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]
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
30 #include <sys/zfs_context.h>
32 #include <sys/spa_impl.h>
33 #include <sys/dsl_pool.h>
34 #include <sys/dsl_scan.h>
35 #include <sys/vdev_impl.h>
38 #include <sys/fs/zfs.h>
43 static kstat_t
*mirror_ksp
= NULL
;
45 typedef struct mirror_stats
{
46 kstat_named_t vdev_mirror_stat_rotating_linear
;
47 kstat_named_t vdev_mirror_stat_rotating_offset
;
48 kstat_named_t vdev_mirror_stat_rotating_seek
;
49 kstat_named_t vdev_mirror_stat_non_rotating_linear
;
50 kstat_named_t vdev_mirror_stat_non_rotating_seek
;
52 kstat_named_t vdev_mirror_stat_preferred_found
;
53 kstat_named_t vdev_mirror_stat_preferred_not_found
;
56 static mirror_stats_t mirror_stats
= {
57 /* New I/O follows directly the last I/O */
58 { "rotating_linear", KSTAT_DATA_UINT64
},
59 /* New I/O is within zfs_vdev_mirror_rotating_seek_offset of the last */
60 { "rotating_offset", KSTAT_DATA_UINT64
},
61 /* New I/O requires random seek */
62 { "rotating_seek", KSTAT_DATA_UINT64
},
63 /* New I/O follows directly the last I/O (nonrot) */
64 { "non_rotating_linear", KSTAT_DATA_UINT64
},
65 /* New I/O requires random seek (nonrot) */
66 { "non_rotating_seek", KSTAT_DATA_UINT64
},
67 /* Preferred child vdev found */
68 { "preferred_found", KSTAT_DATA_UINT64
},
69 /* Preferred child vdev not found or equal load */
70 { "preferred_not_found", KSTAT_DATA_UINT64
},
74 #define MIRROR_STAT(stat) (mirror_stats.stat.value.ui64)
75 #define MIRROR_INCR(stat, val) atomic_add_64(&MIRROR_STAT(stat), val)
76 #define MIRROR_BUMP(stat) MIRROR_INCR(stat, 1)
79 vdev_mirror_stat_init(void)
81 mirror_ksp
= kstat_create("zfs", 0, "vdev_mirror_stats",
82 "misc", KSTAT_TYPE_NAMED
,
83 sizeof (mirror_stats
) / sizeof (kstat_named_t
), KSTAT_FLAG_VIRTUAL
);
84 if (mirror_ksp
!= NULL
) {
85 mirror_ksp
->ks_data
= &mirror_stats
;
86 kstat_install(mirror_ksp
);
91 vdev_mirror_stat_fini(void)
93 if (mirror_ksp
!= NULL
) {
94 kstat_delete(mirror_ksp
);
100 * Virtual device vector for mirroring.
103 typedef struct mirror_child
{
110 uint8_t mc_speculative
;
113 typedef struct mirror_map
{
115 int mm_preferred_cnt
;
117 boolean_t mm_resilvering
;
119 mirror_child_t mm_child
[];
122 static int vdev_mirror_shift
= 21;
125 * The load configuration settings below are tuned by default for
126 * the case where all devices are of the same rotational type.
128 * If there is a mixture of rotating and non-rotating media, setting
129 * zfs_vdev_mirror_non_rotating_seek_inc to 0 may well provide better results
130 * as it will direct more reads to the non-rotating vdevs which are more likely
131 * to have a higher performance.
