2 * raid10.c : Multiple Devices driver for Linux
4 * Copyright (C) 2000-2004 Neil Brown
6 * RAID-10 support for md.
8 * Base on code in raid1.c. See raid1.c for futher copyright information.
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2, or (at your option)
16 * You should have received a copy of the GNU General Public License
17 * (for example /usr/src/linux/COPYING); if not, write to the Free
18 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include <linux/delay.h>
22 #include <linux/blkdev.h>
23 #include <linux/seq_file.h>
25 #include "dm-bio-list.h"
30 * RAID10 provides a combination of RAID0 and RAID1 functionality.
31 * The layout of data is defined by
34 * near_copies (stored in low byte of layout)
35 * far_copies (stored in second byte of layout)
36 * far_offset (stored in bit 16 of layout )
38 * The data to be stored is divided into chunks using chunksize.
39 * Each device is divided into far_copies sections.
40 * In each section, chunks are laid out in a style similar to raid0, but
41 * near_copies copies of each chunk is stored (each on a different drive).
42 * The starting device for each section is offset near_copies from the starting
43 * device of the previous section.
44 * Thus they are (near_copies*far_copies) of each chunk, and each is on a different
46 * near_copies and far_copies must be at least one, and their product is at most
49 * If far_offset is true, then the far_copies are handled a bit differently.
50 * The copies are still in different stripes, but instead of be very far apart
51 * on disk, there are adjacent stripes.
55 * Number of guaranteed r10bios in case of extreme VM load:
57 #define NR_RAID10_BIOS 256
59 static void unplug_slaves(mddev_t
*mddev
);
61 static void allow_barrier(conf_t
*conf
);
62 static void lower_barrier(conf_t
*conf
);
64 static void * r10bio_pool_alloc(gfp_t gfp_flags
, void *data
)
68 int size
= offsetof(struct r10bio_s
, devs
[conf
->copies
]);
70 /* allocate a r10bio with room for raid_disks entries in the bios array */
71 r10_bio
= kzalloc(size
, gfp_flags
);
73 unplug_slaves(conf
->mddev
);
78 static void r10bio_pool_free(void *r10_bio
, void *data
)
83 /* Maximum size of each resync request */
84 #define RESYNC_BLOCK_SIZE (64*1024)
85 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
86 /* amount of memory to reserve for resync requests */
87 #define RESYNC_WINDOW (1024*1024)
88 /* maximum number of concurrent requests, memory permitting */
89 #define RESYNC_DEPTH (32*1024*1024/RESYNC_BLOCK_SIZE)
92 * When performing a resync, we need to read and compare, so
93 * we need as many pages are there are copies.
94 * When performing a recovery, we need 2 bios, one for read,
95 * one for write (we recover only one drive per r10buf)
98 static void * r10buf_pool_alloc(gfp_t gfp_flags
, void *data
)
107 r10_bio
= r10bio_pool_alloc(gfp_flags
, conf
);
109 unplug_slaves(conf
->mddev
);
113 if (test_bit(MD_RECOVERY_SYNC
, &conf
->mddev
->recovery
))
114 nalloc
= conf
->copies
; /* resync */
116 nalloc
= 2; /* recovery */
121 for (j
= nalloc
; j
-- ; ) {
122 bio
= bio_alloc(gfp_flags
, RESYNC_PAGES
);
125 r10_bio
->devs
[j
].bio
= bio
;
128 * Allocate RESYNC_PAGES data pages and attach them
131 for (j
= 0 ; j
< nalloc
; j
++) {
132 bio
= r10_bio
->devs
[j
].bio
;
133 for (i
= 0; i
< RESYNC_PAGES
; i
++) {
134 page
= alloc_page(gfp_flags
);
138 bio
->bi_io_vec
[i
].bv_page
= page
;
146 safe_put_page(bio
->bi_io_vec
[i
-1].bv_page
);
148 for (i
= 0; i
< RESYNC_PAGES
; i
++)
149 safe_put_page(r10_bio
->devs
[j
].bio
->bi_io_vec
[i
].bv_page
);
152 while ( ++j
< nalloc
)
153 bio_put(r10_bio
->devs
[j
].bio
);
154 r10bio_pool_free(r10_bio
, conf
);
158 static void r10buf_pool_free(void *__r10_bio
, void *data
)
162 r10bio_t
*r10bio
= __r10_bio
;
165 for (j
=0; j
< conf
->copies
; j
++) {
166 struct bio
*bio
= r10bio
->devs
[j
].bio
;
168 for (i
= 0; i
< RESYNC_PAGES
; i
++) {
169 safe_put_page(bio
->bi_io_vec
[i
].bv_page
);
170 bio
->bi_io_vec
[i
].bv_page
= NULL
;
175 r10bio_pool_free(r10bio
, conf
);
178 static void put_all_bios(conf_t
*conf
, r10bio_t
*r10_bio
)
182 for (i
= 0; i
< conf
->copies
; i
++) {
183 struct bio
**bio
= & r10_bio
->devs
[i
].bio
;
184 if (*bio
&& *bio
!= IO_BLOCKED
)
190 static void free_r10bio(r10bio_t
*r10_bio
)
192 conf_t
*conf
= mddev_to_conf(r10_bio
->mddev
);
195 * Wake up any possible resync thread that waits for the device
200 put_all_bios(conf
, r10_bio
);
201 mempool_free(r10_bio
, conf
->r10bio_pool
);
204 static void put_buf(r10bio_t
*r10_bio
)
206 conf_t
*conf
= mddev_to_conf(r10_bio
->mddev
);
208 mempool_free(r10_bio
, conf
->r10buf_pool
);
213 static void reschedule_retry(r10bio_t
*r10_bio
)
216 mddev_t
*mddev
= r10_bio
->mddev
;
217 conf_t
*conf
= mddev_to_conf(mddev
);
219 spin_lock_irqsave(&conf
->device_lock
, flags
);
220 list_add(&r10_bio
->retry_list
, &conf
->retry_list
);
222 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
224 /* wake up frozen array... */
225 wake_up(&conf
->wait_barrier
);
227 md_wakeup_thread(mddev
->thread
);
231 * raid_end_bio_io() is called when we have finished servicing a mirrored
232 * operation and are ready to return a success/failure code to the buffer
235 static void raid_end_bio_io(r10bio_t
*r10_bio
)
237 struct bio
*bio
= r10_bio
->master_bio
;
240 test_bit(R10BIO_Uptodate
, &r10_bio
->state
) ? 0 : -EIO
);
241 free_r10bio(r10_bio
);
245 * Update disk head position estimator based on IRQ completion info.
247 static inline void update_head_pos(int slot
, r10bio_t
*r10_bio
)
249 conf_t
*conf
= mddev_to_conf(r10_bio
->mddev
);
251 conf
->mirrors
[r10_bio
->devs
[slot
].devnum
].head_position
=
252 r10_bio
->devs
[slot
].addr
+ (r10_bio
->sectors
);
255 static void raid10_end_read_request(struct bio
*bio
, int error
)
257 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
258 r10bio_t
* r10_bio
= (r10bio_t
*)(bio
->bi_private
);
260 conf_t
*conf
= mddev_to_conf(r10_bio
->mddev
);
263 slot
= r10_bio
->read_slot
;
264 dev
= r10_bio
->devs
[slot
].devnum
;
266 * this branch is our 'one mirror IO has finished' event handler:
268 update_head_pos(slot
, r10_bio
);
272 * Set R10BIO_Uptodate in our master bio, so that
273 * we will return a good error code to the higher
274 * levels even if IO on some other mirrored buffer fails.
276 * The 'master' represents the composite IO operation to
277 * user-side. So if something waits for IO, then it will
278 * wait for the 'master' bio.
280 set_bit(R10BIO_Uptodate
, &r10_bio
->state
);
281 raid_end_bio_io(r10_bio
);
286 char b
[BDEVNAME_SIZE
];
287 if (printk_ratelimit())
288 printk(KERN_ERR
"raid10: %s: rescheduling sector %llu\n",
289 bdevname(conf
->mirrors
[dev
].rdev
->bdev
,b
), (unsigned long long)r10_bio
->sector
);
290 reschedule_retry(r10_bio
);
293 rdev_dec_pending(conf
->mirrors
[dev
].rdev
, conf
->mddev
);
296 static void raid10_end_write_request(struct bio
*bio
, int error
)
298 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
299 r10bio_t
* r10_bio
= (r10bio_t
*)(bio
->bi_private
);
301 conf_t
*conf
= mddev_to_conf(r10_bio
->mddev
);
303 for (slot
= 0; slot
< conf
->copies
; slot
++)
304 if (r10_bio
->devs
[slot
].bio
== bio
)
306 dev
= r10_bio
->devs
[slot
].devnum
;
309 * this branch is our 'one mirror IO has finished' event handler:
312 md_error(r10_bio
->mddev
, conf
->mirrors
[dev
].rdev
);
313 /* an I/O failed, we can't clear the bitmap */
314 set_bit(R10BIO_Degraded
, &r10_bio
->state
);
317 * Set R10BIO_Uptodate in our master bio, so that
318 * we will return a good error code for to the higher
319 * levels even if IO on some other mirrored buffer fails.
