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
;
99 r1_bio
= r1bio_pool_alloc(gfp_flags
, pi
);
104 * Allocate bios : 1 for reading, n-1 for writing
106 for (j
= pi
->raid_disks
; j
-- ; ) {
107 bio
= bio_kmalloc(gfp_flags
, RESYNC_PAGES
);
110 r1_bio
->bios
[j
] = bio
;
113 * Allocate RESYNC_PAGES data pages and attach them to
115 * If this is a user-requested check/repair, allocate
116 * RESYNC_PAGES for each bio.
118 if (test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
))
123 bio
= r1_bio
->bios
[j
];
124 bio
->bi_vcnt
= RESYNC_PAGES
;
126 if (bio_alloc_pages(bio
, gfp_flags
))
129 /* If not user-requests, copy the page pointers to all bios */
130 if (!test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
)) {
131 for (i
=0; i
<RESYNC_PAGES
; i
++)
132 for (j
=1; j
<pi
->raid_disks
; j
++)
133 r1_bio
->bios
[j
]->bi_io_vec
[i
].bv_page
=
134 r1_bio
->bios
[0]->bi_io_vec
[i
].bv_page
;
137 r1_bio
->master_bio
= NULL
;
142 while (++j
< pi
->raid_disks
)
143 bio_put(r1_bio
->bios
[j
]);
144 r1bio_pool_free(r1_bio
, data
);
148 static void r1buf_pool_free(void *__r1_bio
, void *data
)
150 struct pool_info
*pi
= data
;
152 struct r1bio
*r1bio
= __r1_bio
;
154 for (i
= 0; i
< RESYNC_PAGES
; i
++)
155 for (j
= pi
->raid_disks
; j
-- ;) {
157 r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
!=
158 r1bio
->bios
[0]->bi_io_vec
[i
].bv_page
)
159 safe_put_page(r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
);
161 for (i
=0 ; i
< pi
->raid_disks
; i
++)
162 bio_put(r1bio
->bios
[i
]);
164 r1bio_pool_free(r1bio
, data
);
167 static void put_all_bios(struct r1conf
*conf
, struct r1bio
*r1_bio
)
171 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
172 struct bio
**bio
= r1_bio
->bios
+ i
;
173 if (!BIO_SPECIAL(*bio
))
179 static void free_r1bio(struct r1bio
*r1_bio
)
181 struct r1conf
*conf
= r1_bio
->mddev
->private;
183 put_all_bios(conf
, r1_bio
);
184 mempool_free(r1_bio
, conf
->r1bio_pool
);
187 static void put_buf(struct r1bio
*r1_bio
)
189 struct r1conf
*conf
= r1_bio
->mddev
->private;
192 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
193 struct bio
*bio
= r1_bio
->bios
[i
];
195 rdev_dec_pending(conf
->mirrors
[i
].rdev
, r1_bio
->mddev
);
198 mempool_free(r1_bio
, conf
->r1buf_pool
);
203 static void reschedule_retry(struct r1bio
*r1_bio
)
206 struct mddev
*mddev
= r1_bio
->mddev
;
207 struct r1conf
*conf
= mddev
->private;
209 spin_lock_irqsave(&conf
->device_lock
, flags
);
210 list_add(&r1_bio
->retry_list
, &conf
->retry_list
);
212 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
214 wake_up(&conf
->wait_barrier
);
215 md_wakeup_thread(mddev
->thread
);
219 * raid_end_bio_io() is called when we have finished servicing a mirrored
220 * operation and are ready to return a success/failure code to the buffer
223 static void call_bio_endio(struct r1bio
*r1_bio
)
225 struct bio
*bio
= r1_bio
->master_bio
;
227 struct r1conf
*conf
= r1_bio
->mddev
->private;
229 if (bio
->bi_phys_segments
) {
231 spin_lock_irqsave(&conf
->device_lock
, flags
);
232 bio
->bi_phys_segments
--;
233 done
= (bio
->bi_phys_segments
== 0);
234 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
238 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
))
239 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
243 * Wake up any possible resync thread that waits for the device
250 static void raid_end_bio_io(struct r1bio
*r1_bio
)
252 struct bio
*bio
= r1_bio
->master_bio
;
254 /* if nobody has done the final endio yet, do it now */
255 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
256 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
257 (bio_data_dir(bio
) == WRITE
) ? "write" : "read",
258 (unsigned long long) bio
->bi_sector
,
259 (unsigned long long) bio
->bi_sector
+
260 bio_sectors(bio
) - 1);
262 call_bio_endio(r1_bio
);
268 * Update disk head position estimator based on IRQ completion info.
270 static inline void update_head_pos(int disk
, struct r1bio
*r1_bio
)
272 struct r1conf
*conf
= r1_bio
->mddev
->private;
274 conf
->mirrors
[disk
].head_position
=
275 r1_bio
->sector
+ (r1_bio
->sectors
);
279 * Find the disk number which triggered given bio
281 static int find_bio_disk(struct r1bio
*r1_bio
, struct bio
*bio
)
284 struct r1conf
*conf
= r1_bio
->mddev
->private;
285 int raid_disks
= conf
->raid_disks
;
287 for (mirror
= 0; mirror
< raid_disks
* 2; mirror
++)
288 if (r1_bio
->bios
[mirror
] == bio
)
291 BUG_ON(mirror
== raid_disks
* 2);
292 update_head_pos(mirror
, r1_bio
);
297 static void raid1_end_read_request(struct bio
*bio
, int error
)
299 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
300 struct r1bio
*r1_bio
= bio
->bi_private
;
302 struct r1conf
*conf
= r1_bio
->mddev
->private;
304 mirror
= r1_bio
->read_disk
;
306 * this branch is our 'one mirror IO has finished' event handler:
308 update_head_pos(mirror
, r1_bio
);
311 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
313 /* If all other devices have failed, we want to return
314 * the error upwards rather than fail the last device.
315 * Here we redefine "uptodate" to mean "Don't want to retry"
318 spin_lock_irqsave(&conf
->device_lock
, flags
);
319 if (r1_bio
->mddev
->degraded
== conf
->raid_disks
||
320 (r1_bio
->mddev
->degraded
== conf
->raid_disks
-1 &&
321 !test_bit(Faulty
, &conf
->mirrors
[mirror
].rdev
->flags
)))
323 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
327 raid_end_bio_io(r1_bio
);
328 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
, conf
->mddev
);
333 char b
[BDEVNAME_SIZE
];
335 KERN_ERR
"md/raid1:%s: %s: "
336 "rescheduling sector %llu\n",
338 bdevname(conf
->mirrors
[mirror
].rdev
->bdev
,
340 (unsigned long long)r1_bio
->sector
);
341 set_bit(R1BIO_ReadError
, &r1_bio
->state
);
342 reschedule_retry(r1_bio
);
343 /* don't drop the reference on read_disk yet */
347 static void close_write(struct r1bio
*r1_bio
)
349 /* it really is the end of this request */
350 if (test_bit(R1BIO_BehindIO
, &r1_bio
->state
)) {
351 /* free extra copy of the data pages */
352 int i
= r1_bio
->behind_page_count
;
354 safe_put_page(r1_bio
->behind_bvecs
[i
].bv_page
);
355 kfree(r1_bio
->behind_bvecs
);
356 r1_bio
->behind_bvecs
= NULL
;
358 /* clear the bitmap if all writes complete successfully */
359 bitmap_endwrite(r1_bio
->mddev
->bitmap
, r1_bio
->sector
,
361 !test_bit(R1BIO_Degraded
, &r1_bio
->state
),
362 test_bit(R1BIO_BehindIO
, &r1_bio
->state
));
363 md_write_end(r1_bio
->mddev
);
366 static void r1_bio_write_done(struct r1bio
*r1_bio
)
368 if (!atomic_dec_and_test(&r1_bio
->remaining
))
371 if (test_bit(R1BIO_WriteError
, &r1_bio
->state
))
372 reschedule_retry(r1_bio
);
375 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
))
376 reschedule_retry(r1_bio
);
378 raid_end_bio_io(r1_bio
);
382 static void raid1_end_write_request(struct bio
*bio
, int error
)
384 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
385 struct r1bio
*r1_bio
= bio
->bi_private
;
386 int mirror
, behind
= test_bit(R1BIO_BehindIO
, &r1_bio
->state
);
387 struct r1conf
*conf
= r1_bio
->mddev
->private;
388 struct bio
*to_put
= NULL
;
390 mirror
= find_bio_disk(r1_bio
, bio
);
393 * 'one mirror IO has finished' event handler:
396 set_bit(WriteErrorSeen
,
397 &conf
->mirrors
[mirror
].rdev
->flags
);
398 if (!test_and_set_bit(WantReplacement
,
399 &conf
->mirrors
[mirror
].rdev
->flags
))
400 set_bit(MD_RECOVERY_NEEDED
, &
401 conf
->mddev
->recovery
);
403 set_bit(R1BIO_WriteError
, &r1_bio
->state
);
406 * Set R1BIO_Uptodate in our master bio, so that we
407 * will return a good error code for to the higher
408 * levels even if IO on some other mirrored buffer
411 * The 'master' represents the composite IO operation
412 * to user-side. So if something waits for IO, then it
413 * will wait for the 'master' bio.
418 r1_bio
->bios
[mirror
] = NULL
;
420 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
422 /* Maybe we can clear some bad blocks. */
423 if (is_badblock(conf
->mirrors
[mirror
].rdev
,
424 r1_bio
->sector
, r1_bio
->sectors
,
425 &first_bad
, &bad_sectors
)) {
426 r1_bio
->bios
[mirror
] = IO_MADE_GOOD
;
427 set_bit(R1BIO_MadeGood
, &r1_bio
->state
);
432 if (test_bit(WriteMostly
, &conf
->mirrors
[mirror
].rdev
->flags
))
433 atomic_dec(&r1_bio
->behind_remaining
);
436 * In behind mode, we ACK the master bio once the I/O
437 * has safely reached all non-writemostly
438 * disks. Setting the Returned bit ensures that this
439 * gets done only once -- we don't ever want to return
440 * -EIO here, instead we'll wait
442 if (atomic_read(&r1_bio
->behind_remaining
) >= (atomic_read(&r1_bio
->remaining
)-1) &&
443 test_bit(R1BIO_Uptodate
, &r1_bio
->state
)) {
444 /* Maybe we can return now */
445 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
446 struct bio
*mbio
= r1_bio
->master_bio
;
447 pr_debug("raid1: behind end write sectors"
449 (unsigned long long) mbio
->bi_sector
,
450 (unsigned long long) mbio
->bi_sector
+
451 bio_sectors(mbio
) - 1);
452 call_bio_endio(r1_bio
);
456 if (r1_bio
->bios
[mirror
] == NULL
)
457 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
,
461 * Let's see if all mirrored write operations have finished
464 r1_bio_write_done(r1_bio
);
472 * This routine returns the disk from which the requested read should
473 * be done. There is a per-array 'next expected sequential IO' sector
474 * number - if this matches on the next IO then we use the last disk.
475 * There is also a per-disk 'last know head position' sector that is
476 * maintained from IRQ contexts, both the normal and the resync IO
477 * completion handlers update this position correctly. If there is no
478 * perfect sequential match then we pick the disk whose head is closest.
480 * If there are 2 mirrors in the same 2 devices, performance degrades
481 * because position is mirror, not device based.
