2 * raid1.c : Multiple Devices driver for Linux
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
8 * RAID-1 management functions.
10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
12 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
13 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
15 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
16 * bitmapped intelligence in resync:
18 * - bitmap marked during normal i/o
19 * - bitmap used to skip nondirty blocks during sync
21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
22 * - persistent bitmap code
24 * This program is free software; you can redistribute it and/or modify
25 * it under the terms of the GNU General Public License as published by
26 * the Free Software Foundation; either version 2, or (at your option)
29 * You should have received a copy of the GNU General Public License
30 * (for example /usr/src/linux/COPYING); if not, write to the Free
31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
34 #include <linux/slab.h>
35 #include <linux/delay.h>
36 #include <linux/blkdev.h>
37 #include <linux/module.h>
38 #include <linux/seq_file.h>
39 #include <linux/ratelimit.h>
45 * Number of guaranteed r1bios in case of extreme VM load:
47 #define NR_RAID1_BIOS 256
49 /* when we get a read error on a read-only array, we redirect to another
50 * device without failing the first device, or trying to over-write to
51 * correct the read error. To keep track of bad blocks on a per-bio
52 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
54 #define IO_BLOCKED ((struct bio *)1)
55 /* When we successfully write to a known bad-block, we need to remove the
56 * bad-block marking which must be done from process context. So we record
57 * the success by setting devs[n].bio to IO_MADE_GOOD
59 #define IO_MADE_GOOD ((struct bio *)2)
61 #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
63 /* When there are this many requests queue to be written by
64 * the raid1 thread, we become 'congested' to provide back-pressure
67 static int max_queued_requests
= 1024;
69 static void allow_barrier(struct r1conf
*conf
);
70 static void lower_barrier(struct r1conf
*conf
);
72 static void * r1bio_pool_alloc(gfp_t gfp_flags
, void *data
)
74 struct pool_info
*pi
= data
;
75 int size
= offsetof(struct r1bio
, bios
[pi
->raid_disks
]);
77 /* allocate a r1bio with room for raid_disks entries in the bios array */
78 return kzalloc(size
, gfp_flags
);
81 static void r1bio_pool_free(void *r1_bio
, void *data
)
86 #define RESYNC_BLOCK_SIZE (64*1024)
87 //#define RESYNC_BLOCK_SIZE PAGE_SIZE
88 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
89 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
90 #define RESYNC_WINDOW (2048*1024)
92 static void * r1buf_pool_alloc(gfp_t gfp_flags
, void *data
)
94 struct pool_info
*pi
= data
;
100 r1_bio
= r1bio_pool_alloc(gfp_flags
, pi
);
105 * Allocate bios : 1 for reading, n-1 for writing
107 for (j
= pi
->raid_disks
; j
-- ; ) {
108 bio
= bio_kmalloc(gfp_flags
, RESYNC_PAGES
);
111 r1_bio
->bios
[j
] = bio
;
114 * Allocate RESYNC_PAGES data pages and attach them to
116 * If this is a user-requested check/repair, allocate
117 * RESYNC_PAGES for each bio.
119 if (test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
))
124 bio
= r1_bio
->bios
[j
];
125 for (i
= 0; i
< RESYNC_PAGES
; i
++) {
126 page
= alloc_page(gfp_flags
);
130 bio
->bi_io_vec
[i
].bv_page
= page
;
134 /* If not user-requests, copy the page pointers to all bios */
135 if (!test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
)) {
136 for (i
=0; i
<RESYNC_PAGES
; i
++)
137 for (j
=1; j
<pi
->raid_disks
; j
++)
138 r1_bio
->bios
[j
]->bi_io_vec
[i
].bv_page
=
139 r1_bio
->bios
[0]->bi_io_vec
[i
].bv_page
;
142 r1_bio
->master_bio
= NULL
;
147 for (j
=0 ; j
< pi
->raid_disks
; j
++)
148 for (i
=0; i
< r1_bio
->bios
[j
]->bi_vcnt
; i
++)
149 put_page(r1_bio
->bios
[j
]->bi_io_vec
[i
].bv_page
);
152 while (++j
< pi
->raid_disks
)
153 bio_put(r1_bio
->bios
[j
]);
154 r1bio_pool_free(r1_bio
, data
);
158 static void r1buf_pool_free(void *__r1_bio
, void *data
)
160 struct pool_info
*pi
= data
;
162 struct r1bio
*r1bio
= __r1_bio
;
164 for (i
= 0; i
< RESYNC_PAGES
; i
++)
165 for (j
= pi
->raid_disks
; j
-- ;) {
167 r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
!=
168 r1bio
->bios
[0]->bi_io_vec
[i
].bv_page
)
169 safe_put_page(r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
);
171 for (i
=0 ; i
< pi
->raid_disks
; i
++)
172 bio_put(r1bio
->bios
[i
]);
174 r1bio_pool_free(r1bio
, data
);
177 static void put_all_bios(struct r1conf
*conf
, struct r1bio
*r1_bio
)
181 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
182 struct bio
**bio
= r1_bio
->bios
+ i
;
183 if (!BIO_SPECIAL(*bio
))
189 static void free_r1bio(struct r1bio
*r1_bio
)
191 struct r1conf
*conf
= r1_bio
->mddev
->private;
193 put_all_bios(conf
, r1_bio
);
194 mempool_free(r1_bio
, conf
->r1bio_pool
);
197 static void put_buf(struct r1bio
*r1_bio
)
199 struct r1conf
*conf
= r1_bio
->mddev
->private;
202 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
203 struct bio
*bio
= r1_bio
->bios
[i
];
205 rdev_dec_pending(conf
->mirrors
[i
].rdev
, r1_bio
->mddev
);
208 mempool_free(r1_bio
, conf
->r1buf_pool
);
213 static void reschedule_retry(struct r1bio
*r1_bio
)
216 struct mddev
*mddev
= r1_bio
->mddev
;
217 struct r1conf
*conf
= mddev
->private;
219 spin_lock_irqsave(&conf
->device_lock
, flags
);
220 list_add(&r1_bio
->retry_list
, &conf
->retry_list
);
222 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
224 wake_up(&conf
->wait_barrier
);
225 md_wakeup_thread(mddev
->thread
);
229 * raid_end_bio_io() is called when we have finished servicing a mirrored
230 * operation and are ready to return a success/failure code to the buffer
233 static void call_bio_endio(struct r1bio
*r1_bio
)
235 struct bio
*bio
= r1_bio
->master_bio
;
237 struct r1conf
*conf
= r1_bio
->mddev
->private;
239 if (bio
->bi_phys_segments
) {
241 spin_lock_irqsave(&conf
->device_lock
, flags
);
242 bio
->bi_phys_segments
--;
243 done
= (bio
->bi_phys_segments
== 0);
244 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
248 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
))
249 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
253 * Wake up any possible resync thread that waits for the device
260 static void raid_end_bio_io(struct r1bio
*r1_bio
)
262 struct bio
*bio
= r1_bio
->master_bio
;
264 /* if nobody has done the final endio yet, do it now */
265 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
266 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
267 (bio_data_dir(bio
) == WRITE
) ? "write" : "read",
268 (unsigned long long) bio
->bi_sector
,
269 (unsigned long long) bio
->bi_sector
+
270 (bio
->bi_size
>> 9) - 1);
272 call_bio_endio(r1_bio
);
278 * Update disk head position estimator based on IRQ completion info.
280 static inline void update_head_pos(int disk
, struct r1bio
*r1_bio
)
282 struct r1conf
*conf
= r1_bio
->mddev
->private;
284 conf
->mirrors
[disk
].head_position
=
285 r1_bio
->sector
+ (r1_bio
->sectors
);
289 * Find the disk number which triggered given bio
291 static int find_bio_disk(struct r1bio
*r1_bio
, struct bio
*bio
)
294 struct r1conf
*conf
= r1_bio
->mddev
->private;
295 int raid_disks
= conf
->raid_disks
;
297 for (mirror
= 0; mirror
< raid_disks
* 2; mirror
++)
298 if (r1_bio
->bios
[mirror
] == bio
)
301 BUG_ON(mirror
== raid_disks
* 2);
302 update_head_pos(mirror
, r1_bio
);
307 static void raid1_end_read_request(struct bio
*bio
, int error
)
309 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
310 struct r1bio
*r1_bio
= bio
->bi_private
;
312 struct r1conf
*conf
= r1_bio
->mddev
->private;
314 mirror
= r1_bio
->read_disk
;
316 * this branch is our 'one mirror IO has finished' event handler:
318 update_head_pos(mirror
, r1_bio
);
321 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
323 /* If all other devices have failed, we want to return
324 * the error upwards rather than fail the last device.
325 * Here we redefine "uptodate" to mean "Don't want to retry"
328 spin_lock_irqsave(&conf
->device_lock
, flags
);
329 if (r1_bio
->mddev
->degraded
== conf
->raid_disks
||
330 (r1_bio
->mddev
->degraded
== conf
->raid_disks
-1 &&
331 !test_bit(Faulty
, &conf
->mirrors
[mirror
].rdev
->flags
)))
333 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
337 raid_end_bio_io(r1_bio
);
342 char b
[BDEVNAME_SIZE
];
344 KERN_ERR
"md/raid1:%s: %s: "
345 "rescheduling sector %llu\n",
347 bdevname(conf
->mirrors
[mirror
].rdev
->bdev
,
349 (unsigned long long)r1_bio
->sector
);
350 set_bit(R1BIO_ReadError
, &r1_bio
->state
);
351 reschedule_retry(r1_bio
);
354 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
, conf
->mddev
);
357 static void close_write(struct r1bio
*r1_bio
)
359 /* it really is the end of this request */
360 if (test_bit(R1BIO_BehindIO
, &r1_bio
->state
)) {
361 /* free extra copy of the data pages */
362 int i
= r1_bio
->behind_page_count
;
364 safe_put_page(r1_bio
->behind_bvecs
[i
].bv_page
);
365 kfree(r1_bio
->behind_bvecs
);
366 r1_bio
->behind_bvecs
= NULL
;
368 /* clear the bitmap if all writes complete successfully */
369 bitmap_endwrite(r1_bio
->mddev
->bitmap
, r1_bio
->sector
,
371 !test_bit(R1BIO_Degraded
, &r1_bio
->state
),
372 test_bit(R1BIO_BehindIO
, &r1_bio
->state
));
373 md_write_end(r1_bio
->mddev
);
376 static void r1_bio_write_done(struct r1bio
*r1_bio
)
378 if (!atomic_dec_and_test(&r1_bio
->remaining
))
381 if (test_bit(R1BIO_WriteError
, &r1_bio
->state
))
382 reschedule_retry(r1_bio
);
385 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
))
386 reschedule_retry(r1_bio
);
388 raid_end_bio_io(r1_bio
);
392 static void raid1_end_write_request(struct bio
*bio
, int error
)
394 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
395 struct r1bio
*r1_bio
= bio
->bi_private
;
396 int mirror
, behind
= test_bit(R1BIO_BehindIO
, &r1_bio
->state
);
397 struct r1conf
*conf
= r1_bio
->mddev
->private;
398 struct bio
*to_put
= NULL
;
400 mirror
= find_bio_disk(r1_bio
, bio
);
403 * 'one mirror IO has finished' event handler:
406 set_bit(WriteErrorSeen
,
407 &conf
->mirrors
[mirror
].rdev
->flags
);
408 if (!test_and_set_bit(WantReplacement
,
409 &conf
->mirrors
[mirror
].rdev
->flags
))
410 set_bit(MD_RECOVERY_NEEDED
, &
411 conf
->mddev
->recovery
);
413 set_bit(R1BIO_WriteError
, &r1_bio
->state
);
416 * Set R1BIO_Uptodate in our master bio, so that we
417 * will return a good error code for to the higher
418 * levels even if IO on some other mirrored buffer
421 * The 'master' represents the composite IO operation
422 * to user-side. So if something waits for IO, then it
423 * will wait for the 'master' bio.