134 /* Rotating media load calculation configuration. */
135 static int zfs_vdev_mirror_rotating_inc
= 0;
136 static int zfs_vdev_mirror_rotating_seek_inc
= 5;
137 static int zfs_vdev_mirror_rotating_seek_offset
= 1 * 1024 * 1024;
139 /* Non-rotating media load calculation configuration. */
140 static int zfs_vdev_mirror_non_rotating_inc
= 0;
141 static int zfs_vdev_mirror_non_rotating_seek_inc
= 1;
144 vdev_mirror_map_size(int children
)
146 return (offsetof(mirror_map_t
, mm_child
[children
]) +
147 sizeof (int) * children
);
150 static inline mirror_map_t
*
151 vdev_mirror_map_alloc(int children
, boolean_t resilvering
, boolean_t root
)
155 mm
= kmem_zalloc(vdev_mirror_map_size(children
), KM_SLEEP
);
156 mm
->mm_children
= children
;
157 mm
->mm_resilvering
= resilvering
;
159 mm
->mm_preferred
= (int *)((uintptr_t)mm
+
160 offsetof(mirror_map_t
, mm_child
[children
]));
166 vdev_mirror_map_free(zio_t
*zio
)
168 mirror_map_t
*mm
= zio
->io_vsd
;
170 kmem_free(mm
, vdev_mirror_map_size(mm
->mm_children
));
173 static const zio_vsd_ops_t vdev_mirror_vsd_ops
= {
174 .vsd_free
= vdev_mirror_map_free
,
175 .vsd_cksum_report
= zio_vsd_default_cksum_report
179 vdev_mirror_load(mirror_map_t
*mm
, vdev_t
*vd
, uint64_t zio_offset
)
181 uint64_t last_offset
;
185 /* All DVAs have equal weight at the root. */
190 * We don't return INT_MAX if the device is resilvering i.e.
191 * vdev_resilver_txg != 0 as when tested performance was slightly
192 * worse overall when resilvering with compared to without.
195 /* Fix zio_offset for leaf vdevs */
196 if (vd
->vdev_ops
->vdev_op_leaf
)
197 zio_offset
+= VDEV_LABEL_START_SIZE
;
199 /* Standard load based on pending queue length. */
200 load
= vdev_queue_length(vd
);
201 last_offset
= vdev_queue_last_offset(vd
);
203 if (vd
->vdev_nonrot
) {
204 /* Non-rotating media. */
205 if (last_offset
== zio_offset
) {
206 MIRROR_BUMP(vdev_mirror_stat_non_rotating_linear
);
207 return (load
+ zfs_vdev_mirror_non_rotating_inc
);
211 * Apply a seek penalty even for non-rotating devices as
212 * sequential I/O's can be aggregated into fewer operations on
213 * the device, thus avoiding unnecessary per-command overhead
214 * and boosting performance.
216 MIRROR_BUMP(vdev_mirror_stat_non_rotating_seek
);
217 return (load
+ zfs_vdev_mirror_non_rotating_seek_inc
);
220 /* Rotating media I/O's which directly follow the last I/O. */
221 if (last_offset
== zio_offset
) {
222 MIRROR_BUMP(vdev_mirror_stat_rotating_linear
);
223 return (load
+ zfs_vdev_mirror_rotating_inc
);
227 * Apply half the seek increment to I/O's within seek offset
228 * of the last I/O issued to this vdev as they should incur less
229 * of a seek increment.
231 offset_diff
= (int64_t)(last_offset
- zio_offset
);
232 if (ABS(offset_diff
) < zfs_vdev_mirror_rotating_seek_offset
) {
233 MIRROR_BUMP(vdev_mirror_stat_rotating_offset
);
234 return (load
+ (zfs_vdev_mirror_rotating_seek_inc
/ 2));
237 /* Apply the full seek increment to all other I/O's. */
238 MIRROR_BUMP(vdev_mirror_stat_rotating_seek
);
239 return (load
+ zfs_vdev_mirror_rotating_seek_inc
);
243 * Avoid inlining the function to keep vdev_mirror_io_start(), which
244 * is this functions only caller, as small as possible on the stack.