321 * The 'master' represents the composite IO operation to
322 * user-side. So if something waits for IO, then it will
323 * wait for the 'master' bio.
325 set_bit(R10BIO_Uptodate
, &r10_bio
->state
);
327 update_head_pos(slot
, r10_bio
);
331 * Let's see if all mirrored write operations have finished
334 if (atomic_dec_and_test(&r10_bio
->remaining
)) {
335 /* clear the bitmap if all writes complete successfully */
336 bitmap_endwrite(r10_bio
->mddev
->bitmap
, r10_bio
->sector
,
338 !test_bit(R10BIO_Degraded
, &r10_bio
->state
),
340 md_write_end(r10_bio
->mddev
);
341 raid_end_bio_io(r10_bio
);
344 rdev_dec_pending(conf
->mirrors
[dev
].rdev
, conf
->mddev
);
349 * RAID10 layout manager
350 * Aswell as the chunksize and raid_disks count, there are two
351 * parameters: near_copies and far_copies.
352 * near_copies * far_copies must be <= raid_disks.
353 * Normally one of these will be 1.
354 * If both are 1, we get raid0.
355 * If near_copies == raid_disks, we get raid1.
357 * Chunks are layed out in raid0 style with near_copies copies of the
358 * first chunk, followed by near_copies copies of the next chunk and
360 * If far_copies > 1, then after 1/far_copies of the array has been assigned
361 * as described above, we start again with a device offset of near_copies.
362 * So we effectively have another copy of the whole array further down all
363 * the drives, but with blocks on different drives.
364 * With this layout, and block is never stored twice on the one device.
366 * raid10_find_phys finds the sector offset of a given virtual sector
367 * on each device that it is on.
369 * raid10_find_virt does the reverse mapping, from a device and a
370 * sector offset to a virtual address
373 static void raid10_find_phys(conf_t
*conf
, r10bio_t
*r10bio
)
383 /* now calculate first sector/dev */
384 chunk
= r10bio
->sector
>> conf
->chunk_shift
;
385 sector
= r10bio
->sector
& conf
->chunk_mask
;
387 chunk
*= conf
->near_copies
;
389 dev
= sector_div(stripe
, conf
->raid_disks
);
390 if (conf
->far_offset
)
391 stripe
*= conf
->far_copies
;
393 sector
+= stripe
<< conf
->chunk_shift
;
395 /* and calculate all the others */
396 for (n
=0; n
< conf
->near_copies
; n
++) {
399 r10bio
->devs
[slot
].addr
= sector
;
400 r10bio
->devs
[slot
].devnum
= d
;
403 for (f
= 1; f
< conf
->far_copies
; f
++) {
404 d
+= conf
->near_copies
;
405 if (d
>= conf
->raid_disks
)
406 d
-= conf
->raid_disks
;
408 r10bio
->devs
[slot
].devnum
= d
;
409 r10bio
->devs
[slot
].addr
= s
;
413 if (dev
>= conf
->raid_disks
) {
415 sector
+= (conf
->chunk_mask
+ 1);
418 BUG_ON(slot
!= conf
->copies
);
421 static sector_t
raid10_find_virt(conf_t
*conf
, sector_t sector
, int dev
)
423 sector_t offset
, chunk
, vchunk
;
425 offset
= sector
& conf
->chunk_mask
;
426 if (conf
->far_offset
) {
428 chunk
= sector
>> conf
->chunk_shift
;
429 fc
= sector_div(chunk
, conf
->far_copies
);
430 dev
-= fc
* conf
->near_copies
;
432 dev
+= conf
->raid_disks
;
434 while (sector
>= conf
->stride
) {
435 sector
-= conf
->stride
;
436 if (dev
< conf
->near_copies
)
437 dev
+= conf
->raid_disks
- conf
->near_copies
;
439 dev
-= conf
->near_copies
;
441 chunk
= sector
>> conf
->chunk_shift
;
443 vchunk
= chunk
* conf
->raid_disks
+ dev
;
444 sector_div(vchunk
, conf
->near_copies
);
445 return (vchunk
<< conf
->chunk_shift
) + offset
;
449 * raid10_mergeable_bvec -- tell bio layer if a two requests can be merged
451 * @bvm: properties of new bio
452 * @biovec: the request that could be merged to it.
454 * Return amount of bytes we can accept at this offset
455 * If near_copies == raid_disk, there are no striping issues,
456 * but in that case, the function isn't called at all.
458 static int raid10_mergeable_bvec(struct request_queue
*q
,
459 struct bvec_merge_data
*bvm
,
460 struct bio_vec
*biovec
)
462 mddev_t
*mddev
= q
->queuedata
;
463 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
465 unsigned int chunk_sectors
= mddev
->chunk_size
>> 9;
466 unsigned int bio_sectors
= bvm
->bi_size
>> 9;
468 max
= (chunk_sectors
- ((sector
& (chunk_sectors
- 1)) + bio_sectors
)) << 9;
469 if (max
< 0) max
= 0; /* bio_add cannot handle a negative return */
470 if (max
<= biovec
->bv_len
&& bio_sectors
== 0)
471 return biovec
->bv_len
;
477 * This routine returns the disk from which the requested read should
478 * be done. There is a per-array 'next expected sequential IO' sector
479 * number - if this matches on the next IO then we use the last disk.
480 * There is also a per-disk 'last know head position' sector that is
481 * maintained from IRQ contexts, both the normal and the resync IO
482 * completion handlers update this position correctly. If there is no
483 * perfect sequential match then we pick the disk whose head is closest.
485 * If there are 2 mirrors in the same 2 devices, performance degrades
486 * because position is mirror, not device based.
488 * The rdev for the device selected will have nr_pending incremented.
492 * FIXME: possibly should rethink readbalancing and do it differently
493 * depending on near_copies / far_copies geometry.
495 static int read_balance(conf_t
*conf
, r10bio_t
*r10_bio
)
497 const unsigned long this_sector
= r10_bio
->sector
;
498 int disk
, slot
, nslot
;
499 const int sectors
= r10_bio
->sectors
;
500 sector_t new_distance
, current_distance
;
503 raid10_find_phys(conf
, r10_bio
);
506 * Check if we can balance. We can balance on the whole
507 * device if no resync is going on (recovery is ok), or below
508 * the resync window. We take the first readable disk when
509 * above the resync window.
511 if (conf
->mddev
->recovery_cp
< MaxSector
512 && (this_sector
+ sectors
>= conf
->next_resync
)) {
513 /* make sure that disk is operational */
515 disk
= r10_bio
->devs
[slot
].devnum
;
517 while ((rdev
= rcu_dereference(conf
->mirrors
[disk
].rdev
)) == NULL
||
518 r10_bio
->devs
[slot
].bio
== IO_BLOCKED
||
519 !test_bit(In_sync
, &rdev
->flags
)) {
521 if (slot
== conf
->copies
) {
526 disk
= r10_bio
->devs
[slot
].devnum
;
532 /* make sure the disk is operational */
534 disk
= r10_bio
->devs
[slot
].devnum
;
535 while ((rdev
=rcu_dereference(conf
->mirrors
[disk
].rdev
)) == NULL
||
536 r10_bio
->devs
[slot
].bio
== IO_BLOCKED
||
537 !test_bit(In_sync
, &rdev
->flags
)) {
539 if (slot
== conf
->copies
) {
543 disk
= r10_bio
->devs
[slot
].devnum
;
547 current_distance
= abs(r10_bio
->devs
[slot
].addr
-
548 conf
->mirrors
[disk
].head_position
);
550 /* Find the disk whose head is closest,
551 * or - for far > 1 - find the closest to partition beginning */
553 for (nslot
= slot
; nslot
< conf
->copies
; nslot
++) {
554 int ndisk
= r10_bio
->devs
[nslot
].devnum
;
557 if ((rdev
=rcu_dereference(conf
->mirrors
[ndisk
].rdev
)) == NULL
||
558 r10_bio
->devs
[nslot
].bio
== IO_BLOCKED
||
559 !test_bit(In_sync
, &rdev
->flags
))
562 /* This optimisation is debatable, and completely destroys
563 * sequential read speed for 'far copies' arrays. So only
564 * keep it for 'near' arrays, and review those later.
