account for child's outage in parent's missing map */
2626 int s
= (t
== DTL_MISSING
) ? DTL_OUTAGE
: t
;
2628 continue; /* leaf vdevs only */
2629 if (t
== DTL_PARTIAL
)
2630 minref
= 1; /* i.e. non-zero */
2631 else if (vd
->vdev_nparity
!= 0)
2632 minref
= vd
->vdev_nparity
+ 1; /* RAID-Z */
2634 minref
= vd
->vdev_children
; /* any kind of mirror */
2635 space_reftree_create(&reftree
);
2636 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2637 vdev_t
*cvd
= vd
->vdev_child
[c
];
2638 mutex_enter(&cvd
->vdev_dtl_lock
);
2639 space_reftree_add_map(&reftree
, cvd
->vdev_dtl
[s
], 1);
2640 mutex_exit(&cvd
->vdev_dtl_lock
);
2642 space_reftree_generate_map(&reftree
, vd
->vdev_dtl
[t
], minref
);
2643 space_reftree_destroy(&reftree
);
2645 mutex_exit(&vd
->vdev_dtl_lock
);
2649 vdev_dtl_load(vdev_t
*vd
)
2651 spa_t
*spa
= vd
->vdev_spa
;
2652 objset_t
*mos
= spa
->spa_meta_objset
;
2655 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_dtl_object
!= 0) {
2656 ASSERT(vdev_is_concrete(vd
));
2658 error
= space_map_open(&vd
->vdev_dtl_sm
, mos
,
2659 vd
->vdev_dtl_object
, 0, -1ULL, 0);
2662 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2664 mutex_enter(&vd
->vdev_dtl_lock
);
2667 * Now that we've opened the space_map we need to update
2670 space_map_update(vd
->vdev_dtl_sm
);
2672 error
= space_map_load(vd
->vdev_dtl_sm
,
2673 vd
->vdev_dtl
[DTL_MISSING
], SM_ALLOC
);
2674 mutex_exit(&vd
->vdev_dtl_lock
);
2679 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2680 error
= vdev_dtl_load(vd
->vdev_child
[c
]);
2689 vdev_zap_allocation_data(vdev_t
*vd
, dmu_tx_t
*tx
)
2691 spa_t
*spa
= vd
->vdev_spa
;
2692 objset_t
*mos
= spa
->spa_meta_objset
;
2693 vdev_alloc_bias_t alloc_bias
= vd
->vdev_alloc_bias
;
2696 ASSERT(alloc_bias
!= VDEV_BIAS_NONE
);
2699 (alloc_bias
== VDEV_BIAS_LOG
) ? VDEV_ALLOC_BIAS_LOG
:
2700 (alloc_bias
== VDEV_BIAS_SPECIAL
) ? VDEV_ALLOC_BIAS_SPECIAL
:
2701 (alloc_bias
== VDEV_BIAS_DEDUP
) ? VDEV_ALLOC_BIAS_DEDUP
: NULL
;
2703 ASSERT(string
!= NULL
);
2704 VERIFY0(zap_add(mos
, vd
->vdev_top_zap
, VDEV_TOP_ZAP_ALLOCATION_BIAS
,
2705 1, strlen(string
) + 1, string
, tx
));
2707 if (alloc_bias
== VDEV_BIAS_SPECIAL
|| alloc_bias
== VDEV_BIAS_DEDUP
) {
2708 spa_activate_allocation_classes(spa
, tx
);
2713 vdev_destroy_unlink_zap(vdev_t
*vd
, uint64_t zapobj
, dmu_tx_t
*tx
)
2715 spa_t
*spa
= vd
->vdev_spa
;
2717 VERIFY0(zap_destroy(spa
->spa_meta_objset
, zapobj
, tx
));
2718 VERIFY0(zap_remove_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2723 vdev_create_link_zap(vdev_t
*vd
, dmu_tx_t
*tx
)
2725 spa_t
*spa
= vd
->vdev_spa
;
2726 uint64_t zap
= zap_create(spa
->spa_meta_objset
, DMU_OTN_ZAP_METADATA
,
2727 DMU_OT_NONE
, 0, tx
);
2730 VERIFY0(zap_add_int(spa
->spa_meta_objset
, spa
->spa_all_vdev_zaps
,
2737 vdev_construct_zaps(vdev_t
*vd
, dmu_tx_t
*tx
)
2739 if (vd
->vdev_ops
!= &vdev_hole_ops
&&
2740 vd
->vdev_ops
!= &vdev_missing_ops
&&
2741 vd
->vdev_ops
!= &vdev_root_ops
&&
2742 !vd
->vdev_top
->vdev_removing
) {
2743 if (vd
->vdev_ops
->vdev_op_leaf
&& vd
->vdev_leaf_zap
== 0) {
2744 vd
->vdev_leaf_zap
= vdev_create_link_zap(vd
, tx
);
2746 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
== 0) {
2747 vd
->vdev_top_zap
= vdev_create_link_zap(vd
, tx
);
2748 if (vd
->vdev_alloc_bias
!= VDEV_BIAS_NONE
)
2749 vdev_zap_allocation_data(vd
, tx
);
2753 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
2754 vdev_construct_zaps(vd
->vdev_child
[i
], tx
);
2759 vdev_dtl_sync(vdev_t
*vd
, uint64_t txg
)
2761 spa_t
*spa
= vd
->vdev_spa
;
2762 range_tree_t
*rt
= vd
->vdev_dtl
[DTL_MISSING
];
2763 objset_t
*mos
= spa
->spa_meta_objset
;
2764 range_tree_t
*rtsync
;
2766 uint64_t object
= space_map_object(vd
->vdev_dtl_sm
);
2768 ASSERT(vdev_is_concrete(vd
));
2769 ASSERT(vd
->vdev_ops
->vdev_op_leaf
);
2771 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
2773 if (vd
->vdev_detached
|| vd
->vdev_top
->vdev_removing
) {
2774 mutex_enter(&vd
->vdev_dtl_lock
);
2775 space_map_free(vd
->vdev_dtl_sm
, tx
);
2776 space_map_close(vd
->vdev_dtl_sm
);
2777 vd
->vdev_dtl_sm
= NULL
;
2778 mutex_exit(&vd
->vdev_dtl_lock
);
2781 * We only destroy the leaf ZAP for detached leaves or for
2782 * removed log devices. Removed data devices handle leaf ZAP
2783 * cleanup later, once cancellation is no longer possible.
2785 if (vd
->vdev_leaf_zap
!= 0 && (vd
->vdev_detached
||
2786 vd
->vdev_top
->vdev_islog
)) {
2787 vdev_destroy_unlink_zap(vd
, vd
->vdev_leaf_zap
, tx
);
2788 vd
->vdev_leaf_zap
= 0;
2795 if (vd
->vdev_dtl_sm
== NULL
) {
2796 uint64_t new_object
;
2798 new_object
= space_map_alloc(mos
, vdev_dtl_sm_blksz
, tx
);
2799 VERIFY3U(new_object
, !=, 0);
2801 VERIFY0(space_map_open(&vd
->vdev_dtl_sm
, mos
, new_object
,
2803 ASSERT(vd
->vdev_dtl_sm
!= NULL
);
2806 rtsync
= range_tree_create(NULL
, NULL
);
2808 mutex_enter(&vd
->vdev_dtl_lock
);
2809 range_tree_walk(rt
, range_tree_add
, rtsync
);
2810 mutex_exit(&vd
->vdev_dtl_lock
);
2812 space_map_truncate(vd
->vdev_dtl_sm
, vdev_dtl_sm_blksz
, tx
);
2813 space_map_write(vd
->vdev_dtl_sm
, rtsync
, SM_ALLOC
, SM_NO_VDEVID
, tx
);
2814 range_tree_vacate(rtsync
, NULL
, NULL
);
2816 range_tree_destroy(rtsync
);
2819 * If the object for the space map has changed then dirty
2820 * the top level so that we update the config.
2822 if (object
!= space_map_object(vd
->vdev_dtl_sm
)) {
2823 vdev_dbgmsg(vd
, "txg %llu, spa %s, DTL old object %llu, "
2824 "new object %llu", (u_longlong_t
)txg
, spa_name(spa
),
2825 (u_longlong_t
)object
,
2826 (u_longlong_t
)space_map_object(vd
->vdev_dtl_sm
));
2827 vdev_config_dirty(vd
->vdev_top
);
2832 mutex_enter(&vd
->vdev_dtl_lock
);
2833 space_map_update(vd
->vdev_dtl_sm
);
2834 mutex_exit(&vd
->vdev_dtl_lock
);
2838 * Determine whether the specified vdev can be offlined/detached/removed
2839 * without losing data.
2842 vdev_dtl_required(vdev_t
*vd
)
2844 spa_t
*spa
= vd
->vdev_spa
;
2845 vdev_t
*tvd
= vd
->vdev_top
;
2846 uint8_t cant_read
= vd
->vdev_cant_read
;
2849 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
2851 if (vd
== spa
->spa_root_vdev
|| vd
== tvd
)
2855 * Temporarily mark the device as unreadable, and then determine
2856 * whether this results in any DTL outages in the top-level vdev.
2857 * If not, we can safely offline/detach/remove the device.