483 * The rdev for the device selected will have nr_pending incremented.
485 static int read_balance(struct r1conf
*conf
, struct r1bio
*r1_bio
, int *max_sectors
)
487 const sector_t this_sector
= r1_bio
->sector
;
489 int best_good_sectors
;
490 int best_disk
, best_dist_disk
, best_pending_disk
;
494 unsigned int min_pending
;
495 struct md_rdev
*rdev
;
497 int choose_next_idle
;
501 * Check if we can balance. We can balance on the whole
502 * device if no resync is going on, or below the resync window.
503 * We take the first readable disk when above the resync window.
506 sectors
= r1_bio
->sectors
;
509 best_dist
= MaxSector
;
510 best_pending_disk
= -1;
511 min_pending
= UINT_MAX
;
512 best_good_sectors
= 0;
514 choose_next_idle
= 0;
516 if (conf
->mddev
->recovery_cp
< MaxSector
&&
517 (this_sector
+ sectors
>= conf
->next_resync
))
522 for (disk
= 0 ; disk
< conf
->raid_disks
* 2 ; disk
++) {
526 unsigned int pending
;
529 rdev
= rcu_dereference(conf
->mirrors
[disk
].rdev
);
530 if (r1_bio
->bios
[disk
] == IO_BLOCKED
532 || test_bit(Unmerged
, &rdev
->flags
)
533 || test_bit(Faulty
, &rdev
->flags
))
535 if (!test_bit(In_sync
, &rdev
->flags
) &&
536 rdev
->recovery_offset
< this_sector
+ sectors
)
538 if (test_bit(WriteMostly
, &rdev
->flags
)) {
539 /* Don't balance among write-mostly, just
540 * use the first as a last resort */
542 if (is_badblock(rdev
, this_sector
, sectors
,
543 &first_bad
, &bad_sectors
)) {
544 if (first_bad
< this_sector
)
545 /* Cannot use this */
547 best_good_sectors
= first_bad
- this_sector
;
549 best_good_sectors
= sectors
;
554 /* This is a reasonable device to use. It might
557 if (is_badblock(rdev
, this_sector
, sectors
,
558 &first_bad
, &bad_sectors
)) {
559 if (best_dist
< MaxSector
)
560 /* already have a better device */
562 if (first_bad
<= this_sector
) {
563 /* cannot read here. If this is the 'primary'
564 * device, then we must not read beyond
565 * bad_sectors from another device..
567 bad_sectors
-= (this_sector
- first_bad
);
568 if (choose_first
&& sectors
> bad_sectors
)
569 sectors
= bad_sectors
;
570 if (best_good_sectors
> sectors
)
571 best_good_sectors
= sectors
;
574 sector_t good_sectors
= first_bad
- this_sector
;
575 if (good_sectors
> best_good_sectors
) {
576 best_good_sectors
= good_sectors
;
584 best_good_sectors
= sectors
;
586 nonrot
= blk_queue_nonrot(bdev_get_queue(rdev
->bdev
));
587 has_nonrot_disk
|= nonrot
;
588 pending
= atomic_read(&rdev
->nr_pending
);
589 dist
= abs(this_sector
- conf
->mirrors
[disk
].head_position
);
594 /* Don't change to another disk for sequential reads */
595 if (conf
->mirrors
[disk
].next_seq_sect
== this_sector
597 int opt_iosize
= bdev_io_opt(rdev
->bdev
) >> 9;
598 struct raid1_info
*mirror
= &conf
->mirrors
[disk
];
602 * If buffered sequential IO size exceeds optimal
603 * iosize, check if there is idle disk. If yes, choose
604 * the idle disk. read_balance could already choose an
605 * idle disk before noticing it's a sequential IO in
606 * this disk. This doesn't matter because this disk
607 * will idle, next time it will be utilized after the
608 * first disk has IO size exceeds optimal iosize. In
609 * this way, iosize of the first disk will be optimal
610 * iosize at least. iosize of the second disk might be
611 * small, but not a big deal since when the second disk
612 * starts IO, the first disk is likely still busy.
614 if (nonrot
&& opt_iosize
> 0 &&
615 mirror
->seq_start
!= MaxSector
&&
616 mirror
->next_seq_sect
> opt_iosize
&&
617 mirror
->next_seq_sect
- opt_iosize
>=
619 choose_next_idle
= 1;
624 /* If device is idle, use it */
630 if (choose_next_idle
)
633 if (min_pending
> pending
) {
634 min_pending
= pending
;
635 best_pending_disk
= disk
;
638 if (dist
< best_dist
) {
640 best_dist_disk
= disk
;
645 * If all disks are rotational, choose the closest disk. If any disk is
646 * non-rotational, choose the disk with less pending request even the
647 * disk is rotational, which might/might not be optimal for raids with
648 * mixed ratation/non-rotational disks depending on workload.
650 if (best_disk
== -1) {
652 best_disk
= best_pending_disk
;
654 best_disk
= best_dist_disk
;
657 if (best_disk
>= 0) {
658 rdev
= rcu_dereference(conf
->mirrors
[best_disk
].rdev
);
661 atomic_inc(&rdev
->nr_pending
);
662 if (test_bit(Faulty
, &rdev
->flags
)) {
663 /* cannot risk returning a device that failed
664 * before we inc'ed nr_pending
666 rdev_dec_pending(rdev
, conf
->mddev
);
669 sectors
= best_good_sectors
;
671 if (conf
->mirrors
[best_disk
].next_seq_sect
!= this_sector
)
672 conf
->mirrors
[best_disk
].seq_start
= this_sector
;
674 conf
->mirrors
[best_disk
].next_seq_sect
= this_sector
+ sectors
;
677 *max_sectors
= sectors
;
682 static int raid1_mergeable_bvec(struct request_queue
*q
,
683 struct bvec_merge_data
*bvm
,
684 struct bio_vec
*biovec
)
686 struct mddev
*mddev
= q
->queuedata
;
687 struct r1conf
*conf
= mddev
->private;
688 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
689 int max
= biovec
->bv_len
;
691 if (mddev
->merge_check_needed
) {
694 for (disk
= 0; disk
< conf
->raid_disks
* 2; disk
++) {
695 struct md_rdev
*rdev
= rcu_dereference(
696 conf
->mirrors
[disk
].rdev
);
697 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
698 struct request_queue
*q
=
699 bdev_get_queue(rdev
->bdev
);
700 if (q
->merge_bvec_fn
) {
701 bvm
->bi_sector
= sector
+
703 bvm
->bi_bdev
= rdev
->bdev
;
704 max
= min(max
, q
->merge_bvec_fn(
715 int md_raid1_congested(struct mddev
*mddev
, int bits
)
717 struct r1conf
*conf
= mddev
->private;
720 if ((bits
& (1 << BDI_async_congested
)) &&
721 conf
->pending_count
>= max_queued_requests
)
725 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
726 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
727 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
728 struct request_queue
*q
= bdev_get_queue(rdev
->bdev
);
732 /* Note the '|| 1' - when read_balance prefers
733 * non-congested targets, it can be removed
735 if ((bits
& (1<<BDI_async_congested
)) || 1)
736 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
738 ret
&= bdi_congested(&q
->backing_dev_info
, bits
);
744 EXPORT_SYMBOL_GPL(md_raid1_congested
);
746 static int raid1_congested(void *data
, int bits
)
748 struct mddev
*mddev
= data
;
750 return mddev_congested(mddev
, bits
) ||
751 md_raid1_congested(mddev
, bits
);
754 static void flush_pending_writes(struct r1conf
*conf
)
756 /* Any writes that have been queued but are awaiting
757 * bitmap updates get flushed here.
759 spin_lock_irq(&conf
->device_lock
);
761 if (conf
->pending_bio_list
.head
) {
763 bio
= bio_list_get(&conf
->pending_bio_list
);
764 conf
->pending_count
= 0;
765 spin_unlock_irq(&conf
->device_lock
);
766 /* flush any pending bitmap writes to
767 * disk before proceeding w/ I/O */
768 bitmap_unplug(conf
->mddev
->bitmap
);
769 wake_up(&conf
->wait_barrier
);
771 while (bio
) { /* submit pending writes */
772 struct bio
*next
= bio
->bi_next
;
774 if (unlikely((bio
->bi_rw
& REQ_DISCARD
) &&
775 !blk_queue_discard(bdev_get_queue(bio
->bi_bdev
))))
779 generic_make_request(bio
);
783 spin_unlock_irq(&conf
->device_lock
);
787 * Sometimes we need to suspend IO while we do something else,
788 * either some resync/recovery, or reconfigure the array.
789 * To do this we raise a 'barrier'.
790 * The 'barrier' is a counter that can be raised multiple times
791 * to count how many activities are happening which preclude
793 * We can only raise the barrier if there is no pending IO.
794 * i.e. if nr_pending == 0.
795 * We choose only to raise the barrier if no-one is waiting for the
796 * barrier to go down. This means that as soon as an IO request
797 * is ready, no other operations which require a barrier will start
798 * until the IO request has had a chance.
800 * So: regular IO calls 'wait_barrier'. When that returns there
801 * is no backgroup IO happening, It must arrange to call
802 * allow_barrier when it has finished its IO.
803 * backgroup IO calls must call raise_barrier. Once that returns
804 * there is no normal IO happeing. It must arrange to call
805 * lower_barrier when the particular background IO completes.
807 #define RESYNC_DEPTH 32
809 static void raise_barrier(struct r1conf
*conf
)
811 spin_lock_irq(&conf
->resync_lock
);
813 /* Wait until no block IO is waiting */
814 wait_event_lock_irq(conf
->wait_barrier
, !conf
->nr_waiting
,
817 /* block any new IO from starting */
820 /* Now wait for all pending IO to complete */
821 wait_event_lock_irq(conf
->wait_barrier
,
822 !conf
->nr_pending
&& conf
->barrier
< RESYNC_DEPTH
,
825 spin_unlock_irq(&conf
->resync_lock
);
828 static void lower_barrier(struct r1conf
*conf
)
831 BUG_ON(conf
->barrier
<= 0);
832 spin_lock_irqsave(&conf
->resync_lock
, flags
);
834 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
835 wake_up(&conf
->wait_barrier
);
838 static void wait_barrier(struct r1conf
*conf
)
840 spin_lock_irq(&conf
->resync_lock
);
843 /* Wait for the barrier to drop.
844 * However if there are already pending
845 * requests (preventing the barrier from
846 * rising completely), and the
847 * pre-process bio queue isn't empty,
848 * then don't wait, as we need to empty
849 * that queue to get the nr_pending
852 wait_event_lock_irq(conf
->wait_barrier
,
856 !bio_list_empty(current
->bio_list
)),
861 spin_unlock_irq(&conf
->resync_lock
);
864 static void allow_barrier(struct r1conf
*conf
)
867 spin_lock_irqsave(&conf
->resync_lock
, flags
);
869 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
870 wake_up(&conf
->wait_barrier
);
873 static void freeze_array(struct r1conf
*conf
)
875 /* stop syncio and normal IO and wait for everything to
877 * We increment barrier and nr_waiting, and then
878 * wait until nr_pending match nr_queued+1
879 * This is called in the context of one normal IO request
880 * that has failed. Thus any sync request that might be pending
881 * will be blocked by nr_pending, and we need to wait for
882 * pending IO requests to complete or be queued for re-try.
