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
, sector_t start_next_window
,
71 static void lower_barrier(struct r1conf
*conf
);
73 static void * r1bio_pool_alloc(gfp_t gfp_flags
, void *data
)
75 struct pool_info
*pi
= data
;
76 int size
= offsetof(struct r1bio
, bios
[pi
->raid_disks
]);
78 /* allocate a r1bio with room for raid_disks entries in the bios array */
79 return kzalloc(size
, gfp_flags
);
82 static void r1bio_pool_free(void *r1_bio
, void *data
)
87 #define RESYNC_BLOCK_SIZE (64*1024)
88 #define RESYNC_DEPTH 32
89 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
90 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
91 #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
92 #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
93 #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
95 static void * r1buf_pool_alloc(gfp_t gfp_flags
, void *data
)
97 struct pool_info
*pi
= data
;
103 r1_bio
= r1bio_pool_alloc(gfp_flags
, pi
);
108 * Allocate bios : 1 for reading, n-1 for writing
110 for (j
= pi
->raid_disks
; j
-- ; ) {
111 bio
= bio_kmalloc(gfp_flags
, RESYNC_PAGES
);
114 r1_bio
->bios
[j
] = bio
;
117 * Allocate RESYNC_PAGES data pages and attach them to
119 * If this is a user-requested check/repair, allocate
120 * RESYNC_PAGES for each bio.
122 if (test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
))
123 need_pages
= pi
->raid_disks
;
126 for (j
= 0; j
< need_pages
; j
++) {
127 bio
= r1_bio
->bios
[j
];
128 bio
->bi_vcnt
= RESYNC_PAGES
;
130 if (bio_alloc_pages(bio
, gfp_flags
))
133 /* If not user-requests, copy the page pointers to all bios */
134 if (!test_bit(MD_RECOVERY_REQUESTED
, &pi
->mddev
->recovery
)) {
135 for (i
=0; i
<RESYNC_PAGES
; i
++)
136 for (j
=1; j
<pi
->raid_disks
; j
++)
137 r1_bio
->bios
[j
]->bi_io_vec
[i
].bv_page
=
138 r1_bio
->bios
[0]->bi_io_vec
[i
].bv_page
;
141 r1_bio
->master_bio
= NULL
;
149 bio_for_each_segment_all(bv
, r1_bio
->bios
[j
], i
)
150 __free_page(bv
->bv_page
);
154 while (++j
< pi
->raid_disks
)
155 bio_put(r1_bio
->bios
[j
]);
156 r1bio_pool_free(r1_bio
, data
);
160 static void r1buf_pool_free(void *__r1_bio
, void *data
)
162 struct pool_info
*pi
= data
;
164 struct r1bio
*r1bio
= __r1_bio
;
166 for (i
= 0; i
< RESYNC_PAGES
; i
++)
167 for (j
= pi
->raid_disks
; j
-- ;) {
169 r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
!=
170 r1bio
->bios
[0]->bi_io_vec
[i
].bv_page
)
171 safe_put_page(r1bio
->bios
[j
]->bi_io_vec
[i
].bv_page
);
173 for (i
=0 ; i
< pi
->raid_disks
; i
++)
174 bio_put(r1bio
->bios
[i
]);
176 r1bio_pool_free(r1bio
, data
);
179 static void put_all_bios(struct r1conf
*conf
, struct r1bio
*r1_bio
)
183 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
184 struct bio
**bio
= r1_bio
->bios
+ i
;
185 if (!BIO_SPECIAL(*bio
))
191 static void free_r1bio(struct r1bio
*r1_bio
)
193 struct r1conf
*conf
= r1_bio
->mddev
->private;
195 put_all_bios(conf
, r1_bio
);
196 mempool_free(r1_bio
, conf
->r1bio_pool
);
199 static void put_buf(struct r1bio
*r1_bio
)
201 struct r1conf
*conf
= r1_bio
->mddev
->private;
204 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
205 struct bio
*bio
= r1_bio
->bios
[i
];
207 rdev_dec_pending(conf
->mirrors
[i
].rdev
, r1_bio
->mddev
);
210 mempool_free(r1_bio
, conf
->r1buf_pool
);
215 static void reschedule_retry(struct r1bio
*r1_bio
)
218 struct mddev
*mddev
= r1_bio
->mddev
;
219 struct r1conf
*conf
= mddev
->private;
221 spin_lock_irqsave(&conf
->device_lock
, flags
);
222 list_add(&r1_bio
->retry_list
, &conf
->retry_list
);
224 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
226 wake_up(&conf
->wait_barrier
);
227 md_wakeup_thread(mddev
->thread
);
231 * raid_end_bio_io() is called when we have finished servicing a mirrored
232 * operation and are ready to return a success/failure code to the buffer
235 static void call_bio_endio(struct r1bio
*r1_bio
)
237 struct bio
*bio
= r1_bio
->master_bio
;
239 struct r1conf
*conf
= r1_bio
->mddev
->private;
240 sector_t start_next_window
= r1_bio
->start_next_window
;
241 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
243 if (bio
->bi_phys_segments
) {
245 spin_lock_irqsave(&conf
->device_lock
, flags
);
246 bio
->bi_phys_segments
--;
247 done
= (bio
->bi_phys_segments
== 0);
248 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
250 * make_request() might be waiting for
251 * bi_phys_segments to decrease
253 wake_up(&conf
->wait_barrier
);
257 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
))
258 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
262 * Wake up any possible resync thread that waits for the device
265 allow_barrier(conf
, start_next_window
, bi_sector
);
269 static void raid_end_bio_io(struct r1bio
*r1_bio
)
271 struct bio
*bio
= r1_bio
->master_bio
;
273 /* if nobody has done the final endio yet, do it now */
274 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
275 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
276 (bio_data_dir(bio
) == WRITE
) ? "write" : "read",
277 (unsigned long long) bio
->bi_iter
.bi_sector
,
278 (unsigned long long) bio_end_sector(bio
) - 1);
280 call_bio_endio(r1_bio
);
286 * Update disk head position estimator based on IRQ completion info.
288 static inline void update_head_pos(int disk
, struct r1bio
*r1_bio
)
290 struct r1conf
*conf
= r1_bio
->mddev
->private;
292 conf
->mirrors
[disk
].head_position
=
293 r1_bio
->sector
+ (r1_bio
->sectors
);
297 * Find the disk number which triggered given bio
299 static int find_bio_disk(struct r1bio
*r1_bio
, struct bio
*bio
)
302 struct r1conf
*conf
= r1_bio
->mddev
->private;
303 int raid_disks
= conf
->raid_disks
;
305 for (mirror
= 0; mirror
< raid_disks
* 2; mirror
++)
306 if (r1_bio
->bios
[mirror
] == bio
)
309 BUG_ON(mirror
== raid_disks
* 2);
310 update_head_pos(mirror
, r1_bio
);
315 static void raid1_end_read_request(struct bio
*bio
, int error
)
317 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
318 struct r1bio
*r1_bio
= bio
->bi_private
;
320 struct r1conf
*conf
= r1_bio
->mddev
->private;
322 mirror
= r1_bio
->read_disk
;
324 * this branch is our 'one mirror IO has finished' event handler:
326 update_head_pos(mirror
, r1_bio
);
329 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
331 /* If all other devices have failed, we want to return
332 * the error upwards rather than fail the last device.