428 r1_bio
->bios
[mirror
] = NULL
;
430 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
432 /* Maybe we can clear some bad blocks. */
433 if (is_badblock(conf
->mirrors
[mirror
].rdev
,
434 r1_bio
->sector
, r1_bio
->sectors
,
435 &first_bad
, &bad_sectors
)) {
436 r1_bio
->bios
[mirror
] = IO_MADE_GOOD
;
437 set_bit(R1BIO_MadeGood
, &r1_bio
->state
);
442 if (test_bit(WriteMostly
, &conf
->mirrors
[mirror
].rdev
->flags
))
443 atomic_dec(&r1_bio
->behind_remaining
);
446 * In behind mode, we ACK the master bio once the I/O
447 * has safely reached all non-writemostly
448 * disks. Setting the Returned bit ensures that this
449 * gets done only once -- we don't ever want to return
450 * -EIO here, instead we'll wait
452 if (atomic_read(&r1_bio
->behind_remaining
) >= (atomic_read(&r1_bio
->remaining
)-1) &&
453 test_bit(R1BIO_Uptodate
, &r1_bio
->state
)) {
454 /* Maybe we can return now */
455 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
456 struct bio
*mbio
= r1_bio
->master_bio
;
457 pr_debug("raid1: behind end write sectors"
459 (unsigned long long) mbio
->bi_sector
,
460 (unsigned long long) mbio
->bi_sector
+
461 (mbio
->bi_size
>> 9) - 1);
462 call_bio_endio(r1_bio
);
466 if (r1_bio
->bios
[mirror
] == NULL
)
467 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
,
471 * Let's see if all mirrored write operations have finished
474 r1_bio_write_done(r1_bio
);
482 * This routine returns the disk from which the requested read should
483 * be done. There is a per-array 'next expected sequential IO' sector
484 * number - if this matches on the next IO then we use the last disk.
485 * There is also a per-disk 'last know head position' sector that is
486 * maintained from IRQ contexts, both the normal and the resync IO
487 * completion handlers update this position correctly. If there is no
488 * perfect sequential match then we pick the disk whose head is closest.
490 * If there are 2 mirrors in the same 2 devices, performance degrades
491 * because position is mirror, not device based.
493 * The rdev for the device selected will have nr_pending incremented.
495 static int read_balance(struct r1conf
*conf
, struct r1bio
*r1_bio
, int *max_sectors
)
497 const sector_t this_sector
= r1_bio
->sector
;
499 int best_good_sectors
;
500 int best_disk
, best_dist_disk
, best_pending_disk
;
504 unsigned int min_pending
;
505 struct md_rdev
*rdev
;
507 int choose_next_idle
;
511 * Check if we can balance. We can balance on the whole
512 * device if no resync is going on, or below the resync window.
513 * We take the first readable disk when above the resync window.
516 sectors
= r1_bio
->sectors
;
519 best_dist
= MaxSector
;
520 best_pending_disk
= -1;
521 min_pending
= UINT_MAX
;
522 best_good_sectors
= 0;
524 choose_next_idle
= 0;
526 if (conf
->mddev
->recovery_cp
< MaxSector
&&
527 (this_sector
+ sectors
>= conf
->next_resync
))
532 for (disk
= 0 ; disk
< conf
->raid_disks
* 2 ; disk
++) {
536 unsigned int pending
;
539 rdev
= rcu_dereference(conf
->mirrors
[disk
].rdev
);
540 if (r1_bio
->bios
[disk
] == IO_BLOCKED
542 || test_bit(Unmerged
, &rdev
->flags
)
543 || test_bit(Faulty
, &rdev
->flags
))
545 if (!test_bit(In_sync
, &rdev
->flags
) &&
546 rdev
->recovery_offset
< this_sector
+ sectors
)
548 if (test_bit(WriteMostly
, &rdev
->flags
)) {
549 /* Don't balance among write-mostly, just
550 * use the first as a last resort */
552 if (is_badblock(rdev
, this_sector
, sectors
,
553 &first_bad
, &bad_sectors
)) {
554 if (first_bad
< this_sector
)
555 /* Cannot use this */
557 best_good_sectors
= first_bad
- this_sector
;
559 best_good_sectors
= sectors
;
564 /* This is a reasonable device to use. It might
567 if (is_badblock(rdev
, this_sector
, sectors
,
568 &first_bad
, &bad_sectors
)) {
569 if (best_dist
< MaxSector
)
570 /* already have a better device */
572 if (first_bad
<= this_sector
) {
573 /* cannot read here. If this is the 'primary'
574 * device, then we must not read beyond
575 * bad_sectors from another device..
577 bad_sectors
-= (this_sector
- first_bad
);
578 if (choose_first
&& sectors
> bad_sectors
)
579 sectors
= bad_sectors
;
580 if (best_good_sectors
> sectors
)
581 best_good_sectors
= sectors
;
584 sector_t good_sectors
= first_bad
- this_sector
;
585 if (good_sectors
> best_good_sectors
) {
586 best_good_sectors
= good_sectors
;
594 best_good_sectors
= sectors
;
596 nonrot
= blk_queue_nonrot(bdev_get_queue(rdev
->bdev
));
597 has_nonrot_disk
|= nonrot
;
598 pending
= atomic_read(&rdev
->nr_pending
);
599 dist
= abs(this_sector
- conf
->mirrors
[disk
].head_position
);
604 /* Don't change to another disk for sequential reads */
605 if (conf
->mirrors
[disk
].next_seq_sect
== this_sector
607 int opt_iosize
= bdev_io_opt(rdev
->bdev
) >> 9;
608 struct raid1_info
*mirror
= &conf
->mirrors
[disk
];
612 * If buffered sequential IO size exceeds optimal
613 * iosize, check if there is idle disk. If yes, choose
614 * the idle disk. read_balance could already choose an
615 * idle disk before noticing it's a sequential IO in
616 * this disk. This doesn't matter because this disk
617 * will idle, next time it will be utilized after the
618 * first disk has IO size exceeds optimal iosize. In
619 * this way, iosize of the first disk will be optimal
620 * iosize at least. iosize of the second disk might be
621 * small, but not a big deal since when the second disk
622 * starts IO, the first disk is likely still busy.
624 if (nonrot
&& opt_iosize
> 0 &&
625 mirror
->seq_start
!= MaxSector
&&
626 mirror
->next_seq_sect
> opt_iosize
&&
627 mirror
->next_seq_sect
- opt_iosize
>=
629 choose_next_idle
= 1;
634 /* If device is idle, use it */
640 if (choose_next_idle
)
643 if (min_pending
> pending
) {
644 min_pending
= pending
;
645 best_pending_disk
= disk
;
648 if (dist
< best_dist
) {
650 best_dist_disk
= disk
;
655 * If all disks are rotational, choose the closest disk. If any disk is
656 * non-rotational, choose the disk with less pending request even the
657 * disk is rotational, which might/might not be optimal for raids with
658 * mixed ratation/non-rotational disks depending on workload.
660 if (best_disk
== -1) {
662 best_disk
= best_pending_disk
;
664 best_disk
= best_dist_disk
;
667 if (best_disk
>= 0) {
668 rdev
= rcu_dereference(conf
->mirrors
[best_disk
].rdev
);
671 atomic_inc(&rdev
->nr_pending
);
672 if (test_bit(Faulty
, &rdev
->flags
)) {
673 /* cannot risk returning a device that failed
674 * before we inc'ed nr_pending
676 rdev_dec_pending(rdev
, conf
->mddev
);
679 sectors
= best_good_sectors
;
681 if (conf
->mirrors
[best_disk
].next_seq_sect
!= this_sector
)
682 conf
->mirrors
[best_disk
].seq_start
= this_sector
;
684 conf
->mirrors
[best_disk
].next_seq_sect
= this_sector
+ sectors
;
687 *max_sectors
= sectors
;
692 static int raid1_mergeable_bvec(struct request_queue
*q
,
693 struct bvec_merge_data
*bvm
,
694 struct bio_vec
*biovec
)
696 struct mddev
*mddev
= q
->queuedata
;
697 struct r1conf
*conf
= mddev
->private;
698 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
699 int max
= biovec
->bv_len
;
701 if (mddev
->merge_check_needed
) {
704 for (disk
= 0; disk
< conf
->raid_disks
* 2; disk
++) {
705 struct md_rdev
*rdev
= rcu_dereference(
706 conf
->mirrors
[disk
].rdev
);
707 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
708 struct request_queue
*q
=
709 bdev_get_queue(rdev
->bdev
);
710 if (q
->merge_bvec_fn
) {
711 bvm
->bi_sector
= sector
+
713 bvm
->bi_bdev
= rdev
->bdev
;
714 max
= min(max
, q
->merge_bvec_fn(
725 int md_raid1_congested(struct mddev
*mddev
, int bits
)
727 struct r1conf
*conf
= mddev
->private;
730 if ((bits
& (1 << BDI_async_congested
)) &&
731 conf
->pending_count
>= max_queued_requests
)
735 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
736 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
737 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
738 struct request_queue
*q
= bdev_get_queue(rdev
->bdev
);
742 /* Note the '|| 1' - when read_balance prefers
743 * non-congested targets, it can be removed
745 if ((bits
& (1<<BDI_async_congested
)) || 1)
746 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
748 ret
&= bdi_congested(&q
->backing_dev_info
, bits
);
754 EXPORT_SYMBOL_GPL(md_raid1_congested
);
756 static int raid1_congested(void *data
, int bits
)
758 struct mddev
*mddev
= data
;
760 return mddev_congested(mddev
, bits
) ||
761 md_raid1_congested(mddev
, bits
);
764 static void flush_pending_writes(struct r1conf
*conf
)
766 /* Any writes that have been queued but are awaiting
767 * bitmap updates get flushed here.
769 spin_lock_irq(&conf
->device_lock
);
771 if (conf
->pending_bio_list
.head
) {
773 bio
= bio_list_get(&conf
->pending_bio_list
);
774 conf
->pending_count
= 0;
775 spin_unlock_irq(&conf
->device_lock
);
776 /* flush any pending bitmap writes to
777 * disk before proceeding w/ I/O */
778 bitmap_unplug(conf
->mddev
->bitmap
);
779 wake_up(&conf
->wait_barrier
);
781 while (bio
) { /* submit pending writes */
782 struct bio
*next
= bio
->bi_next
;
784 generic_make_request(bio
);
788 spin_unlock_irq(&conf
->device_lock
);
792 * Sometimes we need to suspend IO while we do something else,
793 * either some resync/recovery, or reconfigure the array.
794 * To do this we raise a 'barrier'.
795 * The 'barrier' is a counter that can be raised multiple times
796 * to count how many activities are happening which preclude
798 * We can only raise the barrier if there is no pending IO.
799 * i.e. if nr_pending == 0.
800 * We choose only to raise the barrier if no-one is waiting for the
801 * barrier to go down. This means that as soon as an IO request
802 * is ready, no other operations which require a barrier will start
803 * until the IO request has had a chance.
805 * So: regular IO calls 'wait_barrier'. When that returns there
806 * is no backgroup IO happening, It must arrange to call
807 * allow_barrier when it has finished its IO.