883 * Thus the number queued (nr_queued) plus this request (1)
884 * must match the number of pending IOs (nr_pending) before
887 spin_lock_irq(&conf
->resync_lock
);
890 wait_event_lock_irq_cmd(conf
->wait_barrier
,
891 conf
->nr_pending
== conf
->nr_queued
+1,
893 flush_pending_writes(conf
));
894 spin_unlock_irq(&conf
->resync_lock
);
896 static void unfreeze_array(struct r1conf
*conf
)
898 /* reverse the effect of the freeze */
899 spin_lock_irq(&conf
->resync_lock
);
902 wake_up(&conf
->wait_barrier
);
903 spin_unlock_irq(&conf
->resync_lock
);
907 /* duplicate the data pages for behind I/O
909 static void alloc_behind_pages(struct bio
*bio
, struct r1bio
*r1_bio
)
912 struct bio_vec
*bvec
;
913 struct bio_vec
*bvecs
= kzalloc(bio
->bi_vcnt
* sizeof(struct bio_vec
),
915 if (unlikely(!bvecs
))
918 bio_for_each_segment_all(bvec
, bio
, i
) {
920 bvecs
[i
].bv_page
= alloc_page(GFP_NOIO
);
921 if (unlikely(!bvecs
[i
].bv_page
))
923 memcpy(kmap(bvecs
[i
].bv_page
) + bvec
->bv_offset
,
924 kmap(bvec
->bv_page
) + bvec
->bv_offset
, bvec
->bv_len
);
925 kunmap(bvecs
[i
].bv_page
);
926 kunmap(bvec
->bv_page
);
928 r1_bio
->behind_bvecs
= bvecs
;
929 r1_bio
->behind_page_count
= bio
->bi_vcnt
;
930 set_bit(R1BIO_BehindIO
, &r1_bio
->state
);
934 for (i
= 0; i
< bio
->bi_vcnt
; i
++)
935 if (bvecs
[i
].bv_page
)
936 put_page(bvecs
[i
].bv_page
);
938 pr_debug("%dB behind alloc failed, doing sync I/O\n", bio
->bi_size
);
941 struct raid1_plug_cb
{
942 struct blk_plug_cb cb
;
943 struct bio_list pending
;
947 static void raid1_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
949 struct raid1_plug_cb
*plug
= container_of(cb
, struct raid1_plug_cb
,
951 struct mddev
*mddev
= plug
->cb
.data
;
952 struct r1conf
*conf
= mddev
->private;
955 if (from_schedule
|| current
->bio_list
) {
956 spin_lock_irq(&conf
->device_lock
);
957 bio_list_merge(&conf
->pending_bio_list
, &plug
->pending
);
958 conf
->pending_count
+= plug
->pending_cnt
;
959 spin_unlock_irq(&conf
->device_lock
);
960 wake_up(&conf
->wait_barrier
);
961 md_wakeup_thread(mddev
->thread
);
966 /* we aren't scheduling, so we can do the write-out directly. */
967 bio
= bio_list_get(&plug
->pending
);
968 bitmap_unplug(mddev
->bitmap
);
969 wake_up(&conf
->wait_barrier
);
971 while (bio
) { /* submit pending writes */
972 struct bio
*next
= bio
->bi_next
;
974 generic_make_request(bio
);
980 static void make_request(struct mddev
*mddev
, struct bio
* bio
)
982 struct r1conf
*conf
= mddev
->private;
983 struct raid1_info
*mirror
;
984 struct r1bio
*r1_bio
;
985 struct bio
*read_bio
;
987 struct bitmap
*bitmap
;
989 const int rw
= bio_data_dir(bio
);
990 const unsigned long do_sync
= (bio
->bi_rw
& REQ_SYNC
);
991 const unsigned long do_flush_fua
= (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
));
992 const unsigned long do_discard
= (bio
->bi_rw
993 & (REQ_DISCARD
| REQ_SECURE
));
994 const unsigned long do_same
= (bio
->bi_rw
& REQ_WRITE_SAME
);
995 struct md_rdev
*blocked_rdev
;
996 struct blk_plug_cb
*cb
;
997 struct raid1_plug_cb
*plug
= NULL
;
1003 * Register the new request and wait if the reconstruction
1004 * thread has put up a bar for new requests.
1005 * Continue immediately if no resync is active currently.
1008 md_write_start(mddev
, bio
); /* wait on superblock update early */
1010 if (bio_data_dir(bio
) == WRITE
&&
1011 bio_end_sector(bio
) > mddev
->suspend_lo
&&
1012 bio
->bi_sector
< mddev
->suspend_hi
) {
1013 /* As the suspend_* range is controlled by
1014 * userspace, we want an interruptible
1019 flush_signals(current
);
1020 prepare_to_wait(&conf
->wait_barrier
,
1021 &w
, TASK_INTERRUPTIBLE
);
1022 if (bio_end_sector(bio
) <= mddev
->suspend_lo
||
1023 bio
->bi_sector
>= mddev
->suspend_hi
)
1027 finish_wait(&conf
->wait_barrier
, &w
);
1032 bitmap
= mddev
->bitmap
;
1035 * make_request() can abort the operation when READA is being
1036 * used and no empty request is available.
1039 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1041 r1_bio
->master_bio
= bio
;
1042 r1_bio
->sectors
= bio_sectors(bio
);
1044 r1_bio
->mddev
= mddev
;
1045 r1_bio
->sector
= bio
->bi_sector
;
1047 /* We might need to issue multiple reads to different
1048 * devices if there are bad blocks around, so we keep
1049 * track of the number of reads in bio->bi_phys_segments.
1050 * If this is 0, there is only one r1_bio and no locking
1051 * will be needed when requests complete. If it is
1052 * non-zero, then it is the number of not-completed requests.
1054 bio
->bi_phys_segments
= 0;
1055 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
1059 * read balancing logic:
1064 rdisk
= read_balance(conf
, r1_bio
, &max_sectors
);
1067 /* couldn't find anywhere to read from */
1068 raid_end_bio_io(r1_bio
);
1071 mirror
= conf
->mirrors
+ rdisk
;
1073 if (test_bit(WriteMostly
, &mirror
->rdev
->flags
) &&
1075 /* Reading from a write-mostly device must
1076 * take care not to over-take any writes
1079 wait_event(bitmap
->behind_wait
,
1080 atomic_read(&bitmap
->behind_writes
) == 0);
1082 r1_bio
->read_disk
= rdisk
;
1084 read_bio
= bio_clone_mddev(bio
, GFP_NOIO
, mddev
);
1085 md_trim_bio(read_bio
, r1_bio
->sector
- bio
->bi_sector
,
1088 r1_bio
->bios
[rdisk
] = read_bio
;
1090 read_bio
->bi_sector
= r1_bio
->sector
+ mirror
->rdev
->data_offset
;
1091 read_bio
->bi_bdev
= mirror
->rdev
->bdev
;
1092 read_bio
->bi_end_io
= raid1_end_read_request
;
1093 read_bio
->bi_rw
= READ
| do_sync
;
1094 read_bio
->bi_private
= r1_bio
;
1096 if (max_sectors
< r1_bio
->sectors
) {
1097 /* could not read all from this device, so we will
1098 * need another r1_bio.
1101 sectors_handled
= (r1_bio
->sector
+ max_sectors
1103 r1_bio
->sectors
= max_sectors
;
1104 spin_lock_irq(&conf
->device_lock
);
1105 if (bio
->bi_phys_segments
== 0)
1106 bio
->bi_phys_segments
= 2;
1108 bio
->bi_phys_segments
++;
1109 spin_unlock_irq(&conf
->device_lock
);
1110 /* Cannot call generic_make_request directly
1111 * as that will be queued in __make_request
1112 * and subsequent mempool_alloc might block waiting
1113 * for it. So hand bio over to raid1d.
1115 reschedule_retry(r1_bio
);
1117 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1119 r1_bio
->master_bio
= bio
;
1120 r1_bio
->sectors
= bio_sectors(bio
) - sectors_handled
;
1122 r1_bio
->mddev
= mddev
;
1123 r1_bio
->sector
= bio
->bi_sector
+ sectors_handled
;
1126 generic_make_request(read_bio
);
1133 if (conf
->pending_count
>= max_queued_requests
) {
1134 md_wakeup_thread(mddev
->thread
);
1135 wait_event(conf
->wait_barrier
,
1136 conf
->pending_count
< max_queued_requests
);
1138 /* first select target devices under rcu_lock and
1139 * inc refcount on their rdev. Record them by setting
1141 * If there are known/acknowledged bad blocks on any device on
1142 * which we have seen a write error, we want to avoid writing those
1144 * This potentially requires several writes to write around
1145 * the bad blocks. Each set of writes gets it's own r1bio
1146 * with a set of bios attached.
1149 disks
= conf
->raid_disks
* 2;
1151 blocked_rdev
= NULL
;
1153 max_sectors
= r1_bio
->sectors
;
1154 for (i
= 0; i
< disks
; i
++) {
1155 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1156 if (rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
1157 atomic_inc(&rdev
->nr_pending
);
1158 blocked_rdev
= rdev
;
1161 r1_bio
->bios
[i
] = NULL
;
1162 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
)
1163 || test_bit(Unmerged
, &rdev
->flags
)) {
1164 if (i
< conf
->raid_disks
)
1165 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
1169 atomic_inc(&rdev
->nr_pending
);
1170 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
1175 is_bad
= is_badblock(rdev
, r1_bio
->sector
,
1177 &first_bad
, &bad_sectors
);
1179 /* mustn't write here until the bad block is
1181 set_bit(BlockedBadBlocks
, &rdev
->flags
);
1182 blocked_rdev
= rdev
;
1185 if (is_bad
&& first_bad
<= r1_bio
->sector
) {
1186 /* Cannot write here at all */
1187 bad_sectors
-= (r1_bio
->sector
- first_bad
);
1188 if (bad_sectors
< max_sectors
)
1189 /* mustn't write more than bad_sectors
1190 * to other devices yet
1192 max_sectors
= bad_sectors
;
1193 rdev_dec_pending(rdev
, mddev
);
1194 /* We don't set R1BIO_Degraded as that
1195 * only applies if the disk is
1196 * missing, so it might be re-added,
1197 * and we want to know to recover this
1199 * In this case the device is here,
1200 * and the fact that this chunk is not
1201 * in-sync is recorded in the bad
1207 int good_sectors
= first_bad
- r1_bio
->sector
;
1208 if (good_sectors
< max_sectors
)
1209 max_sectors
= good_sectors
;
1212 r1_bio
->bios
[i
] = bio
;
1216 if (unlikely(blocked_rdev
)) {
1217 /* Wait for this device to become unblocked */
1220 for (j
= 0; j
< i
; j
++)
1221 if (r1_bio
->bios
[j
])
1222 rdev_dec_pending(conf
->mirrors
[j
].rdev
, mddev
);
1224 allow_barrier(conf
);
1225 md_wait_for_blocked_rdev(blocked_rdev
, mddev
);
1230 if (max_sectors
< r1_bio
->sectors
) {
1231 /* We are splitting this write into multiple parts, so
1232 * we need to prepare for allocating another r1_bio.