2859 vd
->vdev_cant_read
= B_TRUE
;
2860 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2861 required
= !vdev_dtl_empty(tvd
, DTL_OUTAGE
);
2862 vd
->vdev_cant_read
= cant_read
;
2863 vdev_dtl_reassess(tvd
, 0, 0, B_FALSE
);
2865 if (!required
&& zio_injection_enabled
)
2866 required
= !!zio_handle_device_injection(vd
, NULL
, ECHILD
);
2872 * Determine if resilver is needed, and if so the txg range.
2875 vdev_resilver_needed(vdev_t
*vd
, uint64_t *minp
, uint64_t *maxp
)
2877 boolean_t needed
= B_FALSE
;
2878 uint64_t thismin
= UINT64_MAX
;
2879 uint64_t thismax
= 0;
2881 if (vd
->vdev_children
== 0) {
2882 mutex_enter(&vd
->vdev_dtl_lock
);
2883 if (!range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
]) &&
2884 vdev_writeable(vd
)) {
2886 thismin
= vdev_dtl_min(vd
);
2887 thismax
= vdev_dtl_max(vd
);
2890 mutex_exit(&vd
->vdev_dtl_lock
);
2892 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2893 vdev_t
*cvd
= vd
->vdev_child
[c
];
2894 uint64_t cmin
, cmax
;
2896 if (vdev_resilver_needed(cvd
, &cmin
, &cmax
)) {
2897 thismin
= MIN(thismin
, cmin
);
2898 thismax
= MAX(thismax
, cmax
);
2904 if (needed
&& minp
) {
2912 * Gets the checkpoint space map object from the vdev's ZAP. On success sm_obj
2913 * will contain either the checkpoint spacemap object or zero if none exists.
2914 * All other errors are returned to the caller.
2917 vdev_checkpoint_sm_object(vdev_t
*vd
, uint64_t *sm_obj
)
2919 ASSERT0(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
));
2921 if (vd
->vdev_top_zap
== 0) {
2926 int error
= zap_lookup(spa_meta_objset(vd
->vdev_spa
), vd
->vdev_top_zap
,
2927 VDEV_TOP_ZAP_POOL_CHECKPOINT_SM
, sizeof (uint64_t), 1, sm_obj
);
2928 if (error
== ENOENT
) {
2937 vdev_load(vdev_t
*vd
)
2942 * Recursively load all children.
2944 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
2945 error
= vdev_load(vd
->vdev_child
[c
]);
2951 vdev_set_deflate_ratio(vd
);
2954 * On spa_load path, grab the allocation bias from our zap
2956 if (vd
== vd
->vdev_top
&& vd
->vdev_top_zap
!= 0) {
2957 spa_t
*spa
= vd
->vdev_spa
;
2960 if (zap_lookup(spa
->spa_meta_objset
, vd
->vdev_top_zap
,
2961 VDEV_TOP_ZAP_ALLOCATION_BIAS
, 1, sizeof (bias_str
),
2963 ASSERT(vd
->vdev_alloc_bias
== VDEV_BIAS_NONE
);
2964 vd
->vdev_alloc_bias
= vdev_derive_alloc_bias(bias_str
);
2969 * If this is a top-level vdev, initialize its metaslabs.
2971 if (vd
== vd
->vdev_top
&& vdev_is_concrete(vd
)) {
2972 vdev_metaslab_group_create(vd
);
2974 if (vd
->vdev_ashift
== 0 || vd
->vdev_asize
== 0) {
2975 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2976 VDEV_AUX_CORRUPT_DATA
);
2977 vdev_dbgmsg(vd
, "vdev_load: invalid size. ashift=%llu, "
2978 "asize=%llu", (u_longlong_t
)vd
->vdev_ashift
,
2979 (u_longlong_t
)vd
->vdev_asize
);
2980 return (SET_ERROR(ENXIO
));
2981 } else if ((error
= vdev_metaslab_init(vd
, 0)) != 0) {
2982 vdev_dbgmsg(vd
, "vdev_load: metaslab_init failed "
2983 "[error=%d]", error
);
2984 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
2985 VDEV_AUX_CORRUPT_DATA
);
2989 uint64_t checkpoint_sm_obj
;
2990 error
= vdev_checkpoint_sm_object(vd
, &checkpoint_sm_obj
);
2991 if (error
== 0 && checkpoint_sm_obj
!= 0) {
2992 objset_t
*mos
= spa_meta_objset(vd
->vdev_spa
);
2993 ASSERT(vd
->vdev_asize
!= 0);
2994 ASSERT3P(vd
->vdev_checkpoint_sm
, ==, NULL
);
2996 if ((error
= space_map_open(&vd
->vdev_checkpoint_sm
,
2997 mos
, checkpoint_sm_obj
, 0, vd
->vdev_asize
,
2998 vd
->vdev_ashift
))) {
2999 vdev_dbgmsg(vd
, "vdev_load: space_map_open "
3000 "failed for checkpoint spacemap (obj %llu) "
3002 (u_longlong_t
)checkpoint_sm_obj
, error
);
3005 ASSERT3P(vd
->vdev_checkpoint_sm
, !=, NULL
);
3006 space_map_update(vd
->vdev_checkpoint_sm
);
3009 * Since the checkpoint_sm contains free entries
3010 * exclusively we can use sm_alloc to indicate the
3011 * cumulative checkpointed space that has been freed.
3013 vd
->vdev_stat
.vs_checkpoint_space
=
3014 -vd
->vdev_checkpoint_sm
->sm_alloc
;
3015 vd
->vdev_spa
->spa_checkpoint_info
.sci_dspace
+=
3016 vd
->vdev_stat
.vs_checkpoint_space
;
3017 } else if (error
!= 0) {
3018 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve "
3019 "checkpoint space map object from vdev ZAP "
3020 "[error=%d]", error
);
3026 * If this is a leaf vdev, load its DTL.
3028 if (vd
->vdev_ops
->vdev_op_leaf
&& (error
= vdev_dtl_load(vd
)) != 0) {
3029 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3030 VDEV_AUX_CORRUPT_DATA
);
3031 vdev_dbgmsg(vd
, "vdev_load: vdev_dtl_load failed "
3032 "[error=%d]", error
);
3036 uint64_t obsolete_sm_object
;
3037 error
= vdev_obsolete_sm_object(vd
, &obsolete_sm_object
);
3038 if (error
== 0 && obsolete_sm_object
!= 0) {
3039 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3040 ASSERT(vd
->vdev_asize
!= 0);
3041 ASSERT3P(vd
->vdev_obsolete_sm
, ==, NULL
);
3043 if ((error
= space_map_open(&vd
->vdev_obsolete_sm
, mos
,
3044 obsolete_sm_object
, 0, vd
->vdev_asize
, 0))) {
3045 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
3046 VDEV_AUX_CORRUPT_DATA
);
3047 vdev_dbgmsg(vd
, "vdev_load: space_map_open failed for "
3048 "obsolete spacemap (obj %llu) [error=%d]",
3049 (u_longlong_t
)obsolete_sm_object
, error
);
3052 space_map_update(vd
->vdev_obsolete_sm
);
3053 } else if (error
!= 0) {
3054 vdev_dbgmsg(vd
, "vdev_load: failed to retrieve obsolete "
3055 "space map object from vdev ZAP [error=%d]", error
);
3063 * The special vdev case is used for hot spares and l2cache devices. Its
3064 * sole purpose it to set the vdev state for the associated vdev. To do this,
3065 * we make sure that we can open the underlying device, then try to read the
3066 * label, and make sure that the label is sane and that it hasn't been
3067 * repurposed to another pool.
3070 vdev_validate_aux(vdev_t
*vd
)
3073 uint64_t guid
, version
;
3076 if (!vdev_readable(vd
))
3079 if ((label
= vdev_label_read_config(vd
, -1ULL)) == NULL
) {
3080 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3081 VDEV_AUX_CORRUPT_DATA
);
3085 if (nvlist_lookup_uint64(label
, ZPOOL_CONFIG_VERSION
, &version
) != 0 ||
3086 !SPA_VERSION_IS_SUPPORTED(version
) ||
3087 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_GUID
, &guid
) != 0 ||
3088 guid
!= vd
->vdev_guid
||
3089 nvlist_lookup_uint64(label
, ZPOOL_CONFIG_POOL_STATE
, &state
) != 0) {
3090 vdev_set_state(vd
, B_TRUE
, VDEV_STATE_CANT_OPEN
,
3091 VDEV_AUX_CORRUPT_DATA
);
3097 * We don't actually check the pool state here. If it's in fact in
3098 * use by another pool, we update this fact on the fly when requested.
3105 * Free the objects used to store this vdev's spacemaps, and the array
3106 * that points to them.