808 * backgroup IO calls must call raise_barrier. Once that returns
809 * there is no normal IO happeing. It must arrange to call
810 * lower_barrier when the particular background IO completes.
812 #define RESYNC_DEPTH 32
814 static void raise_barrier(struct r1conf
*conf
)
816 spin_lock_irq(&conf
->resync_lock
);
818 /* Wait until no block IO is waiting */
819 wait_event_lock_irq(conf
->wait_barrier
, !conf
->nr_waiting
,
820 conf
->resync_lock
, );
822 /* block any new IO from starting */
825 /* Now wait for all pending IO to complete */
826 wait_event_lock_irq(conf
->wait_barrier
,
827 !conf
->nr_pending
&& conf
->barrier
< RESYNC_DEPTH
,
828 conf
->resync_lock
, );
830 spin_unlock_irq(&conf
->resync_lock
);
833 static void lower_barrier(struct r1conf
*conf
)
836 BUG_ON(conf
->barrier
<= 0);
837 spin_lock_irqsave(&conf
->resync_lock
, flags
);
839 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
840 wake_up(&conf
->wait_barrier
);
843 static void wait_barrier(struct r1conf
*conf
)
845 spin_lock_irq(&conf
->resync_lock
);
848 /* Wait for the barrier to drop.
849 * However if there are already pending
850 * requests (preventing the barrier from
851 * rising completely), and the
852 * pre-process bio queue isn't empty,
853 * then don't wait, as we need to empty
854 * that queue to get the nr_pending
857 wait_event_lock_irq(conf
->wait_barrier
,
861 !bio_list_empty(current
->bio_list
)),
867 spin_unlock_irq(&conf
->resync_lock
);
870 static void allow_barrier(struct r1conf
*conf
)
873 spin_lock_irqsave(&conf
->resync_lock
, flags
);
875 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
876 wake_up(&conf
->wait_barrier
);
879 static void freeze_array(struct r1conf
*conf
)
881 /* stop syncio and normal IO and wait for everything to
883 * We increment barrier and nr_waiting, and then
884 * wait until nr_pending match nr_queued+1
885 * This is called in the context of one normal IO request
886 * that has failed. Thus any sync request that might be pending
887 * will be blocked by nr_pending, and we need to wait for
888 * pending IO requests to complete or be queued for re-try.
889 * Thus the number queued (nr_queued) plus this request (1)
890 * must match the number of pending IOs (nr_pending) before
893 spin_lock_irq(&conf
->resync_lock
);
896 wait_event_lock_irq(conf
->wait_barrier
,
897 conf
->nr_pending
== conf
->nr_queued
+1,
899 flush_pending_writes(conf
));
900 spin_unlock_irq(&conf
->resync_lock
);
902 static void unfreeze_array(struct r1conf
*conf
)
904 /* reverse the effect of the freeze */
905 spin_lock_irq(&conf
->resync_lock
);
908 wake_up(&conf
->wait_barrier
);
909 spin_unlock_irq(&conf
->resync_lock
);
913 /* duplicate the data pages for behind I/O
915 static void alloc_behind_pages(struct bio
*bio
, struct r1bio
*r1_bio
)
918 struct bio_vec
*bvec
;
919 struct bio_vec
*bvecs
= kzalloc(bio
->bi_vcnt
* sizeof(struct bio_vec
),
921 if (unlikely(!bvecs
))
924 bio_for_each_segment(bvec
, bio
, i
) {
926 bvecs
[i
].bv_page
= alloc_page(GFP_NOIO
);
927 if (unlikely(!bvecs
[i
].bv_page
))
929 memcpy(kmap(bvecs
[i
].bv_page
) + bvec
->bv_offset
,
930 kmap(bvec
->bv_page
) + bvec
->bv_offset
, bvec
->bv_len
);
931 kunmap(bvecs
[i
].bv_page
);
932 kunmap(bvec
->bv_page
);
934 r1_bio
->behind_bvecs
= bvecs
;
935 r1_bio
->behind_page_count
= bio
->bi_vcnt
;
936 set_bit(R1BIO_BehindIO
, &r1_bio
->state
);
940 for (i
= 0; i
< bio
->bi_vcnt
; i
++)
941 if (bvecs
[i
].bv_page
)
942 put_page(bvecs
[i
].bv_page
);
944 pr_debug("%dB behind alloc failed, doing sync I/O\n", bio
->bi_size
);
947 static void make_request(struct mddev
*mddev
, struct bio
* bio
)
949 struct r1conf
*conf
= mddev
->private;
950 struct raid1_info
*mirror
;
951 struct r1bio
*r1_bio
;
952 struct bio
*read_bio
;
954 struct bitmap
*bitmap
;
956 const int rw
= bio_data_dir(bio
);
957 const unsigned long do_sync
= (bio
->bi_rw
& REQ_SYNC
);
958 const unsigned long do_flush_fua
= (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
));
959 struct md_rdev
*blocked_rdev
;
965 * Register the new request and wait if the reconstruction
966 * thread has put up a bar for new requests.
967 * Continue immediately if no resync is active currently.
970 md_write_start(mddev
, bio
); /* wait on superblock update early */
972 if (bio_data_dir(bio
) == WRITE
&&
973 bio
->bi_sector
+ bio
->bi_size
/512 > mddev
->suspend_lo
&&
974 bio
->bi_sector
< mddev
->suspend_hi
) {
975 /* As the suspend_* range is controlled by
976 * userspace, we want an interruptible
981 flush_signals(current
);
982 prepare_to_wait(&conf
->wait_barrier
,
983 &w
, TASK_INTERRUPTIBLE
);
984 if (bio
->bi_sector
+ bio
->bi_size
/512 <= mddev
->suspend_lo
||
985 bio
->bi_sector
>= mddev
->suspend_hi
)
989 finish_wait(&conf
->wait_barrier
, &w
);
994 bitmap
= mddev
->bitmap
;
997 * make_request() can abort the operation when READA is being
998 * used and no empty request is available.
1001 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1003 r1_bio
->master_bio
= bio
;
1004 r1_bio
->sectors
= bio
->bi_size
>> 9;
1006 r1_bio
->mddev
= mddev
;
1007 r1_bio
->sector
= bio
->bi_sector
;
1009 /* We might need to issue multiple reads to different
1010 * devices if there are bad blocks around, so we keep
1011 * track of the number of reads in bio->bi_phys_segments.
1012 * If this is 0, there is only one r1_bio and no locking
1013 * will be needed when requests complete. If it is
1014 * non-zero, then it is the number of not-completed requests.
1016 bio
->bi_phys_segments
= 0;
1017 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
1021 * read balancing logic:
1026 rdisk
= read_balance(conf
, r1_bio
, &max_sectors
);
1029 /* couldn't find anywhere to read from */
1030 raid_end_bio_io(r1_bio
);
1033 mirror
= conf
->mirrors
+ rdisk
;
1035 if (test_bit(WriteMostly
, &mirror
->rdev
->flags
) &&
1037 /* Reading from a write-mostly device must
1038 * take care not to over-take any writes
1041 wait_event(bitmap
->behind_wait
,
1042 atomic_read(&bitmap
->behind_writes
) == 0);
1044 r1_bio
->read_disk
= rdisk
;
1046 read_bio
= bio_clone_mddev(bio
, GFP_NOIO
, mddev
);
1047 md_trim_bio(read_bio
, r1_bio
->sector
- bio
->bi_sector
,
1050 r1_bio
->bios
[rdisk
] = read_bio
;
1052 read_bio
->bi_sector
= r1_bio
->sector
+ mirror
->rdev
->data_offset
;
1053 read_bio
->bi_bdev
= mirror
->rdev
->bdev
;
1054 read_bio
->bi_end_io
= raid1_end_read_request
;
1055 read_bio
->bi_rw
= READ
| do_sync
;
1056 read_bio
->bi_private
= r1_bio
;
1058 if (max_sectors
< r1_bio
->sectors
) {
1059 /* could not read all from this device, so we will
1060 * need another r1_bio.
1063 sectors_handled
= (r1_bio
->sector
+ max_sectors
1065 r1_bio
->sectors
= max_sectors
;
1066 spin_lock_irq(&conf
->device_lock
);
1067 if (bio
->bi_phys_segments
== 0)
1068 bio
->bi_phys_segments
= 2;
1070 bio
->bi_phys_segments
++;
1071 spin_unlock_irq(&conf
->device_lock
);
1072 /* Cannot call generic_make_request directly
1073 * as that will be queued in __make_request
1074 * and subsequent mempool_alloc might block waiting
1075 * for it. So hand bio over to raid1d.
1077 reschedule_retry(r1_bio
);
1079 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1081 r1_bio
->master_bio
= bio
;
1082 r1_bio
->sectors
= (bio
->bi_size
>> 9) - sectors_handled
;
1084 r1_bio
->mddev
= mddev
;
1085 r1_bio
->sector
= bio
->bi_sector
+ sectors_handled
;
1088 generic_make_request(read_bio
);
1095 if (conf
->pending_count
>= max_queued_requests
) {
1096 md_wakeup_thread(mddev
->thread
);
1097 wait_event(conf
->wait_barrier
,
1098 conf
->pending_count
< max_queued_requests
);
1100 /* first select target devices under rcu_lock and
1101 * inc refcount on their rdev. Record them by setting
1103 * If there are known/acknowledged bad blocks on any device on
1104 * which we have seen a write error, we want to avoid writing those
1106 * This potentially requires several writes to write around
1107 * the bad blocks. Each set of writes gets it's own r1bio
1108 * with a set of bios attached.
1111 disks
= conf
->raid_disks
* 2;
1113 blocked_rdev
= NULL
;
1115 max_sectors
= r1_bio
->sectors
;
1116 for (i
= 0; i
< disks
; i
++) {
1117 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1118 if (rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
1119 atomic_inc(&rdev
->nr_pending
);
1120 blocked_rdev
= rdev
;
1123 r1_bio
->bios
[i
] = NULL
;
1124 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
)
1125 || test_bit(Unmerged
, &rdev
->flags
)) {
1126 if (i
< conf
->raid_disks
)
1127 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
1131 atomic_inc(&rdev
->nr_pending
);
1132 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
1137 is_bad
= is_badblock(rdev
, r1_bio
->sector
,
1139 &first_bad
, &bad_sectors
);
1141 /* mustn't write here until the bad block is
1143 set_bit(BlockedBadBlocks
, &rdev
->flags
);
1144 blocked_rdev
= rdev
;
1147 if (is_bad
&& first_bad
<= r1_bio
->sector
) {
1148 /* Cannot write here at all */
1149 bad_sectors
-= (r1_bio
->sector
- first_bad
);
1150 if (bad_sectors
< max_sectors
)
1151 /* mustn't write more than bad_sectors
1152 * to other devices yet
1154 max_sectors
= bad_sectors
;
1155 rdev_dec_pending(rdev
, mddev
);
1156 /* We don't set R1BIO_Degraded as that
1157 * only applies if the disk is
1158 * missing, so it might be re-added,
1159 * and we want to know to recover this
1161 * In this case the device is here,
1162 * and the fact that this chunk is not
1163 * in-sync is recorded in the bad
1169 int good_sectors
= first_bad
- r1_bio
->sector
;
1170 if (good_sectors
< max_sectors
)
1171 max_sectors
= good_sectors
;
1174 r1_bio
->bios
[i
] = bio
;
1178 if (unlikely(blocked_rdev
)) {
1179 /* Wait for this device to become unblocked */
1182 for (j
= 0; j
< i
; j
++)
1183 if (r1_bio
->bios
[j
])
1184 rdev_dec_pending(conf
->mirrors
[j
].rdev
, mddev
);
1186 allow_barrier(conf
);
1187 md_wait_for_blocked_rdev(blocked_rdev
, mddev
);
1192 if (max_sectors
< r1_bio
->sectors
) {
1193 /* We are splitting this write into multiple parts, so
1194 * we need to prepare for allocating another r1_bio.