1234 r1_bio
->sectors
= max_sectors
;
1235 spin_lock_irq(&conf
->device_lock
);
1236 if (bio
->bi_phys_segments
== 0)
1237 bio
->bi_phys_segments
= 2;
1239 bio
->bi_phys_segments
++;
1240 spin_unlock_irq(&conf
->device_lock
);
1242 sectors_handled
= r1_bio
->sector
+ max_sectors
- bio
->bi_sector
;
1244 atomic_set(&r1_bio
->remaining
, 1);
1245 atomic_set(&r1_bio
->behind_remaining
, 0);
1248 for (i
= 0; i
< disks
; i
++) {
1250 if (!r1_bio
->bios
[i
])
1253 mbio
= bio_clone_mddev(bio
, GFP_NOIO
, mddev
);
1254 md_trim_bio(mbio
, r1_bio
->sector
- bio
->bi_sector
, max_sectors
);
1258 * Not if there are too many, or cannot
1259 * allocate memory, or a reader on WriteMostly
1260 * is waiting for behind writes to flush */
1262 (atomic_read(&bitmap
->behind_writes
)
1263 < mddev
->bitmap_info
.max_write_behind
) &&
1264 !waitqueue_active(&bitmap
->behind_wait
))
1265 alloc_behind_pages(mbio
, r1_bio
);
1267 bitmap_startwrite(bitmap
, r1_bio
->sector
,
1269 test_bit(R1BIO_BehindIO
,
1273 if (r1_bio
->behind_bvecs
) {
1274 struct bio_vec
*bvec
;
1278 * We trimmed the bio, so _all is legit
1280 bio_for_each_segment_all(bvec
, mbio
, j
)
1281 bvec
->bv_page
= r1_bio
->behind_bvecs
[j
].bv_page
;
1282 if (test_bit(WriteMostly
, &conf
->mirrors
[i
].rdev
->flags
))
1283 atomic_inc(&r1_bio
->behind_remaining
);
1286 r1_bio
->bios
[i
] = mbio
;
1288 mbio
->bi_sector
= (r1_bio
->sector
+
1289 conf
->mirrors
[i
].rdev
->data_offset
);
1290 mbio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1291 mbio
->bi_end_io
= raid1_end_write_request
;
1293 WRITE
| do_flush_fua
| do_sync
| do_discard
| do_same
;
1294 mbio
->bi_private
= r1_bio
;
1296 atomic_inc(&r1_bio
->remaining
);
1298 cb
= blk_check_plugged(raid1_unplug
, mddev
, sizeof(*plug
));
1300 plug
= container_of(cb
, struct raid1_plug_cb
, cb
);
1303 spin_lock_irqsave(&conf
->device_lock
, flags
);
1305 bio_list_add(&plug
->pending
, mbio
);
1306 plug
->pending_cnt
++;
1308 bio_list_add(&conf
->pending_bio_list
, mbio
);
1309 conf
->pending_count
++;
1311 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1313 md_wakeup_thread(mddev
->thread
);
1315 /* Mustn't call r1_bio_write_done before this next test,
1316 * as it could result in the bio being freed.
1318 if (sectors_handled
< bio_sectors(bio
)) {
1319 r1_bio_write_done(r1_bio
);
1320 /* We need another r1_bio. It has already been counted
1321 * in bio->bi_phys_segments
1323 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1324 r1_bio
->master_bio
= bio
;
1325 r1_bio
->sectors
= bio_sectors(bio
) - sectors_handled
;
1327 r1_bio
->mddev
= mddev
;
1328 r1_bio
->sector
= bio
->bi_sector
+ sectors_handled
;
1332 r1_bio_write_done(r1_bio
);
1334 /* In case raid1d snuck in to freeze_array */
1335 wake_up(&conf
->wait_barrier
);
1338 static void status(struct seq_file
*seq
, struct mddev
*mddev
)
1340 struct r1conf
*conf
= mddev
->private;
1343 seq_printf(seq
, " [%d/%d] [", conf
->raid_disks
,
1344 conf
->raid_disks
- mddev
->degraded
);
1346 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1347 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1348 seq_printf(seq
, "%s",
1349 rdev
&& test_bit(In_sync
, &rdev
->flags
) ? "U" : "_");
1352 seq_printf(seq
, "]");
1356 static void error(struct mddev
*mddev
, struct md_rdev
*rdev
)
1358 char b
[BDEVNAME_SIZE
];
1359 struct r1conf
*conf
= mddev
->private;
1362 * If it is not operational, then we have already marked it as dead
1363 * else if it is the last working disks, ignore the error, let the
1364 * next level up know.
1365 * else mark the drive as failed
1367 if (test_bit(In_sync
, &rdev
->flags
)
1368 && (conf
->raid_disks
- mddev
->degraded
) == 1) {
1370 * Don't fail the drive, act as though we were just a
1371 * normal single drive.
1372 * However don't try a recovery from this drive as
1373 * it is very likely to fail.
1375 conf
->recovery_disabled
= mddev
->recovery_disabled
;
1378 set_bit(Blocked
, &rdev
->flags
);
1379 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1380 unsigned long flags
;
1381 spin_lock_irqsave(&conf
->device_lock
, flags
);
1383 set_bit(Faulty
, &rdev
->flags
);
1384 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1386 * if recovery is running, make sure it aborts.
1388 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1390 set_bit(Faulty
, &rdev
->flags
);
1391 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1393 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1394 "md/raid1:%s: Operation continuing on %d devices.\n",
1395 mdname(mddev
), bdevname(rdev
->bdev
, b
),
1396 mdname(mddev
), conf
->raid_disks
- mddev
->degraded
);
1399 static void print_conf(struct r1conf
*conf
)
1403 printk(KERN_DEBUG
"RAID1 conf printout:\n");
1405 printk(KERN_DEBUG
"(!conf)\n");
1408 printk(KERN_DEBUG
" --- wd:%d rd:%d\n", conf
->raid_disks
- conf
->mddev
->degraded
,
1412 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1413 char b
[BDEVNAME_SIZE
];
1414 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1416 printk(KERN_DEBUG
" disk %d, wo:%d, o:%d, dev:%s\n",
1417 i
, !test_bit(In_sync
, &rdev
->flags
),
1418 !test_bit(Faulty
, &rdev
->flags
),
1419 bdevname(rdev
->bdev
,b
));
1424 static void close_sync(struct r1conf
*conf
)
1427 allow_barrier(conf
);
1429 mempool_destroy(conf
->r1buf_pool
);
1430 conf
->r1buf_pool
= NULL
;
1433 static int raid1_spare_active(struct mddev
*mddev
)
1436 struct r1conf
*conf
= mddev
->private;
1438 unsigned long flags
;
1441 * Find all failed disks within the RAID1 configuration
1442 * and mark them readable.
1443 * Called under mddev lock, so rcu protection not needed.
1445 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1446 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
1447 struct md_rdev
*repl
= conf
->mirrors
[conf
->raid_disks
+ i
].rdev
;
1449 && repl
->recovery_offset
== MaxSector
1450 && !test_bit(Faulty
, &repl
->flags
)
1451 && !test_and_set_bit(In_sync
, &repl
->flags
)) {
1452 /* replacement has just become active */
1454 !test_and_clear_bit(In_sync
, &rdev
->flags
))
1457 /* Replaced device not technically
1458 * faulty, but we need to be sure
1459 * it gets removed and never re-added
1461 set_bit(Faulty
, &rdev
->flags
);
1462 sysfs_notify_dirent_safe(
1467 && !test_bit(Faulty
, &rdev
->flags
)
1468 && !test_and_set_bit(In_sync
, &rdev
->flags
)) {
1470 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
1473 spin_lock_irqsave(&conf
->device_lock
, flags
);
1474 mddev
->degraded
-= count
;
1475 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1482 static int raid1_add_disk(struct mddev
*mddev
, struct md_rdev
*rdev
)
1484 struct r1conf
*conf
= mddev
->private;
1487 struct raid1_info
*p
;
1489 int last
= conf
->raid_disks
- 1;
1490 struct request_queue
*q
= bdev_get_queue(rdev
->bdev
);
1492 if (mddev
->recovery_disabled
== conf
->recovery_disabled
)
1495 if (rdev
->raid_disk
>= 0)
1496 first
= last
= rdev
->raid_disk
;
1498 if (q
->merge_bvec_fn
) {
1499 set_bit(Unmerged
, &rdev
->flags
);
1500 mddev
->merge_check_needed
= 1;
1503 for (mirror
= first
; mirror
<= last
; mirror
++) {
1504 p
= conf
->mirrors
+mirror
;
1507 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
1508 rdev
->data_offset
<< 9);
1510 p
->head_position
= 0;
1511 rdev
->raid_disk
= mirror
;
1513 /* As all devices are equivalent, we don't need a full recovery
1514 * if this was recently any drive of the array
1516 if (rdev
->saved_raid_disk
< 0)
1518 rcu_assign_pointer(p
->rdev
, rdev
);
1521 if (test_bit(WantReplacement
, &p
->rdev
->flags
) &&
1522 p
[conf
->raid_disks
].rdev
== NULL
) {
1523 /* Add this device as a replacement */
1524 clear_bit(In_sync
, &rdev
->flags
);
1525 set_bit(Replacement
, &rdev
->flags
);
1526 rdev
->raid_disk
= mirror
;
1529 rcu_assign_pointer(p
[conf
->raid_disks
].rdev
, rdev
);
1533 if (err
== 0 && test_bit(Unmerged
, &rdev
->flags
)) {
1534 /* Some requests might not have seen this new
1535 * merge_bvec_fn. We must wait for them to complete
1536 * before merging the device fully.
1537 * First we make sure any code which has tested
1538 * our function has submitted the request, then
1539 * we wait for all outstanding requests to complete.
1541 synchronize_sched();
1542 raise_barrier(conf
);
1543 lower_barrier(conf
);
1544 clear_bit(Unmerged
, &rdev
->flags
);
1546 md_integrity_add_rdev(rdev
, mddev
);
1547 if (blk_queue_discard(bdev_get_queue(rdev
->bdev
)))
1548 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, mddev
->queue
);
1553 static int raid1_remove_disk(struct mddev
*mddev
, struct md_rdev
*rdev
)
1555 struct r1conf
*conf
= mddev
->private;
1557 int number
= rdev
->raid_disk
;
1558 struct raid1_info
*p
= conf
->mirrors
+ number
;
1560 if (rdev
!= p
->rdev
)
1561 p
= conf
->mirrors
+ conf
->raid_disks
+ number
;
1564 if (rdev
== p
->rdev
) {
1565 if (test_bit(In_sync
, &rdev
->flags
) ||
1566 atomic_read(&rdev
->nr_pending
)) {
1570 /* Only remove non-faulty devices if recovery
1573 if (!test_bit(Faulty
, &rdev
->flags
) &&
1574 mddev
->recovery_disabled
!= conf
->recovery_disabled
&&
1575 mddev
->degraded
< conf
->raid_disks
) {
1581 if (atomic_read(&rdev
->nr_pending
)) {
1582 /* lost the race, try later */
1586 } else if (conf
->mirrors
[conf
->raid_disks
+ number
].rdev
) {
1587 /* We just removed a device that is being replaced.
1588 * Move down the replacement. We drain all IO before
1589 * doing this to avoid confusion.