246 noinline
static mirror_map_t
*
247 vdev_mirror_map_init(zio_t
*zio
)
249 mirror_map_t
*mm
= NULL
;
251 vdev_t
*vd
= zio
->io_vd
;
255 dva_t
*dva
= zio
->io_bp
->blk_dva
;
256 spa_t
*spa
= zio
->io_spa
;
257 dsl_scan_t
*scn
= spa
->spa_dsl_pool
->dp_scan
;
258 dva_t dva_copy
[SPA_DVAS_PER_BP
];
260 c
= BP_GET_NDVAS(zio
->io_bp
);
263 * The sequential scrub code sorts and issues all DVAs
264 * of a bp separately. Each of these IOs includes all
265 * original DVA copies so that repairs can be performed
266 * in the event of an error, but we only actually want
267 * to check the first DVA since the others will be
268 * checked by their respective sorted IOs. Only if we
269 * hit an error will we try all DVAs upon retrying.
271 * Note: This check is safe even if the user switches
272 * from a legacy scrub to a sequential one in the middle
273 * of processing, since scn_is_sorted isn't updated until
274 * all outstanding IOs from the previous scrub pass
277 if ((zio
->io_flags
& ZIO_FLAG_SCRUB
) &&
278 !(zio
->io_flags
& ZIO_FLAG_IO_RETRY
) &&
279 dsl_scan_scrubbing(spa
->spa_dsl_pool
) &&
280 scn
->scn_is_sorted
) {
283 c
= BP_GET_NDVAS(zio
->io_bp
);
287 * If we do not trust the pool config, some DVAs might be
288 * invalid or point to vdevs that do not exist. We skip them.
290 if (!spa_trust_config(spa
)) {
291 ASSERT3U(zio
->io_type
, ==, ZIO_TYPE_READ
);
293 for (int i
= 0; i
< c
; i
++) {
294 if (zfs_dva_valid(spa
, &dva
[i
], zio
->io_bp
))
295 dva_copy
[j
++] = dva
[i
];
299 zio
->io_error
= ENXIO
;
308 mm
= vdev_mirror_map_alloc(c
, B_FALSE
, B_TRUE
);
309 for (c
= 0; c
< mm
->mm_children
; c
++) {
310 mc
= &mm
->mm_child
[c
];
312 mc
->mc_vd
= vdev_lookup_top(spa
, DVA_GET_VDEV(&dva
[c
]));
313 mc
->mc_offset
= DVA_GET_OFFSET(&dva
[c
]);
317 * If we are resilvering, then we should handle scrub reads
318 * differently; we shouldn't issue them to the resilvering
319 * device because it might not have those blocks.
321 * We are resilvering iff:
322 * 1) We are a replacing vdev (ie our name is "replacing-1" or
323 * "spare-1" or something like that), and
324 * 2) The pool is currently being resilvered.
326 * We cannot simply check vd->vdev_resilver_txg, because it's
327 * not set in this path.
329 * Nor can we just check our vdev_ops; there are cases (such as
330 * when a user types "zpool replace pool odev spare_dev" and
331 * spare_dev is in the spare list, or when a spare device is
332 * automatically used to replace a DEGRADED device) when
333 * resilvering is complete but both the original vdev and the
334 * spare vdev remain in the pool. That behavior is intentional.
335 * It helps implement the policy that a spare should be
336 * automatically removed from the pool after the user replaces
337 * the device that originally failed.