1196 r1_bio
->sectors
= max_sectors
;
1197 spin_lock_irq(&conf
->device_lock
);
1198 if (bio
->bi_phys_segments
== 0)
1199 bio
->bi_phys_segments
= 2;
1201 bio
->bi_phys_segments
++;
1202 spin_unlock_irq(&conf
->device_lock
);
1204 sectors_handled
= r1_bio
->sector
+ max_sectors
- bio
->bi_sector
;
1206 atomic_set(&r1_bio
->remaining
, 1);
1207 atomic_set(&r1_bio
->behind_remaining
, 0);
1210 for (i
= 0; i
< disks
; i
++) {
1212 if (!r1_bio
->bios
[i
])
1215 mbio
= bio_clone_mddev(bio
, GFP_NOIO
, mddev
);
1216 md_trim_bio(mbio
, r1_bio
->sector
- bio
->bi_sector
, max_sectors
);
1220 * Not if there are too many, or cannot
1221 * allocate memory, or a reader on WriteMostly
1222 * is waiting for behind writes to flush */
1224 (atomic_read(&bitmap
->behind_writes
)
1225 < mddev
->bitmap_info
.max_write_behind
) &&
1226 !waitqueue_active(&bitmap
->behind_wait
))
1227 alloc_behind_pages(mbio
, r1_bio
);
1229 bitmap_startwrite(bitmap
, r1_bio
->sector
,
1231 test_bit(R1BIO_BehindIO
,
1235 if (r1_bio
->behind_bvecs
) {
1236 struct bio_vec
*bvec
;
1239 /* Yes, I really want the '__' version so that
1240 * we clear any unused pointer in the io_vec, rather
1241 * than leave them unchanged. This is important
1242 * because when we come to free the pages, we won't
1243 * know the original bi_idx, so we just free
1246 __bio_for_each_segment(bvec
, mbio
, j
, 0)
1247 bvec
->bv_page
= r1_bio
->behind_bvecs
[j
].bv_page
;
1248 if (test_bit(WriteMostly
, &conf
->mirrors
[i
].rdev
->flags
))
1249 atomic_inc(&r1_bio
->behind_remaining
);
1252 r1_bio
->bios
[i
] = mbio
;
1254 mbio
->bi_sector
= (r1_bio
->sector
+
1255 conf
->mirrors
[i
].rdev
->data_offset
);
1256 mbio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1257 mbio
->bi_end_io
= raid1_end_write_request
;
1258 mbio
->bi_rw
= WRITE
| do_flush_fua
| do_sync
;
1259 mbio
->bi_private
= r1_bio
;
1261 atomic_inc(&r1_bio
->remaining
);
1262 spin_lock_irqsave(&conf
->device_lock
, flags
);
1263 bio_list_add(&conf
->pending_bio_list
, mbio
);
1264 conf
->pending_count
++;
1265 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1266 if (!mddev_check_plugged(mddev
))
1267 md_wakeup_thread(mddev
->thread
);
1269 /* Mustn't call r1_bio_write_done before this next test,
1270 * as it could result in the bio being freed.
1272 if (sectors_handled
< (bio
->bi_size
>> 9)) {
1273 r1_bio_write_done(r1_bio
);
1274 /* We need another r1_bio. It has already been counted
1275 * in bio->bi_phys_segments
1277 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1278 r1_bio
->master_bio
= bio
;
1279 r1_bio
->sectors
= (bio
->bi_size
>> 9) - sectors_handled
;
1281 r1_bio
->mddev
= mddev
;
1282 r1_bio
->sector
= bio
->bi_sector
+ sectors_handled
;
1286 r1_bio_write_done(r1_bio
);
1288 /* In case raid1d snuck in to freeze_array */
1289 wake_up(&conf
->wait_barrier
);
1292 static void status(struct seq_file
*seq
, struct mddev
*mddev
)
1294 struct r1conf
*conf
= mddev
->private;
1297 seq_printf(seq
, " [%d/%d] [", conf
->raid_disks
,
1298 conf
->raid_disks
- mddev
->degraded
);
1300 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1301 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1302 seq_printf(seq
, "%s",
1303 rdev
&& test_bit(In_sync
, &rdev
->flags
) ? "U" : "_");
1306 seq_printf(seq
, "]");
1310 static void error(struct mddev
*mddev
, struct md_rdev
*rdev
)
1312 char b
[BDEVNAME_SIZE
];
1313 struct r1conf
*conf
= mddev
->private;
1316 * If it is not operational, then we have already marked it as dead
1317 * else if it is the last working disks, ignore the error, let the
1318 * next level up know.
1319 * else mark the drive as failed
1321 if (test_bit(In_sync
, &rdev
->flags
)
1322 && (conf
->raid_disks
- mddev
->degraded
) == 1) {
1324 * Don't fail the drive, act as though we were just a
1325 * normal single drive.
1326 * However don't try a recovery from this drive as
1327 * it is very likely to fail.
1329 conf
->recovery_disabled
= mddev
->recovery_disabled
;
1332 set_bit(Blocked
, &rdev
->flags
);
1333 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1334 unsigned long flags
;
1335 spin_lock_irqsave(&conf
->device_lock
, flags
);
1337 set_bit(Faulty
, &rdev
->flags
);
1338 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1340 * if recovery is running, make sure it aborts.
1342 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1344 set_bit(Faulty
, &rdev
->flags
);
1345 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1347 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1348 "md/raid1:%s: Operation continuing on %d devices.\n",
1349 mdname(mddev
), bdevname(rdev
->bdev
, b
),
1350 mdname(mddev
), conf
->raid_disks
- mddev
->degraded
);
1353 static void print_conf(struct r1conf
*conf
)
1357 printk(KERN_DEBUG
"RAID1 conf printout:\n");
1359 printk(KERN_DEBUG
"(!conf)\n");
1362 printk(KERN_DEBUG
" --- wd:%d rd:%d\n", conf
->raid_disks
- conf
->mddev
->degraded
,
1366 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1367 char b
[BDEVNAME_SIZE
];
1368 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1370 printk(KERN_DEBUG
" disk %d, wo:%d, o:%d, dev:%s\n",
1371 i
, !test_bit(In_sync
, &rdev
->flags
),
1372 !test_bit(Faulty
, &rdev
->flags
),
1373 bdevname(rdev
->bdev
,b
));
1378 static void close_sync(struct r1conf
*conf
)
1381 allow_barrier(conf
);
1383 mempool_destroy(conf
->r1buf_pool
);
1384 conf
->r1buf_pool
= NULL
;
1387 static int raid1_spare_active(struct mddev
*mddev
)
1390 struct r1conf
*conf
= mddev
->private;
1392 unsigned long flags
;
1395 * Find all failed disks within the RAID1 configuration
1396 * and mark them readable.
1397 * Called under mddev lock, so rcu protection not needed.
1399 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1400 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
1401 struct md_rdev
*repl
= conf
->mirrors
[conf
->raid_disks
+ i
].rdev
;
1403 && repl
->recovery_offset
== MaxSector
1404 && !test_bit(Faulty
, &repl
->flags
)
1405 && !test_and_set_bit(In_sync
, &repl
->flags
)) {
1406 /* replacement has just become active */
1408 !test_and_clear_bit(In_sync
, &rdev
->flags
))
1411 /* Replaced device not technically
1412 * faulty, but we need to be sure
1413 * it gets removed and never re-added
1415 set_bit(Faulty
, &rdev
->flags
);
1416 sysfs_notify_dirent_safe(
1421 && !test_bit(Faulty
, &rdev
->flags
)
1422 && !test_and_set_bit(In_sync
, &rdev
->flags
)) {
1424 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
1427 spin_lock_irqsave(&conf
->device_lock
, flags
);
1428 mddev
->degraded
-= count
;
1429 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1436 static int raid1_add_disk(struct mddev
*mddev
, struct md_rdev
*rdev
)
1438 struct r1conf
*conf
= mddev
->private;
1441 struct raid1_info
*p
;
1443 int last
= conf
->raid_disks
- 1;
1444 struct request_queue
*q
= bdev_get_queue(rdev
->bdev
);
1446 if (mddev
->recovery_disabled
== conf
->recovery_disabled
)
1449 if (rdev
->raid_disk
>= 0)
1450 first
= last
= rdev
->raid_disk
;
1452 if (q
->merge_bvec_fn
) {
1453 set_bit(Unmerged
, &rdev
->flags
);
1454 mddev
->merge_check_needed
= 1;
1457 for (mirror
= first
; mirror
<= last
; mirror
++) {
1458 p
= conf
->mirrors
+mirror
;
1461 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
1462 rdev
->data_offset
<< 9);
1464 p
->head_position
= 0;
1465 rdev
->raid_disk
= mirror
;
1467 /* As all devices are equivalent, we don't need a full recovery
1468 * if this was recently any drive of the array
1470 if (rdev
->saved_raid_disk
< 0)
1472 rcu_assign_pointer(p
->rdev
, rdev
);
1475 if (test_bit(WantReplacement
, &p
->rdev
->flags
) &&
1476 p
[conf
->raid_disks
].rdev
== NULL
) {
1477 /* Add this device as a replacement */
1478 clear_bit(In_sync
, &rdev
->flags
);
1479 set_bit(Replacement
, &rdev
->flags
);
1480 rdev
->raid_disk
= mirror
;
1483 rcu_assign_pointer(p
[conf
->raid_disks
].rdev
, rdev
);
1487 if (err
== 0 && test_bit(Unmerged
, &rdev
->flags
)) {
1488 /* Some requests might not have seen this new
1489 * merge_bvec_fn. We must wait for them to complete
1490 * before merging the device fully.
1491 * First we make sure any code which has tested
1492 * our function has submitted the request, then
1493 * we wait for all outstanding requests to complete.
1495 synchronize_sched();
1496 raise_barrier(conf
);
1497 lower_barrier(conf
);
1498 clear_bit(Unmerged
, &rdev
->flags
);
1500 md_integrity_add_rdev(rdev
, mddev
);
1505 static int raid1_remove_disk(struct mddev
*mddev
, struct md_rdev
*rdev
)
1507 struct r1conf
*conf
= mddev
->private;
1509 int number
= rdev
->raid_disk
;
1510 struct raid1_info
*p
= conf
->mirrors
+ number
;
1512 if (rdev
!= p
->rdev
)
1513 p
= conf
->mirrors
+ conf
->raid_disks
+ number
;
1516 if (rdev
== p
->rdev
) {
1517 if (test_bit(In_sync
, &rdev
->flags
) ||
1518 atomic_read(&rdev
->nr_pending
)) {
1522 /* Only remove non-faulty devices if recovery
1525 if (!test_bit(Faulty
, &rdev
->flags
) &&
1526 mddev
->recovery_disabled
!= conf
->recovery_disabled
&&
1527 mddev
->degraded
< conf
->raid_disks
) {
1533 if (atomic_read(&rdev
->nr_pending
)) {
1534 /* lost the race, try later */
1538 } else if (conf
->mirrors
[conf
->raid_disks
+ number
].rdev
) {
1539 /* We just removed a device that is being replaced.
1540 * Move down the replacement. We drain all IO before
1541 * doing this to avoid confusion.