566 if (conf
->near_copies
> 1 && !atomic_read(&rdev
->nr_pending
)) {
572 /* for far > 1 always use the lowest address */
573 if (conf
->far_copies
> 1)
574 new_distance
= r10_bio
->devs
[nslot
].addr
;
576 new_distance
= abs(r10_bio
->devs
[nslot
].addr
-
577 conf
->mirrors
[ndisk
].head_position
);
578 if (new_distance
< current_distance
) {
579 current_distance
= new_distance
;
586 r10_bio
->read_slot
= slot
;
587 /* conf->next_seq_sect = this_sector + sectors;*/
589 if (disk
>= 0 && (rdev
=rcu_dereference(conf
->mirrors
[disk
].rdev
))!= NULL
)
590 atomic_inc(&conf
->mirrors
[disk
].rdev
->nr_pending
);
598 static void unplug_slaves(mddev_t
*mddev
)
600 conf_t
*conf
= mddev_to_conf(mddev
);
604 for (i
=0; i
<mddev
->raid_disks
; i
++) {
605 mdk_rdev_t
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
606 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
) && atomic_read(&rdev
->nr_pending
)) {
607 struct request_queue
*r_queue
= bdev_get_queue(rdev
->bdev
);
609 atomic_inc(&rdev
->nr_pending
);
614 rdev_dec_pending(rdev
, mddev
);
621 static void raid10_unplug(struct request_queue
*q
)
623 mddev_t
*mddev
= q
->queuedata
;
625 unplug_slaves(q
->queuedata
);
626 md_wakeup_thread(mddev
->thread
);
629 static int raid10_congested(void *data
, int bits
)
631 mddev_t
*mddev
= data
;
632 conf_t
*conf
= mddev_to_conf(mddev
);
636 for (i
= 0; i
< mddev
->raid_disks
&& ret
== 0; i
++) {
637 mdk_rdev_t
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
638 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
639 struct request_queue
*q
= bdev_get_queue(rdev
->bdev
);
641 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
648 static int flush_pending_writes(conf_t
*conf
)
650 /* Any writes that have been queued but are awaiting
651 * bitmap updates get flushed here.
652 * We return 1 if any requests were actually submitted.
656 spin_lock_irq(&conf
->device_lock
);
658 if (conf
->pending_bio_list
.head
) {
660 bio
= bio_list_get(&conf
->pending_bio_list
);
661 blk_remove_plug(conf
->mddev
->queue
);
662 spin_unlock_irq(&conf
->device_lock
);
663 /* flush any pending bitmap writes to disk
664 * before proceeding w/ I/O */
665 bitmap_unplug(conf
->mddev
->bitmap
);
667 while (bio
) { /* submit pending writes */
668 struct bio
*next
= bio
->bi_next
;
670 generic_make_request(bio
);
675 spin_unlock_irq(&conf
->device_lock
);
679 * Sometimes we need to suspend IO while we do something else,
680 * either some resync/recovery, or reconfigure the array.
681 * To do this we raise a 'barrier'.
682 * The 'barrier' is a counter that can be raised multiple times
683 * to count how many activities are happening which preclude
685 * We can only raise the barrier if there is no pending IO.
686 * i.e. if nr_pending == 0.
687 * We choose only to raise the barrier if no-one is waiting for the
688 * barrier to go down. This means that as soon as an IO request
689 * is ready, no other operations which require a barrier will start
690 * until the IO request has had a chance.
692 * So: regular IO calls 'wait_barrier'. When that returns there
693 * is no backgroup IO happening, It must arrange to call
694 * allow_barrier when it has finished its IO.
695 * backgroup IO calls must call raise_barrier. Once that returns
696 * there is no normal IO happeing. It must arrange to call
697 * lower_barrier when the particular background IO completes.
700 static void raise_barrier(conf_t
*conf
, int force
)
702 BUG_ON(force
&& !conf
->barrier
);
703 spin_lock_irq(&conf
->resync_lock
);
705 /* Wait until no block IO is waiting (unless 'force') */
706 wait_event_lock_irq(conf
->wait_barrier
, force
|| !conf
->nr_waiting
,
708 raid10_unplug(conf
->mddev
->queue
));
710 /* block any new IO from starting */
713 /* No wait for all pending IO to complete */
714 wait_event_lock_irq(conf
->wait_barrier
,
715 !conf
->nr_pending
&& conf
->barrier
< RESYNC_DEPTH
,
717 raid10_unplug(conf
->mddev
->queue
));
719 spin_unlock_irq(&conf
->resync_lock
);
722 static void lower_barrier(conf_t
*conf
)
725 spin_lock_irqsave(&conf
->resync_lock
, flags
);
727 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
728 wake_up(&conf
->wait_barrier
);
731 static void wait_barrier(conf_t
*conf
)
733 spin_lock_irq(&conf
->resync_lock
);
736 wait_event_lock_irq(conf
->wait_barrier
, !conf
->barrier
,
738 raid10_unplug(conf
->mddev
->queue
));
742 spin_unlock_irq(&conf
->resync_lock
);
745 static void allow_barrier(conf_t
*conf
)
748 spin_lock_irqsave(&conf
->resync_lock
, flags
);
750 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
751 wake_up(&conf
->wait_barrier
);
754 static void freeze_array(conf_t
*conf
)
756 /* stop syncio and normal IO and wait for everything to
758 * We increment barrier and nr_waiting, and then
759 * wait until nr_pending match nr_queued+1
760 * This is called in the context of one normal IO request
761 * that has failed. Thus any sync request that might be pending
762 * will be blocked by nr_pending, and we need to wait for
763 * pending IO requests to complete or be queued for re-try.
764 * Thus the number queued (nr_queued) plus this request (1)
765 * must match the number of pending IOs (nr_pending) before
768 spin_lock_irq(&conf
->resync_lock
);
771 wait_event_lock_irq(conf
->wait_barrier
,
772 conf
->nr_pending
== conf
->nr_queued
+1,
774 ({ flush_pending_writes(conf
);
775 raid10_unplug(conf
->mddev
->queue
); }));
776 spin_unlock_irq(&conf
->resync_lock
);
779 static void unfreeze_array(conf_t
*conf
)
781 /* reverse the effect of the freeze */
782 spin_lock_irq(&conf
->resync_lock
);
785 wake_up(&conf
->wait_barrier
);
786 spin_unlock_irq(&conf
->resync_lock
);
789 static int make_request(struct request_queue
*q
, struct bio
* bio
)
791 mddev_t
*mddev
= q
->queuedata
;
792 conf_t
*conf
= mddev_to_conf(mddev
);
793 mirror_info_t
*mirror
;
795 struct bio
*read_bio
;
798 int chunk_sects
= conf
->chunk_mask
+ 1;
799 const int rw
= bio_data_dir(bio
);
800 const int do_sync
= bio_sync(bio
);
803 mdk_rdev_t
*blocked_rdev
;
805 if (unlikely(bio_barrier(bio
))) {
806 bio_endio(bio
, -EOPNOTSUPP
);
810 /* If this request crosses a chunk boundary, we need to
811 * split it. This will only happen for 1 PAGE (or less) requests.
813 if (unlikely( (bio
->bi_sector
& conf
->chunk_mask
) + (bio
->bi_size
>> 9)
815 conf
->near_copies
< conf
->raid_disks
)) {
817 /* Sanity check -- queue functions should prevent this happening */
818 if (bio
->bi_vcnt
!= 1 ||
821 /* This is a one page bio that upper layers
822 * refuse to split for us, so we need to split it.
825 chunk_sects
- (bio
->bi_sector
& (chunk_sects
- 1)) );
826 if (make_request(q
, &bp
->bio1
))
827 generic_make_request(&bp
->bio1
);
828 if (make_request(q
, &bp
->bio2
))
829 generic_make_request(&bp
->bio2
);
831 bio_pair_release(bp
);
834 printk("raid10_make_request bug: can't convert block across chunks"
835 " or bigger than %dk %llu %d\n", chunk_sects
/2,
836 (unsigned long long)bio
->bi_sector
, bio
->bi_size
>> 10);
842 md_write_start(mddev
, bio
);
845 * Register the new request and wait if the reconstruction
846 * thread has put up a bar for new requests.
847 * Continue immediately if no resync is active currently.