333 * Here we redefine "uptodate" to mean "Don't want to retry"
336 spin_lock_irqsave(&conf
->device_lock
, flags
);
337 if (r1_bio
->mddev
->degraded
== conf
->raid_disks
||
338 (r1_bio
->mddev
->degraded
== conf
->raid_disks
-1 &&
339 !test_bit(Faulty
, &conf
->mirrors
[mirror
].rdev
->flags
)))
341 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
345 raid_end_bio_io(r1_bio
);
346 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
, conf
->mddev
);
351 char b
[BDEVNAME_SIZE
];
353 KERN_ERR
"md/raid1:%s: %s: "
354 "rescheduling sector %llu\n",
356 bdevname(conf
->mirrors
[mirror
].rdev
->bdev
,
358 (unsigned long long)r1_bio
->sector
);
359 set_bit(R1BIO_ReadError
, &r1_bio
->state
);
360 reschedule_retry(r1_bio
);
361 /* don't drop the reference on read_disk yet */
365 static void close_write(struct r1bio
*r1_bio
)
367 /* it really is the end of this request */
368 if (test_bit(R1BIO_BehindIO
, &r1_bio
->state
)) {
369 /* free extra copy of the data pages */
370 int i
= r1_bio
->behind_page_count
;
372 safe_put_page(r1_bio
->behind_bvecs
[i
].bv_page
);
373 kfree(r1_bio
->behind_bvecs
);
374 r1_bio
->behind_bvecs
= NULL
;
376 /* clear the bitmap if all writes complete successfully */
377 bitmap_endwrite(r1_bio
->mddev
->bitmap
, r1_bio
->sector
,
379 !test_bit(R1BIO_Degraded
, &r1_bio
->state
),
380 test_bit(R1BIO_BehindIO
, &r1_bio
->state
));
381 md_write_end(r1_bio
->mddev
);
384 static void r1_bio_write_done(struct r1bio
*r1_bio
)
386 if (!atomic_dec_and_test(&r1_bio
->remaining
))
389 if (test_bit(R1BIO_WriteError
, &r1_bio
->state
))
390 reschedule_retry(r1_bio
);
393 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
))
394 reschedule_retry(r1_bio
);
396 raid_end_bio_io(r1_bio
);
400 static void raid1_end_write_request(struct bio
*bio
, int error
)
402 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
403 struct r1bio
*r1_bio
= bio
->bi_private
;
404 int mirror
, behind
= test_bit(R1BIO_BehindIO
, &r1_bio
->state
);
405 struct r1conf
*conf
= r1_bio
->mddev
->private;
406 struct bio
*to_put
= NULL
;
408 mirror
= find_bio_disk(r1_bio
, bio
);
411 * 'one mirror IO has finished' event handler:
414 set_bit(WriteErrorSeen
,
415 &conf
->mirrors
[mirror
].rdev
->flags
);
416 if (!test_and_set_bit(WantReplacement
,
417 &conf
->mirrors
[mirror
].rdev
->flags
))
418 set_bit(MD_RECOVERY_NEEDED
, &
419 conf
->mddev
->recovery
);
421 set_bit(R1BIO_WriteError
, &r1_bio
->state
);
424 * Set R1BIO_Uptodate in our master bio, so that we
425 * will return a good error code for to the higher
426 * levels even if IO on some other mirrored buffer
429 * The 'master' represents the composite IO operation
430 * to user-side. So if something waits for IO, then it
431 * will wait for the 'master' bio.
436 r1_bio
->bios
[mirror
] = NULL
;
439 * Do not set R1BIO_Uptodate if the current device is
440 * rebuilding or Faulty. This is because we cannot use
441 * such device for properly reading the data back (we could
442 * potentially use it, if the current write would have felt
443 * before rdev->recovery_offset, but for simplicity we don't
446 if (test_bit(In_sync
, &conf
->mirrors
[mirror
].rdev
->flags
) &&
447 !test_bit(Faulty
, &conf
->mirrors
[mirror
].rdev
->flags
))
448 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
450 /* Maybe we can clear some bad blocks. */
451 if (is_badblock(conf
->mirrors
[mirror
].rdev
,
452 r1_bio
->sector
, r1_bio
->sectors
,
453 &first_bad
, &bad_sectors
)) {
454 r1_bio
->bios
[mirror
] = IO_MADE_GOOD
;
455 set_bit(R1BIO_MadeGood
, &r1_bio
->state
);
460 if (test_bit(WriteMostly
, &conf
->mirrors
[mirror
].rdev
->flags
))
461 atomic_dec(&r1_bio
->behind_remaining
);
464 * In behind mode, we ACK the master bio once the I/O
465 * has safely reached all non-writemostly
466 * disks. Setting the Returned bit ensures that this
467 * gets done only once -- we don't ever want to return
468 * -EIO here, instead we'll wait
470 if (atomic_read(&r1_bio
->behind_remaining
) >= (atomic_read(&r1_bio
->remaining
)-1) &&
471 test_bit(R1BIO_Uptodate
, &r1_bio
->state
)) {
472 /* Maybe we can return now */
473 if (!test_and_set_bit(R1BIO_Returned
, &r1_bio
->state
)) {
474 struct bio
*mbio
= r1_bio
->master_bio
;
475 pr_debug("raid1: behind end write sectors"
477 (unsigned long long) mbio
->bi_iter
.bi_sector
,
478 (unsigned long long) bio_end_sector(mbio
) - 1);
479 call_bio_endio(r1_bio
);
483 if (r1_bio
->bios
[mirror
] == NULL
)
484 rdev_dec_pending(conf
->mirrors
[mirror
].rdev
,
488 * Let's see if all mirrored write operations have finished
491 r1_bio_write_done(r1_bio
);
498 * This routine returns the disk from which the requested read should
499 * be done. There is a per-array 'next expected sequential IO' sector
500 * number - if this matches on the next IO then we use the last disk.
501 * There is also a per-disk 'last know head position' sector that is
502 * maintained from IRQ contexts, both the normal and the resync IO
503 * completion handlers update this position correctly. If there is no
504 * perfect sequential match then we pick the disk whose head is closest.
506 * If there are 2 mirrors in the same 2 devices, performance degrades
507 * because position is mirror, not device based.
509 * The rdev for the device selected will have nr_pending incremented.
511 static int read_balance(struct r1conf
*conf
, struct r1bio
*r1_bio
, int *max_sectors
)
513 const sector_t this_sector
= r1_bio
->sector
;
515 int best_good_sectors
;
516 int best_disk
, best_dist_disk
, best_pending_disk
;
520 unsigned int min_pending
;
521 struct md_rdev
*rdev
;
523 int choose_next_idle
;
527 * Check if we can balance. We can balance on the whole
528 * device if no resync is going on, or below the resync window.
529 * We take the first readable disk when above the resync window.