1543 struct md_rdev
*repl
=
1544 conf
->mirrors
[conf
->raid_disks
+ number
].rdev
;
1545 raise_barrier(conf
);
1546 clear_bit(Replacement
, &repl
->flags
);
1548 conf
->mirrors
[conf
->raid_disks
+ number
].rdev
= NULL
;
1549 lower_barrier(conf
);
1550 clear_bit(WantReplacement
, &rdev
->flags
);
1552 clear_bit(WantReplacement
, &rdev
->flags
);
1553 err
= md_integrity_register(mddev
);
1562 static void end_sync_read(struct bio
*bio
, int error
)
1564 struct r1bio
*r1_bio
= bio
->bi_private
;
1566 update_head_pos(r1_bio
->read_disk
, r1_bio
);
1569 * we have read a block, now it needs to be re-written,
1570 * or re-read if the read failed.
1571 * We don't do much here, just schedule handling by raid1d
1573 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1574 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1576 if (atomic_dec_and_test(&r1_bio
->remaining
))
1577 reschedule_retry(r1_bio
);
1580 static void end_sync_write(struct bio
*bio
, int error
)
1582 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1583 struct r1bio
*r1_bio
= bio
->bi_private
;
1584 struct mddev
*mddev
= r1_bio
->mddev
;
1585 struct r1conf
*conf
= mddev
->private;
1590 mirror
= find_bio_disk(r1_bio
, bio
);
1593 sector_t sync_blocks
= 0;
1594 sector_t s
= r1_bio
->sector
;
1595 long sectors_to_go
= r1_bio
->sectors
;
1596 /* make sure these bits doesn't get cleared. */
1598 bitmap_end_sync(mddev
->bitmap
, s
,
1601 sectors_to_go
-= sync_blocks
;
1602 } while (sectors_to_go
> 0);
1603 set_bit(WriteErrorSeen
,
1604 &conf
->mirrors
[mirror
].rdev
->flags
);
1605 if (!test_and_set_bit(WantReplacement
,
1606 &conf
->mirrors
[mirror
].rdev
->flags
))
1607 set_bit(MD_RECOVERY_NEEDED
, &
1609 set_bit(R1BIO_WriteError
, &r1_bio
->state
);
1610 } else if (is_badblock(conf
->mirrors
[mirror
].rdev
,
1613 &first_bad
, &bad_sectors
) &&
1614 !is_badblock(conf
->mirrors
[r1_bio
->read_disk
].rdev
,
1617 &first_bad
, &bad_sectors
)
1619 set_bit(R1BIO_MadeGood
, &r1_bio
->state
);
1621 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
1622 int s
= r1_bio
->sectors
;
1623 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
1624 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
1625 reschedule_retry(r1_bio
);
1628 md_done_sync(mddev
, s
, uptodate
);
1633 static int r1_sync_page_io(struct md_rdev
*rdev
, sector_t sector
,
1634 int sectors
, struct page
*page
, int rw
)
1636 if (sync_page_io(rdev
, sector
, sectors
<< 9, page
, rw
, false))
1640 set_bit(WriteErrorSeen
, &rdev
->flags
);
1641 if (!test_and_set_bit(WantReplacement
,
1643 set_bit(MD_RECOVERY_NEEDED
, &
1644 rdev
->mddev
->recovery
);
1646 /* need to record an error - either for the block or the device */
1647 if (!rdev_set_badblocks(rdev
, sector
, sectors
, 0))
1648 md_error(rdev
->mddev
, rdev
);
1652 static int fix_sync_read_error(struct r1bio
*r1_bio
)
1654 /* Try some synchronous reads of other devices to get
1655 * good data, much like with normal read errors. Only
1656 * read into the pages we already have so we don't
1657 * need to re-issue the read request.
1658 * We don't need to freeze the array, because being in an
1659 * active sync request, there is no normal IO, and
1660 * no overlapping syncs.
1661 * We don't need to check is_badblock() again as we
1662 * made sure that anything with a bad block in range
1663 * will have bi_end_io clear.
1665 struct mddev
*mddev
= r1_bio
->mddev
;
1666 struct r1conf
*conf
= mddev
->private;
1667 struct bio
*bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1668 sector_t sect
= r1_bio
->sector
;
1669 int sectors
= r1_bio
->sectors
;
1674 int d
= r1_bio
->read_disk
;
1676 struct md_rdev
*rdev
;
1679 if (s
> (PAGE_SIZE
>>9))
1682 if (r1_bio
->bios
[d
]->bi_end_io
== end_sync_read
) {
1683 /* No rcu protection needed here devices
1684 * can only be removed when no resync is
1685 * active, and resync is currently active
1687 rdev
= conf
->mirrors
[d
].rdev
;
1688 if (sync_page_io(rdev
, sect
, s
<<9,
1689 bio
->bi_io_vec
[idx
].bv_page
,
1696 if (d
== conf
->raid_disks
* 2)
1698 } while (!success
&& d
!= r1_bio
->read_disk
);
1701 char b
[BDEVNAME_SIZE
];
1703 /* Cannot read from anywhere, this block is lost.
1704 * Record a bad block on each device. If that doesn't
1705 * work just disable and interrupt the recovery.
1706 * Don't fail devices as that won't really help.
1708 printk(KERN_ALERT
"md/raid1:%s: %s: unrecoverable I/O read error"
1709 " for block %llu\n",
1711 bdevname(bio
->bi_bdev
, b
),
1712 (unsigned long long)r1_bio
->sector
);
1713 for (d
= 0; d
< conf
->raid_disks
* 2; d
++) {
1714 rdev
= conf
->mirrors
[d
].rdev
;
1715 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
1717 if (!rdev_set_badblocks(rdev
, sect
, s
, 0))
1721 conf
->recovery_disabled
=
1722 mddev
->recovery_disabled
;
1723 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1724 md_done_sync(mddev
, r1_bio
->sectors
, 0);
1736 /* write it back and re-read */
1737 while (d
!= r1_bio
->read_disk
) {
1739 d
= conf
->raid_disks
* 2;
1741 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1743 rdev
= conf
->mirrors
[d
].rdev
;
1744 if (r1_sync_page_io(rdev
, sect
, s
,
1745 bio
->bi_io_vec
[idx
].bv_page
,
1747 r1_bio
->bios
[d
]->bi_end_io
= NULL
;
1748 rdev_dec_pending(rdev
, mddev
);
1752 while (d
!= r1_bio
->read_disk
) {
1754 d
= conf
->raid_disks
* 2;
1756 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1758 rdev
= conf
->mirrors
[d
].rdev
;
1759 if (r1_sync_page_io(rdev
, sect
, s
,
1760 bio
->bi_io_vec
[idx
].bv_page
,
1762 atomic_add(s
, &rdev
->corrected_errors
);
1768 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1769 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1773 static int process_checks(struct r1bio
*r1_bio
)
1775 /* We have read all readable devices. If we haven't
1776 * got the block, then there is no hope left.
1777 * If we have, then we want to do a comparison
1778 * and skip the write if everything is the same.
1779 * If any blocks failed to read, then we need to
1780 * attempt an over-write
1782 struct mddev
*mddev
= r1_bio
->mddev
;
1783 struct r1conf
*conf
= mddev
->private;
1788 for (primary
= 0; primary
< conf
->raid_disks
* 2; primary
++)
1789 if (r1_bio
->bios
[primary
]->bi_end_io
== end_sync_read
&&
1790 test_bit(BIO_UPTODATE
, &r1_bio
->bios
[primary
]->bi_flags
)) {
1791 r1_bio
->bios
[primary
]->bi_end_io
= NULL
;
1792 rdev_dec_pending(conf
->mirrors
[primary
].rdev
, mddev
);
1795 r1_bio
->read_disk
= primary
;
1796 vcnt
= (r1_bio
->sectors
+ PAGE_SIZE
/ 512 - 1) >> (PAGE_SHIFT
- 9);
1797 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
1799 struct bio
*pbio
= r1_bio
->bios
[primary
];
1800 struct bio
*sbio
= r1_bio
->bios
[i
];
1803 if (r1_bio
->bios
[i
]->bi_end_io
!= end_sync_read
)
1806 if (test_bit(BIO_UPTODATE
, &sbio
->bi_flags
)) {
1807 for (j
= vcnt
; j
-- ; ) {
1809 p
= pbio
->bi_io_vec
[j
].bv_page
;
1810 s
= sbio
->bi_io_vec
[j
].bv_page
;
1811 if (memcmp(page_address(p
),
1813 sbio
->bi_io_vec
[j
].bv_len
))
1819 mddev
->resync_mismatches
+= r1_bio
->sectors
;
1820 if (j
< 0 || (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)
1821 && test_bit(BIO_UPTODATE
, &sbio
->bi_flags
))) {
1822 /* No need to write to this device. */
1823 sbio
->bi_end_io
= NULL
;
1824 rdev_dec_pending(conf
->mirrors
[i
].rdev
, mddev
);
1827 /* fixup the bio for reuse */
1828 sbio
->bi_vcnt
= vcnt
;
1829 sbio
->bi_size
= r1_bio
->sectors
<< 9;
1831 sbio
->bi_phys_segments
= 0;
1832 sbio
->bi_flags
&= ~(BIO_POOL_MASK
- 1);
1833 sbio
->bi_flags
|= 1 << BIO_UPTODATE
;
1834 sbio
->bi_next
= NULL
;
1835 sbio
->bi_sector
= r1_bio
->sector
+
1836 conf
->mirrors
[i
].rdev
->data_offset
;
1837 sbio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1838 size
= sbio
->bi_size
;
1839 for (j
= 0; j
< vcnt
; j
++) {
1841 bi
= &sbio
->bi_io_vec
[j
];
1843 if (size
> PAGE_SIZE
)
1844 bi
->bv_len
= PAGE_SIZE
;
1848 memcpy(page_address(bi
->bv_page
),
1849 page_address(pbio
->bi_io_vec
[j
].bv_page
),
1856 static void sync_request_write(struct mddev
*mddev
, struct r1bio
*r1_bio
)
1858 struct r1conf
*conf
= mddev
->private;
1860 int disks
= conf
->raid_disks
* 2;
1861 struct bio
*bio
, *wbio
;
1863 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1865 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
))
1866 /* ouch - failed to read all of that. */
1867 if (!fix_sync_read_error(r1_bio
))
1870 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
1871 if (process_checks(r1_bio
) < 0)
1876 atomic_set(&r1_bio
->remaining
, 1);
1877 for (i
= 0; i
< disks
; i
++) {
1878 wbio
= r1_bio
->bios
[i
];
1879 if (wbio
->bi_end_io
== NULL
||
1880 (wbio
->bi_end_io
== end_sync_read
&&
1881 (i
== r1_bio
->read_disk
||
1882 !test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))))
1885 wbio
->bi_rw
= WRITE
;
1886 wbio
->bi_end_io
= end_sync_write
;
1887 atomic_inc(&r1_bio
->remaining
);
1888 md_sync_acct(conf
->mirrors
[i
].rdev
->bdev
, wbio
->bi_size
>> 9);
1890 generic_make_request(wbio
);
1893 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
1894 /* if we're here, all write(s) have completed, so clean up */
1895 int s
= r1_bio
->sectors
;
1896 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
1897 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
1898 reschedule_retry(r1_bio
);
1901 md_done_sync(mddev
, s
, 1);
1907 * This is a kernel thread which:
1909 * 1. Retries failed read operations on working mirrors.
1910 * 2. Updates the raid superblock when problems encounter.