851 cpu
= part_stat_lock();
852 part_stat_inc(cpu
, &mddev
->gendisk
->part0
, ios
[rw
]);
853 part_stat_add(cpu
, &mddev
->gendisk
->part0
, sectors
[rw
],
857 r10_bio
= mempool_alloc(conf
->r10bio_pool
, GFP_NOIO
);
859 r10_bio
->master_bio
= bio
;
860 r10_bio
->sectors
= bio
->bi_size
>> 9;
862 r10_bio
->mddev
= mddev
;
863 r10_bio
->sector
= bio
->bi_sector
;
868 * read balancing logic:
870 int disk
= read_balance(conf
, r10_bio
);
871 int slot
= r10_bio
->read_slot
;
873 raid_end_bio_io(r10_bio
);
876 mirror
= conf
->mirrors
+ disk
;
878 read_bio
= bio_clone(bio
, GFP_NOIO
);
880 r10_bio
->devs
[slot
].bio
= read_bio
;
882 read_bio
->bi_sector
= r10_bio
->devs
[slot
].addr
+
883 mirror
->rdev
->data_offset
;
884 read_bio
->bi_bdev
= mirror
->rdev
->bdev
;
885 read_bio
->bi_end_io
= raid10_end_read_request
;
886 read_bio
->bi_rw
= READ
| do_sync
;
887 read_bio
->bi_private
= r10_bio
;
889 generic_make_request(read_bio
);
896 /* first select target devices under rcu_lock and
897 * inc refcount on their rdev. Record them by setting
900 raid10_find_phys(conf
, r10_bio
);
904 for (i
= 0; i
< conf
->copies
; i
++) {
905 int d
= r10_bio
->devs
[i
].devnum
;
906 mdk_rdev_t
*rdev
= rcu_dereference(conf
->mirrors
[d
].rdev
);
907 if (rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
908 atomic_inc(&rdev
->nr_pending
);
912 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
913 atomic_inc(&rdev
->nr_pending
);
914 r10_bio
->devs
[i
].bio
= bio
;
916 r10_bio
->devs
[i
].bio
= NULL
;
917 set_bit(R10BIO_Degraded
, &r10_bio
->state
);
922 if (unlikely(blocked_rdev
)) {
923 /* Have to wait for this device to get unblocked, then retry */
927 for (j
= 0; j
< i
; j
++)
928 if (r10_bio
->devs
[j
].bio
) {
929 d
= r10_bio
->devs
[j
].devnum
;
930 rdev_dec_pending(conf
->mirrors
[d
].rdev
, mddev
);
933 md_wait_for_blocked_rdev(blocked_rdev
, mddev
);
938 atomic_set(&r10_bio
->remaining
, 0);
941 for (i
= 0; i
< conf
->copies
; i
++) {
943 int d
= r10_bio
->devs
[i
].devnum
;
944 if (!r10_bio
->devs
[i
].bio
)
947 mbio
= bio_clone(bio
, GFP_NOIO
);
948 r10_bio
->devs
[i
].bio
= mbio
;
950 mbio
->bi_sector
= r10_bio
->devs
[i
].addr
+
951 conf
->mirrors
[d
].rdev
->data_offset
;
952 mbio
->bi_bdev
= conf
->mirrors
[d
].rdev
->bdev
;
953 mbio
->bi_end_io
= raid10_end_write_request
;
954 mbio
->bi_rw
= WRITE
| do_sync
;
955 mbio
->bi_private
= r10_bio
;
957 atomic_inc(&r10_bio
->remaining
);
958 bio_list_add(&bl
, mbio
);
961 if (unlikely(!atomic_read(&r10_bio
->remaining
))) {
962 /* the array is dead */
964 raid_end_bio_io(r10_bio
);
968 bitmap_startwrite(mddev
->bitmap
, bio
->bi_sector
, r10_bio
->sectors
, 0);
969 spin_lock_irqsave(&conf
->device_lock
, flags
);
970 bio_list_merge(&conf
->pending_bio_list
, &bl
);
971 blk_plug_device(mddev
->queue
);
972 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
974 /* In case raid10d snuck in to freeze_array */
975 wake_up(&conf
->wait_barrier
);
978 md_wakeup_thread(mddev
->thread
);
983 static void status(struct seq_file
*seq
, mddev_t
*mddev
)
985 conf_t
*conf
= mddev_to_conf(mddev
);
988 if (conf
->near_copies
< conf
->raid_disks
)
989 seq_printf(seq
, " %dK chunks", mddev
->chunk_size
/1024);
990 if (conf
->near_copies
> 1)
991 seq_printf(seq
, " %d near-copies", conf
->near_copies
);
992 if (conf
->far_copies
> 1) {
993 if (conf
->far_offset
)
994 seq_printf(seq
, " %d offset-copies", conf
->far_copies
);
996 seq_printf(seq
, " %d far-copies", conf
->far_copies
);
998 seq_printf(seq
, " [%d/%d] [", conf
->raid_disks
,
999 conf
->raid_disks
- mddev
->degraded
);
1000 for (i
= 0; i
< conf
->raid_disks
; i
++)
1001 seq_printf(seq
, "%s",
1002 conf
->mirrors
[i
].rdev
&&
1003 test_bit(In_sync
, &conf
->mirrors
[i
].rdev
->flags
) ? "U" : "_");
1004 seq_printf(seq
, "]");
1007 static void error(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1009 char b
[BDEVNAME_SIZE
];
1010 conf_t
*conf
= mddev_to_conf(mddev
);
1013 * If it is not operational, then we have already marked it as dead
1014 * else if it is the last working disks, ignore the error, let the
1015 * next level up know.
1016 * else mark the drive as failed
1018 if (test_bit(In_sync
, &rdev
->flags
)
1019 && conf
->raid_disks
-mddev
->degraded
== 1)
1021 * Don't fail the drive, just return an IO error.
1022 * The test should really be more sophisticated than
1023 * "working_disks == 1", but it isn't critical, and
1024 * can wait until we do more sophisticated "is the drive
1025 * really dead" tests...
1028 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1029 unsigned long flags
;
1030 spin_lock_irqsave(&conf
->device_lock
, flags
);
1032 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1034 * if recovery is running, make sure it aborts.
1036 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1038 set_bit(Faulty
, &rdev
->flags
);
1039 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1040 printk(KERN_ALERT
"raid10: Disk failure on %s, disabling device.\n"
1041 "raid10: Operation continuing on %d devices.\n",
1042 bdevname(rdev
->bdev
,b
), conf
->raid_disks
- mddev
->degraded
);
1045 static void print_conf(conf_t
*conf
)
1050 printk("RAID10 conf printout:\n");
1052 printk("(!conf)\n");
1055 printk(" --- wd:%d rd:%d\n", conf
->raid_disks
- conf
->mddev
->degraded
,
1058 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1059 char b
[BDEVNAME_SIZE
];
1060 tmp
= conf
->mirrors
+ i
;
1062 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
1063 i
, !test_bit(In_sync
, &tmp
->rdev
->flags
),
1064 !test_bit(Faulty
, &tmp
->rdev
->flags
),
1065 bdevname(tmp
->rdev
->bdev
,b
));
1069 static void close_sync(conf_t
*conf
)
1072 allow_barrier(conf
);
1074 mempool_destroy(conf
->r10buf_pool
);
1075 conf
->r10buf_pool
= NULL
;
1078 /* check if there are enough drives for
1079 * every block to appear on atleast one
1081 static int enough(conf_t
*conf
)
1086 int n
= conf
->copies
;
1089 if (conf
->mirrors
[first
].rdev
)
1091 first
= (first
+1) % conf
->raid_disks
;
1095 } while (first
!= 0);
1099 static int raid10_spare_active(mddev_t
*mddev
)
1102 conf_t
*conf
= mddev
->private;
1106 * Find all non-in_sync disks within the RAID10 configuration
1107 * and mark them in_sync
1109 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1110 tmp
= conf
->mirrors
+ i
;
1112 && !test_bit(Faulty
, &tmp
->rdev
->flags
)
1113 && !test_and_set_bit(In_sync
, &tmp
->rdev
->flags
)) {
1114 unsigned long flags
;
1115 spin_lock_irqsave(&conf
->device_lock
, flags
);
1117 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1126 static int raid10_add_disk(mddev_t
*mddev
, mdk_rdev_t
*rdev
)
1128 conf_t
*conf
= mddev
->private;
1133 int last
= mddev
->raid_disks
- 1;
1135 if (mddev
->recovery_cp
< MaxSector
)
1136 /* only hot-add to in-sync arrays, as recovery is
1137 * very different from resync
1143 if (rdev
->raid_disk
>= 0)
1144 first
= last
= rdev
->raid_disk
;
1146 if (rdev
->saved_raid_disk
>= 0 &&
1147 rdev
->saved_raid_disk
>= first
&&
1148 conf
->mirrors
[rdev
->saved_raid_disk
].rdev
== NULL
)
1149 mirror
= rdev
->saved_raid_disk
;
1152 for ( ; mirror
<= last
; mirror
++)
1153 if ( !(p
=conf
->mirrors
+mirror
)->rdev
) {
1155 blk_queue_stack_limits(mddev
->queue
,
1156 rdev
->bdev
->bd_disk
->queue
);
1157 /* as we don't honour merge_bvec_fn, we must never risk
1158 * violating it, so limit ->max_sector to one PAGE, as
1159 * a one page request is never in violation.
1161 if (rdev
->bdev
->bd_disk
->queue
->merge_bvec_fn
&&
1162 mddev
->queue
->max_sectors
> (PAGE_SIZE
>>9))
1163 mddev
->queue
->max_sectors
= (PAGE_SIZE
>>9);
1165 p
->head_position
= 0;
1166 rdev
->raid_disk
= mirror
;
1168 if (rdev
->saved_raid_disk
!= mirror
)
1170 rcu_assign_pointer(p
->rdev
, rdev
);
1178 static int raid10_remove_disk(mddev_t
*mddev
, int number
)
1180 conf_t
*conf
= mddev
->private;
1183 mirror_info_t
*p
= conf
->mirrors
+ number
;
1188 if (test_bit(In_sync
, &rdev
->flags
) ||
1189 atomic_read(&rdev
->nr_pending
)) {
1193 /* Only remove faulty devices in recovery
1196 if (!test_bit(Faulty
, &rdev
->flags
) &&
1203 if (atomic_read(&rdev
->nr_pending
)) {
1204 /* lost the race, try later */
1216 static void end_sync_read(struct bio
*bio
, int error
)
1218 r10bio_t
* r10_bio
= (r10bio_t
*)(bio
->bi_private
);
1219 conf_t
*conf
= mddev_to_conf(r10_bio
->mddev
);
1222 for (i
=0; i
<conf
->copies
; i
++)
1223 if (r10_bio
->devs
[i
].bio
== bio
)
1225 BUG_ON(i
== conf
->copies
);
1226 update_head_pos(i
, r10_bio
);
1227 d
= r10_bio
->devs
[i
].devnum
;
1229 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1230 set_bit(R10BIO_Uptodate
, &r10_bio
->state
);
1232 atomic_add(r10_bio
->sectors
,
1233 &conf
->mirrors
[d
].rdev
->corrected_errors
);
1234 if (!test_bit(MD_RECOVERY_SYNC
, &conf
->mddev
->recovery
))
1235 md_error(r10_bio
->mddev
,
1236 conf
->mirrors
[d
].rdev
);
1239 /* for reconstruct, we always reschedule after a read.