3109 vdev_destroy_spacemaps(vdev_t
*vd
, dmu_tx_t
*tx
)
3111 if (vd
->vdev_ms_array
== 0)
3114 objset_t
*mos
= vd
->vdev_spa
->spa_meta_objset
;
3115 uint64_t array_count
= vd
->vdev_asize
>> vd
->vdev_ms_shift
;
3116 size_t array_bytes
= array_count
* sizeof (uint64_t);
3117 uint64_t *smobj_array
= kmem_alloc(array_bytes
, KM_SLEEP
);
3118 VERIFY0(dmu_read(mos
, vd
->vdev_ms_array
, 0,
3119 array_bytes
, smobj_array
, 0));
3121 for (uint64_t i
= 0; i
< array_count
; i
++) {
3122 uint64_t smobj
= smobj_array
[i
];
3126 space_map_free_obj(mos
, smobj
, tx
);
3129 kmem_free(smobj_array
, array_bytes
);
3130 VERIFY0(dmu_object_free(mos
, vd
->vdev_ms_array
, tx
));
3131 vd
->vdev_ms_array
= 0;
3135 vdev_remove_empty_log(vdev_t
*vd
, uint64_t txg
)
3137 spa_t
*spa
= vd
->vdev_spa
;
3139 ASSERT(vd
->vdev_islog
);
3140 ASSERT(vd
== vd
->vdev_top
);
3141 ASSERT3U(txg
, ==, spa_syncing_txg(spa
));
3143 if (vd
->vdev_ms
!= NULL
) {
3144 metaslab_group_t
*mg
= vd
->vdev_mg
;
3146 metaslab_group_histogram_verify(mg
);
3147 metaslab_class_histogram_verify(mg
->mg_class
);
3149 for (int m
= 0; m
< vd
->vdev_ms_count
; m
++) {
3150 metaslab_t
*msp
= vd
->vdev_ms
[m
];
3152 if (msp
== NULL
|| msp
->ms_sm
== NULL
)
3155 mutex_enter(&msp
->ms_lock
);
3157 * If the metaslab was not loaded when the vdev
3158 * was removed then the histogram accounting may
3159 * not be accurate. Update the histogram information
3160 * here so that we ensure that the metaslab group
3161 * and metaslab class are up-to-date.
3163 metaslab_group_histogram_remove(mg
, msp
);
3165 VERIFY0(space_map_allocated(msp
->ms_sm
));
3166 space_map_close(msp
->ms_sm
);
3168 mutex_exit(&msp
->ms_lock
);
3171 if (vd
->vdev_checkpoint_sm
!= NULL
) {
3172 ASSERT(spa_has_checkpoint(spa
));
3173 space_map_close(vd
->vdev_checkpoint_sm
);
3174 vd
->vdev_checkpoint_sm
= NULL
;
3177 metaslab_group_histogram_verify(mg
);
3178 metaslab_class_histogram_verify(mg
->mg_class
);
3180 for (int i
= 0; i
< RANGE_TREE_HISTOGRAM_SIZE
; i
++)
3181 ASSERT0(mg
->mg_histogram
[i
]);
3184 dmu_tx_t
*tx
= dmu_tx_create_assigned(spa_get_dsl(spa
), txg
);
3186 vdev_destroy_spacemaps(vd
, tx
);
3187 if (vd
->vdev_top_zap
!= 0) {
3188 vdev_destroy_unlink_zap(vd
, vd
->vdev_top_zap
, tx
);
3189 vd
->vdev_top_zap
= 0;
3196 vdev_sync_done(vdev_t
*vd
, uint64_t txg
)
3199 boolean_t reassess
= !txg_list_empty(&vd
->vdev_ms_list
, TXG_CLEAN(txg
));
3201 ASSERT(vdev_is_concrete(vd
));
3203 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, TXG_CLEAN(txg
))))
3204 metaslab_sync_done(msp
, txg
);
3207 metaslab_sync_reassess(vd
->vdev_mg
);
3211 vdev_sync(vdev_t
*vd
, uint64_t txg
)
3213 spa_t
*spa
= vd
->vdev_spa
;
3218 if (range_tree_space(vd
->vdev_obsolete_segments
) > 0) {
3221 ASSERT(vd
->vdev_removing
||
3222 vd
->vdev_ops
== &vdev_indirect_ops
);
3224 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3225 vdev_indirect_sync_obsolete(vd
, tx
);
3229 * If the vdev is indirect, it can't have dirty
3230 * metaslabs or DTLs.
3232 if (vd
->vdev_ops
== &vdev_indirect_ops
) {
3233 ASSERT(txg_list_empty(&vd
->vdev_ms_list
, txg
));
3234 ASSERT(txg_list_empty(&vd
->vdev_dtl_list
, txg
));
3239 ASSERT(vdev_is_concrete(vd
));
3241 if (vd
->vdev_ms_array
== 0 && vd
->vdev_ms_shift
!= 0 &&
3242 !vd
->vdev_removing
) {
3243 ASSERT(vd
== vd
->vdev_top
);
3244 ASSERT0(vd
->vdev_indirect_config
.vic_mapping_object
);
3245 tx
= dmu_tx_create_assigned(spa
->spa_dsl_pool
, txg
);
3246 vd
->vdev_ms_array
= dmu_object_alloc(spa
->spa_meta_objset
,
3247 DMU_OT_OBJECT_ARRAY
, 0, DMU_OT_NONE
, 0, tx
);
3248 ASSERT(vd
->vdev_ms_array
!= 0);
3249 vdev_config_dirty(vd
);
3253 while ((msp
= txg_list_remove(&vd
->vdev_ms_list
, txg
)) != NULL
) {
3254 metaslab_sync(msp
, txg
);
3255 (void) txg_list_add(&vd
->vdev_ms_list
, msp
, TXG_CLEAN(txg
));
3258 while ((lvd
= txg_list_remove(&vd
->vdev_dtl_list
, txg
)) != NULL
)
3259 vdev_dtl_sync(lvd
, txg
);
3262 * If this is an empty log device being removed, destroy the
3263 * metadata associated with it.
3265 if (vd
->vdev_islog
&& vd
->vdev_stat
.vs_alloc
== 0 && vd
->vdev_removing
)
3266 vdev_remove_empty_log(vd
, txg
);
3268 (void) txg_list_add(&spa
->spa_vdev_txg_list
, vd
, TXG_CLEAN(txg
));
3272 vdev_psize_to_asize(vdev_t
*vd
, uint64_t psize
)
3274 return (vd
->vdev_ops
->vdev_op_asize(vd
, psize
));
3278 * Mark the given vdev faulted. A faulted vdev behaves as if the device could
3279 * not be opened, and no I/O is attempted.
3282 vdev_fault(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3286 spa_vdev_state_enter(spa
, SCL_NONE
);
3288 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3289 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3291 if (!vd
->vdev_ops
->vdev_op_leaf
)
3292 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3297 * If user did a 'zpool offline -f' then make the fault persist across
3300 if (aux
== VDEV_AUX_EXTERNAL_PERSIST
) {
3302 * There are two kinds of forced faults: temporary and
3303 * persistent. Temporary faults go away at pool import, while
3304 * persistent faults stay set. Both types of faults can be
3305 * cleared with a zpool clear.
3307 * We tell if a vdev is persistently faulted by looking at the
3308 * ZPOOL_CONFIG_AUX_STATE nvpair. If it's set to "external" at
3309 * import then it's a persistent fault. Otherwise, it's
3310 * temporary. We get ZPOOL_CONFIG_AUX_STATE set to "external"
3311 * by setting vd.vdev_stat.vs_aux to VDEV_AUX_EXTERNAL. This
3312 * tells vdev_config_generate() (which gets run later) to set
3313 * ZPOOL_CONFIG_AUX_STATE to "external" in the nvlist.
3315 vd
->vdev_stat
.vs_aux
= VDEV_AUX_EXTERNAL
;
3316 vd
->vdev_tmpoffline
= B_FALSE
;
3317 aux
= VDEV_AUX_EXTERNAL
;
3319 vd
->vdev_tmpoffline
= B_TRUE
;
3323 * We don't directly use the aux state here, but if we do a
3324 * vdev_reopen(), we need this value to be present to remember why we
3327 vd
->vdev_label_aux
= aux
;
3330 * Faulted state takes precedence over degraded.
3332 vd
->vdev_delayed_close
= B_FALSE
;
3333 vd
->vdev_faulted
= 1ULL;
3334 vd
->vdev_degraded
= 0ULL;
3335 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_FAULTED
, aux
);
3338 * If this device has the only valid copy of the data, then
3339 * back off and simply mark the vdev as degraded instead.
3341 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&& vdev_dtl_required(vd
)) {
3342 vd
->vdev_degraded
= 1ULL;
3343 vd
->vdev_faulted
= 0ULL;
3346 * If we reopen the device and it's not dead, only then do we
3351 if (vdev_readable(vd
))
3352 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, aux
);
3355 return (spa_vdev_state_exit(spa
, vd
, 0));
3359 * Mark the given vdev degraded. A degraded vdev is purely an indication to the
3360 * user that something is wrong. The vdev continues to operate as normal as far
3361 * as I/O is concerned.
3364 vdev_degrade(spa_t
*spa
, uint64_t guid
, vdev_aux_t aux
)
3368 spa_vdev_state_enter(spa
, SCL_NONE
);
3370 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3371 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3373 if (!vd
->vdev_ops
->vdev_op_leaf
)
3374 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3377 * If the vdev is already faulted, then don't do anything.