1911 * 3. Performs writes following reads for array synchronising.
1914 static void fix_read_error(struct r1conf
*conf
, int read_disk
,
1915 sector_t sect
, int sectors
)
1917 struct mddev
*mddev
= conf
->mddev
;
1923 struct md_rdev
*rdev
;
1925 if (s
> (PAGE_SIZE
>>9))
1929 /* Note: no rcu protection needed here
1930 * as this is synchronous in the raid1d thread
1931 * which is the thread that might remove
1932 * a device. If raid1d ever becomes multi-threaded....
1937 rdev
= conf
->mirrors
[d
].rdev
;
1939 (test_bit(In_sync
, &rdev
->flags
) ||
1940 (!test_bit(Faulty
, &rdev
->flags
) &&
1941 rdev
->recovery_offset
>= sect
+ s
)) &&
1942 is_badblock(rdev
, sect
, s
,
1943 &first_bad
, &bad_sectors
) == 0 &&
1944 sync_page_io(rdev
, sect
, s
<<9,
1945 conf
->tmppage
, READ
, false))
1949 if (d
== conf
->raid_disks
* 2)
1952 } while (!success
&& d
!= read_disk
);
1955 /* Cannot read from anywhere - mark it bad */
1956 struct md_rdev
*rdev
= conf
->mirrors
[read_disk
].rdev
;
1957 if (!rdev_set_badblocks(rdev
, sect
, s
, 0))
1958 md_error(mddev
, rdev
);
1961 /* write it back and re-read */
1963 while (d
!= read_disk
) {
1965 d
= conf
->raid_disks
* 2;
1967 rdev
= conf
->mirrors
[d
].rdev
;
1969 test_bit(In_sync
, &rdev
->flags
))
1970 r1_sync_page_io(rdev
, sect
, s
,
1971 conf
->tmppage
, WRITE
);
1974 while (d
!= read_disk
) {
1975 char b
[BDEVNAME_SIZE
];
1977 d
= conf
->raid_disks
* 2;
1979 rdev
= conf
->mirrors
[d
].rdev
;
1981 test_bit(In_sync
, &rdev
->flags
)) {
1982 if (r1_sync_page_io(rdev
, sect
, s
,
1983 conf
->tmppage
, READ
)) {
1984 atomic_add(s
, &rdev
->corrected_errors
);
1986 "md/raid1:%s: read error corrected "
1987 "(%d sectors at %llu on %s)\n",
1989 (unsigned long long)(sect
+
1991 bdevname(rdev
->bdev
, b
));
2000 static void bi_complete(struct bio
*bio
, int error
)
2002 complete((struct completion
*)bio
->bi_private
);
2005 static int submit_bio_wait(int rw
, struct bio
*bio
)
2007 struct completion event
;
2010 init_completion(&event
);
2011 bio
->bi_private
= &event
;
2012 bio
->bi_end_io
= bi_complete
;
2013 submit_bio(rw
, bio
);
2014 wait_for_completion(&event
);
2016 return test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
2019 static int narrow_write_error(struct r1bio
*r1_bio
, int i
)
2021 struct mddev
*mddev
= r1_bio
->mddev
;
2022 struct r1conf
*conf
= mddev
->private;
2023 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
2025 struct bio_vec
*vec
;
2027 /* bio has the data to be written to device 'i' where
2028 * we just recently had a write error.
2029 * We repeatedly clone the bio and trim down to one block,
2030 * then try the write. Where the write fails we record
2032 * It is conceivable that the bio doesn't exactly align with
2033 * blocks. We must handle this somehow.
2035 * We currently own a reference on the rdev.
2041 int sect_to_write
= r1_bio
->sectors
;
2044 if (rdev
->badblocks
.shift
< 0)
2047 block_sectors
= 1 << rdev
->badblocks
.shift
;
2048 sector
= r1_bio
->sector
;
2049 sectors
= ((sector
+ block_sectors
)
2050 & ~(sector_t
)(block_sectors
- 1))
2053 if (test_bit(R1BIO_BehindIO
, &r1_bio
->state
)) {
2054 vcnt
= r1_bio
->behind_page_count
;
2055 vec
= r1_bio
->behind_bvecs
;
2057 while (vec
[idx
].bv_page
== NULL
)
2060 vcnt
= r1_bio
->master_bio
->bi_vcnt
;
2061 vec
= r1_bio
->master_bio
->bi_io_vec
;
2062 idx
= r1_bio
->master_bio
->bi_idx
;
2064 while (sect_to_write
) {
2066 if (sectors
> sect_to_write
)
2067 sectors
= sect_to_write
;
2068 /* Write at 'sector' for 'sectors'*/
2070 wbio
= bio_alloc_mddev(GFP_NOIO
, vcnt
, mddev
);
2071 memcpy(wbio
->bi_io_vec
, vec
, vcnt
* sizeof(struct bio_vec
));
2072 wbio
->bi_sector
= r1_bio
->sector
;
2073 wbio
->bi_rw
= WRITE
;
2074 wbio
->bi_vcnt
= vcnt
;
2075 wbio
->bi_size
= r1_bio
->sectors
<< 9;
2078 md_trim_bio(wbio
, sector
- r1_bio
->sector
, sectors
);
2079 wbio
->bi_sector
+= rdev
->data_offset
;
2080 wbio
->bi_bdev
= rdev
->bdev
;
2081 if (submit_bio_wait(WRITE
, wbio
) == 0)
2083 ok
= rdev_set_badblocks(rdev
, sector
,
2088 sect_to_write
-= sectors
;
2090 sectors
= block_sectors
;
2095 static void handle_sync_write_finished(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2098 int s
= r1_bio
->sectors
;
2099 for (m
= 0; m
< conf
->raid_disks
* 2 ; m
++) {
2100 struct md_rdev
*rdev
= conf
->mirrors
[m
].rdev
;
2101 struct bio
*bio
= r1_bio
->bios
[m
];
2102 if (bio
->bi_end_io
== NULL
)
2104 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
2105 test_bit(R1BIO_MadeGood
, &r1_bio
->state
)) {
2106 rdev_clear_badblocks(rdev
, r1_bio
->sector
, s
, 0);
2108 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
2109 test_bit(R1BIO_WriteError
, &r1_bio
->state
)) {
2110 if (!rdev_set_badblocks(rdev
, r1_bio
->sector
, s
, 0))
2111 md_error(conf
->mddev
, rdev
);
2115 md_done_sync(conf
->mddev
, s
, 1);
2118 static void handle_write_finished(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2121 for (m
= 0; m
< conf
->raid_disks
* 2 ; m
++)
2122 if (r1_bio
->bios
[m
] == IO_MADE_GOOD
) {
2123 struct md_rdev
*rdev
= conf
->mirrors
[m
].rdev
;
2124 rdev_clear_badblocks(rdev
,
2126 r1_bio
->sectors
, 0);
2127 rdev_dec_pending(rdev
, conf
->mddev
);
2128 } else if (r1_bio
->bios
[m
] != NULL
) {
2129 /* This drive got a write error. We need to
2130 * narrow down and record precise write
2133 if (!narrow_write_error(r1_bio
, m
)) {
2134 md_error(conf
->mddev
,
2135 conf
->mirrors
[m
].rdev
);
2136 /* an I/O failed, we can't clear the bitmap */
2137 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
2139 rdev_dec_pending(conf
->mirrors
[m
].rdev
,
2142 if (test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2143 close_write(r1_bio
);
2144 raid_end_bio_io(r1_bio
);
2147 static void handle_read_error(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2151 struct mddev
*mddev
= conf
->mddev
;
2153 char b
[BDEVNAME_SIZE
];
2154 struct md_rdev
*rdev
;
2156 clear_bit(R1BIO_ReadError
, &r1_bio
->state
);
2157 /* we got a read error. Maybe the drive is bad. Maybe just
2158 * the block and we can fix it.
2159 * We freeze all other IO, and try reading the block from
2160 * other devices. When we find one, we re-write
2161 * and check it that fixes the read error.
2162 * This is all done synchronously while the array is
2165 if (mddev
->ro
== 0) {
2167 fix_read_error(conf
, r1_bio
->read_disk
,
2168 r1_bio
->sector
, r1_bio
->sectors
);
2169 unfreeze_array(conf
);
2171 md_error(mddev
, conf
->mirrors
[r1_bio
->read_disk
].rdev
);
2173 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
2174 bdevname(bio
->bi_bdev
, b
);
2176 disk
= read_balance(conf
, r1_bio
, &max_sectors
);
2178 printk(KERN_ALERT
"md/raid1:%s: %s: unrecoverable I/O"
2179 " read error for block %llu\n",
2180 mdname(mddev
), b
, (unsigned long long)r1_bio
->sector
);
2181 raid_end_bio_io(r1_bio
);
2183 const unsigned long do_sync
2184 = r1_bio
->master_bio
->bi_rw
& REQ_SYNC
;
2186 r1_bio
->bios
[r1_bio
->read_disk
] =
2187 mddev
->ro
? IO_BLOCKED
: NULL
;
2190 r1_bio
->read_disk
= disk
;
2191 bio
= bio_clone_mddev(r1_bio
->master_bio
, GFP_NOIO
, mddev
);
2192 md_trim_bio(bio
, r1_bio
->sector
- bio
->bi_sector
, max_sectors
);
2193 r1_bio
->bios
[r1_bio
->read_disk
] = bio
;
2194 rdev
= conf
->mirrors
[disk
].rdev
;
2195 printk_ratelimited(KERN_ERR
2196 "md/raid1:%s: redirecting sector %llu"
2197 " to other mirror: %s\n",
2199 (unsigned long long)r1_bio
->sector
,
2200 bdevname(rdev
->bdev
, b
));
2201 bio
->bi_sector
= r1_bio
->sector
+ rdev
->data_offset
;
2202 bio
->bi_bdev
= rdev
->bdev
;
2203 bio
->bi_end_io
= raid1_end_read_request
;
2204 bio
->bi_rw
= READ
| do_sync
;
2205 bio
->bi_private
= r1_bio
;
2206 if (max_sectors
< r1_bio
->sectors
) {
2207 /* Drat - have to split this up more */
2208 struct bio
*mbio
= r1_bio
->master_bio
;
2209 int sectors_handled
= (r1_bio
->sector
+ max_sectors
2211 r1_bio
->sectors
= max_sectors
;
2212 spin_lock_irq(&conf
->device_lock
);
2213 if (mbio
->bi_phys_segments
== 0)
2214 mbio
->bi_phys_segments
= 2;
2216 mbio
->bi_phys_segments
++;
2217 spin_unlock_irq(&conf
->device_lock
);
2218 generic_make_request(bio
);
2221 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
2223 r1_bio
->master_bio
= mbio
;
2224 r1_bio
->sectors
= (mbio
->bi_size
>> 9)
2227 set_bit(R1BIO_ReadError
, &r1_bio
->state
);
2228 r1_bio
->mddev
= mddev
;
2229 r1_bio
->sector
= mbio
->bi_sector
+ sectors_handled
;
2233 generic_make_request(bio
);
2237 static void raid1d(struct mddev
*mddev
)
2239 struct r1bio
*r1_bio
;
2240 unsigned long flags
;
2241 struct r1conf
*conf
= mddev
->private;
2242 struct list_head
*head
= &conf
->retry_list
;
2243 struct blk_plug plug
;
2245 md_check_recovery(mddev
);
2247 blk_start_plug(&plug
);
2250 flush_pending_writes(conf
);
2252 spin_lock_irqsave(&conf
->device_lock
, flags
);
2253 if (list_empty(head
)) {
2254 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2257 r1_bio
= list_entry(head
->prev
, struct r1bio
, retry_list
);
2258 list_del(head
->prev
);
2260 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2262 mddev
= r1_bio
->mddev
;
2263 conf
= mddev
->private;
2264 if (test_bit(R1BIO_IsSync
, &r1_bio
->state
)) {
2265 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
2266 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2267 handle_sync_write_finished(conf
, r1_bio
);
2269 sync_request_write(mddev
, r1_bio
);
2270 } else if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
2271 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2272 handle_write_finished(conf
, r1_bio
);
2273 else if (test_bit(R1BIO_ReadError
, &r1_bio
->state
))
2274 handle_read_error(conf
, r1_bio
);
2276 /* just a partial read to be scheduled from separate
2279 generic_make_request(r1_bio
->bios
[r1_bio
->read_disk
]);
2282 if (mddev
->flags
& ~(1<<MD_CHANGE_PENDING
))
2283 md_check_recovery(mddev
);
2285 blk_finish_plug(&plug
);
2289 static int init_resync(struct r1conf
*conf
)
2293 buffs
= RESYNC_WINDOW
/ RESYNC_BLOCK_SIZE
;
2294 BUG_ON(conf
->r1buf_pool
);
2295 conf
->r1buf_pool
= mempool_create(buffs
, r1buf_pool_alloc
, r1buf_pool_free
,
2297 if (!conf
->r1buf_pool
)
2299 conf
->next_resync
= 0;
2304 * perform a "sync" on one "block"
2306 * We need to make sure that no normal I/O request - particularly write
2307 * requests - conflict with active sync requests.