532 sectors
= r1_bio
->sectors
;
535 best_dist
= MaxSector
;
536 best_pending_disk
= -1;
537 min_pending
= UINT_MAX
;
538 best_good_sectors
= 0;
540 choose_next_idle
= 0;
542 choose_first
= (conf
->mddev
->recovery_cp
< this_sector
+ sectors
);
544 for (disk
= 0 ; disk
< conf
->raid_disks
* 2 ; disk
++) {
548 unsigned int pending
;
551 rdev
= rcu_dereference(conf
->mirrors
[disk
].rdev
);
552 if (r1_bio
->bios
[disk
] == IO_BLOCKED
554 || test_bit(Unmerged
, &rdev
->flags
)
555 || test_bit(Faulty
, &rdev
->flags
))
557 if (!test_bit(In_sync
, &rdev
->flags
) &&
558 rdev
->recovery_offset
< this_sector
+ sectors
)
560 if (test_bit(WriteMostly
, &rdev
->flags
)) {
561 /* Don't balance among write-mostly, just
562 * use the first as a last resort */
564 if (is_badblock(rdev
, this_sector
, sectors
,
565 &first_bad
, &bad_sectors
)) {
566 if (first_bad
< this_sector
)
567 /* Cannot use this */
569 best_good_sectors
= first_bad
- this_sector
;
571 best_good_sectors
= sectors
;
576 /* This is a reasonable device to use. It might
579 if (is_badblock(rdev
, this_sector
, sectors
,
580 &first_bad
, &bad_sectors
)) {
581 if (best_dist
< MaxSector
)
582 /* already have a better device */
584 if (first_bad
<= this_sector
) {
585 /* cannot read here. If this is the 'primary'
586 * device, then we must not read beyond
587 * bad_sectors from another device..
589 bad_sectors
-= (this_sector
- first_bad
);
590 if (choose_first
&& sectors
> bad_sectors
)
591 sectors
= bad_sectors
;
592 if (best_good_sectors
> sectors
)
593 best_good_sectors
= sectors
;
596 sector_t good_sectors
= first_bad
- this_sector
;
597 if (good_sectors
> best_good_sectors
) {
598 best_good_sectors
= good_sectors
;
606 best_good_sectors
= sectors
;
608 nonrot
= blk_queue_nonrot(bdev_get_queue(rdev
->bdev
));
609 has_nonrot_disk
|= nonrot
;
610 pending
= atomic_read(&rdev
->nr_pending
);
611 dist
= abs(this_sector
- conf
->mirrors
[disk
].head_position
);
616 /* Don't change to another disk for sequential reads */
617 if (conf
->mirrors
[disk
].next_seq_sect
== this_sector
619 int opt_iosize
= bdev_io_opt(rdev
->bdev
) >> 9;
620 struct raid1_info
*mirror
= &conf
->mirrors
[disk
];
624 * If buffered sequential IO size exceeds optimal
625 * iosize, check if there is idle disk. If yes, choose
626 * the idle disk. read_balance could already choose an
627 * idle disk before noticing it's a sequential IO in
628 * this disk. This doesn't matter because this disk
629 * will idle, next time it will be utilized after the
630 * first disk has IO size exceeds optimal iosize. In
631 * this way, iosize of the first disk will be optimal
632 * iosize at least. iosize of the second disk might be
633 * small, but not a big deal since when the second disk
634 * starts IO, the first disk is likely still busy.
636 if (nonrot
&& opt_iosize
> 0 &&
637 mirror
->seq_start
!= MaxSector
&&
638 mirror
->next_seq_sect
> opt_iosize
&&
639 mirror
->next_seq_sect
- opt_iosize
>=
641 choose_next_idle
= 1;
646 /* If device is idle, use it */
652 if (choose_next_idle
)
655 if (min_pending
> pending
) {
656 min_pending
= pending
;
657 best_pending_disk
= disk
;
660 if (dist
< best_dist
) {
662 best_dist_disk
= disk
;
667 * If all disks are rotational, choose the closest disk. If any disk is
668 * non-rotational, choose the disk with less pending request even the
669 * disk is rotational, which might/might not be optimal for raids with
670 * mixed ratation/non-rotational disks depending on workload.
672 if (best_disk
== -1) {
674 best_disk
= best_pending_disk
;
676 best_disk
= best_dist_disk
;
679 if (best_disk
>= 0) {
680 rdev
= rcu_dereference(conf
->mirrors
[best_disk
].rdev
);
683 atomic_inc(&rdev
->nr_pending
);
684 if (test_bit(Faulty
, &rdev
->flags
)) {
685 /* cannot risk returning a device that failed
686 * before we inc'ed nr_pending
688 rdev_dec_pending(rdev
, conf
->mddev
);
691 sectors
= best_good_sectors
;
693 if (conf
->mirrors
[best_disk
].next_seq_sect
!= this_sector
)
694 conf
->mirrors
[best_disk
].seq_start
= this_sector
;
696 conf
->mirrors
[best_disk
].next_seq_sect
= this_sector
+ sectors
;
699 *max_sectors
= sectors
;
704 static int raid1_mergeable_bvec(struct mddev
*mddev
,
705 struct bvec_merge_data
*bvm
,
706 struct bio_vec
*biovec
)
708 struct r1conf
*conf
= mddev
->private;
709 sector_t sector
= bvm
->bi_sector
+ get_start_sect(bvm
->bi_bdev
);
710 int max
= biovec
->bv_len
;
712 if (mddev
->merge_check_needed
) {
715 for (disk
= 0; disk
< conf
->raid_disks
* 2; disk
++) {
716 struct md_rdev
*rdev
= rcu_dereference(
717 conf
->mirrors
[disk
].rdev
);
718 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
719 struct request_queue
*q
=
720 bdev_get_queue(rdev
->bdev
);
721 if (q
->merge_bvec_fn
) {
722 bvm
->bi_sector
= sector
+
724 bvm
->bi_bdev
= rdev
->bdev
;
725 max
= min(max
, q
->merge_bvec_fn(
736 static int raid1_congested(struct mddev
*mddev
, int bits
)
738 struct r1conf
*conf
= mddev
->private;
741 if ((bits
& (1 << BDI_async_congested
)) &&
742 conf
->pending_count
>= max_queued_requests
)
746 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
747 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
748 if (rdev
&& !test_bit(Faulty
, &rdev
->flags
)) {
749 struct request_queue
*q
= bdev_get_queue(rdev
->bdev
);
753 /* Note the '|| 1' - when read_balance prefers
754 * non-congested targets, it can be removed
756 if ((bits
& (1<<BDI_async_congested
)) || 1)
757 ret
|= bdi_congested(&q
->backing_dev_info
, bits
);
759 ret
&= bdi_congested(&q
->backing_dev_info
, bits
);
766 static void flush_pending_writes(struct r1conf
*conf
)
768 /* Any writes that have been queued but are awaiting
769 * bitmap updates get flushed here.