3379 if (vd
->vdev_faulted
|| vd
->vdev_degraded
)
3380 return (spa_vdev_state_exit(spa
, NULL
, 0));
3382 vd
->vdev_degraded
= 1ULL;
3383 if (!vdev_is_dead(vd
))
3384 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
,
3387 return (spa_vdev_state_exit(spa
, vd
, 0));
3391 * Online the given vdev.
3393 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached
3394 * spare device should be detached when the device finishes resilvering.
3395 * Second, the online should be treated like a 'test' online case, so no FMA
3396 * events are generated if the device fails to open.
3399 vdev_online(spa_t
*spa
, uint64_t guid
, uint64_t flags
, vdev_state_t
*newstate
)
3401 vdev_t
*vd
, *tvd
, *pvd
, *rvd
= spa
->spa_root_vdev
;
3402 boolean_t wasoffline
;
3403 vdev_state_t oldstate
;
3405 spa_vdev_state_enter(spa
, SCL_NONE
);
3407 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3408 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3410 if (!vd
->vdev_ops
->vdev_op_leaf
)
3411 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3413 wasoffline
= (vd
->vdev_offline
|| vd
->vdev_tmpoffline
);
3414 oldstate
= vd
->vdev_state
;
3417 vd
->vdev_offline
= B_FALSE
;
3418 vd
->vdev_tmpoffline
= B_FALSE
;
3419 vd
->vdev_checkremove
= !!(flags
& ZFS_ONLINE_CHECKREMOVE
);
3420 vd
->vdev_forcefault
= !!(flags
& ZFS_ONLINE_FORCEFAULT
);
3422 /* XXX - L2ARC 1.0 does not support expansion */
3423 if (!vd
->vdev_aux
) {
3424 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3425 pvd
->vdev_expanding
= !!((flags
& ZFS_ONLINE_EXPAND
) ||
3426 spa
->spa_autoexpand
);
3430 vd
->vdev_checkremove
= vd
->vdev_forcefault
= B_FALSE
;
3432 if (!vd
->vdev_aux
) {
3433 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
3434 pvd
->vdev_expanding
= B_FALSE
;
3438 *newstate
= vd
->vdev_state
;
3439 if ((flags
& ZFS_ONLINE_UNSPARE
) &&
3440 !vdev_is_dead(vd
) && vd
->vdev_parent
&&
3441 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3442 vd
->vdev_parent
->vdev_child
[0] == vd
)
3443 vd
->vdev_unspare
= B_TRUE
;
3445 if ((flags
& ZFS_ONLINE_EXPAND
) || spa
->spa_autoexpand
) {
3447 /* XXX - L2ARC 1.0 does not support expansion */
3449 return (spa_vdev_state_exit(spa
, vd
, ENOTSUP
));
3450 spa_async_request(spa
, SPA_ASYNC_CONFIG_UPDATE
);
3454 (oldstate
< VDEV_STATE_DEGRADED
&&
3455 vd
->vdev_state
>= VDEV_STATE_DEGRADED
))
3456 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_ONLINE
);
3458 return (spa_vdev_state_exit(spa
, vd
, 0));
3462 vdev_offline_locked(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3466 uint64_t generation
;
3467 metaslab_group_t
*mg
;
3470 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3472 if ((vd
= spa_lookup_by_guid(spa
, guid
, B_TRUE
)) == NULL
)
3473 return (spa_vdev_state_exit(spa
, NULL
, ENODEV
));
3475 if (!vd
->vdev_ops
->vdev_op_leaf
)
3476 return (spa_vdev_state_exit(spa
, NULL
, ENOTSUP
));
3480 generation
= spa
->spa_config_generation
+ 1;
3483 * If the device isn't already offline, try to offline it.
3485 if (!vd
->vdev_offline
) {
3487 * If this device has the only valid copy of some data,
3488 * don't allow it to be offlined. Log devices are always
3491 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3492 vdev_dtl_required(vd
))
3493 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3496 * If the top-level is a slog and it has had allocations
3497 * then proceed. We check that the vdev's metaslab group
3498 * is not NULL since it's possible that we may have just
3499 * added this vdev but not yet initialized its metaslabs.
3501 if (tvd
->vdev_islog
&& mg
!= NULL
) {
3503 * Prevent any future allocations.
3505 metaslab_group_passivate(mg
);
3506 (void) spa_vdev_state_exit(spa
, vd
, 0);
3508 error
= spa_reset_logs(spa
);
3511 * If the log device was successfully reset but has
3512 * checkpointed data, do not offline it.
3515 tvd
->vdev_checkpoint_sm
!= NULL
) {
3516 ASSERT3U(tvd
->vdev_checkpoint_sm
->sm_alloc
,
3518 error
= ZFS_ERR_CHECKPOINT_EXISTS
;
3521 spa_vdev_state_enter(spa
, SCL_ALLOC
);
3524 * Check to see if the config has changed.
3526 if (error
|| generation
!= spa
->spa_config_generation
) {
3527 metaslab_group_activate(mg
);
3529 return (spa_vdev_state_exit(spa
,
3531 (void) spa_vdev_state_exit(spa
, vd
, 0);
3534 ASSERT0(tvd
->vdev_stat
.vs_alloc
);
3538 * Offline this device and reopen its top-level vdev.
3539 * If the top-level vdev is a log device then just offline
3540 * it. Otherwise, if this action results in the top-level
3541 * vdev becoming unusable, undo it and fail the request.
3543 vd
->vdev_offline
= B_TRUE
;
3546 if (!tvd
->vdev_islog
&& vd
->vdev_aux
== NULL
&&
3547 vdev_is_dead(tvd
)) {
3548 vd
->vdev_offline
= B_FALSE
;
3550 return (spa_vdev_state_exit(spa
, NULL
, EBUSY
));
3554 * Add the device back into the metaslab rotor so that
3555 * once we online the device it's open for business.
3557 if (tvd
->vdev_islog
&& mg
!= NULL
)
3558 metaslab_group_activate(mg
);
3561 vd
->vdev_tmpoffline
= !!(flags
& ZFS_OFFLINE_TEMPORARY
);
3563 return (spa_vdev_state_exit(spa
, vd
, 0));
3567 vdev_offline(spa_t
*spa
, uint64_t guid
, uint64_t flags
)
3571 mutex_enter(&spa
->spa_vdev_top_lock
);
3572 error
= vdev_offline_locked(spa
, guid
, flags
);
3573 mutex_exit(&spa
->spa_vdev_top_lock
);
3579 * Clear the error counts associated with this vdev. Unlike vdev_online() and
3580 * vdev_offline(), we assume the spa config is locked. We also clear all
3581 * children. If 'vd' is NULL, then the user wants to clear all vdevs.
3584 vdev_clear(spa_t
*spa
, vdev_t
*vd
)
3586 vdev_t
*rvd
= spa
->spa_root_vdev
;
3588 ASSERT(spa_config_held(spa
, SCL_STATE_ALL
, RW_WRITER
) == SCL_STATE_ALL
);
3593 vd
->vdev_stat
.vs_read_errors
= 0;
3594 vd
->vdev_stat
.vs_write_errors
= 0;
3595 vd
->vdev_stat
.vs_checksum_errors
= 0;
3596 vd
->vdev_stat
.vs_slow_ios
= 0;
3598 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3599 vdev_clear(spa
, vd
->vdev_child
[c
]);
3602 * It makes no sense to "clear" an indirect vdev.
3604 if (!vdev_is_concrete(vd
))
3608 * If we're in the FAULTED state or have experienced failed I/O, then
3609 * clear the persistent state and attempt to reopen the device. We
3610 * also mark the vdev config dirty, so that the new faulted state is
3611 * written out to disk.
3613 if (vd
->vdev_faulted
|| vd
->vdev_degraded
||
3614 !vdev_readable(vd
) || !vdev_writeable(vd
)) {
3616 * When reopening in response to a clear event, it may be due to
3617 * a fmadm repair request. In this case, if the device is
3618 * still broken, we want to still post the ereport again.
3620 vd
->vdev_forcefault
= B_TRUE
;
3622 vd
->vdev_faulted
= vd
->vdev_degraded
= 0ULL;
3623 vd
->vdev_cant_read
= B_FALSE
;
3624 vd
->vdev_cant_write
= B_FALSE
;
3625 vd
->vdev_stat
.vs_aux
= 0;
3627 vdev_reopen(vd
== rvd
? rvd
: vd
->vdev_top
);
3629 vd
->vdev_forcefault
= B_FALSE
;
3631 if (vd
!= rvd
&& vdev_writeable(vd
->vdev_top
))
3632 vdev_state_dirty(vd
->vdev_top
);
3634 if (vd
->vdev_aux
== NULL
&& !vdev_is_dead(vd
)) {
3635 if (dsl_scan_resilvering(spa
->spa_dsl_pool
) &&
3636 spa_feature_is_enabled(spa
,
3637 SPA_FEATURE_RESILVER_DEFER
))
3638 vdev_set_deferred_resilver(spa
, vd
);
3640 spa_async_request(spa
, SPA_ASYNC_RESILVER
);
3643 spa_event_notify(spa
, vd
, NULL
, ESC_ZFS_VDEV_CLEAR
);
3647 * When clearing a FMA-diagnosed fault, we always want to
3648 * unspare the device, as we assume that the original spare was
3649 * done in response to the FMA fault.