2309 * This is achieved by tracking pending requests and a 'barrier' concept
2310 * that can be installed to exclude normal IO requests.
2313 static sector_t
sync_request(struct mddev
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
2315 struct r1conf
*conf
= mddev
->private;
2316 struct r1bio
*r1_bio
;
2318 sector_t max_sector
, nr_sectors
;
2322 int write_targets
= 0, read_targets
= 0;
2323 sector_t sync_blocks
;
2324 int still_degraded
= 0;
2325 int good_sectors
= RESYNC_SECTORS
;
2326 int min_bad
= 0; /* number of sectors that are bad in all devices */
2328 if (!conf
->r1buf_pool
)
2329 if (init_resync(conf
))
2332 max_sector
= mddev
->dev_sectors
;
2333 if (sector_nr
>= max_sector
) {
2334 /* If we aborted, we need to abort the
2335 * sync on the 'current' bitmap chunk (there will
2336 * only be one in raid1 resync.
2337 * We can find the current addess in mddev->curr_resync
2339 if (mddev
->curr_resync
< max_sector
) /* aborted */
2340 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
2342 else /* completed sync */
2345 bitmap_close_sync(mddev
->bitmap
);
2350 if (mddev
->bitmap
== NULL
&&
2351 mddev
->recovery_cp
== MaxSector
&&
2352 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
2353 conf
->fullsync
== 0) {
2355 return max_sector
- sector_nr
;
2357 /* before building a request, check if we can skip these blocks..
2358 * This call the bitmap_start_sync doesn't actually record anything
2360 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
2361 !conf
->fullsync
&& !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
2362 /* We can skip this block, and probably several more */
2367 * If there is non-resync activity waiting for a turn,
2368 * and resync is going fast enough,
2369 * then let it though before starting on this new sync request.
2371 if (!go_faster
&& conf
->nr_waiting
)
2372 msleep_interruptible(1000);
2374 bitmap_cond_end_sync(mddev
->bitmap
, sector_nr
);
2375 r1_bio
= mempool_alloc(conf
->r1buf_pool
, GFP_NOIO
);
2376 raise_barrier(conf
);
2378 conf
->next_resync
= sector_nr
;
2382 * If we get a correctably read error during resync or recovery,
2383 * we might want to read from a different device. So we
2384 * flag all drives that could conceivably be read from for READ,
2385 * and any others (which will be non-In_sync devices) for WRITE.
2386 * If a read fails, we try reading from something else for which READ
2390 r1_bio
->mddev
= mddev
;
2391 r1_bio
->sector
= sector_nr
;
2393 set_bit(R1BIO_IsSync
, &r1_bio
->state
);
2395 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
2396 struct md_rdev
*rdev
;
2397 bio
= r1_bio
->bios
[i
];
2399 /* take from bio_init */
2400 bio
->bi_next
= NULL
;
2401 bio
->bi_flags
&= ~(BIO_POOL_MASK
-1);
2402 bio
->bi_flags
|= 1 << BIO_UPTODATE
;
2406 bio
->bi_phys_segments
= 0;
2408 bio
->bi_end_io
= NULL
;
2409 bio
->bi_private
= NULL
;
2411 rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
2413 test_bit(Faulty
, &rdev
->flags
)) {
2414 if (i
< conf
->raid_disks
)
2416 } else if (!test_bit(In_sync
, &rdev
->flags
)) {
2418 bio
->bi_end_io
= end_sync_write
;
2421 /* may need to read from here */
2422 sector_t first_bad
= MaxSector
;
2425 if (is_badblock(rdev
, sector_nr
, good_sectors
,
2426 &first_bad
, &bad_sectors
)) {
2427 if (first_bad
> sector_nr
)
2428 good_sectors
= first_bad
- sector_nr
;
2430 bad_sectors
-= (sector_nr
- first_bad
);
2432 min_bad
> bad_sectors
)
2433 min_bad
= bad_sectors
;
2436 if (sector_nr
< first_bad
) {
2437 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2445 bio
->bi_end_io
= end_sync_read
;
2447 } else if (!test_bit(WriteErrorSeen
, &rdev
->flags
) &&
2448 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
2449 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
2451 * The device is suitable for reading (InSync),
2452 * but has bad block(s) here. Let's try to correct them,
2453 * if we are doing resync or repair. Otherwise, leave
2454 * this device alone for this sync request.
2457 bio
->bi_end_io
= end_sync_write
;
2461 if (bio
->bi_end_io
) {
2462 atomic_inc(&rdev
->nr_pending
);
2463 bio
->bi_sector
= sector_nr
+ rdev
->data_offset
;
2464 bio
->bi_bdev
= rdev
->bdev
;
2465 bio
->bi_private
= r1_bio
;
2471 r1_bio
->read_disk
= disk
;
2473 if (read_targets
== 0 && min_bad
> 0) {
2474 /* These sectors are bad on all InSync devices, so we
2475 * need to mark them bad on all write targets
2478 for (i
= 0 ; i
< conf
->raid_disks
* 2 ; i
++)
2479 if (r1_bio
->bios
[i
]->bi_end_io
== end_sync_write
) {
2480 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
2481 ok
= rdev_set_badblocks(rdev
, sector_nr
,
2485 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2490 /* Cannot record the badblocks, so need to
2492 * If there are multiple read targets, could just
2493 * fail the really bad ones ???
2495 conf
->recovery_disabled
= mddev
->recovery_disabled
;
2496 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2502 if (min_bad
> 0 && min_bad
< good_sectors
) {
2503 /* only resync enough to reach the next bad->good
2505 good_sectors
= min_bad
;
2508 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && read_targets
> 0)
2509 /* extra read targets are also write targets */
2510 write_targets
+= read_targets
-1;
2512 if (write_targets
== 0 || read_targets
== 0) {
2513 /* There is nowhere to write, so all non-sync
2514 * drives must be failed - so we are finished
2518 max_sector
= sector_nr
+ min_bad
;
2519 rv
= max_sector
- sector_nr
;
2525 if (max_sector
> mddev
->resync_max
)
2526 max_sector
= mddev
->resync_max
; /* Don't do IO beyond here */
2527 if (max_sector
> sector_nr
+ good_sectors
)
2528 max_sector
= sector_nr
+ good_sectors
;
2533 int len
= PAGE_SIZE
;
2534 if (sector_nr
+ (len
>>9) > max_sector
)
2535 len
= (max_sector
- sector_nr
) << 9;
2538 if (sync_blocks
== 0) {
2539 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
,
2540 &sync_blocks
, still_degraded
) &&
2542 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2544 BUG_ON(sync_blocks
< (PAGE_SIZE
>>9));
2545 if ((len
>> 9) > sync_blocks
)
2546 len
= sync_blocks
<<9;
2549 for (i
= 0 ; i
< conf
->raid_disks
* 2; i
++) {
2550 bio
= r1_bio
->bios
[i
];
2551 if (bio
->bi_end_io
) {
2552 page
= bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
;
2553 if (bio_add_page(bio
, page
, len
, 0) == 0) {
2555 bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
= page
;
2558 bio
= r1_bio
->bios
[i
];
2559 if (bio
->bi_end_io
==NULL
)
2561 /* remove last page from this bio */
2563 bio
->bi_size
-= len
;
2564 bio
->bi_flags
&= ~(1<< BIO_SEG_VALID
);
2570 nr_sectors
+= len
>>9;
2571 sector_nr
+= len
>>9;
2572 sync_blocks
-= (len
>>9);
2573 } while (r1_bio
->bios
[disk
]->bi_vcnt
< RESYNC_PAGES
);
2575 r1_bio
->sectors
= nr_sectors
;
2577 /* For a user-requested sync, we read all readable devices and do a
2580 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
2581 atomic_set(&r1_bio
->remaining
, read_targets
);
2582 for (i
= 0; i
< conf
->raid_disks
* 2 && read_targets
; i
++) {
2583 bio
= r1_bio
->bios
[i
];
2584 if (bio
->bi_end_io
== end_sync_read
) {
2586 md_sync_acct(bio
->bi_bdev
, nr_sectors
);
2587 generic_make_request(bio
);
2591 atomic_set(&r1_bio
->remaining
, 1);
2592 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
2593 md_sync_acct(bio
->bi_bdev
, nr_sectors
);
2594 generic_make_request(bio
);
2600 static sector_t
raid1_size(struct mddev
*mddev
, sector_t sectors
, int raid_disks
)
2605 return mddev
->dev_sectors
;
2608 static struct r1conf
*setup_conf(struct mddev
*mddev
)
2610 struct r1conf
*conf
;
2612 struct raid1_info
*disk
;
2613 struct md_rdev
*rdev
;
2616 conf
= kzalloc(sizeof(struct r1conf
), GFP_KERNEL
);
2620 conf
->mirrors
= kzalloc(sizeof(struct raid1_info
)
2621 * mddev
->raid_disks
* 2,
2626 conf
->tmppage
= alloc_page(GFP_KERNEL
);
2630 conf
->poolinfo
= kzalloc(sizeof(*conf
->poolinfo
), GFP_KERNEL
);
2631 if (!conf
->poolinfo
)
2633 conf
->poolinfo
->raid_disks
= mddev
->raid_disks
* 2;
2634 conf
->r1bio_pool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
2637 if (!conf
->r1bio_pool
)
2640 conf
->poolinfo
->mddev
= mddev
;
2643 spin_lock_init(&conf
->device_lock
);
2644 rdev_for_each(rdev
, mddev
) {
2645 struct request_queue
*q
;
2646 int disk_idx
= rdev
->raid_disk
;
2647 if (disk_idx
>= mddev
->raid_disks
2650 if (test_bit(Replacement
, &rdev
->flags
))
2651 disk
= conf
->mirrors
+ conf
->raid_disks
+ disk_idx
;
2653 disk
= conf
->mirrors
+ disk_idx
;
2658 q
= bdev_get_queue(rdev
->bdev
);
2659 if (q
->merge_bvec_fn
)
2660 mddev
->merge_check_needed
= 1;
2662 disk
->head_position
= 0;
2663 disk
->seq_start
= MaxSector
;
2665 conf
->raid_disks
= mddev
->raid_disks
;
2666 conf
->mddev
= mddev
;
2667 INIT_LIST_HEAD(&conf
->retry_list
);
2669 spin_lock_init(&conf
->resync_lock
);
2670 init_waitqueue_head(&conf
->wait_barrier
);
2672 bio_list_init(&conf
->pending_bio_list
);
2673 conf