1591 struct md_rdev
*repl
=
1592 conf
->mirrors
[conf
->raid_disks
+ number
].rdev
;
1593 raise_barrier(conf
);
1594 clear_bit(Replacement
, &repl
->flags
);
1596 conf
->mirrors
[conf
->raid_disks
+ number
].rdev
= NULL
;
1597 lower_barrier(conf
);
1598 clear_bit(WantReplacement
, &rdev
->flags
);
1600 clear_bit(WantReplacement
, &rdev
->flags
);
1601 err
= md_integrity_register(mddev
);
1610 static void end_sync_read(struct bio
*bio
, int error
)
1612 struct r1bio
*r1_bio
= bio
->bi_private
;
1614 update_head_pos(r1_bio
->read_disk
, r1_bio
);
1617 * we have read a block, now it needs to be re-written,
1618 * or re-read if the read failed.
1619 * We don't do much here, just schedule handling by raid1d
1621 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1622 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1624 if (atomic_dec_and_test(&r1_bio
->remaining
))
1625 reschedule_retry(r1_bio
);
1628 static void end_sync_write(struct bio
*bio
, int error
)
1630 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1631 struct r1bio
*r1_bio
= bio
->bi_private
;
1632 struct mddev
*mddev
= r1_bio
->mddev
;
1633 struct r1conf
*conf
= mddev
->private;
1638 mirror
= find_bio_disk(r1_bio
, bio
);
1641 sector_t sync_blocks
= 0;
1642 sector_t s
= r1_bio
->sector
;
1643 long sectors_to_go
= r1_bio
->sectors
;
1644 /* make sure these bits doesn't get cleared. */
1646 bitmap_end_sync(mddev
->bitmap
, s
,
1649 sectors_to_go
-= sync_blocks
;
1650 } while (sectors_to_go
> 0);
1651 set_bit(WriteErrorSeen
,
1652 &conf
->mirrors
[mirror
].rdev
->flags
);
1653 if (!test_and_set_bit(WantReplacement
,
1654 &conf
->mirrors
[mirror
].rdev
->flags
))
1655 set_bit(MD_RECOVERY_NEEDED
, &
1657 set_bit(R1BIO_WriteError
, &r1_bio
->state
);
1658 } else if (is_badblock(conf
->mirrors
[mirror
].rdev
,
1661 &first_bad
, &bad_sectors
) &&
1662 !is_badblock(conf
->mirrors
[r1_bio
->read_disk
].rdev
,
1665 &first_bad
, &bad_sectors
)
1667 set_bit(R1BIO_MadeGood
, &r1_bio
->state
);
1669 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
1670 int s
= r1_bio
->sectors
;
1671 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
1672 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
1673 reschedule_retry(r1_bio
);
1676 md_done_sync(mddev
, s
, uptodate
);
1681 static int r1_sync_page_io(struct md_rdev
*rdev
, sector_t sector
,
1682 int sectors
, struct page
*page
, int rw
)
1684 if (sync_page_io(rdev
, sector
, sectors
<< 9, page
, rw
, false))
1688 set_bit(WriteErrorSeen
, &rdev
->flags
);
1689 if (!test_and_set_bit(WantReplacement
,
1691 set_bit(MD_RECOVERY_NEEDED
, &
1692 rdev
->mddev
->recovery
);
1694 /* need to record an error - either for the block or the device */
1695 if (!rdev_set_badblocks(rdev
, sector
, sectors
, 0))
1696 md_error(rdev
->mddev
, rdev
);
1700 static int fix_sync_read_error(struct r1bio
*r1_bio
)
1702 /* Try some synchronous reads of other devices to get
1703 * good data, much like with normal read errors. Only
1704 * read into the pages we already have so we don't
1705 * need to re-issue the read request.
1706 * We don't need to freeze the array, because being in an
1707 * active sync request, there is no normal IO, and
1708 * no overlapping syncs.
1709 * We don't need to check is_badblock() again as we
1710 * made sure that anything with a bad block in range
1711 * will have bi_end_io clear.
1713 struct mddev
*mddev
= r1_bio
->mddev
;
1714 struct r1conf
*conf
= mddev
->private;
1715 struct bio
*bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1716 sector_t sect
= r1_bio
->sector
;
1717 int sectors
= r1_bio
->sectors
;
1722 int d
= r1_bio
->read_disk
;
1724 struct md_rdev
*rdev
;
1727 if (s
> (PAGE_SIZE
>>9))
1730 if (r1_bio
->bios
[d
]->bi_end_io
== end_sync_read
) {
1731 /* No rcu protection needed here devices
1732 * can only be removed when no resync is
1733 * active, and resync is currently active
1735 rdev
= conf
->mirrors
[d
].rdev
;
1736 if (sync_page_io(rdev
, sect
, s
<<9,
1737 bio
->bi_io_vec
[idx
].bv_page
,
1744 if (d
== conf
->raid_disks
* 2)
1746 } while (!success
&& d
!= r1_bio
->read_disk
);
1749 char b
[BDEVNAME_SIZE
];
1751 /* Cannot read from anywhere, this block is lost.
1752 * Record a bad block on each device. If that doesn't
1753 * work just disable and interrupt the recovery.
1754 * Don't fail devices as that won't really help.
1756 printk(KERN_ALERT
"md/raid1:%s: %s: unrecoverable I/O read error"
1757 " for block %llu\n",
1759 bdevname(bio
->bi_bdev
, b
),
1760 (unsigned long long)r1_bio
->sector
);
1761 for (d
= 0; d
< conf
->raid_disks
* 2; d
++) {
1762 rdev
= conf
->mirrors
[d
].rdev
;
1763 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
1765 if (!rdev_set_badblocks(rdev
, sect
, s
, 0))
1769 conf
->recovery_disabled
=
1770 mddev
->recovery_disabled
;
1771 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1772 md_done_sync(mddev
, r1_bio
->sectors
, 0);
1784 /* write it back and re-read */
1785 while (d
!= r1_bio
->read_disk
) {
1787 d
= conf
->raid_disks
* 2;
1789 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1791 rdev
= conf
->mirrors
[d
].rdev
;
1792 if (r1_sync_page_io(rdev
, sect
, s
,
1793 bio
->bi_io_vec
[idx
].bv_page
,
1795 r1_bio
->bios
[d
]->bi_end_io
= NULL
;
1796 rdev_dec_pending(rdev
, mddev
);
1800 while (d
!= r1_bio
->read_disk
) {
1802 d
= conf
->raid_disks
* 2;
1804 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1806 rdev
= conf
->mirrors
[d
].rdev
;
1807 if (r1_sync_page_io(rdev
, sect
, s
,
1808 bio
->bi_io_vec
[idx
].bv_page
,
1810 atomic_add(s
, &rdev
->corrected_errors
);
1816 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1817 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1821 static int process_checks(struct r1bio
*r1_bio
)
1823 /* We have read all readable devices. If we haven't
1824 * got the block, then there is no hope left.
1825 * If we have, then we want to do a comparison
1826 * and skip the write if everything is the same.
1827 * If any blocks failed to read, then we need to
1828 * attempt an over-write
1830 struct mddev
*mddev
= r1_bio
->mddev
;
1831 struct r1conf
*conf
= mddev
->private;
1836 for (primary
= 0; primary
< conf
->raid_disks
* 2; primary
++)
1837 if (r1_bio
->bios
[primary
]->bi_end_io
== end_sync_read
&&
1838 test_bit(BIO_UPTODATE
, &r1_bio
->bios
[primary
]->bi_flags
)) {
1839 r1_bio
->bios
[primary
]->bi_end_io
= NULL
;
1840 rdev_dec_pending(conf
->mirrors
[primary
].rdev
, mddev
);
1843 r1_bio
->read_disk
= primary
;
1844 vcnt
= (r1_bio
->sectors
+ PAGE_SIZE
/ 512 - 1) >> (PAGE_SHIFT
- 9);
1845 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
1847 struct bio
*pbio
= r1_bio
->bios
[primary
];
1848 struct bio
*sbio
= r1_bio
->bios
[i
];
1851 if (sbio
->bi_end_io
!= end_sync_read
)
1854 if (test_bit(BIO_UPTODATE
, &sbio
->bi_flags
)) {
1855 for (j
= vcnt
; j
-- ; ) {
1857 p
= pbio
->bi_io_vec
[j
].bv_page
;
1858 s
= sbio
->bi_io_vec
[j
].bv_page
;
1859 if (memcmp(page_address(p
),
1861 sbio
->bi_io_vec
[j
].bv_len
))
1867 atomic64_add(r1_bio
->sectors
, &mddev
->resync_mismatches
);
1868 if (j
< 0 || (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)
1869 && test_bit(BIO_UPTODATE
, &sbio
->bi_flags
))) {
1870 /* No need to write to this device. */
1871 sbio
->bi_end_io
= NULL
;
1872 rdev_dec_pending(conf
->mirrors
[i
].rdev
, mddev
);
1875 /* fixup the bio for reuse */
1877 sbio
->bi_vcnt
= vcnt
;
1878 sbio
->bi_size
= r1_bio
->sectors
<< 9;
1879 sbio
->bi_sector
= r1_bio
->sector
+
1880 conf
->mirrors
[i
].rdev
->data_offset
;
1881 sbio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1882 sbio
->bi_end_io
= end_sync_read
;
1883 sbio
->bi_private
= r1_bio
;
1885 size
= sbio
->bi_size
;
1886 for (j
= 0; j
< vcnt
; j
++) {
1888 bi
= &sbio
->bi_io_vec
[j
];
1890 if (size
> PAGE_SIZE
)
1891 bi
->bv_len
= PAGE_SIZE
;
1897 bio_copy_data(sbio
, pbio
);
1902 static void sync_request_write(struct mddev
*mddev
, struct r1bio
*r1_bio
)
1904 struct r1conf
*conf
= mddev
->private;
1906 int disks
= conf
->raid_disks
* 2;
1907 struct bio
*bio
, *wbio
;
1909 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1911 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
))
1912 /* ouch - failed to read all of that. */
1913 if (!fix_sync_read_error(r1_bio
))
1916 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
1917 if (process_checks(r1_bio
) < 0)
1922 atomic_set(&r1_bio
->remaining
, 1);
1923 for (i
= 0; i
< disks
; i
++) {
1924 wbio
= r1_bio
->bios
[i
];
1925 if (wbio
->bi_end_io
== NULL
||
1926 (wbio
->bi_end_io
== end_sync_read
&&
1927 (i
== r1_bio
->read_disk
||
1928 !test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))))
1931 wbio
->bi_rw
= WRITE
;
1932 wbio
->bi_end_io
= end_sync_write
;
1933 atomic_inc(&r1_bio
->remaining
);
1934 md_sync_acct(conf
->mirrors
[i
].rdev
->bdev
, bio_sectors(wbio
));
1936 generic_make_request(wbio
);
1939 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
1940 /* if we're here, all write(s) have completed, so clean up */
1941 int s
= r1_bio
->sectors
;
1942 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
1943 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
1944 reschedule_retry(r1_bio
);
1947 md_done_sync(mddev
, s
, 1);
1953 * This is a kernel thread which:
1955 * 1. Retries failed read operations on working mirrors.
1956 * 2. Updates the raid superblock when problems encounter.
1957 * 3. Performs writes following reads for array synchronising.