1240 * for resync, only after all reads
1242 rdev_dec_pending(conf
->mirrors
[d
].rdev
, conf
->mddev
);
1243 if (test_bit(R10BIO_IsRecover
, &r10_bio
->state
) ||
1244 atomic_dec_and_test(&r10_bio
->remaining
)) {
1245 /* we have read all the blocks,
1246 * do the comparison in process context in raid10d
1248 reschedule_retry(r10_bio
);
1252 static void end_sync_write(struct bio
*bio
, int error
)
1254 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1255 r10bio_t
* r10_bio
= (r10bio_t
*)(bio
->bi_private
);
1256 mddev_t
*mddev
= r10_bio
->mddev
;
1257 conf_t
*conf
= mddev_to_conf(mddev
);
1260 for (i
= 0; i
< conf
->copies
; i
++)
1261 if (r10_bio
->devs
[i
].bio
== bio
)
1263 d
= r10_bio
->devs
[i
].devnum
;
1266 md_error(mddev
, conf
->mirrors
[d
].rdev
);
1268 update_head_pos(i
, r10_bio
);
1270 rdev_dec_pending(conf
->mirrors
[d
].rdev
, mddev
);
1271 while (atomic_dec_and_test(&r10_bio
->remaining
)) {
1272 if (r10_bio
->master_bio
== NULL
) {
1273 /* the primary of several recovery bios */
1274 sector_t s
= r10_bio
->sectors
;
1276 md_done_sync(mddev
, s
, 1);
1279 r10bio_t
*r10_bio2
= (r10bio_t
*)r10_bio
->master_bio
;
1287 * Note: sync and recover and handled very differently for raid10
1288 * This code is for resync.
1289 * For resync, we read through virtual addresses and read all blocks.
1290 * If there is any error, we schedule a write. The lowest numbered
1291 * drive is authoritative.
1292 * However requests come for physical address, so we need to map.
1293 * For every physical address there are raid_disks/copies virtual addresses,
1294 * which is always are least one, but is not necessarly an integer.
1295 * This means that a physical address can span multiple chunks, so we may
1296 * have to submit multiple io requests for a single sync request.
1299 * We check if all blocks are in-sync and only write to blocks that
1302 static void sync_request_write(mddev_t
*mddev
, r10bio_t
*r10_bio
)
1304 conf_t
*conf
= mddev_to_conf(mddev
);
1306 struct bio
*tbio
, *fbio
;
1308 atomic_set(&r10_bio
->remaining
, 1);
1310 /* find the first device with a block */
1311 for (i
=0; i
<conf
->copies
; i
++)
1312 if (test_bit(BIO_UPTODATE
, &r10_bio
->devs
[i
].bio
->bi_flags
))
1315 if (i
== conf
->copies
)
1319 fbio
= r10_bio
->devs
[i
].bio
;
1321 /* now find blocks with errors */
1322 for (i
=0 ; i
< conf
->copies
; i
++) {
1324 int vcnt
= r10_bio
->sectors
>> (PAGE_SHIFT
-9);
1326 tbio
= r10_bio
->devs
[i
].bio
;
1328 if (tbio
->bi_end_io
!= end_sync_read
)
1332 if (test_bit(BIO_UPTODATE
, &r10_bio
->devs
[i
].bio
->bi_flags
)) {
1333 /* We know that the bi_io_vec layout is the same for
1334 * both 'first' and 'i', so we just compare them.
1335 * All vec entries are PAGE_SIZE;
1337 for (j
= 0; j
< vcnt
; j
++)
1338 if (memcmp(page_address(fbio
->bi_io_vec
[j
].bv_page
),
1339 page_address(tbio
->bi_io_vec
[j
].bv_page
),
1344 mddev
->resync_mismatches
+= r10_bio
->sectors
;
1346 if (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
))
1347 /* Don't fix anything. */
1349 /* Ok, we need to write this bio
1350 * First we need to fixup bv_offset, bv_len and
1351 * bi_vecs, as the read request might have corrupted these
1353 tbio
->bi_vcnt
= vcnt
;
1354 tbio
->bi_size
= r10_bio
->sectors
<< 9;
1356 tbio
->bi_phys_segments
= 0;
1357 tbio
->bi_flags
&= ~(BIO_POOL_MASK
- 1);
1358 tbio
->bi_flags
|= 1 << BIO_UPTODATE
;
1359 tbio
->bi_next
= NULL
;
1360 tbio
->bi_rw
= WRITE
;
1361 tbio
->bi_private
= r10_bio
;
1362 tbio
->bi_sector
= r10_bio
->devs
[i
].addr
;
1364 for (j
=0; j
< vcnt
; j
++) {
1365 tbio
->bi_io_vec
[j
].bv_offset
= 0;
1366 tbio
->bi_io_vec
[j
].bv_len
= PAGE_SIZE
;
1368 memcpy(page_address(tbio
->bi_io_vec
[j
].bv_page
),
1369 page_address(fbio
->bi_io_vec
[j
].bv_page
),
1372 tbio
->bi_end_io
= end_sync_write
;
1374 d
= r10_bio
->devs
[i
].devnum
;
1375 atomic_inc(&conf
->mirrors
[d
].rdev
->nr_pending
);
1376 atomic_inc(&r10_bio
->remaining
);
1377 md_sync_acct(conf
->mirrors
[d
].rdev
->bdev
, tbio
->bi_size
>> 9);
1379 tbio
->bi_sector
+= conf
->mirrors
[d
].rdev
->data_offset
;
1380 tbio
->bi_bdev
= conf
->mirrors
[d
].rdev
->bdev
;
1381 generic_make_request(tbio
);
1385 if (atomic_dec_and_test(&r10_bio
->remaining
)) {
1386 md_done_sync(mddev
, r10_bio
->sectors
, 1);
1392 * Now for the recovery code.
1393 * Recovery happens across physical sectors.
1394 * We recover all non-is_sync drives by finding the virtual address of
1395 * each, and then choose a working drive that also has that virt address.
1396 * There is a separate r10_bio for each non-in_sync drive.
1397 * Only the first two slots are in use. The first for reading,
1398 * The second for writing.
1402 static void recovery_request_write(mddev_t
*mddev
, r10bio_t
*r10_bio
)
1404 conf_t
*conf
= mddev_to_conf(mddev
);
1406 struct bio
*bio
, *wbio
;
1409 /* move the pages across to the second bio
1410 * and submit the write request
1412 bio
= r10_bio
->devs
[0].bio
;
1413 wbio
= r10_bio
->devs
[1].bio
;
1414 for (i
=0; i
< wbio
->bi_vcnt
; i
++) {
1415 struct page
*p
= bio
->bi_io_vec
[i
].bv_page
;
1416 bio
->bi_io_vec
[i
].bv_page
= wbio
->bi_io_vec
[i
].bv_page
;
1417 wbio
->bi_io_vec
[i
].bv_page
= p
;
1419 d
= r10_bio
->devs
[1].devnum
;
1421 atomic_inc(&conf
->mirrors
[d
].rdev
->nr_pending
);
1422 md_sync_acct(conf
->mirrors
[d
].rdev
->bdev
, wbio
->bi_size
>> 9);
1423 if (test_bit(R10BIO_Uptodate
, &r10_bio
->state
))
1424 generic_make_request(wbio
);
1426 bio_endio(wbio
, -EIO
);
1431 * This is a kernel thread which:
1433 * 1. Retries failed read operations on working mirrors.
1434 * 2. Updates the raid superblock when problems encounter.
1435 * 3. Performs writes following reads for array synchronising.