339 * If a spa load is in progress, then spa_dsl_pool may be
340 * uninitialized. But we shouldn't be resilvering during a spa
343 boolean_t replacing
= (vd
->vdev_ops
== &vdev_replacing_ops
||
344 vd
->vdev_ops
== &vdev_spare_ops
) &&
345 spa_load_state(vd
->vdev_spa
) == SPA_LOAD_NONE
&&
346 dsl_scan_resilvering(vd
->vdev_spa
->spa_dsl_pool
);
347 mm
= vdev_mirror_map_alloc(vd
->vdev_children
, replacing
,
349 for (c
= 0; c
< mm
->mm_children
; c
++) {
350 mc
= &mm
->mm_child
[c
];
351 mc
->mc_vd
= vd
->vdev_child
[c
];
352 mc
->mc_offset
= zio
->io_offset
;
357 zio
->io_vsd_ops
= &vdev_mirror_vsd_ops
;
362 vdev_mirror_open(vdev_t
*vd
, uint64_t *asize
, uint64_t *max_asize
,
368 if (vd
->vdev_children
== 0) {
369 vd
->vdev_stat
.vs_aux
= VDEV_AUX_BAD_LABEL
;
370 return (SET_ERROR(EINVAL
));
373 vdev_open_children(vd
);
375 for (int c
= 0; c
< vd
->vdev_children
; c
++) {
376 vdev_t
*cvd
= vd
->vdev_child
[c
];
378 if (cvd
->vdev_open_error
) {
379 lasterror
= cvd
->vdev_open_error
;
384 *asize
= MIN(*asize
- 1, cvd
->vdev_asize
- 1) + 1;
385 *max_asize
= MIN(*max_asize
- 1, cvd
->vdev_max_asize
- 1) + 1;
386 *ashift
= MAX(*ashift
, cvd
->vdev_ashift
);
389 if (numerrors
== vd
->vdev_children
) {
390 if (vdev_children_are_offline(vd
))
391 vd
->vdev_stat
.vs_aux
= VDEV_AUX_CHILDREN_OFFLINE
;
393 vd
->vdev_stat
.vs_aux
= VDEV_AUX_NO_REPLICAS
;
401 vdev_mirror_close(vdev_t
*vd
)
403 for (int c
= 0; c
< vd
->vdev_children
; c
++)
404 vdev_close(vd
->vdev_child
[c
]);
408 vdev_mirror_child_done(zio_t
*zio
)
410 mirror_child_t
*mc
= zio
->io_private
;
412 mc
->mc_error
= zio
->io_error
;
418 vdev_mirror_scrub_done(zio_t
*zio
)
420 mirror_child_t
*mc
= zio
->io_private
;
422 if (zio
->io_error
== 0) {
424 zio_link_t
*zl
= NULL
;
426 mutex_enter(&zio
->io_lock
);
427 while ((pio
= zio_walk_parents(zio
, &zl
)) != NULL
) {
428 mutex_enter(&pio
->io_lock
);
429 ASSERT3U(zio
->io_size
, >=, pio
->io_size
);
430 abd_copy(pio
->io_abd
, zio
->io_abd
, pio
->io_size
);
431 mutex_exit(&pio
->io_lock
);
433 mutex_exit(&zio
->io_lock
);
436 abd_free(zio
->io_abd
);
438 mc
->mc_error
= zio
->io_error
;
444 * Check the other, lower-index DVAs to see if they're on the same
445 * vdev as the child we picked. If they are, use them since they
446 * are likely to have been allocated from the primary metaslab in
447 * use at the time, and hence are more likely to have locality with
451 vdev_mirror_dva_select(zio_t
*zio
, int p
)
453 dva_t
*dva
= zio
->io_bp
->blk_dva
;
454 mirror_map_t
*mm
= zio
->io_vsd
;
458 preferred
= mm
->mm_preferred
[p
];
459 for (p
--; p
>= 0; p
--) {
460 c
= mm
->mm_preferred
[p
];
461 if (DVA_GET_VDEV(&dva
[c
]) == DVA_GET_VDEV(&dva
[preferred
]))
468 vdev_mirror_preferred_child_randomize(zio_t
*zio
)
470 mirror_map_t
*mm
= zio
->io_vsd
;
474 p
= spa_get_random(mm
->mm_preferred_cnt
);
475 return (vdev_mirror_dva_select(zio
, p
));
479 * To ensure we don't always favour the first matching vdev,
480 * which could lead to wear leveling issues on SSD's, we
481 * use the I/O offset as a pseudo random seed into the vdevs
482 * which have the lowest load.
484 p
= (zio
->io_offset
>> vdev_mirror_shift
) % mm
->mm_preferred_cnt
;
485 return (mm
->mm_preferred
[p
]);
489 * Try to find a vdev whose DTL doesn't contain the block we want to read
490 * prefering vdevs based on determined load.