3651 if (!vdev_is_dead(vd
) && vd
->vdev_parent
!= NULL
&&
3652 vd
->vdev_parent
->vdev_ops
== &vdev_spare_ops
&&
3653 vd
->vdev_parent
->vdev_child
[0] == vd
)
3654 vd
->vdev_unspare
= B_TRUE
;
3658 vdev_is_dead(vdev_t
*vd
)
3661 * Holes and missing devices are always considered "dead".
3662 * This simplifies the code since we don't have to check for
3663 * these types of devices in the various code paths.
3664 * Instead we rely on the fact that we skip over dead devices
3665 * before issuing I/O to them.
3667 return (vd
->vdev_state
< VDEV_STATE_DEGRADED
||
3668 vd
->vdev_ops
== &vdev_hole_ops
||
3669 vd
->vdev_ops
== &vdev_missing_ops
);
3673 vdev_readable(vdev_t
*vd
)
3675 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_read
);
3679 vdev_writeable(vdev_t
*vd
)
3681 return (!vdev_is_dead(vd
) && !vd
->vdev_cant_write
&&
3682 vdev_is_concrete(vd
));
3686 vdev_allocatable(vdev_t
*vd
)
3688 uint64_t state
= vd
->vdev_state
;
3691 * We currently allow allocations from vdevs which may be in the
3692 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
3693 * fails to reopen then we'll catch it later when we're holding
3694 * the proper locks. Note that we have to get the vdev state
3695 * in a local variable because although it changes atomically,
3696 * we're asking two separate questions about it.
3698 return (!(state
< VDEV_STATE_DEGRADED
&& state
!= VDEV_STATE_CLOSED
) &&
3699 !vd
->vdev_cant_write
&& vdev_is_concrete(vd
) &&
3700 vd
->vdev_mg
->mg_initialized
);
3704 vdev_accessible(vdev_t
*vd
, zio_t
*zio
)
3706 ASSERT(zio
->io_vd
== vd
);
3708 if (vdev_is_dead(vd
) || vd
->vdev_remove_wanted
)
3711 if (zio
->io_type
== ZIO_TYPE_READ
)
3712 return (!vd
->vdev_cant_read
);
3714 if (zio
->io_type
== ZIO_TYPE_WRITE
)
3715 return (!vd
->vdev_cant_write
);
3721 vdev_get_child_stat(vdev_t
*cvd
, vdev_stat_t
*vs
, vdev_stat_t
*cvs
)
3724 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3725 vs
->vs_ops
[t
] += cvs
->vs_ops
[t
];
3726 vs
->vs_bytes
[t
] += cvs
->vs_bytes
[t
];
3729 cvs
->vs_scan_removing
= cvd
->vdev_removing
;
3733 * Get extended stats
3736 vdev_get_child_stat_ex(vdev_t
*cvd
, vdev_stat_ex_t
*vsx
, vdev_stat_ex_t
*cvsx
)
3739 for (t
= 0; t
< ZIO_TYPES
; t
++) {
3740 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_disk_histo
[0]); b
++)
3741 vsx
->vsx_disk_histo
[t
][b
] += cvsx
->vsx_disk_histo
[t
][b
];
3743 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_total_histo
[0]); b
++) {
3744 vsx
->vsx_total_histo
[t
][b
] +=
3745 cvsx
->vsx_total_histo
[t
][b
];
3749 for (t
= 0; t
< ZIO_PRIORITY_NUM_QUEUEABLE
; t
++) {
3750 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_queue_histo
[0]); b
++) {
3751 vsx
->vsx_queue_histo
[t
][b
] +=
3752 cvsx
->vsx_queue_histo
[t
][b
];
3754 vsx
->vsx_active_queue
[t
] += cvsx
->vsx_active_queue
[t
];
3755 vsx
->vsx_pend_queue
[t
] += cvsx
->vsx_pend_queue
[t
];
3757 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_ind_histo
[0]); b
++)
3758 vsx
->vsx_ind_histo
[t
][b
] += cvsx
->vsx_ind_histo
[t
][b
];
3760 for (b
= 0; b
< ARRAY_SIZE(vsx
->vsx_agg_histo
[0]); b
++)
3761 vsx
->vsx_agg_histo
[t
][b
] += cvsx
->vsx_agg_histo
[t
][b
];
3767 vdev_is_spacemap_addressable(vdev_t
*vd
)
3770 * Assuming 47 bits of the space map entry dedicated for the entry's
3771 * offset (see description in space_map.h), we calculate the maximum
3772 * address that can be described by a space map entry for the given
3775 uint64_t shift
= vd
->vdev_ashift
+ 47;
3777 if (shift
>= 63) /* detect potential overflow */
3780 return (vd
->vdev_asize
< (1ULL << shift
));
3784 * Get statistics for the given vdev.
3787 vdev_get_stats_ex_impl(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3791 * If we're getting stats on the root vdev, aggregate the I/O counts
3792 * over all top-level vdevs (i.e. the direct children of the root).
3794 if (!vd
->vdev_ops
->vdev_op_leaf
) {
3796 memset(vs
->vs_ops
, 0, sizeof (vs
->vs_ops
));
3797 memset(vs
->vs_bytes
, 0, sizeof (vs
->vs_bytes
));
3800 memset(vsx
, 0, sizeof (*vsx
));
3802 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
3803 vdev_t
*cvd
= vd
->vdev_child
[c
];
3804 vdev_stat_t
*cvs
= &cvd
->vdev_stat
;
3805 vdev_stat_ex_t
*cvsx
= &cvd
->vdev_stat_ex
;
3807 vdev_get_stats_ex_impl(cvd
, cvs
, cvsx
);
3809 vdev_get_child_stat(cvd
, vs
, cvs
);
3811 vdev_get_child_stat_ex(cvd
, vsx
, cvsx
);
3816 * We're a leaf. Just copy our ZIO active queue stats in. The
3817 * other leaf stats are updated in vdev_stat_update().
3822 memcpy(vsx
, &vd
->vdev_stat_ex
, sizeof (vd
->vdev_stat_ex
));
3824 for (t
= 0; t
< ARRAY_SIZE(vd
->vdev_queue
.vq_class
); t
++) {
3825 vsx
->vsx_active_queue
[t
] =
3826 vd
->vdev_queue
.vq_class
[t
].vqc_active
;
3827 vsx
->vsx_pend_queue
[t
] = avl_numnodes(
3828 &vd
->vdev_queue
.vq_class
[t
].vqc_queued_tree
);
3834 vdev_get_stats_ex(vdev_t
*vd
, vdev_stat_t
*vs
, vdev_stat_ex_t
*vsx
)
3836 vdev_t
*tvd
= vd
->vdev_top
;
3837 mutex_enter(&vd
->vdev_stat_lock
);
3839 bcopy(&vd
->vdev_stat
, vs
, sizeof (*vs
));
3840 vs
->vs_timestamp
= gethrtime() - vs
->vs_timestamp
;
3841 vs
->vs_state
= vd
->vdev_state
;
3842 vs
->vs_rsize
= vdev_get_min_asize(vd
);
3843 if (vd
->vdev_ops
->vdev_op_leaf
)
3844 vs
->vs_rsize
+= VDEV_LABEL_START_SIZE
+
3845 VDEV_LABEL_END_SIZE
;
3847 * Report expandable space on top-level, non-auxillary devices
3848 * only. The expandable space is reported in terms of metaslab
3849 * sized units since that determines how much space the pool
3852 if (vd
->vdev_aux
== NULL
&& tvd
!= NULL
) {
3853 vs
->vs_esize
= P2ALIGN(
3854 vd
->vdev_max_asize
- vd
->vdev_asize
,
3855 1ULL << tvd
->vdev_ms_shift
);
3857 if (vd
->vdev_aux
== NULL
&& vd
== vd
->vdev_top
&&
3858 vdev_is_concrete(vd
)) {
3859 vs
->vs_fragmentation
= (vd
->vdev_mg
!= NULL
) ?
3860 vd
->vdev_mg
->mg_fragmentation
: 0;
3862 if (vd
->vdev_ops
->vdev_op_leaf
)
3863 vs
->vs_resilver_deferred
= vd
->vdev_resilver_deferred
;
3866 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_READER
) != 0);
3867 vdev_get_stats_ex_impl(vd
, vs
, vsx
);
3868 mutex_exit(&vd
->vdev_stat_lock
);
3872 vdev_get_stats(vdev_t
*vd
, vdev_stat_t
*vs
)
3874 return (vdev_get_stats_ex(vd
, vs
, NULL
));
3878 vdev_clear_stats(vdev_t
*vd
)
3880 mutex_enter(&vd
->vdev_stat_lock
);
3881 vd
->vdev_stat
.vs_space
= 0;
3882 vd
->vdev_stat
.vs_dspace
= 0;
3883 vd
->vdev_stat
.vs_alloc
= 0;
3884 mutex_exit(&vd
->vdev_stat_lock
);
3888 vdev_scan_stat_init(vdev_t
*vd
)
3890 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3892 for (int c
= 0; c
< vd
->vdev_children
; c
++)
3893 vdev_scan_stat_init(vd
->vdev_child
[c
]);
3895 mutex_enter(&vd
->vdev_stat_lock
);
3896 vs
->vs_scan_processed
= 0;
3897 mutex_exit(&vd
->vdev_stat_lock
);
3901 vdev_stat_update(zio_t
*zio
, uint64_t psize
)
3903 spa_t
*spa
= zio
->io_spa
;
3904 vdev_t
*rvd
= spa
->spa_root_vdev
;
3905 vdev_t
*vd
= zio
->io_vd
? zio
->io_vd
: rvd
;
3907 uint64_t txg
= zio
->io_txg
;
3908 vdev_stat_t
*vs
= &vd
->vdev_stat
;
3909 vdev_stat_ex_t
*vsx
= &vd
->vdev_stat_ex
;
3910 zio_type_t type
= zio
->io_type
;
3911 int flags
= zio
->io_flags
;
3914 * If this i/o is a gang leader, it didn't do any actual work.