1438 static void fix_read_error(conf_t
*conf
, mddev_t
*mddev
, r10bio_t
*r10_bio
)
1440 int sect
= 0; /* Offset from r10_bio->sector */
1441 int sectors
= r10_bio
->sectors
;
1445 int sl
= r10_bio
->read_slot
;
1449 if (s
> (PAGE_SIZE
>>9))
1454 int d
= r10_bio
->devs
[sl
].devnum
;
1455 rdev
= rcu_dereference(conf
->mirrors
[d
].rdev
);
1457 test_bit(In_sync
, &rdev
->flags
)) {
1458 atomic_inc(&rdev
->nr_pending
);
1460 success
= sync_page_io(rdev
->bdev
,
1461 r10_bio
->devs
[sl
].addr
+
1462 sect
+ rdev
->data_offset
,
1464 conf
->tmppage
, READ
);
1465 rdev_dec_pending(rdev
, mddev
);
1471 if (sl
== conf
->copies
)
1473 } while (!success
&& sl
!= r10_bio
->read_slot
);
1477 /* Cannot read from anywhere -- bye bye array */
1478 int dn
= r10_bio
->devs
[r10_bio
->read_slot
].devnum
;
1479 md_error(mddev
, conf
->mirrors
[dn
].rdev
);
1484 /* write it back and re-read */
1486 while (sl
!= r10_bio
->read_slot
) {
1491 d
= r10_bio
->devs
[sl
].devnum
;
1492 rdev
= rcu_dereference(conf
->mirrors
[d
].rdev
);
1494 test_bit(In_sync
, &rdev
->flags
)) {
1495 atomic_inc(&rdev
->nr_pending
);
1497 atomic_add(s
, &rdev
->corrected_errors
);
1498 if (sync_page_io(rdev
->bdev
,
1499 r10_bio
->devs
[sl
].addr
+
1500 sect
+ rdev
->data_offset
,
1501 s
<<9, conf
->tmppage
, WRITE
)
1503 /* Well, this device is dead */
1504 md_error(mddev
, rdev
);
1505 rdev_dec_pending(rdev
, mddev
);
1510 while (sl
!= r10_bio
->read_slot
) {
1515 d
= r10_bio
->devs
[sl
].devnum
;
1516 rdev
= rcu_dereference(conf
->mirrors
[d
].rdev
);
1518 test_bit(In_sync
, &rdev
->flags
)) {
1519 char b
[BDEVNAME_SIZE
];
1520 atomic_inc(&rdev
->nr_pending
);
1522 if (sync_page_io(rdev
->bdev
,
1523 r10_bio
->devs
[sl
].addr
+
1524 sect
+ rdev
->data_offset
,
1525 s
<<9, conf
->tmppage
, READ
) == 0)
1526 /* Well, this device is dead */
1527 md_error(mddev
, rdev
);
1530 "raid10:%s: read error corrected"
1531 " (%d sectors at %llu on %s)\n",
1533 (unsigned long long)(sect
+
1535 bdevname(rdev
->bdev
, b
));
1537 rdev_dec_pending(rdev
, mddev
);
1548 static void raid10d(mddev_t
*mddev
)
1552 unsigned long flags
;
1553 conf_t
*conf
= mddev_to_conf(mddev
);
1554 struct list_head
*head
= &conf
->retry_list
;
1558 md_check_recovery(mddev
);
1561 char b
[BDEVNAME_SIZE
];
1563 unplug
+= flush_pending_writes(conf
);
1565 spin_lock_irqsave(&conf
->device_lock
, flags
);
1566 if (list_empty(head
)) {
1567 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1570 r10_bio
= list_entry(head
->prev
, r10bio_t
, retry_list
);
1571 list_del(head
->prev
);
1573 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1575 mddev
= r10_bio
->mddev
;
1576 conf
= mddev_to_conf(mddev
);
1577 if (test_bit(R10BIO_IsSync
, &r10_bio
->state
)) {
1578 sync_request_write(mddev
, r10_bio
);
1580 } else if (test_bit(R10BIO_IsRecover
, &r10_bio
->state
)) {
1581 recovery_request_write(mddev
, r10_bio
);
1585 /* we got a read error. Maybe the drive is bad. Maybe just
1586 * the block and we can fix it.
1587 * We freeze all other IO, and try reading the block from
1588 * other devices. When we find one, we re-write
1589 * and check it that fixes the read error.
1590 * This is all done synchronously while the array is
1593 if (mddev
->ro
== 0) {
1595 fix_read_error(conf
, mddev
, r10_bio
);
1596 unfreeze_array(conf
);
1599 bio
= r10_bio
->devs
[r10_bio
->read_slot
].bio
;
1600 r10_bio
->devs
[r10_bio
->read_slot
].bio
=
1601 mddev
->ro
? IO_BLOCKED
: NULL
;
1602 mirror
= read_balance(conf
, r10_bio
);
1604 printk(KERN_ALERT
"raid10: %s: unrecoverable I/O"
1605 " read error for block %llu\n",
1606 bdevname(bio
->bi_bdev
,b
),
1607 (unsigned long long)r10_bio
->sector
);
1608 raid_end_bio_io(r10_bio
);
1611 const int do_sync
= bio_sync(r10_bio
->master_bio
);
1613 rdev
= conf
->mirrors
[mirror
].rdev
;
1614 if (printk_ratelimit())
1615 printk(KERN_ERR
"raid10: %s: redirecting sector %llu to"
1616 " another mirror\n",
1617 bdevname(rdev
->bdev
,b
),
1618 (unsigned long long)r10_bio
->sector
);
1619 bio
= bio_clone(r10_bio
->master_bio
, GFP_NOIO
);
1620 r10_bio
->devs
[r10_bio
->read_slot
].bio
= bio
;
1621 bio
->bi_sector
= r10_bio
->devs
[r10_bio
->read_slot
].addr
1622 + rdev
->data_offset
;
1623 bio
->bi_bdev
= rdev
->bdev
;
1624 bio
->bi_rw
= READ
| do_sync
;
1625 bio
->bi_private
= r10_bio
;
1626 bio
->bi_end_io
= raid10_end_read_request
;
1628 generic_make_request(bio
);
1633 unplug_slaves(mddev
);
1637 static int init_resync(conf_t
*conf
)
1641 buffs
= RESYNC_WINDOW
/ RESYNC_BLOCK_SIZE
;
1642 BUG_ON(conf
->r10buf_pool
);
1643 conf
->r10buf_pool
= mempool_create(buffs
, r10buf_pool_alloc
, r10buf_pool_free
, conf
);
1644 if (!conf
->r10buf_pool
)
1646 conf
->next_resync
= 0;
1651 * perform a "sync" on one "block"
1653 * We need to make sure that no normal I/O request - particularly write
1654 * requests - conflict with active sync requests.
1656 * This is achieved by tracking pending requests and a 'barrier' concept
1657 * that can be installed to exclude normal IO requests.
1659 * Resync and recovery are handled very differently.
1660 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
1662 * For resync, we iterate over virtual addresses, read all copies,
1663 * and update if there are differences. If only one copy is live,
1665 * For recovery, we iterate over physical addresses, read a good
1666 * value for each non-in_sync drive, and over-write.
1668 * So, for recovery we may have several outstanding complex requests for a
1669 * given address, one for each out-of-sync device. We model this by allocating
1670 * a number of r10_bio structures, one for each out-of-sync device.
1671 * As we setup these structures, we collect all bio's together into a list
1672 * which we then process collectively to add pages, and then process again
1673 * to pass to generic_make_request.
1675 * The r10_bio structures are linked using a borrowed master_bio pointer.
1676 * This link is counted in ->remaining. When the r10_bio that points to NULL
1677 * has its remaining count decremented to 0, the whole complex operation
1682 static sector_t
sync_request(mddev_t
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
1684 conf_t
*conf
= mddev_to_conf(mddev
);
1686 struct bio
*biolist
= NULL
, *bio
;
1687 sector_t max_sector
, nr_sectors
;
1693 sector_t sectors_skipped
= 0;
1694 int chunks_skipped
= 0;
1696 if (!conf
->r10buf_pool
)
1697 if (init_resync(conf
))
1701 max_sector
= mddev
->dev_sectors
;
1702 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
1703 max_sector
= mddev
->resync_max_sectors
;
1704 if (sector_nr
>= max_sector
) {
1705 /* If we aborted, we need to abort the
1706 * sync on the 'current' bitmap chucks (there can
1707 * be several when recovering multiple devices).
1708 * as we may have started syncing it but not finished.
1709 * We can find the current address in
1710 * mddev->curr_resync, but for recovery,
1711 * we need to convert that to several
1712 * virtual addresses.
1714 if (mddev
->curr_resync
< max_sector
) { /* aborted */
1715 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))
1716 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
1718 else for (i
=0; i
<conf
->raid_disks
; i
++) {
1720 raid10_find_virt(conf
, mddev
->curr_resync
, i
);
1721 bitmap_end_sync(mddev
->bitmap
, sect
,
1724 } else /* completed sync */
1727 bitmap_close_sync(mddev
->bitmap
);
1730 return sectors_skipped
;
1732 if (chunks_skipped
>= conf
->raid_disks
) {
1733 /* if there has been nothing to do on any drive,
1734 * then there is nothing to do at all..
1737 return (max_sector
- sector_nr
) + sectors_skipped
;
1740 if (max_sector
> mddev
->resync_max
)
1741 max_sector
= mddev
->resync_max
; /* Don't do IO beyond here */
1743 /* make sure whole request will fit in a chunk - if chunks
1746 if (conf
->near_copies
< conf
->raid_disks
&&
1747 max_sector
> (sector_nr
| conf
->chunk_mask
))
1748 max_sector
= (sector_nr
| conf
->chunk_mask
) + 1;
1750 * If there is non-resync activity waiting for us then
1751 * put in a delay to throttle resync.
1753 if (!go_faster
&& conf
->nr_waiting
)
1754 msleep_interruptible(1000);
1756 /* Again, very different code for resync and recovery.
1757 * Both must result in an r10bio with a list of bios that
1758 * have bi_end_io, bi_sector, bi_bdev set,
1759 * and bi_private set to the r10bio.
1760 * For recovery, we may actually create several r10bios
1761 * with 2 bios in each, that correspond to the bios in the main one.