771 spin_lock_irq(&conf
->device_lock
);
773 if (conf
->pending_bio_list
.head
) {
775 bio
= bio_list_get(&conf
->pending_bio_list
);
776 conf
->pending_count
= 0;
777 spin_unlock_irq(&conf
->device_lock
);
778 /* flush any pending bitmap writes to
779 * disk before proceeding w/ I/O */
780 bitmap_unplug(conf
->mddev
->bitmap
);
781 wake_up(&conf
->wait_barrier
);
783 while (bio
) { /* submit pending writes */
784 struct bio
*next
= bio
->bi_next
;
786 if (unlikely((bio
->bi_rw
& REQ_DISCARD
) &&
787 !blk_queue_discard(bdev_get_queue(bio
->bi_bdev
))))
791 generic_make_request(bio
);
795 spin_unlock_irq(&conf
->device_lock
);
799 * Sometimes we need to suspend IO while we do something else,
800 * either some resync/recovery, or reconfigure the array.
801 * To do this we raise a 'barrier'.
802 * The 'barrier' is a counter that can be raised multiple times
803 * to count how many activities are happening which preclude
805 * We can only raise the barrier if there is no pending IO.
806 * i.e. if nr_pending == 0.
807 * We choose only to raise the barrier if no-one is waiting for the
808 * barrier to go down. This means that as soon as an IO request
809 * is ready, no other operations which require a barrier will start
810 * until the IO request has had a chance.
812 * So: regular IO calls 'wait_barrier'. When that returns there
813 * is no backgroup IO happening, It must arrange to call
814 * allow_barrier when it has finished its IO.
815 * backgroup IO calls must call raise_barrier. Once that returns
816 * there is no normal IO happeing. It must arrange to call
817 * lower_barrier when the particular background IO completes.
819 static void raise_barrier(struct r1conf
*conf
, sector_t sector_nr
)
821 spin_lock_irq(&conf
->resync_lock
);
823 /* Wait until no block IO is waiting */
824 wait_event_lock_irq(conf
->wait_barrier
, !conf
->nr_waiting
,
827 /* block any new IO from starting */
829 conf
->next_resync
= sector_nr
;
831 /* For these conditions we must wait:
832 * A: while the array is in frozen state
833 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
834 * the max count which allowed.
835 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
836 * next resync will reach to the window which normal bios are
838 * D: while there are any active requests in the current window.
840 wait_event_lock_irq(conf
->wait_barrier
,
841 !conf
->array_frozen
&&
842 conf
->barrier
< RESYNC_DEPTH
&&
843 conf
->current_window_requests
== 0 &&
844 (conf
->start_next_window
>=
845 conf
->next_resync
+ RESYNC_SECTORS
),
849 spin_unlock_irq(&conf
->resync_lock
);
852 static void lower_barrier(struct r1conf
*conf
)
855 BUG_ON(conf
->barrier
<= 0);
856 spin_lock_irqsave(&conf
->resync_lock
, flags
);
859 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
860 wake_up(&conf
->wait_barrier
);
863 static bool need_to_wait_for_sync(struct r1conf
*conf
, struct bio
*bio
)
867 if (conf
->array_frozen
|| !bio
)
869 else if (conf
->barrier
&& bio_data_dir(bio
) == WRITE
) {
870 if ((conf
->mddev
->curr_resync_completed
871 >= bio_end_sector(bio
)) ||
872 (conf
->next_resync
+ NEXT_NORMALIO_DISTANCE
873 <= bio
->bi_iter
.bi_sector
))
882 static sector_t
wait_barrier(struct r1conf
*conf
, struct bio
*bio
)
886 spin_lock_irq(&conf
->resync_lock
);
887 if (need_to_wait_for_sync(conf
, bio
)) {
889 /* Wait for the barrier to drop.
890 * However if there are already pending
891 * requests (preventing the barrier from
892 * rising completely), and the
893 * per-process bio queue isn't empty,
894 * then don't wait, as we need to empty
895 * that queue to allow conf->start_next_window
898 wait_event_lock_irq(conf
->wait_barrier
,
899 !conf
->array_frozen
&&
901 ((conf
->start_next_window
<
902 conf
->next_resync
+ RESYNC_SECTORS
) &&
904 !bio_list_empty(current
->bio_list
))),
909 if (bio
&& bio_data_dir(bio
) == WRITE
) {
910 if (bio
->bi_iter
.bi_sector
>=
911 conf
->mddev
->curr_resync_completed
) {
912 if (conf
->start_next_window
== MaxSector
)
913 conf
->start_next_window
=
915 NEXT_NORMALIO_DISTANCE
;
917 if ((conf
->start_next_window
+ NEXT_NORMALIO_DISTANCE
)
918 <= bio
->bi_iter
.bi_sector
)
919 conf
->next_window_requests
++;
921 conf
->current_window_requests
++;
922 sector
= conf
->start_next_window
;
927 spin_unlock_irq(&conf
->resync_lock
);
931 static void allow_barrier(struct r1conf
*conf
, sector_t start_next_window
,
936 spin_lock_irqsave(&conf
->resync_lock
, flags
);
938 if (start_next_window
) {
939 if (start_next_window
== conf
->start_next_window
) {
940 if (conf
->start_next_window
+ NEXT_NORMALIO_DISTANCE
942 conf
->next_window_requests
--;
944 conf
->current_window_requests
--;
946 conf
->current_window_requests
--;
948 if (!conf
->current_window_requests
) {
949 if (conf
->next_window_requests
) {
950 conf
->current_window_requests
=
951 conf
->next_window_requests
;
952 conf
->next_window_requests
= 0;
953 conf
->start_next_window
+=
954 NEXT_NORMALIO_DISTANCE
;
956 conf
->start_next_window
= MaxSector
;
959 spin_unlock_irqrestore(&conf
->resync_lock
, flags
);
960 wake_up(&conf
->wait_barrier
);
963 static void freeze_array(struct r1conf
*conf
, int extra
)
965 /* stop syncio and normal IO and wait for everything to
967 * We wait until nr_pending match nr_queued+extra
968 * This is called in the context of one normal IO request
969 * that has failed. Thus any sync request that might be pending
970 * will be blocked by nr_pending, and we need to wait for
971 * pending IO requests to complete or be queued for re-try.