492 * Try to find a child whose DTL doesn't contain the block we want to read.
493 * If we can't, try the read on any vdev we haven't already tried.
496 vdev_mirror_child_select(zio_t
*zio
)
498 mirror_map_t
*mm
= zio
->io_vsd
;
499 uint64_t txg
= zio
->io_txg
;
502 ASSERT(zio
->io_bp
== NULL
|| BP_PHYSICAL_BIRTH(zio
->io_bp
) == txg
);
504 lowest_load
= INT_MAX
;
505 mm
->mm_preferred_cnt
= 0;
506 for (c
= 0; c
< mm
->mm_children
; c
++) {
509 mc
= &mm
->mm_child
[c
];
510 if (mc
->mc_tried
|| mc
->mc_skipped
)
513 if (mc
->mc_vd
== NULL
|| !vdev_readable(mc
->mc_vd
)) {
514 mc
->mc_error
= SET_ERROR(ENXIO
);
515 mc
->mc_tried
= 1; /* don't even try */
520 if (vdev_dtl_contains(mc
->mc_vd
, DTL_MISSING
, txg
, 1)) {
521 mc
->mc_error
= SET_ERROR(ESTALE
);
523 mc
->mc_speculative
= 1;
527 mc
->mc_load
= vdev_mirror_load(mm
, mc
->mc_vd
, mc
->mc_offset
);
528 if (mc
->mc_load
> lowest_load
)
531 if (mc
->mc_load
< lowest_load
) {
532 lowest_load
= mc
->mc_load
;
533 mm
->mm_preferred_cnt
= 0;
535 mm
->mm_preferred
[mm
->mm_preferred_cnt
] = c
;
536 mm
->mm_preferred_cnt
++;
539 if (mm
->mm_preferred_cnt
== 1) {
540 MIRROR_BUMP(vdev_mirror_stat_preferred_found
);
541 return (mm
->mm_preferred
[0]);
544 if (mm
->mm_preferred_cnt
> 1) {
545 MIRROR_BUMP(vdev_mirror_stat_preferred_not_found
);
546 return (vdev_mirror_preferred_child_randomize(zio
));
550 * Every device is either missing or has this txg in its DTL.
551 * Look for any child we haven't already tried before giving up.
553 for (c
= 0; c
< mm
->mm_children
; c
++) {
554 if (!mm
->mm_child
[c
].mc_tried
)
559 * Every child failed. There's no place left to look.
565 vdev_mirror_io_start(zio_t
*zio
)
571 mm
= vdev_mirror_map_init(zio
);
574 ASSERT(!spa_trust_config(zio
->io_spa
));
575 ASSERT(zio
->io_type
== ZIO_TYPE_READ
);
580 if (zio
->io_type
== ZIO_TYPE_READ
) {
581 if (zio
->io_bp
!= NULL
&&
582 (zio
->io_flags
& ZIO_FLAG_SCRUB
) && !mm
->mm_resilvering
) {
584 * For scrubbing reads (if we can verify the
585 * checksum here, as indicated by io_bp being
586 * non-NULL) we need to allocate a read buffer for
587 * each child and issue reads to all children. If
588 * any child succeeds, it will copy its data into
589 * zio->io_data in vdev_mirror_scrub_done.
591 for (c
= 0; c
< mm
->mm_children
; c
++) {
592 mc
= &mm
->mm_child
[c
];
593 zio_nowait(zio_vdev_child_io(zio
, zio
->io_bp
,
594 mc
->mc_vd
, mc
->mc_offset
,
595 abd_alloc_sametype(zio
->io_abd
,
596 zio
->io_size
), zio
->io_size
,
597 zio
->io_type
, zio
->io_priority
, 0,
598 vdev_mirror_scrub_done
, mc
));
604 * For normal reads just pick one child.