->pending_count
= 0;
2674 conf
->recovery_disabled
= mddev
->recovery_disabled
- 1;
2677 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
2679 disk
= conf
->mirrors
+ i
;
2681 if (i
< conf
->raid_disks
&&
2682 disk
[conf
->raid_disks
].rdev
) {
2683 /* This slot has a replacement. */
2685 /* No original, just make the replacement
2686 * a recovering spare
2689 disk
[conf
->raid_disks
].rdev
;
2690 disk
[conf
->raid_disks
].rdev
= NULL
;
2691 } else if (!test_bit(In_sync
, &disk
->rdev
->flags
))
2692 /* Original is not in_sync - bad */
2697 !test_bit(In_sync
, &disk
->rdev
->flags
)) {
2698 disk
->head_position
= 0;
2700 (disk
->rdev
->saved_raid_disk
< 0))
2706 conf
->thread
= md_register_thread(raid1d
, mddev
, "raid1");
2707 if (!conf
->thread
) {
2709 "md/raid1:%s: couldn't allocate thread\n",
2718 if (conf
->r1bio_pool
)
2719 mempool_destroy(conf
->r1bio_pool
);
2720 kfree(conf
->mirrors
);
2721 safe_put_page(conf
->tmppage
);
2722 kfree(conf
->poolinfo
);
2725 return ERR_PTR(err
);
2728 static int stop(struct mddev
*mddev
);
2729 static int run(struct mddev
*mddev
)
2731 struct r1conf
*conf
;
2733 struct md_rdev
*rdev
;
2736 if (mddev
->level
!= 1) {
2737 printk(KERN_ERR
"md/raid1:%s: raid level not set to mirroring (%d)\n",
2738 mdname(mddev
), mddev
->level
);
2741 if (mddev
->reshape_position
!= MaxSector
) {
2742 printk(KERN_ERR
"md/raid1:%s: reshape_position set but not supported\n",
2747 * copy the already verified devices into our private RAID1
2748 * bookkeeping area. [whatever we allocate in run(),
2749 * should be freed in stop()]
2751 if (mddev
->private == NULL
)
2752 conf
= setup_conf(mddev
);
2754 conf
= mddev
->private;
2757 return PTR_ERR(conf
);
2759 rdev_for_each(rdev
, mddev
) {
2760 if (!mddev
->gendisk
)
2762 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
2763 rdev
->data_offset
<< 9);
2766 mddev
->degraded
= 0;
2767 for (i
=0; i
< conf
->raid_disks
; i
++)
2768 if (conf
->mirrors
[i
].rdev
== NULL
||
2769 !test_bit(In_sync
, &conf
->mirrors
[i
].rdev
->flags
) ||
2770 test_bit(Faulty
, &conf
->mirrors
[i
].rdev
->flags
))
2773 if (conf
->raid_disks
- mddev
->degraded
== 1)
2774 mddev
->recovery_cp
= MaxSector
;
2776 if (mddev
->recovery_cp
!= MaxSector
)
2777 printk(KERN_NOTICE
"md/raid1:%s: not clean"
2778 " -- starting background reconstruction\n",
2781 "md/raid1:%s: active with %d out of %d mirrors\n",
2782 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
2786 * Ok, everything is just fine now
2788 mddev
->thread
= conf
->thread
;
2789 conf
->thread
= NULL
;
2790 mddev
->private = conf
;
2792 md_set_array_sectors(mddev
, raid1_size(mddev
, 0, 0));
2795 mddev
->queue
->backing_dev_info
.congested_fn
= raid1_congested
;
2796 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
2797 blk_queue_merge_bvec(mddev
->queue
, raid1_mergeable_bvec
);
2800 ret
= md_integrity_register(mddev
);
2806 static int stop(struct mddev
*mddev
)
2808 struct r1conf
*conf
= mddev
->private;
2809 struct bitmap
*bitmap
= mddev
->bitmap
;
2811 /* wait for behind writes to complete */
2812 if (bitmap
&& atomic_read(&bitmap
->behind_writes
) > 0) {
2813 printk(KERN_INFO
"md/raid1:%s: behind writes in progress - waiting to stop.\n",
2815 /* need to kick something here to make sure I/O goes? */
2816 wait_event(bitmap
->behind_wait
,
2817 atomic_read(&bitmap
->behind_writes
) == 0);
2820 raise_barrier(conf
);
2821 lower_barrier(conf
);
2823 md_unregister_thread(&mddev
->thread
);
2824 if (conf
->r1bio_pool
)
2825 mempool_destroy(conf
->r1bio_pool
);
2826 kfree(conf
->mirrors
);
2827 kfree(conf
->poolinfo
);
2829 mddev
->private = NULL
;
2833 static int raid1_resize(struct mddev
*mddev
, sector_t sectors
)
2835 /* no resync is happening, and there is enough space
2836 * on all devices, so we can resize.
2837 * We need to make sure resync covers any new space.
2838 * If the array is shrinking we should possibly wait until
2839 * any io in the removed space completes, but it hardly seems
2842 sector_t newsize
= raid1_size(mddev
, sectors
, 0);
2843 if (mddev
->external_size
&&
2844 mddev
->array_sectors
> newsize
)
2846 if (mddev
->bitmap
) {
2847 int ret
= bitmap_resize(mddev
->bitmap
, newsize
, 0, 0);
2851 md_set_array_sectors(mddev
, newsize
);
2852 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
2853 revalidate_disk(mddev
->gendisk
);
2854 if (sectors
> mddev
->dev_sectors
&&
2855 mddev
->recovery_cp
> mddev
->dev_sectors
) {
2856 mddev
->recovery_cp
= mddev
->dev_sectors
;
2857 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2859 mddev
->dev_sectors
= sectors
;
2860 mddev
->resync_max_sectors
= sectors
;
2864 static int raid1_reshape(struct mddev
*mddev
)
2867 * 1/ resize the r1bio_pool
2868 * 2/ resize conf->mirrors
2870 * We allocate a new r1bio_pool if we can.
2871 * Then raise a device barrier and wait until all IO stops.
2872 * Then resize conf->mirrors and swap in the new r1bio pool.
2874 * At the same time, we "pack" the devices so that all the missing
2875 * devices have the higher raid_disk numbers.
2877 mempool_t
*newpool
, *oldpool
;
2878 struct pool_info
*newpoolinfo
;
2879 struct raid1_info
*newmirrors
;
2880 struct r1conf
*conf
= mddev
->private;
2881 int cnt
, raid_disks
;
2882 unsigned long flags
;
2885 /* Cannot change chunk_size, layout, or level */
2886 if (mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
||
2887 mddev
->layout
!= mddev
->new_layout
||
2888 mddev
->level
!= mddev
->new_level
) {
2889 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2890 mddev
->new_layout
= mddev
->layout
;
2891 mddev
->new_level
= mddev
->level
;
2895 err
= md_allow_write(mddev
);
2899 raid_disks
= mddev
->raid_disks
+ mddev
->delta_disks
;
2901 if (raid_disks
< conf
->raid_disks
) {
2903 for (d
= 0; d
< conf
->raid_disks
; d
++)
2904 if (conf
->mirrors
[d
].rdev
)
2906 if (cnt
> raid_disks
)
2910 newpoolinfo
= kmalloc(sizeof(*newpoolinfo
), GFP_KERNEL
);
2913 newpoolinfo
->mddev
= mddev
;
2914 newpoolinfo
->raid_disks
= raid_disks
* 2;
2916 newpool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
2917 r1bio_pool_free
, newpoolinfo
);
2922 newmirrors
= kzalloc(sizeof(struct raid1_info
) * raid_disks
* 2,
2926 mempool_destroy(newpool
);
2930 raise_barrier(conf
);
2932 /* ok, everything is stopped */
2933 oldpool
= conf
->r1bio_pool
;
2934 conf
->r1bio_pool
= newpool
;
2936 for (d
= d2
= 0; d
< conf
->raid_disks
; d
++) {
2937 struct md_rdev
*rdev
= conf
->mirrors
[d
].rdev
;
2938 if (rdev
&& rdev
->raid_disk
!= d2
) {
2939 sysfs_unlink_rdev(mddev
, rdev
);
2940 rdev
->raid_disk
= d2
;
2941 sysfs_unlink_rdev(mddev
, rdev
);
2942 if (sysfs_link_rdev(mddev
, rdev
))
2944 "md/raid1:%s: cannot register rd%d\n",
2945 mdname(mddev
), rdev
->raid_disk
);
2948 newmirrors
[d2
++].rdev
= rdev
;
2950 kfree(conf
->mirrors
);
2951 conf
->mirrors
= newmirrors
;
2952 kfree(conf
->poolinfo
);
2953 conf
->poolinfo
= newpoolinfo
;
2955 spin_lock_irqsave(&conf
->device_lock
, flags
);
2956 mddev
->degraded
+= (raid_disks
- conf
->raid_disks
);
2957 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2958 conf
->raid_disks
= mddev
->raid_disks
= raid_disks
;
2959 mddev
->delta_disks
= 0;
2961 lower_barrier(conf
);
2963 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2964 md_wakeup_thread(mddev
->thread
);
2966 mempool_destroy(oldpool
);
2970 static void raid1_quiesce(struct mddev
*mddev
, int state
)
2972 struct r1conf
*conf
= mddev
->private;
2975 case 2: /* wake for suspend */
2976 wake_up(&conf
->wait_barrier
);
2979 raise_barrier(conf
);
2982 lower_barrier(conf
);
2987 static void *raid1_takeover(struct mddev
*mddev
)
2989 /* raid1 can take over:
2990 * raid5 with 2 devices, any layout or chunk size
2992 if (mddev
->level
== 5 && mddev
->raid_disks
== 2) {
2993 struct r1conf
*conf
;
2994 mddev
->new_level
= 1;
2995 mddev
->new_layout
= 0;
2996 mddev
->new_chunk_sectors
= 0;
2997 conf
= setup_conf(mddev
);
3002 return ERR_PTR(-EINVAL
);
3005 static struct md_personality raid1_personality
=
3009 .owner
= THIS_MODULE
,
3010 .make_request
= make_request
,
3014 .error_handler
= error
,
3015 .hot_add_disk
= raid1_add_disk
,
3016 .hot_remove_disk
= raid1_remove_disk
,
3017 .spare_active
= raid1_spare_active
,
3018 .sync_request
= sync_request
,
3019 .resize
= raid1_resize
,
3021 .check_reshape
= raid1_reshape
,
3022 .quiesce
= raid1_quiesce
,
3023 .takeover
= raid1_takeover
,
3026 static int __init
raid_init(void)
3028 return register_md_personality(&raid1_personality
);
3031 static void raid_exit(void)
3033 unregister_md_personality(&raid1_personality
);
3036 module_init(raid_init
);
3037 module_exit(raid_exit
);
3038 MODULE_LICENSE("GPL");
3039 MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3040 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3041 MODULE_ALIAS("md-raid1");
3042 MODULE_ALIAS("md-level-1");
3044 module_param(max_queued_requests
, int, S_IRUGO
|S_IWUSR
);