972 * Thus the number queued (nr_queued) plus this request (extra)
973 * must match the number of pending IOs (nr_pending) before
976 spin_lock_irq(&conf
->resync_lock
);
977 conf
->array_frozen
= 1;
978 wait_event_lock_irq_cmd(conf
->wait_barrier
,
979 conf
->nr_pending
== conf
->nr_queued
+extra
,
981 flush_pending_writes(conf
));
982 spin_unlock_irq(&conf
->resync_lock
);
984 static void unfreeze_array(struct r1conf
*conf
)
986 /* reverse the effect of the freeze */
987 spin_lock_irq(&conf
->resync_lock
);
988 conf
->array_frozen
= 0;
989 wake_up(&conf
->wait_barrier
);
990 spin_unlock_irq(&conf
->resync_lock
);
993 /* duplicate the data pages for behind I/O
995 static void alloc_behind_pages(struct bio
*bio
, struct r1bio
*r1_bio
)
998 struct bio_vec
*bvec
;
999 struct bio_vec
*bvecs
= kzalloc(bio
->bi_vcnt
* sizeof(struct bio_vec
),
1001 if (unlikely(!bvecs
))
1004 bio_for_each_segment_all(bvec
, bio
, i
) {
1006 bvecs
[i
].bv_page
= alloc_page(GFP_NOIO
);
1007 if (unlikely(!bvecs
[i
].bv_page
))
1009 memcpy(kmap(bvecs
[i
].bv_page
) + bvec
->bv_offset
,
1010 kmap(bvec
->bv_page
) + bvec
->bv_offset
, bvec
->bv_len
);
1011 kunmap(bvecs
[i
].bv_page
);
1012 kunmap(bvec
->bv_page
);
1014 r1_bio
->behind_bvecs
= bvecs
;
1015 r1_bio
->behind_page_count
= bio
->bi_vcnt
;
1016 set_bit(R1BIO_BehindIO
, &r1_bio
->state
);
1020 for (i
= 0; i
< bio
->bi_vcnt
; i
++)
1021 if (bvecs
[i
].bv_page
)
1022 put_page(bvecs
[i
].bv_page
);
1024 pr_debug("%dB behind alloc failed, doing sync I/O\n",
1025 bio
->bi_iter
.bi_size
);
1028 struct raid1_plug_cb
{
1029 struct blk_plug_cb cb
;
1030 struct bio_list pending
;
1034 static void raid1_unplug(struct blk_plug_cb
*cb
, bool from_schedule
)
1036 struct raid1_plug_cb
*plug
= container_of(cb
, struct raid1_plug_cb
,
1038 struct mddev
*mddev
= plug
->cb
.data
;
1039 struct r1conf
*conf
= mddev
->private;
1042 if (from_schedule
|| current
->bio_list
) {
1043 spin_lock_irq(&conf
->device_lock
);
1044 bio_list_merge(&conf
->pending_bio_list
, &plug
->pending
);
1045 conf
->pending_count
+= plug
->pending_cnt
;
1046 spin_unlock_irq(&conf
->device_lock
);
1047 wake_up(&conf
->wait_barrier
);
1048 md_wakeup_thread(mddev
->thread
);
1053 /* we aren't scheduling, so we can do the write-out directly. */
1054 bio
= bio_list_get(&plug
->pending
);
1055 bitmap_unplug(mddev
->bitmap
);
1056 wake_up(&conf
->wait_barrier
);
1058 while (bio
) { /* submit pending writes */
1059 struct bio
*next
= bio
->bi_next
;
1060 bio
->bi_next
= NULL
;
1061 if (unlikely((bio
->bi_rw
& REQ_DISCARD
) &&
1062 !blk_queue_discard(bdev_get_queue(bio
->bi_bdev
))))
1063 /* Just ignore it */
1066 generic_make_request(bio
);
1072 static void make_request(struct mddev
*mddev
, struct bio
* bio
)
1074 struct r1conf
*conf
= mddev
->private;
1075 struct raid1_info
*mirror
;
1076 struct r1bio
*r1_bio
;
1077 struct bio
*read_bio
;
1079 struct bitmap
*bitmap
;
1080 unsigned long flags
;
1081 const int rw
= bio_data_dir(bio
);
1082 const unsigned long do_sync
= (bio
->bi_rw
& REQ_SYNC
);
1083 const unsigned long do_flush_fua
= (bio
->bi_rw
& (REQ_FLUSH
| REQ_FUA
));
1084 const unsigned long do_discard
= (bio
->bi_rw
1085 & (REQ_DISCARD
| REQ_SECURE
));
1086 const unsigned long do_same
= (bio
->bi_rw
& REQ_WRITE_SAME
);
1087 struct md_rdev
*blocked_rdev
;
1088 struct blk_plug_cb
*cb
;
1089 struct raid1_plug_cb
*plug
= NULL
;
1091 int sectors_handled
;
1093 sector_t start_next_window
;
1096 * Register the new request and wait if the reconstruction
1097 * thread has put up a bar for new requests.
1098 * Continue immediately if no resync is active currently.
1101 md_write_start(mddev
, bio
); /* wait on superblock update early */
1103 if (bio_data_dir(bio
) == WRITE
&&
1104 bio_end_sector(bio
) > mddev
->suspend_lo
&&
1105 bio
->bi_iter
.bi_sector
< mddev
->suspend_hi
) {
1106 /* As the suspend_* range is controlled by
1107 * userspace, we want an interruptible
1112 flush_signals(current
);
1113 prepare_to_wait(&conf
->wait_barrier
,
1114 &w
, TASK_INTERRUPTIBLE
);
1115 if (bio_end_sector(bio
) <= mddev
->suspend_lo
||
1116 bio
->bi_iter
.bi_sector
>= mddev
->suspend_hi
)
1120 finish_wait(&conf
->wait_barrier
, &w
);
1123 start_next_window
= wait_barrier(conf
, bio
);
1125 bitmap
= mddev
->bitmap
;
1128 * make_request() can abort the operation when READA is being
1129 * used and no empty request is available.
1132 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1134 r1_bio
->master_bio
= bio
;
1135 r1_bio
->sectors
= bio_sectors(bio
);
1137 r1_bio
->mddev
= mddev
;
1138 r1_bio
->sector
= bio
->bi_iter
.bi_sector
;
1140 /* We might need to issue multiple reads to different
1141 * devices if there are bad blocks around, so we keep
1142 * track of the number of reads in bio->bi_phys_segments.
1143 * If this is 0, there is only one r1_bio and no locking
1144 * will be needed when requests complete. If it is
1145 * non-zero, then it is the number of not-completed requests.
1147 bio
->bi_phys_segments
= 0;
1148 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
1152 * read balancing logic:
1157 rdisk
= read_balance(conf
, r1_bio
, &max_sectors
);
1160 /* couldn't find anywhere to read from */
1161 raid_end_bio_io(r1_bio
);
1164 mirror
= conf
->mirrors
+ rdisk
;
1166 if (test_bit(WriteMostly
, &mirror
->rdev
->flags
) &&
1168 /* Reading from a write-mostly device must
1169 * take care not to over-take any writes
1172 wait_event(bitmap
->behind_wait
,
1173 atomic_read(&bitmap
->behind_writes
) == 0);
1175 r1_bio
->read_disk
= rdisk
;
1176 r1_bio
->start_next_window
= 0;
1178 read_bio
= bio_clone_mddev(bio
, GFP_NOIO
, mddev
);
1179 bio_trim(read_bio
, r1_bio
->sector
- bio
->bi_iter
.bi_sector
,
1182 r1_bio
->bios
[rdisk
] = read_bio
;
1184 read_bio
->bi_iter
.bi_sector
= r1_bio
->sector
+
1185 mirror
->rdev
->data_offset
;
1186 read_bio
->bi_bdev
= mirror
->rdev
->bdev
;
1187 read_bio
->bi_end_io
= raid1_end_read_request
;
1188 read_bio
->bi_rw
= READ
| do_sync
;
1189 read_bio
->bi_private
= r1_bio
;
1191 if (max_sectors
< r1_bio
->sectors
) {
1192 /* could not read all from this device, so we will
1193 * need another r1_bio.