606 c
= vdev_mirror_child_select(zio
);
609 ASSERT(zio
->io_type
== ZIO_TYPE_WRITE
);
612 * Writes go to all children.
615 children
= mm
->mm_children
;
619 mc
= &mm
->mm_child
[c
];
620 zio_nowait(zio_vdev_child_io(zio
, zio
->io_bp
,
621 mc
->mc_vd
, mc
->mc_offset
, zio
->io_abd
, zio
->io_size
,
622 zio
->io_type
, zio
->io_priority
, 0,
623 vdev_mirror_child_done
, mc
));
631 vdev_mirror_worst_error(mirror_map_t
*mm
)
633 int error
[2] = { 0, 0 };
635 for (int c
= 0; c
< mm
->mm_children
; c
++) {
636 mirror_child_t
*mc
= &mm
->mm_child
[c
];
637 int s
= mc
->mc_speculative
;
638 error
[s
] = zio_worst_error(error
[s
], mc
->mc_error
);
641 return (error
[0] ? error
[0] : error
[1]);
645 vdev_mirror_io_done(zio_t
*zio
)
647 mirror_map_t
*mm
= zio
->io_vsd
;
651 int unexpected_errors
= 0;
656 for (c
= 0; c
< mm
->mm_children
; c
++) {
657 mc
= &mm
->mm_child
[c
];
662 } else if (mc
->mc_tried
) {
667 if (zio
->io_type
== ZIO_TYPE_WRITE
) {
669 * XXX -- for now, treat partial writes as success.
671 * Now that we support write reallocation, it would be better
672 * to treat partial failure as real failure unless there are
673 * no non-degraded top-level vdevs left, and not update DTLs
674 * if we intend to reallocate.
677 if (good_copies
!= mm
->mm_children
) {
679 * Always require at least one good copy.
681 * For ditto blocks (io_vd == NULL), require
682 * all copies to be good.
684 * XXX -- for replacing vdevs, there's no great answer.
685 * If the old device is really dead, we may not even
686 * be able to access it -- so we only want to
687 * require good writes to the new device. But if
688 * the new device turns out to be flaky, we want
689 * to be able to detach it -- which requires all
690 * writes to the old device to have succeeded.
692 if (good_copies
== 0 || zio
->io_vd
== NULL
)
693 zio
->io_error
= vdev_mirror_worst_error(mm
);
698 ASSERT(zio
->io_type
== ZIO_TYPE_READ
);
701 * If we don't have a good copy yet, keep trying other children.
704 if (good_copies
== 0 && (c
= vdev_mirror_child_select(zio
)) != -1) {
705 ASSERT(c
>= 0 && c
< mm
->mm_children
);
706 mc
= &mm
->mm_child
[c
];
707 zio_vdev_io_redone(zio
);
708 zio_nowait(zio_vdev_child_io(zio
, zio
->io_bp
,
709 mc
->mc_vd
, mc
->mc_offset
, zio
->io_abd
, zio
->io_size
,
710 ZIO_TYPE_READ
, zio
->io_priority
, 0,
711 vdev_mirror_child_done
, mc
));
716 if (good_copies
== 0) {
717 zio
->io_error
= vdev_mirror_worst_error(mm
);
718 ASSERT(zio
->io_error
!= 0);
721 if (good_copies
&& spa_writeable(zio
->io_spa
) &&
722 (unexpected_errors
||
723 (zio
->io_flags
& ZIO_FLAG_RESILVER
) ||
724 ((zio
->io_flags
& ZIO_FLAG_SCRUB
) && mm
->mm_resilvering
))) {
726 * Use the good data we have in hand to repair damaged children.
728 for (c
= 0; c
< mm
->mm_children
; c
++) {
730 * Don't rewrite known good children.
731 * Not only is it unnecessary, it could
732 * actually be harmful: if the system lost
733 * power while rewriting the only good copy,
734 * there would be no good copies left!