3916 if (zio
->io_gang_tree
)
3919 if (zio
->io_error
== 0) {
3921 * If this is a root i/o, don't count it -- we've already
3922 * counted the top-level vdevs, and vdev_get_stats() will
3923 * aggregate them when asked. This reduces contention on
3924 * the root vdev_stat_lock and implicitly handles blocks
3925 * that compress away to holes, for which there is no i/o.
3926 * (Holes never create vdev children, so all the counters
3927 * remain zero, which is what we want.)
3929 * Note: this only applies to successful i/o (io_error == 0)
3930 * because unlike i/o counts, errors are not additive.
3931 * When reading a ditto block, for example, failure of
3932 * one top-level vdev does not imply a root-level error.
3937 ASSERT(vd
== zio
->io_vd
);
3939 if (flags
& ZIO_FLAG_IO_BYPASS
)
3942 mutex_enter(&vd
->vdev_stat_lock
);
3944 if (flags
& ZIO_FLAG_IO_REPAIR
) {
3945 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
3946 dsl_scan_phys_t
*scn_phys
=
3947 &spa
->spa_dsl_pool
->dp_scan
->scn_phys
;
3948 uint64_t *processed
= &scn_phys
->scn_processed
;
3951 if (vd
->vdev_ops
->vdev_op_leaf
)
3952 atomic_add_64(processed
, psize
);
3953 vs
->vs_scan_processed
+= psize
;
3956 if (flags
& ZIO_FLAG_SELF_HEAL
)
3957 vs
->vs_self_healed
+= psize
;
3961 * The bytes/ops/histograms are recorded at the leaf level and
3962 * aggregated into the higher level vdevs in vdev_get_stats().
3964 if (vd
->vdev_ops
->vdev_op_leaf
&&
3965 (zio
->io_priority
< ZIO_PRIORITY_NUM_QUEUEABLE
)) {
3968 vs
->vs_bytes
[type
] += psize
;
3970 if (flags
& ZIO_FLAG_DELEGATED
) {
3971 vsx
->vsx_agg_histo
[zio
->io_priority
]
3972 [RQ_HISTO(zio
->io_size
)]++;
3974 vsx
->vsx_ind_histo
[zio
->io_priority
]
3975 [RQ_HISTO(zio
->io_size
)]++;
3978 if (zio
->io_delta
&& zio
->io_delay
) {
3979 vsx
->vsx_queue_histo
[zio
->io_priority
]
3980 [L_HISTO(zio
->io_delta
- zio
->io_delay
)]++;
3981 vsx
->vsx_disk_histo
[type
]
3982 [L_HISTO(zio
->io_delay
)]++;
3983 vsx
->vsx_total_histo
[type
]
3984 [L_HISTO(zio
->io_delta
)]++;
3988 mutex_exit(&vd
->vdev_stat_lock
);
3992 if (flags
& ZIO_FLAG_SPECULATIVE
)
3996 * If this is an I/O error that is going to be retried, then ignore the
3997 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as
3998 * hard errors, when in reality they can happen for any number of
3999 * innocuous reasons (bus resets, MPxIO link failure, etc).
4001 if (zio
->io_error
== EIO
&&
4002 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
))
4006 * Intent logs writes won't propagate their error to the root
4007 * I/O so don't mark these types of failures as pool-level
4010 if (zio
->io_vd
== NULL
&& (zio
->io_flags
& ZIO_FLAG_DONT_PROPAGATE
))
4013 mutex_enter(&vd
->vdev_stat_lock
);
4014 if (type
== ZIO_TYPE_READ
&& !vdev_is_dead(vd
)) {
4015 if (zio
->io_error
== ECKSUM
)
4016 vs
->vs_checksum_errors
++;
4018 vs
->vs_read_errors
++;
4020 if (type
== ZIO_TYPE_WRITE
&& !vdev_is_dead(vd
))
4021 vs
->vs_write_errors
++;
4022 mutex_exit(&vd
->vdev_stat_lock
);
4024 if (spa
->spa_load_state
== SPA_LOAD_NONE
&&
4025 type
== ZIO_TYPE_WRITE
&& txg
!= 0 &&
4026 (!(flags
& ZIO_FLAG_IO_REPAIR
) ||
4027 (flags
& ZIO_FLAG_SCAN_THREAD
) ||
4028 spa
->spa_claiming
)) {
4030 * This is either a normal write (not a repair), or it's
4031 * a repair induced by the scrub thread, or it's a repair
4032 * made by zil_claim() during spa_load() in the first txg.
4033 * In the normal case, we commit the DTL change in the same
4034 * txg as the block was born. In the scrub-induced repair
4035 * case, we know that scrubs run in first-pass syncing context,
4036 * so we commit the DTL change in spa_syncing_txg(spa).
4037 * In the zil_claim() case, we commit in spa_first_txg(spa).
4039 * We currently do not make DTL entries for failed spontaneous
4040 * self-healing writes triggered by normal (non-scrubbing)
4041 * reads, because we have no transactional context in which to
4042 * do so -- and it's not clear that it'd be desirable anyway.
4044 if (vd
->vdev_ops
->vdev_op_leaf
) {
4045 uint64_t commit_txg
= txg
;
4046 if (flags
& ZIO_FLAG_SCAN_THREAD
) {
4047 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4048 ASSERT(spa_sync_pass(spa
) == 1);
4049 vdev_dtl_dirty(vd
, DTL_SCRUB
, txg
, 1);
4050 commit_txg
= spa_syncing_txg(spa
);
4051 } else if (spa
->spa_claiming
) {
4052 ASSERT(flags
& ZIO_FLAG_IO_REPAIR
);
4053 commit_txg
= spa_first_txg(spa
);
4055 ASSERT(commit_txg
>= spa_syncing_txg(spa
));
4056 if (vdev_dtl_contains(vd
, DTL_MISSING
, txg
, 1))
4058 for (pvd
= vd
; pvd
!= rvd
; pvd
= pvd
->vdev_parent
)
4059 vdev_dtl_dirty(pvd
, DTL_PARTIAL
, txg
, 1);
4060 vdev_dirty(vd
->vdev_top
, VDD_DTL
, vd
, commit_txg
);
4063 vdev_dtl_dirty(vd
, DTL_MISSING
, txg
, 1);
4068 vdev_deflated_space(vdev_t
*vd
, int64_t space
)
4070 ASSERT((space
& (SPA_MINBLOCKSIZE
-1)) == 0);
4071 ASSERT(vd
->vdev_deflate_ratio
!= 0 || vd
->vdev_isl2cache
);
4073 return ((space
>> SPA_MINBLOCKSHIFT
) * vd
->vdev_deflate_ratio
);
4077 * Update the in-core space usage stats for this vdev and the root vdev.
4080 vdev_space_update(vdev_t
*vd
, int64_t alloc_delta
, int64_t defer_delta
,
4081 int64_t space_delta
)
4083 int64_t dspace_delta
;
4084 spa_t
*spa
= vd
->vdev_spa
;
4085 vdev_t
*rvd
= spa
->spa_root_vdev
;
4087 ASSERT(vd
== vd
->vdev_top
);
4090 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
4091 * factor. We must calculate this here and not at the root vdev
4092 * because the root vdev's psize-to-asize is simply the max of its
4093 * childrens', thus not accurate enough for us.
4095 dspace_delta
= vdev_deflated_space(vd
, space_delta
);
4097 mutex_enter(&vd
->vdev_stat_lock
);
4098 vd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4099 vd
->vdev_stat
.vs_space
+= space_delta
;
4100 vd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4101 mutex_exit(&vd
->vdev_stat_lock
);
4103 /* every class but log contributes to root space stats */
4104 if (vd
->vdev_mg
!= NULL
&& !vd
->vdev_islog
) {
4105 mutex_enter(&rvd
->vdev_stat_lock
);
4106 rvd
->vdev_stat
.vs_alloc
+= alloc_delta
;
4107 rvd
->vdev_stat
.vs_space
+= space_delta
;
4108 rvd
->vdev_stat
.vs_dspace
+= dspace_delta
;
4109 mutex_exit(&rvd
->vdev_stat_lock
);
4111 /* Note: metaslab_class_space_update moved to metaslab_space_update */
4115 * Mark a top-level vdev's config as dirty, placing it on the dirty list
4116 * so that it will be written out next time the vdev configuration is synced.