1196 sectors_handled
= (r1_bio
->sector
+ max_sectors
1197 - bio
->bi_iter
.bi_sector
);
1198 r1_bio
->sectors
= max_sectors
;
1199 spin_lock_irq(&conf
->device_lock
);
1200 if (bio
->bi_phys_segments
== 0)
1201 bio
->bi_phys_segments
= 2;
1203 bio
->bi_phys_segments
++;
1204 spin_unlock_irq(&conf
->device_lock
);
1205 /* Cannot call generic_make_request directly
1206 * as that will be queued in __make_request
1207 * and subsequent mempool_alloc might block waiting
1208 * for it. So hand bio over to raid1d.
1210 reschedule_retry(r1_bio
);
1212 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1214 r1_bio
->master_bio
= bio
;
1215 r1_bio
->sectors
= bio_sectors(bio
) - sectors_handled
;
1217 r1_bio
->mddev
= mddev
;
1218 r1_bio
->sector
= bio
->bi_iter
.bi_sector
+
1222 generic_make_request(read_bio
);
1229 if (conf
->pending_count
>= max_queued_requests
) {
1230 md_wakeup_thread(mddev
->thread
);
1231 wait_event(conf
->wait_barrier
,
1232 conf
->pending_count
< max_queued_requests
);
1234 /* first select target devices under rcu_lock and
1235 * inc refcount on their rdev. Record them by setting
1237 * If there are known/acknowledged bad blocks on any device on
1238 * which we have seen a write error, we want to avoid writing those
1240 * This potentially requires several writes to write around
1241 * the bad blocks. Each set of writes gets it's own r1bio
1242 * with a set of bios attached.
1245 disks
= conf
->raid_disks
* 2;
1247 r1_bio
->start_next_window
= start_next_window
;
1248 blocked_rdev
= NULL
;
1250 max_sectors
= r1_bio
->sectors
;
1251 for (i
= 0; i
< disks
; i
++) {
1252 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1253 if (rdev
&& unlikely(test_bit(Blocked
, &rdev
->flags
))) {
1254 atomic_inc(&rdev
->nr_pending
);
1255 blocked_rdev
= rdev
;
1258 r1_bio
->bios
[i
] = NULL
;
1259 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
)
1260 || test_bit(Unmerged
, &rdev
->flags
)) {
1261 if (i
< conf
->raid_disks
)
1262 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
1266 atomic_inc(&rdev
->nr_pending
);
1267 if (test_bit(WriteErrorSeen
, &rdev
->flags
)) {
1272 is_bad
= is_badblock(rdev
, r1_bio
->sector
,
1274 &first_bad
, &bad_sectors
);
1276 /* mustn't write here until the bad block is
1278 set_bit(BlockedBadBlocks
, &rdev
->flags
);
1279 blocked_rdev
= rdev
;
1282 if (is_bad
&& first_bad
<= r1_bio
->sector
) {
1283 /* Cannot write here at all */
1284 bad_sectors
-= (r1_bio
->sector
- first_bad
);
1285 if (bad_sectors
< max_sectors
)
1286 /* mustn't write more than bad_sectors
1287 * to other devices yet
1289 max_sectors
= bad_sectors
;
1290 rdev_dec_pending(rdev
, mddev
);
1291 /* We don't set R1BIO_Degraded as that
1292 * only applies if the disk is
1293 * missing, so it might be re-added,
1294 * and we want to know to recover this
1296 * In this case the device is here,
1297 * and the fact that this chunk is not
1298 * in-sync is recorded in the bad
1304 int good_sectors
= first_bad
- r1_bio
->sector
;
1305 if (good_sectors
< max_sectors
)
1306 max_sectors
= good_sectors
;
1309 r1_bio
->bios
[i
] = bio
;
1313 if (unlikely(blocked_rdev
)) {
1314 /* Wait for this device to become unblocked */
1316 sector_t old
= start_next_window
;
1318 for (j
= 0; j
< i
; j
++)
1319 if (r1_bio
->bios
[j
])
1320 rdev_dec_pending(conf
->mirrors
[j
].rdev
, mddev
);
1322 allow_barrier(conf
, start_next_window
, bio
->bi_iter
.bi_sector
);
1323 md_wait_for_blocked_rdev(blocked_rdev
, mddev
);
1324 start_next_window
= wait_barrier(conf
, bio
);
1326 * We must make sure the multi r1bios of bio have
1327 * the same value of bi_phys_segments
1329 if (bio
->bi_phys_segments
&& old
&&
1330 old
!= start_next_window
)
1331 /* Wait for the former r1bio(s) to complete */
1332 wait_event(conf
->wait_barrier
,
1333 bio
->bi_phys_segments
== 1);
1337 if (max_sectors
< r1_bio
->sectors
) {
1338 /* We are splitting this write into multiple parts, so
1339 * we need to prepare for allocating another r1_bio.
1341 r1_bio
->sectors
= max_sectors
;
1342 spin_lock_irq(&conf
->device_lock
);
1343 if (bio
->bi_phys_segments
== 0)
1344 bio
->bi_phys_segments
= 2;
1346 bio
->bi_phys_segments
++;
1347 spin_unlock_irq(&conf
->device_lock
);
1349 sectors_handled
= r1_bio
->sector
+ max_sectors
- bio
->bi_iter
.bi_sector
;
1351 atomic_set(&r1_bio
->remaining
, 1);
1352 atomic_set(&r1_bio
->behind_remaining
, 0);
1355 for (i
= 0; i
< disks
; i
++) {
1357 if (!r1_bio
->bios
[i
])
1360 mbio
= bio_clone_mddev(bio
, GFP_NOIO
, mddev
);
1361 bio_trim(mbio
, r1_bio
->sector
- bio
->bi_iter
.bi_sector
, max_sectors
);
1365 * Not if there are too many, or cannot
1366 * allocate memory, or a reader on WriteMostly
1367 * is waiting for behind writes to flush */
1369 (atomic_read(&bitmap
->behind_writes
)
1370 < mddev
->bitmap_info
.max_write_behind
) &&
1371 !waitqueue_active(&bitmap
->behind_wait
))
1372 alloc_behind_pages(mbio
, r1_bio
);
1374 bitmap_startwrite(bitmap
, r1_bio
->sector
,
1376 test_bit(R1BIO_BehindIO
,
1380 if (r1_bio
->behind_bvecs
) {
1381 struct bio_vec
*bvec
;
1385 * We trimmed the bio, so _all is legit
1387 bio_for_each_segment_all(bvec
, mbio
, j
)
1388 bvec
->bv_page
= r1_bio
->behind_bvecs
[j
].bv_page
;
1389 if (test_bit(WriteMostly
, &conf
->mirrors
[i
].rdev
->flags
))
1390 atomic_inc(&r1_bio
->behind_remaining
);
1393 r1_bio
->bios
[i
] = mbio
;
1395 mbio
->bi_iter
.bi_sector
= (r1_bio
->sector
+
1396 conf
->mirrors
[i
].rdev
->data_offset
);
1397 mbio
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1398 mbio
->bi_end_io
= raid1_end_write_request
;
1400 WRITE
| do_flush_fua
| do_sync
| do_discard
| do_same
;
1401 mbio
->bi_private
= r1_bio
;
1403 atomic_inc(&r1_bio
->remaining
);
1405 cb
= blk_check_plugged(raid1_unplug
, mddev
, sizeof(*plug
));
1407 plug
= container_of(cb
, struct raid1_plug_cb
, cb
);
1410 spin_lock_irqsave(&conf
->device_lock
, flags
);
1412 bio_list_add(&plug
->pending
, mbio
);
1413 plug
->pending_cnt
++;
1415 bio_list_add(&conf
->pending_bio_list
, mbio
);
1416 conf
->pending_count
++;
1418 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1420 md_wakeup_thread(mddev
->thread
);
1422 /* Mustn't call r1_bio_write_done before this next test,
1423 * as it could result in the bio being freed.