1960 static void fix_read_error(struct r1conf
*conf
, int read_disk
,
1961 sector_t sect
, int sectors
)
1963 struct mddev
*mddev
= conf
->mddev
;
1969 struct md_rdev
*rdev
;
1971 if (s
> (PAGE_SIZE
>>9))
1975 /* Note: no rcu protection needed here
1976 * as this is synchronous in the raid1d thread
1977 * which is the thread that might remove
1978 * a device. If raid1d ever becomes multi-threaded....
1983 rdev
= conf
->mirrors
[d
].rdev
;
1985 (test_bit(In_sync
, &rdev
->flags
) ||
1986 (!test_bit(Faulty
, &rdev
->flags
) &&
1987 rdev
->recovery_offset
>= sect
+ s
)) &&
1988 is_badblock(rdev
, sect
, s
,
1989 &first_bad
, &bad_sectors
) == 0 &&
1990 sync_page_io(rdev
, sect
, s
<<9,
1991 conf
->tmppage
, READ
, false))
1995 if (d
== conf
->raid_disks
* 2)
1998 } while (!success
&& d
!= read_disk
);
2001 /* Cannot read from anywhere - mark it bad */
2002 struct md_rdev
*rdev
= conf
->mirrors
[read_disk
].rdev
;
2003 if (!rdev_set_badblocks(rdev
, sect
, s
, 0))
2004 md_error(mddev
, rdev
);
2007 /* write it back and re-read */
2009 while (d
!= read_disk
) {
2011 d
= conf
->raid_disks
* 2;
2013 rdev
= conf
->mirrors
[d
].rdev
;
2015 test_bit(In_sync
, &rdev
->flags
))
2016 r1_sync_page_io(rdev
, sect
, s
,
2017 conf
->tmppage
, WRITE
);
2020 while (d
!= read_disk
) {
2021 char b
[BDEVNAME_SIZE
];
2023 d
= conf
->raid_disks
* 2;
2025 rdev
= conf
->mirrors
[d
].rdev
;
2027 test_bit(In_sync
, &rdev
->flags
)) {
2028 if (r1_sync_page_io(rdev
, sect
, s
,
2029 conf
->tmppage
, READ
)) {
2030 atomic_add(s
, &rdev
->corrected_errors
);
2032 "md/raid1:%s: read error corrected "
2033 "(%d sectors at %llu on %s)\n",
2035 (unsigned long long)(sect
+
2037 bdevname(rdev
->bdev
, b
));
2046 static int narrow_write_error(struct r1bio
*r1_bio
, int i
)
2048 struct mddev
*mddev
= r1_bio
->mddev
;
2049 struct r1conf
*conf
= mddev
->private;
2050 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
2052 /* bio has the data to be written to device 'i' where
2053 * we just recently had a write error.
2054 * We repeatedly clone the bio and trim down to one block,
2055 * then try the write. Where the write fails we record
2057 * It is conceivable that the bio doesn't exactly align with
2058 * blocks. We must handle this somehow.
2060 * We currently own a reference on the rdev.
2066 int sect_to_write
= r1_bio
->sectors
;
2069 if (rdev
->badblocks
.shift
< 0)
2072 block_sectors
= 1 << rdev
->badblocks
.shift
;
2073 sector
= r1_bio
->sector
;
2074 sectors
= ((sector
+ block_sectors
)
2075 & ~(sector_t
)(block_sectors
- 1))
2078 while (sect_to_write
) {
2080 if (sectors
> sect_to_write
)
2081 sectors
= sect_to_write
;
2082 /* Write at 'sector' for 'sectors'*/
2084 if (test_bit(R1BIO_BehindIO
, &r1_bio
->state
)) {
2085 unsigned vcnt
= r1_bio
->behind_page_count
;
2086 struct bio_vec
*vec
= r1_bio
->behind_bvecs
;
2088 while (!vec
->bv_page
) {
2093 wbio
= bio_alloc_mddev(GFP_NOIO
, vcnt
, mddev
);
2094 memcpy(wbio
->bi_io_vec
, vec
, vcnt
* sizeof(struct bio_vec
));
2096 wbio
->bi_vcnt
= vcnt
;
2098 wbio
= bio_clone_mddev(r1_bio
->master_bio
, GFP_NOIO
, mddev
);
2101 wbio
->bi_rw
= WRITE
;
2102 wbio
->bi_sector
= r1_bio
->sector
;
2103 wbio
->bi_size
= r1_bio
->sectors
<< 9;
2105 md_trim_bio(wbio
, sector
- r1_bio
->sector
, sectors
);
2106 wbio
->bi_sector
+= rdev
->data_offset
;
2107 wbio
->bi_bdev
= rdev
->bdev
;
2108 if (submit_bio_wait(WRITE
, wbio
) == 0)
2110 ok
= rdev_set_badblocks(rdev
, sector
,
2115 sect_to_write
-= sectors
;
2117 sectors
= block_sectors
;
2122 static void handle_sync_write_finished(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2125 int s
= r1_bio
->sectors
;
2126 for (m
= 0; m
< conf
->raid_disks
* 2 ; m
++) {
2127 struct md_rdev
*rdev
= conf
->mirrors
[m
].rdev
;
2128 struct bio
*bio
= r1_bio
->bios
[m
];
2129 if (bio
->bi_end_io
== NULL
)
2131 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
2132 test_bit(R1BIO_MadeGood
, &r1_bio
->state
)) {
2133 rdev_clear_badblocks(rdev
, r1_bio
->sector
, s
, 0);
2135 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
2136 test_bit(R1BIO_WriteError
, &r1_bio
->state
)) {
2137 if (!rdev_set_badblocks(rdev
, r1_bio
->sector
, s
, 0))
2138 md_error(conf
->mddev
, rdev
);
2142 md_done_sync(conf
->mddev
, s
, 1);
2145 static void handle_write_finished(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2148 for (m
= 0; m
< conf
->raid_disks
* 2 ; m
++)
2149 if (r1_bio
->bios
[m
] == IO_MADE_GOOD
) {
2150 struct md_rdev
*rdev
= conf
->mirrors
[m
].rdev
;
2151 rdev_clear_badblocks(rdev
,
2153 r1_bio
->sectors
, 0);
2154 rdev_dec_pending(rdev
, conf
->mddev
);
2155 } else if (r1_bio
->bios
[m
] != NULL
) {
2156 /* This drive got a write error. We need to
2157 * narrow down and record precise write
2160 if (!narrow_write_error(r1_bio
, m
)) {
2161 md_error(conf
->mddev
,
2162 conf
->mirrors
[m
].rdev
);
2163 /* an I/O failed, we can't clear the bitmap */
2164 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
2166 rdev_dec_pending(conf
->mirrors
[m
].rdev
,
2169 if (test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2170 close_write(r1_bio
);
2171 raid_end_bio_io(r1_bio
);
2174 static void handle_read_error(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2178 struct mddev
*mddev
= conf
->mddev
;
2180 char b
[BDEVNAME_SIZE
];
2181 struct md_rdev
*rdev
;
2183 clear_bit(R1BIO_ReadError
, &r1_bio
->state
);
2184 /* we got a read error. Maybe the drive is bad. Maybe just
2185 * the block and we can fix it.
2186 * We freeze all other IO, and try reading the block from
2187 * other devices. When we find one, we re-write
2188 * and check it that fixes the read error.
2189 * This is all done synchronously while the array is
2192 if (mddev
->ro
== 0) {
2194 fix_read_error(conf
, r1_bio
->read_disk
,
2195 r1_bio
->sector
, r1_bio
->sectors
);
2196 unfreeze_array(conf
);
2198 md_error(mddev
, conf
->mirrors
[r1_bio
->read_disk
].rdev
);
2199 rdev_dec_pending(conf
->mirrors
[r1_bio
->read_disk
].rdev
, conf
->mddev
);
2201 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
2202 bdevname(bio
->bi_bdev
, b
);
2204 disk
= read_balance(conf
, r1_bio
, &max_sectors
);
2206 printk(KERN_ALERT
"md/raid1:%s: %s: unrecoverable I/O"
2207 " read error for block %llu\n",
2208 mdname(mddev
), b
, (unsigned long long)r1_bio
->sector
);
2209 raid_end_bio_io(r1_bio
);
2211 const unsigned long do_sync
2212 = r1_bio
->master_bio
->bi_rw
& REQ_SYNC
;
2214 r1_bio
->bios
[r1_bio
->read_disk
] =
2215 mddev
->ro
? IO_BLOCKED
: NULL
;
2218 r1_bio
->read_disk
= disk
;
2219 bio
= bio_clone_mddev(r1_bio
->master_bio
, GFP_NOIO
, mddev
);
2220 md_trim_bio(bio
, r1_bio
->sector
- bio
->bi_sector
, max_sectors
);
2221 r1_bio
->bios
[r1_bio
->read_disk
] = bio
;
2222 rdev
= conf
->mirrors
[disk
].rdev
;
2223 printk_ratelimited(KERN_ERR
2224 "md/raid1:%s: redirecting sector %llu"
2225 " to other mirror: %s\n",
2227 (unsigned long long)r1_bio
->sector
,
2228 bdevname(rdev
->bdev
, b
));
2229 bio
->bi_sector
= r1_bio
->sector
+ rdev
->data_offset
;
2230 bio
->bi_bdev
= rdev
->bdev
;
2231 bio
->bi_end_io
= raid1_end_read_request
;
2232 bio
->bi_rw
= READ
| do_sync
;
2233 bio
->bi_private
= r1_bio
;
2234 if (max_sectors
< r1_bio
->sectors
) {
2235 /* Drat - have to split this up more */
2236 struct bio
*mbio
= r1_bio
->master_bio
;
2237 int sectors_handled
= (r1_bio
->sector
+ max_sectors
2239 r1_bio
->sectors
= max_sectors
;
2240 spin_lock_irq(&conf
->device_lock
);
2241 if (mbio
->bi_phys_segments
== 0)
2242 mbio
->bi_phys_segments
= 2;
2244 mbio
->bi_phys_segments
++;
2245 spin_unlock_irq(&conf
->device_lock
);
2246 generic_make_request(bio
);
2249 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
2251 r1_bio
->master_bio
= mbio
;
2252 r1_bio
->sectors
= bio_sectors(mbio
) - sectors_handled
;
2254 set_bit(R1BIO_ReadError
, &r1_bio
->state
);
2255 r1_bio
->mddev
= mddev
;
2256 r1_bio
->sector
= mbio
->bi_sector
+ sectors_handled
;
2260 generic_make_request(bio
);
2264 static void raid1d(struct md_thread
*thread
)
2266 struct mddev
*mddev
= thread
->mddev
;
2267 struct r1bio
*r1_bio
;
2268 unsigned long flags
;
2269 struct r1conf
*conf
= mddev
->private;
2270 struct list_head
*head
= &conf
->retry_list
;
2271 struct blk_plug plug
;
2273 md_check_recovery(mddev
);
2275 blk_start_plug(&plug
);
2278 flush_pending_writes(conf
);
2280 spin_lock_irqsave(&conf
->device_lock
, flags
);
2281 if (list_empty(head
)) {
2282 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2285 r1_bio
= list_entry(head
->prev
, struct r1bio
, retry_list
);
2286 list_del(head
->prev
);
2288 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2290 mddev
= r1_bio
->mddev
;
2291 conf
= mddev
->private;
2292 if (test_bit(R1BIO_IsSync
, &r1_bio
->state
)) {
2293 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
2294 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2295 handle_sync_write_finished(conf
, r1_bio
);
2297 sync_request_write(mddev
, r1_bio
);
2298 } else if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
2299 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2300 handle_write_finished(conf
, r1_bio
);
2301 else if (test_bit(R1BIO_ReadError
, &r1_bio
->state
))
2302 handle_read_error(conf
, r1_bio
);
2304 /* just a partial read to be scheduled from separate
2307 generic_make_request(r1_bio
->bios
[r1_bio
->read_disk
]);
2310 if (mddev
->flags
& ~(1<<MD_CHANGE_PENDING
))
2311 md_check_recovery(mddev
);
2313 blk_finish_plug(&plug
);
2317 static int init_resync(struct r1conf
*conf
)
2321 buffs
= RESYNC_WINDOW
/ RESYNC_BLOCK_SIZE
;
2322 BUG_ON(conf
->r1buf_pool
);
2323 conf
->r1buf_pool
= mempool_create(buffs
, r1buf_pool_alloc
, r1buf_pool_free
,
2325 if (!conf
->r1buf_pool
)
2327 conf
->next_resync
= 0;
2332 * perform a "sync" on one "block"
2334 * We need to make sure that no normal I/O request - particularly write
2335 * requests - conflict with active sync requests.