1762 * In this case, the subordinate r10bios link back through a
1763 * borrowed master_bio pointer, and the counter in the master
1764 * includes a ref from each subordinate.
1766 /* First, we decide what to do and set ->bi_end_io
1767 * To end_sync_read if we want to read, and
1768 * end_sync_write if we will want to write.
1771 max_sync
= RESYNC_PAGES
<< (PAGE_SHIFT
-9);
1772 if (!test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
)) {
1773 /* recovery... the complicated one */
1777 for (i
=0 ; i
<conf
->raid_disks
; i
++)
1778 if (conf
->mirrors
[i
].rdev
&&
1779 !test_bit(In_sync
, &conf
->mirrors
[i
].rdev
->flags
)) {
1780 int still_degraded
= 0;
1781 /* want to reconstruct this device */
1782 r10bio_t
*rb2
= r10_bio
;
1783 sector_t sect
= raid10_find_virt(conf
, sector_nr
, i
);
1785 /* Unless we are doing a full sync, we only need
1786 * to recover the block if it is set in the bitmap
1788 must_sync
= bitmap_start_sync(mddev
->bitmap
, sect
,
1790 if (sync_blocks
< max_sync
)
1791 max_sync
= sync_blocks
;
1794 /* yep, skip the sync_blocks here, but don't assume
1795 * that there will never be anything to do here
1797 chunks_skipped
= -1;
1801 r10_bio
= mempool_alloc(conf
->r10buf_pool
, GFP_NOIO
);
1802 raise_barrier(conf
, rb2
!= NULL
);
1803 atomic_set(&r10_bio
->remaining
, 0);
1805 r10_bio
->master_bio
= (struct bio
*)rb2
;
1807 atomic_inc(&rb2
->remaining
);
1808 r10_bio
->mddev
= mddev
;
1809 set_bit(R10BIO_IsRecover
, &r10_bio
->state
);
1810 r10_bio
->sector
= sect
;
1812 raid10_find_phys(conf
, r10_bio
);
1813 /* Need to check if this section will still be
1816 for (j
=0; j
<conf
->copies
;j
++) {
1817 int d
= r10_bio
->devs
[j
].devnum
;
1818 if (conf
->mirrors
[d
].rdev
== NULL
||
1819 test_bit(Faulty
, &conf
->mirrors
[d
].rdev
->flags
)) {
1824 must_sync
= bitmap_start_sync(mddev
->bitmap
, sect
,
1825 &sync_blocks
, still_degraded
);
1827 for (j
=0; j
<conf
->copies
;j
++) {
1828 int d
= r10_bio
->devs
[j
].devnum
;
1829 if (conf
->mirrors
[d
].rdev
&&
1830 test_bit(In_sync
, &conf
->mirrors
[d
].rdev
->flags
)) {
1831 /* This is where we read from */
1832 bio
= r10_bio
->devs
[0].bio
;
1833 bio
->bi_next
= biolist
;
1835 bio
->bi_private
= r10_bio
;
1836 bio
->bi_end_io
= end_sync_read
;
1838 bio
->bi_sector
= r10_bio
->devs
[j
].addr
+
1839 conf
->mirrors
[d
].rdev
->data_offset
;
1840 bio
->bi_bdev
= conf
->mirrors
[d
].rdev
->bdev
;
1841 atomic_inc(&conf
->mirrors
[d
].rdev
->nr_pending
);
1842 atomic_inc(&r10_bio
->remaining
);
1843 /* and we write to 'i' */
1845 for (k
=0; k
<conf
->copies
; k
++)
1846 if (r10_bio
->devs
[k
].devnum
== i
)
1848 BUG_ON(k
== conf
->copies
);
1849 bio
= r10_bio
->devs
[1].bio
;
1850 bio
->bi_next
= biolist
;
1852 bio
->bi_private
= r10_bio
;
1853 bio
->bi_end_io
= end_sync_write
;
1855 bio
->bi_sector
= r10_bio
->devs
[k
].addr
+
1856 conf
->mirrors
[i
].rdev
->data_offset
;
1857 bio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1859 r10_bio
->devs
[0].devnum
= d
;
1860 r10_bio
->devs
[1].devnum
= i
;
1865 if (j
== conf
->copies
) {
1866 /* Cannot recover, so abort the recovery */
1869 atomic_dec(&rb2
->remaining
);
1871 if (!test_and_set_bit(MD_RECOVERY_INTR
,
1873 printk(KERN_INFO
"raid10: %s: insufficient working devices for recovery.\n",
1878 if (biolist
== NULL
) {
1880 r10bio_t
*rb2
= r10_bio
;
1881 r10_bio
= (r10bio_t
*) rb2
->master_bio
;
1882 rb2
->master_bio
= NULL
;
1888 /* resync. Schedule a read for every block at this virt offset */
1891 bitmap_cond_end_sync(mddev
->bitmap
, sector_nr
);
1893 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
,
1894 &sync_blocks
, mddev
->degraded
) &&
1895 !conf
->fullsync
&& !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
1896 /* We can skip this block */
1898 return sync_blocks
+ sectors_skipped
;
1900 if (sync_blocks
< max_sync
)
1901 max_sync
= sync_blocks
;
1902 r10_bio
= mempool_alloc(conf
->r10buf_pool
, GFP_NOIO
);
1904 r10_bio
->mddev
= mddev
;
1905 atomic_set(&r10_bio
->remaining
, 0);
1906 raise_barrier(conf
, 0);
1907 conf
->next_resync
= sector_nr
;
1909 r10_bio
->master_bio
= NULL
;
1910 r10_bio
->sector
= sector_nr
;
1911 set_bit(R10BIO_IsSync
, &r10_bio
->state
);
1912 raid10_find_phys(conf
, r10_bio
);
1913 r10_bio
->sectors
= (sector_nr
| conf
->chunk_mask
) - sector_nr
+1;
1915 for (i
=0; i
<conf
->copies
; i
++) {
1916 int d
= r10_bio
->devs
[i
].devnum
;
1917 bio
= r10_bio
->devs
[i
].bio
;
1918 bio
->bi_end_io
= NULL
;
1919 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1920 if (conf
->mirrors
[d
].rdev
== NULL
||
1921 test_bit(Faulty
, &conf
->mirrors
[d
].rdev
->flags
))
1923 atomic_inc(&conf
->mirrors
[d
].rdev
->nr_pending
);
1924 atomic_inc(&r10_bio
->remaining
);
1925 bio
->bi_next
= biolist
;
1927 bio
->bi_private
= r10_bio
;
1928 bio
->bi_end_io
= end_sync_read
;
1930 bio
->bi_sector
= r10_bio
->devs
[i
].addr
+
1931 conf
->mirrors
[d
].rdev
->data_offset
;
1932 bio
->bi_bdev
= conf
->mirrors
[d
].rdev
->bdev
;
1937 for (i
=0; i
<conf
->copies
; i
++) {
1938 int d
= r10_bio
->devs
[i
].devnum
;
1939 if (r10_bio
->devs
[i
].bio
->bi_end_io
)
1940 rdev_dec_pending(conf
->mirrors
[d
].rdev
, mddev
);
1948 for (bio
= biolist
; bio
; bio
=bio
->bi_next
) {
1950 bio
->bi_flags
&= ~(BIO_POOL_MASK
- 1);
1952 bio
->bi_flags
|= 1 << BIO_UPTODATE
;
1955 bio
->bi_phys_segments
= 0;
1960 if (sector_nr
+ max_sync
< max_sector
)
1961 max_sector
= sector_nr
+ max_sync
;
1964 int len
= PAGE_SIZE
;
1966 if (sector_nr
+ (len
>>9) > max_sector
)
1967 len
= (max_sector
- sector_nr
) << 9;
1970 for (bio
= biolist
; bio
; bio
=bio
->bi_next
) {
1971 page
= bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
;
1972 if (bio_add_page(bio
, page
, len
, 0) == 0) {
1975 bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
= page
;
1976 for (bio2
= biolist
; bio2
&& bio2
!= bio
; bio2
= bio2
->bi_next
) {
1977 /* remove last page from this bio */
1979 bio2
->bi_size
-= len
;
1980 bio2
->bi_flags
&= ~(1<< BIO_SEG_VALID
);
1986 nr_sectors
+= len
>>9;
1987 sector_nr
+= len
>>9;
1988 } while (biolist
->bi_vcnt
< RESYNC_PAGES
);
1990 r10_bio
->sectors
= nr_sectors
;
1994 biolist
= biolist
->bi_next
;
1996 bio
->bi_next
= NULL
;
1997 r10_bio
= bio
->bi_private
;
1998 r10_bio
->sectors
= nr_sectors
;
2000 if (bio
->bi_end_io
== end_sync_read
) {
2001 md_sync_acct(bio
->bi_bdev
, nr_sectors
);
2002 generic_make_request(bio
);
2006 if (sectors_skipped
)
2007 /* pretend they weren't skipped, it makes
2008 * no important difference in this case
2010 md_done_sync(mddev
, sectors_skipped
, 1);
2012 return sectors_skipped
+ nr_sectors
;
2014 /* There is nowhere to write, so all non-sync
2015 * drives must be failed, so try the next chunk...