1425 if (sectors_handled
< bio_sectors(bio
)) {
1426 r1_bio_write_done(r1_bio
);
1427 /* We need another r1_bio. It has already been counted
1428 * in bio->bi_phys_segments
1430 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
1431 r1_bio
->master_bio
= bio
;
1432 r1_bio
->sectors
= bio_sectors(bio
) - sectors_handled
;
1434 r1_bio
->mddev
= mddev
;
1435 r1_bio
->sector
= bio
->bi_iter
.bi_sector
+ sectors_handled
;
1439 r1_bio_write_done(r1_bio
);
1441 /* In case raid1d snuck in to freeze_array */
1442 wake_up(&conf
->wait_barrier
);
1445 static void status(struct seq_file
*seq
, struct mddev
*mddev
)
1447 struct r1conf
*conf
= mddev
->private;
1450 seq_printf(seq
, " [%d/%d] [", conf
->raid_disks
,
1451 conf
->raid_disks
- mddev
->degraded
);
1453 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1454 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1455 seq_printf(seq
, "%s",
1456 rdev
&& test_bit(In_sync
, &rdev
->flags
) ? "U" : "_");
1459 seq_printf(seq
, "]");
1462 static void error(struct mddev
*mddev
, struct md_rdev
*rdev
)
1464 char b
[BDEVNAME_SIZE
];
1465 struct r1conf
*conf
= mddev
->private;
1468 * If it is not operational, then we have already marked it as dead
1469 * else if it is the last working disks, ignore the error, let the
1470 * next level up know.
1471 * else mark the drive as failed
1473 if (test_bit(In_sync
, &rdev
->flags
)
1474 && (conf
->raid_disks
- mddev
->degraded
) == 1) {
1476 * Don't fail the drive, act as though we were just a
1477 * normal single drive.
1478 * However don't try a recovery from this drive as
1479 * it is very likely to fail.
1481 conf
->recovery_disabled
= mddev
->recovery_disabled
;
1484 set_bit(Blocked
, &rdev
->flags
);
1485 if (test_and_clear_bit(In_sync
, &rdev
->flags
)) {
1486 unsigned long flags
;
1487 spin_lock_irqsave(&conf
->device_lock
, flags
);
1489 set_bit(Faulty
, &rdev
->flags
);
1490 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1492 set_bit(Faulty
, &rdev
->flags
);
1494 * if recovery is running, make sure it aborts.
1496 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1497 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
1499 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1500 "md/raid1:%s: Operation continuing on %d devices.\n",
1501 mdname(mddev
), bdevname(rdev
->bdev
, b
),
1502 mdname(mddev
), conf
->raid_disks
- mddev
->degraded
);
1505 static void print_conf(struct r1conf
*conf
)
1509 printk(KERN_DEBUG
"RAID1 conf printout:\n");
1511 printk(KERN_DEBUG
"(!conf)\n");
1514 printk(KERN_DEBUG
" --- wd:%d rd:%d\n", conf
->raid_disks
- conf
->mddev
->degraded
,
1518 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1519 char b
[BDEVNAME_SIZE
];
1520 struct md_rdev
*rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
1522 printk(KERN_DEBUG
" disk %d, wo:%d, o:%d, dev:%s\n",
1523 i
, !test_bit(In_sync
, &rdev
->flags
),
1524 !test_bit(Faulty
, &rdev
->flags
),
1525 bdevname(rdev
->bdev
,b
));
1530 static void close_sync(struct r1conf
*conf
)
1532 wait_barrier(conf
, NULL
);
1533 allow_barrier(conf
, 0, 0);
1535 mempool_destroy(conf
->r1buf_pool
);
1536 conf
->r1buf_pool
= NULL
;
1538 spin_lock_irq(&conf
->resync_lock
);
1539 conf
->next_resync
= 0;
1540 conf
->start_next_window
= MaxSector
;
1541 conf
->current_window_requests
+=
1542 conf
->next_window_requests
;
1543 conf
->next_window_requests
= 0;
1544 spin_unlock_irq(&conf
->resync_lock
);
1547 static int raid1_spare_active(struct mddev
*mddev
)
1550 struct r1conf
*conf
= mddev
->private;
1552 unsigned long flags
;
1555 * Find all failed disks within the RAID1 configuration
1556 * and mark them readable.
1557 * Called under mddev lock, so rcu protection not needed.
1559 for (i
= 0; i
< conf
->raid_disks
; i
++) {
1560 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
1561 struct md_rdev
*repl
= conf
->mirrors
[conf
->raid_disks
+ i
].rdev
;
1563 && repl
->recovery_offset
== MaxSector
1564 && !test_bit(Faulty
, &repl
->flags
)
1565 && !test_and_set_bit(In_sync
, &repl
->flags
)) {
1566 /* replacement has just become active */
1568 !test_and_clear_bit(In_sync
, &rdev
->flags
))
1571 /* Replaced device not technically
1572 * faulty, but we need to be sure
1573 * it gets removed and never re-added
1575 set_bit(Faulty
, &rdev
->flags
);
1576 sysfs_notify_dirent_safe(
1581 && rdev
->recovery_offset
== MaxSector
1582 && !test_bit(Faulty
, &rdev
->flags
)
1583 && !test_and_set_bit(In_sync
, &rdev
->flags
)) {
1585 sysfs_notify_dirent_safe(rdev
->sysfs_state
);
1588 spin_lock_irqsave(&conf
->device_lock
, flags
);
1589 mddev
->degraded
-= count
;
1590 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
1596 static int raid1_add_disk(struct mddev
*mddev
, struct md_rdev
*rdev
)
1598 struct r1conf
*conf
= mddev
->private;
1601 struct raid1_info
*p
;
1603 int last
= conf
->raid_disks
- 1;
1604 struct request_queue
*q
= bdev_get_queue(rdev
->bdev
);
1606 if (mddev
->recovery_disabled
== conf
->recovery_disabled
)
1609 if (rdev
->raid_disk
>= 0)
1610 first
= last
= rdev
->raid_disk
;
1612 if (q
->merge_bvec_fn
) {
1613 set_bit(Unmerged
, &rdev
->flags
);
1614 mddev
->merge_check_needed
= 1;
1617 for (mirror
= first
; mirror
<= last
; mirror
++) {
1618 p
= conf
->mirrors
+mirror
;
1622 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
1623 rdev
->data_offset
<< 9);
1625 p
->head_position
= 0;
1626 rdev
->raid_disk
= mirror
;
1628 /* As all devices are equivalent, we don't need a full recovery
1629 * if this was recently any drive of the array
1631 if (rdev
->saved_raid_disk
< 0)
1633 rcu_assign_pointer(p
->rdev
, rdev
);
1636 if (test_bit(WantReplacement
, &p
->rdev
->flags
) &&
1637 p
[conf
->raid_disks
].rdev
== NULL
) {
1638 /* Add this device as a replacement */
1639 clear_bit(In_sync
, &rdev
->flags
);
1640 set_bit(Replacement
, &rdev
->flags
);
1641 rdev
->raid_disk
= mirror
;
1644 rcu_assign_pointer(p
[conf
->raid_disks
].rdev
, rdev
);
1648 if (err
== 0 && test_bit(Unmerged
, &rdev
->flags
)) {
1649 /* Some requests might not have seen this new
1650 * merge_bvec_fn. We must wait for them to complete
1651 * before merging the device fully.