736 mc
= &mm
->mm_child
[c
];
738 if (mc
->mc_error
== 0) {
742 * We didn't try this child. We need to
744 * 1. it's a scrub (in which case we have
745 * tried everything that was healthy)
747 * 2. it's an indirect vdev (in which case
748 * it could point to any other vdev, which
749 * might have a bad DTL)
751 * 3. the DTL indicates that this data is
752 * missing from this vdev
754 if (!(zio
->io_flags
& ZIO_FLAG_SCRUB
) &&
755 mc
->mc_vd
->vdev_ops
!= &vdev_indirect_ops
&&
756 !vdev_dtl_contains(mc
->mc_vd
, DTL_PARTIAL
,
759 mc
->mc_error
= SET_ERROR(ESTALE
);
762 zio_nowait(zio_vdev_child_io(zio
, zio
->io_bp
,
763 mc
->mc_vd
, mc
->mc_offset
,
764 zio
->io_abd
, zio
->io_size
,
765 ZIO_TYPE_WRITE
, ZIO_PRIORITY_ASYNC_WRITE
,
766 ZIO_FLAG_IO_REPAIR
| (unexpected_errors
?
767 ZIO_FLAG_SELF_HEAL
: 0), NULL
, NULL
));
773 vdev_mirror_state_change(vdev_t
*vd
, int faulted
, int degraded
)
775 if (faulted
== vd
->vdev_children
) {
776 if (vdev_children_are_offline(vd
)) {
777 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_OFFLINE
,
778 VDEV_AUX_CHILDREN_OFFLINE
);
780 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_CANT_OPEN
,
781 VDEV_AUX_NO_REPLICAS
);
783 } else if (degraded
+ faulted
!= 0) {
784 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_DEGRADED
, VDEV_AUX_NONE
);
786 vdev_set_state(vd
, B_FALSE
, VDEV_STATE_HEALTHY
, VDEV_AUX_NONE
);
790 vdev_ops_t vdev_mirror_ops
= {
794 vdev_mirror_io_start
,
796 vdev_mirror_state_change
,
802 VDEV_TYPE_MIRROR
, /* name of this vdev type */
803 B_FALSE
/* not a leaf vdev */
806 vdev_ops_t vdev_replacing_ops
= {
810 vdev_mirror_io_start
,
812 vdev_mirror_state_change
,
818 VDEV_TYPE_REPLACING
, /* name of this vdev type */
819 B_FALSE
/* not a leaf vdev */
822 vdev_ops_t vdev_spare_ops
= {
826 vdev_mirror_io_start
,
828 vdev_mirror_state_change
,
834 VDEV_TYPE_SPARE
, /* name of this vdev type */
835 B_FALSE
/* not a leaf vdev */
840 module_param(zfs_vdev_mirror_rotating_inc
, int, 0644);
841 MODULE_PARM_DESC(zfs_vdev_mirror_rotating_inc
,
842 "Rotating media load increment for non-seeking I/O's");
844 module_param(zfs_vdev_mirror_rotating_seek_inc
, int, 0644);
845 MODULE_PARM_DESC(zfs_vdev_mirror_rotating_seek_inc
,
846 "Rotating media load increment for seeking I/O's");
848 module_param(zfs_vdev_mirror_rotating_seek_offset
, int, 0644);
850 MODULE_PARM_DESC(zfs_vdev_mirror_rotating_seek_offset
,
851 "Offset in bytes from the last I/O which "
852 "triggers a reduced rotating media seek increment");
854 module_param(zfs_vdev_mirror_non_rotating_inc
, int, 0644);
855 MODULE_PARM_DESC(zfs_vdev_mirror_non_rotating_inc
,
856 "Non-rotating media load increment for non-seeking I/O's");
858 module_param(zfs_vdev_mirror_non_rotating_seek_inc
, int, 0644);
859 MODULE_PARM_DESC(zfs_vdev_mirror_non_rotating_seek_inc
,
860 "Non-rotating media load increment for seeking I/O's");