2337 * This is achieved by tracking pending requests and a 'barrier' concept
2338 * that can be installed to exclude normal IO requests.
2341 static sector_t
sync_request(struct mddev
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
2343 struct r1conf
*conf
= mddev
->private;
2344 struct r1bio
*r1_bio
;
2346 sector_t max_sector
, nr_sectors
;
2350 int write_targets
= 0, read_targets
= 0;
2351 sector_t sync_blocks
;
2352 int still_degraded
= 0;
2353 int good_sectors
= RESYNC_SECTORS
;
2354 int min_bad
= 0; /* number of sectors that are bad in all devices */
2356 if (!conf
->r1buf_pool
)
2357 if (init_resync(conf
))
2360 max_sector
= mddev
->dev_sectors
;
2361 if (sector_nr
>= max_sector
) {
2362 /* If we aborted, we need to abort the
2363 * sync on the 'current' bitmap chunk (there will
2364 * only be one in raid1 resync.
2365 * We can find the current addess in mddev->curr_resync
2367 if (mddev
->curr_resync
< max_sector
) /* aborted */
2368 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
2370 else /* completed sync */
2373 bitmap_close_sync(mddev
->bitmap
);
2378 if (mddev
->bitmap
== NULL
&&
2379 mddev
->recovery_cp
== MaxSector
&&
2380 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
2381 conf
->fullsync
== 0) {
2383 return max_sector
- sector_nr
;
2385 /* before building a request, check if we can skip these blocks..
2386 * This call the bitmap_start_sync doesn't actually record anything
2388 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
2389 !conf
->fullsync
&& !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
2390 /* We can skip this block, and probably several more */
2395 * If there is non-resync activity waiting for a turn,
2396 * and resync is going fast enough,
2397 * then let it though before starting on this new sync request.
2399 if (!go_faster
&& conf
->nr_waiting
)
2400 msleep_interruptible(1000);
2402 bitmap_cond_end_sync(mddev
->bitmap
, sector_nr
);
2403 r1_bio
= mempool_alloc(conf
->r1buf_pool
, GFP_NOIO
);
2404 raise_barrier(conf
);
2406 conf
->next_resync
= sector_nr
;
2410 * If we get a correctably read error during resync or recovery,
2411 * we might want to read from a different device. So we
2412 * flag all drives that could conceivably be read from for READ,
2413 * and any others (which will be non-In_sync devices) for WRITE.
2414 * If a read fails, we try reading from something else for which READ
2418 r1_bio
->mddev
= mddev
;
2419 r1_bio
->sector
= sector_nr
;
2421 set_bit(R1BIO_IsSync
, &r1_bio
->state
);
2423 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
2424 struct md_rdev
*rdev
;
2425 bio
= r1_bio
->bios
[i
];
2428 rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
2430 test_bit(Faulty
, &rdev
->flags
)) {
2431 if (i
< conf
->raid_disks
)
2433 } else if (!test_bit(In_sync
, &rdev
->flags
)) {
2435 bio
->bi_end_io
= end_sync_write
;
2438 /* may need to read from here */
2439 sector_t first_bad
= MaxSector
;
2442 if (is_badblock(rdev
, sector_nr
, good_sectors
,
2443 &first_bad
, &bad_sectors
)) {
2444 if (first_bad
> sector_nr
)
2445 good_sectors
= first_bad
- sector_nr
;
2447 bad_sectors
-= (sector_nr
- first_bad
);
2449 min_bad
> bad_sectors
)
2450 min_bad
= bad_sectors
;
2453 if (sector_nr
< first_bad
) {
2454 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2462 bio
->bi_end_io
= end_sync_read
;
2464 } else if (!test_bit(WriteErrorSeen
, &rdev
->flags
) &&
2465 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
2466 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
2468 * The device is suitable for reading (InSync),
2469 * but has bad block(s) here. Let's try to correct them,
2470 * if we are doing resync or repair. Otherwise, leave
2471 * this device alone for this sync request.
2474 bio
->bi_end_io
= end_sync_write
;
2478 if (bio
->bi_end_io
) {
2479 atomic_inc(&rdev
->nr_pending
);
2480 bio
->bi_sector
= sector_nr
+ rdev
->data_offset
;
2481 bio
->bi_bdev
= rdev
->bdev
;
2482 bio
->bi_private
= r1_bio
;
2488 r1_bio
->read_disk
= disk
;
2490 if (read_targets
== 0 && min_bad
> 0) {
2491 /* These sectors are bad on all InSync devices, so we
2492 * need to mark them bad on all write targets
2495 for (i
= 0 ; i
< conf
->raid_disks
* 2 ; i
++)
2496 if (r1_bio
->bios
[i
]->bi_end_io
== end_sync_write
) {
2497 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
2498 ok
= rdev_set_badblocks(rdev
, sector_nr
,
2502 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2507 /* Cannot record the badblocks, so need to
2509 * If there are multiple read targets, could just
2510 * fail the really bad ones ???
2512 conf
->recovery_disabled
= mddev
->recovery_disabled
;
2513 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2519 if (min_bad
> 0 && min_bad
< good_sectors
) {
2520 /* only resync enough to reach the next bad->good
2522 good_sectors
= min_bad
;
2525 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && read_targets
> 0)
2526 /* extra read targets are also write targets */
2527 write_targets
+= read_targets
-1;
2529 if (write_targets
== 0 || read_targets
== 0) {
2530 /* There is nowhere to write, so all non-sync
2531 * drives must be failed - so we are finished
2535 max_sector
= sector_nr
+ min_bad
;
2536 rv
= max_sector
- sector_nr
;
2542 if (max_sector
> mddev
->resync_max
)
2543 max_sector
= mddev
->resync_max
; /* Don't do IO beyond here */
2544 if (max_sector
> sector_nr
+ good_sectors
)
2545 max_sector
= sector_nr
+ good_sectors
;
2550 int len
= PAGE_SIZE
;
2551 if (sector_nr
+ (len
>>9) > max_sector
)
2552 len
= (max_sector
- sector_nr
) << 9;
2555 if (sync_blocks
== 0) {
2556 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
,
2557 &sync_blocks
, still_degraded
) &&
2559 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2561 BUG_ON(sync_blocks
< (PAGE_SIZE
>>9));
2562 if ((len
>> 9) > sync_blocks
)
2563 len
= sync_blocks
<<9;
2566 for (i
= 0 ; i
< conf
->raid_disks
* 2; i
++) {
2567 bio
= r1_bio
->bios
[i
];
2568 if (bio
->bi_end_io
) {
2569 page
= bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
;
2570 if (bio_add_page(bio
, page
, len
, 0) == 0) {
2572 bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
= page
;
2575 bio
= r1_bio
->bios
[i
];
2576 if (bio
->bi_end_io
==NULL
)
2578 /* remove last page from this bio */
2580 bio
->bi_size
-= len
;
2581 bio
->bi_flags
&= ~(1<< BIO_SEG_VALID
);
2587 nr_sectors
+= len
>>9;
2588 sector_nr
+= len
>>9;
2589 sync_blocks
-= (len
>>9);
2590 } while (r1_bio
->bios
[disk
]->bi_vcnt
< RESYNC_PAGES
);
2592 r1_bio
->sectors
= nr_sectors
;
2594 /* For a user-requested sync, we read all readable devices and do a
2597 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
2598 atomic_set(&r1_bio
->remaining
, read_targets
);
2599 for (i
= 0; i
< conf
->raid_disks
* 2 && read_targets
; i
++) {
2600 bio
= r1_bio
->bios
[i
];
2601 if (bio
->bi_end_io
== end_sync_read
) {
2603 md_sync_acct(bio
->bi_bdev
, nr_sectors
);
2604 generic_make_request(bio
);
2608 atomic_set(&r1_bio
->remaining
, 1);
2609 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
2610 md_sync_acct(bio
->bi_bdev
, nr_sectors
);
2611 generic_make_request(bio
);
2617 static sector_t
raid1_size(struct mddev
*mddev
, sector_t sectors
, int raid_disks
)
2622 return mddev
->dev_sectors
;
2625 static struct r1conf
*setup_conf(struct mddev
*mddev
)
2627 struct r1conf
*conf
;
2629 struct raid1_info
*disk
;
2630 struct md_rdev
*rdev
;
2633 conf
= kzalloc(sizeof(struct r1conf
), GFP_KERNEL
);
2637 conf
->mirrors
= kzalloc(sizeof(struct raid1_info
)
2638 * mddev
->raid_disks
* 2,
2643 conf
->tmppage
= alloc_page(GFP_KERNEL
);
2647 conf
->poolinfo
= kzalloc(sizeof(*conf
->poolinfo
), GFP_KERNEL
);
2648 if (!conf
->poolinfo
)
2650 conf
->poolinfo
->raid_disks
= mddev
->raid_disks
* 2;
2651 conf
->r1bio_pool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
2654 if (!conf
->r1bio_pool
)
2657 conf
->poolinfo
->mddev
= mddev
;
2660 spin_lock_init(&conf
->device_lock
);
2661 rdev_for_each(rdev
, mddev
) {
2662 struct request_queue
*q
;
2663 int disk_idx
= rdev
->raid_disk
;
2664 if (disk_idx
>= mddev
->raid_disks
2667 if (test_bit(Replacement
, &rdev
->flags
))
2668 disk
= conf
->mirrors
+ mddev
->raid_disks
+ disk_idx
;
2670 disk
= conf
->mirrors
+ disk_idx
;
2675 q
= bdev_get_queue(rdev
->bdev
);
2676 if (q
->merge_bvec_fn
)
2677 mddev
->merge_check_needed
= 1;
2679 disk
->head_position
= 0;
2680 disk
->seq_start
= MaxSector
;
2682 conf
->raid_disks
= mddev
->raid_disks
;
2683 conf
->mddev
= mddev
;
2684 INIT_LIST_HEAD(&conf
->retry_list
);
2686 spin_lock_init(&conf
->resync_lock
);
2687 init_waitqueue_head(&conf
->wait_barrier
);
2689 bio_list_init(&conf
->pending_bio_list
);
2690 conf
->pending_count
= 0;
2691 conf
->recovery_disabled
= mddev
->recovery_disabled
- 1;
2694 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
2696 disk
= conf
->mirrors
+ i
;
2698 if (i
< conf
->raid_disks
&&
2699 disk
[conf
->raid_disks
].rdev
) {
2700 /* This slot has a replacement. */