4117 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
4120 vdev_config_dirty(vdev_t
*vd
)
4122 spa_t
*spa
= vd
->vdev_spa
;
4123 vdev_t
*rvd
= spa
->spa_root_vdev
;
4126 ASSERT(spa_writeable(spa
));
4129 * If this is an aux vdev (as with l2cache and spare devices), then we
4130 * update the vdev config manually and set the sync flag.
4132 if (vd
->vdev_aux
!= NULL
) {
4133 spa_aux_vdev_t
*sav
= vd
->vdev_aux
;
4137 for (c
= 0; c
< sav
->sav_count
; c
++) {
4138 if (sav
->sav_vdevs
[c
] == vd
)
4142 if (c
== sav
->sav_count
) {
4144 * We're being removed. There's nothing more to do.
4146 ASSERT(sav
->sav_sync
== B_TRUE
);
4150 sav
->sav_sync
= B_TRUE
;
4152 if (nvlist_lookup_nvlist_array(sav
->sav_config
,
4153 ZPOOL_CONFIG_L2CACHE
, &aux
, &naux
) != 0) {
4154 VERIFY(nvlist_lookup_nvlist_array(sav
->sav_config
,
4155 ZPOOL_CONFIG_SPARES
, &aux
, &naux
) == 0);
4161 * Setting the nvlist in the middle if the array is a little
4162 * sketchy, but it will work.
4164 nvlist_free(aux
[c
]);
4165 aux
[c
] = vdev_config_generate(spa
, vd
, B_TRUE
, 0);
4171 * The dirty list is protected by the SCL_CONFIG lock. The caller
4172 * must either hold SCL_CONFIG as writer, or must be the sync thread
4173 * (which holds SCL_CONFIG as reader). There's only one sync thread,
4174 * so this is sufficient to ensure mutual exclusion.
4176 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4177 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4178 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4181 for (c
= 0; c
< rvd
->vdev_children
; c
++)
4182 vdev_config_dirty(rvd
->vdev_child
[c
]);
4184 ASSERT(vd
== vd
->vdev_top
);
4186 if (!list_link_active(&vd
->vdev_config_dirty_node
) &&
4187 vdev_is_concrete(vd
)) {
4188 list_insert_head(&spa
->spa_config_dirty_list
, vd
);
4194 vdev_config_clean(vdev_t
*vd
)
4196 spa_t
*spa
= vd
->vdev_spa
;
4198 ASSERT(spa_config_held(spa
, SCL_CONFIG
, RW_WRITER
) ||
4199 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4200 spa_config_held(spa
, SCL_CONFIG
, RW_READER
)));
4202 ASSERT(list_link_active(&vd
->vdev_config_dirty_node
));
4203 list_remove(&spa
->spa_config_dirty_list
, vd
);
4207 * Mark a top-level vdev's state as dirty, so that the next pass of
4208 * spa_sync() can convert this into vdev_config_dirty(). We distinguish
4209 * the state changes from larger config changes because they require
4210 * much less locking, and are often needed for administrative actions.
4213 vdev_state_dirty(vdev_t
*vd
)
4215 spa_t
*spa
= vd
->vdev_spa
;
4217 ASSERT(spa_writeable(spa
));
4218 ASSERT(vd
== vd
->vdev_top
);
4221 * The state list is protected by the SCL_STATE lock. The caller
4222 * must either hold SCL_STATE as writer, or must be the sync thread
4223 * (which holds SCL_STATE as reader). There's only one sync thread,
4224 * so this is sufficient to ensure mutual exclusion.
4226 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4227 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4228 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4230 if (!list_link_active(&vd
->vdev_state_dirty_node
) &&
4231 vdev_is_concrete(vd
))
4232 list_insert_head(&spa
->spa_state_dirty_list
, vd
);
4236 vdev_state_clean(vdev_t
*vd
)
4238 spa_t
*spa
= vd
->vdev_spa
;
4240 ASSERT(spa_config_held(spa
, SCL_STATE
, RW_WRITER
) ||
4241 (dsl_pool_sync_context(spa_get_dsl(spa
)) &&
4242 spa_config_held(spa
, SCL_STATE
, RW_READER
)));
4244 ASSERT(list_link_active(&vd
->vdev_state_dirty_node
));
4245 list_remove(&spa
->spa_state_dirty_list
, vd
);
4249 * Propagate vdev state up from children to parent.
4252 vdev_propagate_state(vdev_t
*vd
)
4254 spa_t
*spa
= vd
->vdev_spa
;
4255 vdev_t
*rvd
= spa
->spa_root_vdev
;
4256 int degraded
= 0, faulted
= 0;
4260 if (vd
->vdev_children
> 0) {
4261 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4262 child
= vd
->vdev_child
[c
];
4265 * Don't factor holes or indirect vdevs into the
4268 if (!vdev_is_concrete(child
))
4271 if (!vdev_readable(child
) ||
4272 (!vdev_writeable(child
) && spa_writeable(spa
))) {
4274 * Root special: if there is a top-level log
4275 * device, treat the root vdev as if it were
4278 if (child
->vdev_islog
&& vd
== rvd
)
4282 } else if (child
->vdev_state
<= VDEV_STATE_DEGRADED
) {
4286 if (child
->vdev_stat
.vs_aux
== VDEV_AUX_CORRUPT_DATA
)
4290 vd
->vdev_ops
->vdev_op_state_change(vd
, faulted
, degraded
);
4293 * Root special: if there is a top-level vdev that cannot be
4294 * opened due to corrupted metadata, then propagate the root
4295 * vdev's aux state as 'corrupt' rather than 'insufficient
4298 if (corrupted
&& vd
== rvd
&&
4299 rvd
->vdev_state
== VDEV_STATE_CANT_OPEN
)
4300 vdev_set_state(rvd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
4301 VDEV_AUX_CORRUPT_DATA
);
4304 if (vd
->vdev_parent
)
4305 vdev_propagate_state(vd
->vdev_parent
);
4309 * Set a vdev's state. If this is during an open, we don't update the parent
4310 * state, because we're in the process of opening children depth-first.
4311 * Otherwise, we propagate the change to the parent.
4313 * If this routine places a device in a faulted state, an appropriate ereport is
4317 vdev_set_state(vdev_t
*vd
, boolean_t isopen
, vdev_state_t state
, vdev_aux_t aux
)
4319 uint64_t save_state
;
4320 spa_t
*spa
= vd
->vdev_spa
;
4322 if (state
== vd
->vdev_state
) {
4324 * Since vdev_offline() code path is already in an offline
4325 * state we can miss a statechange event to OFFLINE. Check
4326 * the previous state to catch this condition.
4328 if (vd
->vdev_ops
->vdev_op_leaf
&&
4329 (state
== VDEV_STATE_OFFLINE
) &&
4330 (vd
->vdev_prevstate
>= VDEV_STATE_FAULTED
)) {
4331 /* post an offline state change */
4332 zfs_post_state_change(spa
, vd
, vd
->vdev_prevstate
);
4334 vd
->vdev_stat
.vs_aux
= aux
;
4338 save_state
= vd
->vdev_state
;
4340 vd
->vdev_state
= state
;
4341 vd
->vdev_stat
.vs_aux
= aux
;
4344 * If we are setting the vdev state to anything but an open state, then
4345 * always close the underlying device unless the device has requested
4346 * a delayed close (i.e. we're about to remove or fault the device).
4347 * Otherwise, we keep accessible but invalid devices open forever.
4348 * We don't call vdev_close() itself, because that implies some extra
4349 * checks (offline, etc) that we don't want here. This is limited to
4350 * leaf devices, because otherwise closing the device will affect other
4353 if (!vd
->vdev_delayed_close
&& vdev_is_dead(vd
) &&
4354 vd
->vdev_ops
->vdev_op_leaf
)
4355 vd
->vdev_ops
->vdev_op_close(vd
);
4357 if (vd
->vdev_removed
&&
4358 state
== VDEV_STATE_CANT_OPEN
&&
4359 (aux
== VDEV_AUX_OPEN_FAILED
|| vd
->vdev_checkremove
)) {
4361 * If the previous state is set to VDEV_STATE_REMOVED, then this
4362 * device was previously marked removed and someone attempted to
4363 * reopen it. If this failed due to a nonexistent device, then
4364 * keep the device in the REMOVED state. We also let this be if
4365 * it is one of our special test online cases, which is only
4366 * attempting to online the device and shouldn't generate an FMA
4369 vd
->vdev_state
= VDEV_STATE_REMOVED
;
4370 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NONE
;
4371 } else if (state
== VDEV_STATE_REMOVED
) {
4372 vd
->vdev_removed
= B_TRUE
;
4373 } else if (state
== VDEV_STATE_CANT_OPEN
) {
4375 * If we fail to open a vdev during an import or recovery, we
4376 * mark it as "not available", which signifies that it was
4377 * never there to begin with. Failure to open such a device
4378 * is not considered an error.
4380 if ((spa_load_state(spa
) == SPA_LOAD_IMPORT
||
4381 spa_load_state(spa
) == SPA_LOAD_RECOVER
) &&
4382 vd
->vdev_ops
->vdev_op_leaf
)
4383 vd
->vdev_not_present
= 1;
4386 * Post the appropriate ereport. If the 'prevstate' field is
4387 * set to something other than VDEV_STATE_UNKNOWN, it indicates
4388 * that this is part of a vdev_reopen(). In this case, we don't
4389 * want to post the ereport if the device was already in the
4390 * CANT_OPEN state beforehand.