2017 if (sector_nr
+ max_sync
< max_sector
)
2018 max_sector
= sector_nr
+ max_sync
;
2020 sectors_skipped
+= (max_sector
- sector_nr
);
2022 sector_nr
= max_sector
;
2027 raid10_size(mddev_t
*mddev
, sector_t sectors
, int raid_disks
)
2030 conf_t
*conf
= mddev_to_conf(mddev
);
2033 raid_disks
= mddev
->raid_disks
;
2035 sectors
= mddev
->dev_sectors
;
2037 size
= sectors
>> conf
->chunk_shift
;
2038 sector_div(size
, conf
->far_copies
);
2039 size
= size
* raid_disks
;
2040 sector_div(size
, conf
->near_copies
);
2042 return size
<< conf
->chunk_shift
;
2045 static int run(mddev_t
*mddev
)
2049 mirror_info_t
*disk
;
2052 sector_t stride
, size
;
2054 if (mddev
->chunk_size
< PAGE_SIZE
) {
2055 printk(KERN_ERR
"md/raid10: chunk size must be "
2056 "at least PAGE_SIZE(%ld).\n", PAGE_SIZE
);
2060 nc
= mddev
->layout
& 255;
2061 fc
= (mddev
->layout
>> 8) & 255;
2062 fo
= mddev
->layout
& (1<<16);
2063 if ((nc
*fc
) <2 || (nc
*fc
) > mddev
->raid_disks
||
2064 (mddev
->layout
>> 17)) {
2065 printk(KERN_ERR
"raid10: %s: unsupported raid10 layout: 0x%8x\n",
2066 mdname(mddev
), mddev
->layout
);
2070 * copy the already verified devices into our private RAID10
2071 * bookkeeping area. [whatever we allocate in run(),
2072 * should be freed in stop()]
2074 conf
= kzalloc(sizeof(conf_t
), GFP_KERNEL
);
2075 mddev
->private = conf
;
2077 printk(KERN_ERR
"raid10: couldn't allocate memory for %s\n",
2081 conf
->mirrors
= kzalloc(sizeof(struct mirror_info
)*mddev
->raid_disks
,
2083 if (!conf
->mirrors
) {
2084 printk(KERN_ERR
"raid10: couldn't allocate memory for %s\n",
2089 conf
->tmppage
= alloc_page(GFP_KERNEL
);
2093 conf
->mddev
= mddev
;
2094 conf
->raid_disks
= mddev
->raid_disks
;
2095 conf
->near_copies
= nc
;
2096 conf
->far_copies
= fc
;
2097 conf
->copies
= nc
*fc
;
2098 conf
->far_offset
= fo
;
2099 conf
->chunk_mask
= (sector_t
)(mddev
->chunk_size
>>9)-1;
2100 conf
->chunk_shift
= ffz(~mddev
->chunk_size
) - 9;
2101 size
= mddev
->dev_sectors
>> conf
->chunk_shift
;
2102 sector_div(size
, fc
);
2103 size
= size
* conf
->raid_disks
;
2104 sector_div(size
, nc
);
2105 /* 'size' is now the number of chunks in the array */
2106 /* calculate "used chunks per device" in 'stride' */
2107 stride
= size
* conf
->copies
;
2109 /* We need to round up when dividing by raid_disks to
2110 * get the stride size.
2112 stride
+= conf
->raid_disks
- 1;
2113 sector_div(stride
, conf
->raid_disks
);
2114 mddev
->dev_sectors
= stride
<< conf
->chunk_shift
;
2119 sector_div(stride
, fc
);
2120 conf
->stride
= stride
<< conf
->chunk_shift
;
2122 conf
->r10bio_pool
= mempool_create(NR_RAID10_BIOS
, r10bio_pool_alloc
,
2123 r10bio_pool_free
, conf
);
2124 if (!conf
->r10bio_pool
) {
2125 printk(KERN_ERR
"raid10: couldn't allocate memory for %s\n",
2130 spin_lock_init(&conf
->device_lock
);
2131 mddev
->queue
->queue_lock
= &conf
->device_lock
;
2133 list_for_each_entry(rdev
, &mddev
->disks
, same_set
) {
2134 disk_idx
= rdev
->raid_disk
;
2135 if (disk_idx
>= mddev
->raid_disks
2138 disk
= conf
->mirrors
+ disk_idx
;
2142 blk_queue_stack_limits(mddev
->queue
,
2143 rdev
->bdev
->bd_disk
->queue
);
2144 /* as we don't honour merge_bvec_fn, we must never risk
2145 * violating it, so limit ->max_sector to one PAGE, as
2146 * a one page request is never in violation.
2148 if (rdev
->bdev
->bd_disk
->queue
->merge_bvec_fn
&&
2149 mddev
->queue
->max_sectors
> (PAGE_SIZE
>>9))
2150 mddev
->queue
->max_sectors
= (PAGE_SIZE
>>9);
2152 disk
->head_position
= 0;
2154 INIT_LIST_HEAD(&conf
->retry_list
);
2156 spin_lock_init(&conf
->resync_lock
);
2157 init_waitqueue_head(&conf
->wait_barrier
);
2159 /* need to check that every block has at least one working mirror */
2160 if (!enough(conf
)) {
2161 printk(KERN_ERR
"raid10: not enough operational mirrors for %s\n",
2166 mddev
->degraded
= 0;
2167 for (i
= 0; i
< conf
->raid_disks
; i
++) {
2169 disk
= conf
->mirrors
+ i
;
2172 !test_bit(In_sync
, &disk
->rdev
->flags
)) {
2173 disk
->head_position
= 0;
2181 mddev
->thread
= md_register_thread(raid10d
, mddev
, "%s_raid10");
2182 if (!mddev
->thread
) {
2184 "raid10: couldn't allocate thread for %s\n",
2190 "raid10: raid set %s active with %d out of %d devices\n",
2191 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
2194 * Ok, everything is just fine now
2196 md_set_array_sectors(mddev
, raid10_size(mddev
, 0, 0));
2197 mddev
->resync_max_sectors
= raid10_size(mddev
, 0, 0);
2199 mddev
->queue
->unplug_fn
= raid10_unplug
;
2200 mddev
->queue
->backing_dev_info
.congested_fn
= raid10_congested
;
2201 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
2203 /* Calculate max read-ahead size.
2204 * We need to readahead at least twice a whole stripe....
2208 int stripe
= conf
->raid_disks
* (mddev
->chunk_size
/ PAGE_SIZE
);
2209 stripe
/= conf
->near_copies
;
2210 if (mddev
->queue
->backing_dev_info
.ra_pages
< 2* stripe
)
2211 mddev
->queue
->backing_dev_info
.ra_pages
= 2* stripe
;
2214 if (conf
->near_copies
< mddev
->raid_disks
)
2215 blk_queue_merge_bvec(mddev
->queue
, raid10_mergeable_bvec
);
2219 if (conf
->r10bio_pool
)
2220 mempool_destroy(conf
->r10bio_pool
);
2221 safe_put_page(conf
->tmppage
);
2222 kfree(conf
->mirrors
);
2224 mddev
->private = NULL
;
2229 static int stop(mddev_t
*mddev
)
2231 conf_t
*conf
= mddev_to_conf(mddev
);
2233 raise_barrier(conf
, 0);
2234 lower_barrier(conf
);
2236 md_unregister_thread(mddev
->thread
);
2237 mddev
->thread
= NULL
;
2238 blk_sync_queue(mddev
->queue
); /* the unplug fn references 'conf'*/
2239 if (conf
->r10bio_pool
)
2240 mempool_destroy(conf
->r10bio_pool
);
2241 kfree(conf
->mirrors
);
2243 mddev
->private = NULL
;
2247 static void raid10_quiesce(mddev_t
*mddev
, int state
)
2249 conf_t
*conf
= mddev_to_conf(mddev
);
2253 raise_barrier(conf
, 0);
2256 lower_barrier(conf
);
2259 if (mddev
->thread
) {
2261 mddev
->thread
->timeout
= mddev
->bitmap
->daemon_sleep
* HZ
;
2263 mddev
->thread
->timeout
= MAX_SCHEDULE_TIMEOUT
;
2264 md_wakeup_thread(mddev
->thread
);
2268 static struct mdk_personality raid10_personality
=
2272 .owner
= THIS_MODULE
,
2273 .make_request
= make_request
,
2277 .error_handler
= error
,
2278 .hot_add_disk
= raid10_add_disk
,
2279 .hot_remove_disk
= raid10_remove_disk
,
2280 .spare_active
= raid10_spare_active
,
2281 .sync_request
= sync_request
,
2282 .quiesce
= raid10_quiesce
,
2283 .size
= raid10_size
,
2286 static int __init
raid_init(void)
2288 return register_md_personality(&raid10_personality
);
2291 static void raid_exit(void)
2293 unregister_md_personality(&raid10_personality
);
2296 module_init(raid_init
);
2297 module_exit(raid_exit
);
2298 MODULE_LICENSE("GPL");
2299 MODULE_ALIAS("md-personality-9"); /* RAID10 */
2300 MODULE_ALIAS("md-raid10");
2301 MODULE_ALIAS("md-level-10");