2702 /* No original, just make the replacement
2703 * a recovering spare
2706 disk
[conf
->raid_disks
].rdev
;
2707 disk
[conf
->raid_disks
].rdev
= NULL
;
2708 } else if (!test_bit(In_sync
, &disk
->rdev
->flags
))
2709 /* Original is not in_sync - bad */
2714 !test_bit(In_sync
, &disk
->rdev
->flags
)) {
2715 disk
->head_position
= 0;
2717 (disk
->rdev
->saved_raid_disk
< 0))
2723 conf
->thread
= md_register_thread(raid1d
, mddev
, "raid1");
2724 if (!conf
->thread
) {
2726 "md/raid1:%s: couldn't allocate thread\n",
2735 if (conf
->r1bio_pool
)
2736 mempool_destroy(conf
->r1bio_pool
);
2737 kfree(conf
->mirrors
);
2738 safe_put_page(conf
->tmppage
);
2739 kfree(conf
->poolinfo
);
2742 return ERR_PTR(err
);
2745 static int stop(struct mddev
*mddev
);
2746 static int run(struct mddev
*mddev
)
2748 struct r1conf
*conf
;
2750 struct md_rdev
*rdev
;
2752 bool discard_supported
= false;
2754 if (mddev
->level
!= 1) {
2755 printk(KERN_ERR
"md/raid1:%s: raid level not set to mirroring (%d)\n",
2756 mdname(mddev
), mddev
->level
);
2759 if (mddev
->reshape_position
!= MaxSector
) {
2760 printk(KERN_ERR
"md/raid1:%s: reshape_position set but not supported\n",
2765 * copy the already verified devices into our private RAID1
2766 * bookkeeping area. [whatever we allocate in run(),
2767 * should be freed in stop()]
2769 if (mddev
->private == NULL
)
2770 conf
= setup_conf(mddev
);
2772 conf
= mddev
->private;
2775 return PTR_ERR(conf
);
2778 blk_queue_max_write_same_sectors(mddev
->queue
,
2779 mddev
->chunk_sectors
);
2780 rdev_for_each(rdev
, mddev
) {
2781 if (!mddev
->gendisk
)
2783 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
2784 rdev
->data_offset
<< 9);
2785 if (blk_queue_discard(bdev_get_queue(rdev
->bdev
)))
2786 discard_supported
= true;
2789 mddev
->degraded
= 0;
2790 for (i
=0; i
< conf
->raid_disks
; i
++)
2791 if (conf
->mirrors
[i
].rdev
== NULL
||
2792 !test_bit(In_sync
, &conf
->mirrors
[i
].rdev
->flags
) ||
2793 test_bit(Faulty
, &conf
->mirrors
[i
].rdev
->flags
))
2796 if (conf
->raid_disks
- mddev
->degraded
== 1)
2797 mddev
->recovery_cp
= MaxSector
;
2799 if (mddev
->recovery_cp
!= MaxSector
)
2800 printk(KERN_NOTICE
"md/raid1:%s: not clean"
2801 " -- starting background reconstruction\n",
2804 "md/raid1:%s: active with %d out of %d mirrors\n",
2805 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
2809 * Ok, everything is just fine now
2811 mddev
->thread
= conf
->thread
;
2812 conf
->thread
= NULL
;
2813 mddev
->private = conf
;
2815 md_set_array_sectors(mddev
, raid1_size(mddev
, 0, 0));
2818 mddev
->queue
->backing_dev_info
.congested_fn
= raid1_congested
;
2819 mddev
->queue
->backing_dev_info
.congested_data
= mddev
;
2820 blk_queue_merge_bvec(mddev
->queue
, raid1_mergeable_bvec
);
2822 if (discard_supported
)
2823 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
,
2826 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
,
2830 ret
= md_integrity_register(mddev
);
2836 static int stop(struct mddev
*mddev
)
2838 struct r1conf
*conf
= mddev
->private;
2839 struct bitmap
*bitmap
= mddev
->bitmap
;
2841 /* wait for behind writes to complete */
2842 if (bitmap
&& atomic_read(&bitmap
->behind_writes
) > 0) {
2843 printk(KERN_INFO
"md/raid1:%s: behind writes in progress - waiting to stop.\n",
2845 /* need to kick something here to make sure I/O goes? */
2846 wait_event(bitmap
->behind_wait
,
2847 atomic_read(&bitmap
->behind_writes
) == 0);
2850 raise_barrier(conf
);
2851 lower_barrier(conf
);
2853 md_unregister_thread(&mddev
->thread
);
2854 if (conf
->r1bio_pool
)
2855 mempool_destroy(conf
->r1bio_pool
);
2856 kfree(conf
->mirrors
);
2857 kfree(conf
->poolinfo
);
2859 mddev
->private = NULL
;
2863 static int raid1_resize(struct mddev
*mddev
, sector_t sectors
)
2865 /* no resync is happening, and there is enough space
2866 * on all devices, so we can resize.
2867 * We need to make sure resync covers any new space.
2868 * If the array is shrinking we should possibly wait until
2869 * any io in the removed space completes, but it hardly seems
2872 sector_t newsize
= raid1_size(mddev
, sectors
, 0);
2873 if (mddev
->external_size
&&
2874 mddev
->array_sectors
> newsize
)
2876 if (mddev
->bitmap
) {
2877 int ret
= bitmap_resize(mddev
->bitmap
, newsize
, 0, 0);
2881 md_set_array_sectors(mddev
, newsize
);
2882 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
2883 revalidate_disk(mddev
->gendisk
);
2884 if (sectors
> mddev
->dev_sectors
&&
2885 mddev
->recovery_cp
> mddev
->dev_sectors
) {
2886 mddev
->recovery_cp
= mddev
->dev_sectors
;
2887 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2889 mddev
->dev_sectors
= sectors
;
2890 mddev
->resync_max_sectors
= sectors
;
2894 static int raid1_reshape(struct mddev
*mddev
)
2897 * 1/ resize the r1bio_pool
2898 * 2/ resize conf->mirrors
2900 * We allocate a new r1bio_pool if we can.
2901 * Then raise a device barrier and wait until all IO stops.
2902 * Then resize conf->mirrors and swap in the new r1bio pool.
2904 * At the same time, we "pack" the devices so that all the missing
2905 * devices have the higher raid_disk numbers.
2907 mempool_t
*newpool
, *oldpool
;
2908 struct pool_info
*newpoolinfo
;
2909 struct raid1_info
*newmirrors
;
2910 struct r1conf
*conf
= mddev
->private;
2911 int cnt
, raid_disks
;
2912 unsigned long flags
;
2915 /* Cannot change chunk_size, layout, or level */
2916 if (mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
||
2917 mddev
->layout
!= mddev
->new_layout
||
2918 mddev
->level
!= mddev
->new_level
) {
2919 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
2920 mddev
->new_layout
= mddev
->layout
;
2921 mddev
->new_level
= mddev
->level
;
2925 err
= md_allow_write(mddev
);
2929 raid_disks
= mddev
->raid_disks
+ mddev
->delta_disks
;
2931 if (raid_disks
< conf
->raid_disks
) {
2933 for (d
= 0; d
< conf
->raid_disks
; d
++)
2934 if (conf
->mirrors
[d
].rdev
)
2936 if (cnt
> raid_disks
)
2940 newpoolinfo
= kmalloc(sizeof(*newpoolinfo
), GFP_KERNEL
);
2943 newpoolinfo
->mddev
= mddev
;
2944 newpoolinfo
->raid_disks
= raid_disks
* 2;
2946 newpool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
2947 r1bio_pool_free
, newpoolinfo
);
2952 newmirrors
= kzalloc(sizeof(struct raid1_info
) * raid_disks
* 2,
2956 mempool_destroy(newpool
);
2960 raise_barrier(conf
);
2962 /* ok, everything is stopped */
2963 oldpool
= conf
->r1bio_pool
;
2964 conf
->r1bio_pool
= newpool
;
2966 for (d
= d2
= 0; d
< conf
->raid_disks
; d
++) {
2967 struct md_rdev
*rdev
= conf
->mirrors
[d
].rdev
;
2968 if (rdev
&& rdev
->raid_disk
!= d2
) {
2969 sysfs_unlink_rdev(mddev
, rdev
);
2970 rdev
->raid_disk
= d2
;
2971 sysfs_unlink_rdev(mddev
, rdev
);
2972 if (sysfs_link_rdev(mddev
, rdev
))
2974 "md/raid1:%s: cannot register rd%d\n",
2975 mdname(mddev
), rdev
->raid_disk
);
2978 newmirrors
[d2
++].rdev
= rdev
;
2980 kfree(conf
->mirrors
);
2981 conf
->mirrors
= newmirrors
;
2982 kfree(conf
->poolinfo
);
2983 conf
->poolinfo
= newpoolinfo
;
2985 spin_lock_irqsave(&conf
->device_lock
, flags
);
2986 mddev
->degraded
+= (raid_disks
- conf
->raid_disks
);
2987 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2988 conf
->raid_disks
= mddev
->raid_disks
= raid_disks
;
2989 mddev
->delta_disks
= 0;
2991 lower_barrier(conf
);
2993 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
2994 md_wakeup_thread(mddev
->thread
);
2996 mempool_destroy(oldpool
);
3000 static void raid1_quiesce(struct mddev
*mddev
, int state
)
3002 struct r1conf
*conf
= mddev
->private;
3005 case 2: /* wake for suspend */
3006 wake_up(&conf
->wait_barrier
);
3009 raise_barrier(conf
);
3012 lower_barrier(conf
);
3017 static void *raid1_takeover(struct mddev
*mddev
)
3019 /* raid1 can take over:
3020 * raid5 with 2 devices, any layout or chunk size
3022 if (mddev
->level
== 5 && mddev
->raid_disks
== 2) {
3023 struct r1conf
*conf
;
3024 mddev
->new_level
= 1;
3025 mddev
->new_layout
= 0;
3026 mddev
->new_chunk_sectors
= 0;
3027 conf
= setup_conf(mddev
);
3032 return ERR_PTR(-EINVAL
);
3035 static struct md_personality raid1_personality
=
3039 .owner
= THIS_MODULE
,
3040 .make_request
= make_request
,
3044 .error_handler
= error
,
3045 .hot_add_disk
= raid1_add_disk
,
3046 .hot_remove_disk
= raid1_remove_disk
,
3047 .spare_active
= raid1_spare_active
,
3048 .sync_request
= sync_request
,
3049 .resize
= raid1_resize
,
3051 .check_reshape
= raid1_reshape
,
3052 .quiesce
= raid1_quiesce
,
3053 .takeover
= raid1_takeover
,
3056 static int __init
raid_init(void)
3058 return register_md_personality(&raid1_personality
);
3061 static void raid_exit(void)
3063 unregister_md_personality(&raid1_personality
);
3066 module_init(raid_init
);
3067 module_exit(raid_exit
);
3068 MODULE_LICENSE("GPL");
3069 MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3070 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3071 MODULE_ALIAS("md-raid1");
3072 MODULE_ALIAS("md-level-1");
3074 module_param(max_queued_requests
, int, S_IRUGO
|S_IWUSR
);