4392 * If the 'checkremove' flag is set, then this is an attempt to
4393 * online the device in response to an insertion event. If we
4394 * hit this case, then we have detected an insertion event for a
4395 * faulted or offline device that wasn't in the removed state.
4396 * In this scenario, we don't post an ereport because we are
4397 * about to replace the device, or attempt an online with
4398 * vdev_forcefault, which will generate the fault for us.
4400 if ((vd
->vdev_prevstate
!= state
|| vd
->vdev_forcefault
) &&
4401 !vd
->vdev_not_present
&& !vd
->vdev_checkremove
&&
4402 vd
!= spa
->spa_root_vdev
) {
4406 case VDEV_AUX_OPEN_FAILED
:
4407 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED
;
4409 case VDEV_AUX_CORRUPT_DATA
:
4410 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA
;
4412 case VDEV_AUX_NO_REPLICAS
:
4413 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS
;
4415 case VDEV_AUX_BAD_GUID_SUM
:
4416 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM
;
4418 case VDEV_AUX_TOO_SMALL
:
4419 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL
;
4421 case VDEV_AUX_BAD_LABEL
:
4422 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL
;
4424 case VDEV_AUX_BAD_ASHIFT
:
4425 class = FM_EREPORT_ZFS_DEVICE_BAD_ASHIFT
;
4428 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN
;
4431 zfs_ereport_post(class, spa
, vd
, NULL
, NULL
,
4435 /* Erase any notion of persistent removed state */
4436 vd
->vdev_removed
= B_FALSE
;
4438 vd
->vdev_removed
= B_FALSE
;
4442 * Notify ZED of any significant state-change on a leaf vdev.
4445 if (vd
->vdev_ops
->vdev_op_leaf
) {
4446 /* preserve original state from a vdev_reopen() */
4447 if ((vd
->vdev_prevstate
!= VDEV_STATE_UNKNOWN
) &&
4448 (vd
->vdev_prevstate
!= vd
->vdev_state
) &&
4449 (save_state
<= VDEV_STATE_CLOSED
))
4450 save_state
= vd
->vdev_prevstate
;
4452 /* filter out state change due to initial vdev_open */
4453 if (save_state
> VDEV_STATE_CLOSED
)
4454 zfs_post_state_change(spa
, vd
, save_state
);
4457 if (!isopen
&& vd
->vdev_parent
)
4458 vdev_propagate_state(vd
->vdev_parent
);
4462 vdev_children_are_offline(vdev_t
*vd
)
4464 ASSERT(!vd
->vdev_ops
->vdev_op_leaf
);
4466 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++) {
4467 if (vd
->vdev_child
[i
]->vdev_state
!= VDEV_STATE_OFFLINE
)
4475 * Check the vdev configuration to ensure that it's capable of supporting
4476 * a root pool. We do not support partial configuration.
4479 vdev_is_bootable(vdev_t
*vd
)
4481 if (!vd
->vdev_ops
->vdev_op_leaf
) {
4482 const char *vdev_type
= vd
->vdev_ops
->vdev_op_type
;
4484 if (strcmp(vdev_type
, VDEV_TYPE_MISSING
) == 0 ||
4485 strcmp(vdev_type
, VDEV_TYPE_INDIRECT
) == 0) {
4490 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4491 if (!vdev_is_bootable(vd
->vdev_child
[c
]))
4498 vdev_is_concrete(vdev_t
*vd
)
4500 vdev_ops_t
*ops
= vd
->vdev_ops
;
4501 if (ops
== &vdev_indirect_ops
|| ops
== &vdev_hole_ops
||
4502 ops
== &vdev_missing_ops
|| ops
== &vdev_root_ops
) {
4510 * Determine if a log device has valid content. If the vdev was
4511 * removed or faulted in the MOS config then we know that
4512 * the content on the log device has already been written to the pool.
4515 vdev_log_state_valid(vdev_t
*vd
)
4517 if (vd
->vdev_ops
->vdev_op_leaf
&& !vd
->vdev_faulted
&&
4521 for (int c
= 0; c
< vd
->vdev_children
; c
++)
4522 if (vdev_log_state_valid(vd
->vdev_child
[c
]))
4529 * Expand a vdev if possible.
4532 vdev_expand(vdev_t
*vd
, uint64_t txg
)
4534 ASSERT(vd
->vdev_top
== vd
);
4535 ASSERT(spa_config_held(vd
->vdev_spa
, SCL_ALL
, RW_WRITER
) == SCL_ALL
);
4536 ASSERT(vdev_is_concrete(vd
));
4538 vdev_set_deflate_ratio(vd
);
4540 if ((vd
->vdev_asize
>> vd
->vdev_ms_shift
) > vd
->vdev_ms_count
&&
4541 vdev_is_concrete(vd
)) {
4542 vdev_metaslab_group_create(vd
);
4543 VERIFY(vdev_metaslab_init(vd
, txg
) == 0);
4544 vdev_config_dirty(vd
);
4552 vdev_split(vdev_t
*vd
)
4554 vdev_t
*cvd
, *pvd
= vd
->vdev_parent
;
4556 vdev_remove_child(pvd
, vd
);
4557 vdev_compact_children(pvd
);
4559 cvd
= pvd
->vdev_child
[0];
4560 if (pvd
->vdev_children
== 1) {
4561 vdev_remove_parent(cvd
);
4562 cvd
->vdev_splitting
= B_TRUE
;
4564 vdev_propagate_state(cvd
);
4568 vdev_deadman(vdev_t
*vd
, char *tag
)
4570 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
4571 vdev_t
*cvd
= vd
->vdev_child
[c
];
4573 vdev_deadman(cvd
, tag
);
4576 if (vd
->vdev_ops
->vdev_op_leaf
) {
4577 vdev_queue_t
*vq
= &vd
->vdev_queue
;
4579 mutex_enter(&vq
->vq_lock
);
4580 if (avl_numnodes(&vq
->vq_active_tree
) > 0) {
4581 spa_t
*spa
= vd
->vdev_spa
;
4585 zfs_dbgmsg("slow vdev: %s has %d active IOs",
4586 vd
->vdev_path
, avl_numnodes(&vq
->vq_active_tree
));
4589 * Look at the head of all the pending queues,
4590 * if any I/O has been outstanding for longer than
4591 * the spa_deadman_synctime invoke the deadman logic.
4593 fio
= avl_first(&vq
->vq_active_tree
);
4594 delta
= gethrtime() - fio
->io_timestamp
;
4595 if (delta
> spa_deadman_synctime(spa
))
4596 zio_deadman(fio
, tag
);
4598 mutex_exit(&vq
->vq_lock
);
4603 vdev_set_deferred_resilver(spa_t
*spa
, vdev_t
*vd
)
4605 for (uint64_t i
= 0; i
< vd
->vdev_children
; i
++)
4606 vdev_set_deferred_resilver(spa
, vd
->vdev_child
[i
]);
4608 if (!vd
->vdev_ops
->vdev_op_leaf
|| !vdev_writeable(vd
) ||
4609 range_tree_is_empty(vd
->vdev_dtl
[DTL_MISSING
])) {
4613 vd
->vdev_resilver_deferred
= B_TRUE
;
4614 spa
->spa_resilver_deferred
= B_TRUE
;
4617 #if defined(_KERNEL)
4618 EXPORT_SYMBOL(vdev_fault
);
4619 EXPORT_SYMBOL(vdev_degrade
);
4620 EXPORT_SYMBOL(vdev_online
);
4621 EXPORT_SYMBOL(vdev_offline
);
4622 EXPORT_SYMBOL(vdev_clear
);
4624 module_param(vdev_max_ms_count
, int, 0644);
4625 MODULE_PARM_DESC(vdev_max_ms_count
,
4626 "Target number of metaslabs per top-level vdev");
4628 module_param(vdev_min_ms_count
, int, 0644);
4629 MODULE_PARM_DESC(vdev_min_ms_count
,
4630 "Minimum number of metaslabs per top-level vdev");
4632 module_param(vdev_ms_count_limit
, int, 0644);
4633 MODULE_PARM_DESC(vdev_ms_count_limit
,
4634 "Practical upper limit of total metaslabs per top-level vdev");
4636 module_param(zfs_slow_io_events_per_second
, uint
, 0644);
4637 MODULE_PARM_DESC(zfs_slow_io_events_per_second
,
4638 "Rate limit slow IO (delay) events to this many per second");
4640 module_param(zfs_checksum_events_per_second
, uint
, 0644);
4641 MODULE_PARM_DESC(zfs_checksum_events_per_second
, "Rate limit checksum events "
4642 "to this many checksum errors per second (do not set below zed"
4645 module_param(zfs_scan_ignore_errors
, int, 0644);
4646 MODULE_PARM_DESC(zfs_scan_ignore_errors
,
4647 "Ignore errors during resilver/scrub");
4649 module_param(vdev_validate_skip
, int, 0644);
4650 MODULE_PARM_DESC(vdev_validate_skip
,
4651 "Bypass vdev_validate()");