1652 * First we make sure any code which has tested
1653 * our function has submitted the request, then
1654 * we wait for all outstanding requests to complete.
1656 synchronize_sched();
1657 freeze_array(conf
, 0);
1658 unfreeze_array(conf
);
1659 clear_bit(Unmerged
, &rdev
->flags
);
1661 md_integrity_add_rdev(rdev
, mddev
);
1662 if (mddev
->queue
&& blk_queue_discard(bdev_get_queue(rdev
->bdev
)))
1663 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, mddev
->queue
);
1668 static int raid1_remove_disk(struct mddev
*mddev
, struct md_rdev
*rdev
)
1670 struct r1conf
*conf
= mddev
->private;
1672 int number
= rdev
->raid_disk
;
1673 struct raid1_info
*p
= conf
->mirrors
+ number
;
1675 if (rdev
!= p
->rdev
)
1676 p
= conf
->mirrors
+ conf
->raid_disks
+ number
;
1679 if (rdev
== p
->rdev
) {
1680 if (test_bit(In_sync
, &rdev
->flags
) ||
1681 atomic_read(&rdev
->nr_pending
)) {
1685 /* Only remove non-faulty devices if recovery
1688 if (!test_bit(Faulty
, &rdev
->flags
) &&
1689 mddev
->recovery_disabled
!= conf
->recovery_disabled
&&
1690 mddev
->degraded
< conf
->raid_disks
) {
1696 if (atomic_read(&rdev
->nr_pending
)) {
1697 /* lost the race, try later */
1701 } else if (conf
->mirrors
[conf
->raid_disks
+ number
].rdev
) {
1702 /* We just removed a device that is being replaced.
1703 * Move down the replacement. We drain all IO before
1704 * doing this to avoid confusion.
1706 struct md_rdev
*repl
=
1707 conf
->mirrors
[conf
->raid_disks
+ number
].rdev
;
1708 freeze_array(conf
, 0);
1709 clear_bit(Replacement
, &repl
->flags
);
1711 conf
->mirrors
[conf
->raid_disks
+ number
].rdev
= NULL
;
1712 unfreeze_array(conf
);
1713 clear_bit(WantReplacement
, &rdev
->flags
);
1715 clear_bit(WantReplacement
, &rdev
->flags
);
1716 err
= md_integrity_register(mddev
);
1724 static void end_sync_read(struct bio
*bio
, int error
)
1726 struct r1bio
*r1_bio
= bio
->bi_private
;
1728 update_head_pos(r1_bio
->read_disk
, r1_bio
);
1731 * we have read a block, now it needs to be re-written,
1732 * or re-read if the read failed.
1733 * We don't do much here, just schedule handling by raid1d
1735 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1736 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1738 if (atomic_dec_and_test(&r1_bio
->remaining
))
1739 reschedule_retry(r1_bio
);
1742 static void end_sync_write(struct bio
*bio
, int error
)
1744 int uptodate
= test_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1745 struct r1bio
*r1_bio
= bio
->bi_private
;
1746 struct mddev
*mddev
= r1_bio
->mddev
;
1747 struct r1conf
*conf
= mddev
->private;
1752 mirror
= find_bio_disk(r1_bio
, bio
);
1755 sector_t sync_blocks
= 0;
1756 sector_t s
= r1_bio
->sector
;
1757 long sectors_to_go
= r1_bio
->sectors
;
1758 /* make sure these bits doesn't get cleared. */
1760 bitmap_end_sync(mddev
->bitmap
, s
,
1763 sectors_to_go
-= sync_blocks
;
1764 } while (sectors_to_go
> 0);
1765 set_bit(WriteErrorSeen
,
1766 &conf
->mirrors
[mirror
].rdev
->flags
);
1767 if (!test_and_set_bit(WantReplacement
,
1768 &conf
->mirrors
[mirror
].rdev
->flags
))
1769 set_bit(MD_RECOVERY_NEEDED
, &
1771 set_bit(R1BIO_WriteError
, &r1_bio
->state
);
1772 } else if (is_badblock(conf
->mirrors
[mirror
].rdev
,
1775 &first_bad
, &bad_sectors
) &&
1776 !is_badblock(conf
->mirrors
[r1_bio
->read_disk
].rdev
,
1779 &first_bad
, &bad_sectors
)
1781 set_bit(R1BIO_MadeGood
, &r1_bio
->state
);
1783 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
1784 int s
= r1_bio
->sectors
;
1785 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
1786 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
1787 reschedule_retry(r1_bio
);
1790 md_done_sync(mddev
, s
, uptodate
);
1795 static int r1_sync_page_io(struct md_rdev
*rdev
, sector_t sector
,
1796 int sectors
, struct page
*page
, int rw
)
1798 if (sync_page_io(rdev
, sector
, sectors
<< 9, page
, rw
, false))
1802 set_bit(WriteErrorSeen
, &rdev
->flags
);
1803 if (!test_and_set_bit(WantReplacement
,
1805 set_bit(MD_RECOVERY_NEEDED
, &
1806 rdev
->mddev
->recovery
);
1808 /* need to record an error - either for the block or the device */
1809 if (!rdev_set_badblocks(rdev
, sector
, sectors
, 0))
1810 md_error(rdev
->mddev
, rdev
);
1814 static int fix_sync_read_error(struct r1bio
*r1_bio
)
1816 /* Try some synchronous reads of other devices to get
1817 * good data, much like with normal read errors. Only
1818 * read into the pages we already have so we don't
1819 * need to re-issue the read request.
1820 * We don't need to freeze the array, because being in an
1821 * active sync request, there is no normal IO, and
1822 * no overlapping syncs.
1823 * We don't need to check is_badblock() again as we
1824 * made sure that anything with a bad block in range
1825 * will have bi_end_io clear.
1827 struct mddev
*mddev
= r1_bio
->mddev
;
1828 struct r1conf
*conf
= mddev
->private;
1829 struct bio
*bio
= r1_bio
->bios
[r1_bio
->read_disk
];
1830 sector_t sect
= r1_bio
->sector
;
1831 int sectors
= r1_bio
->sectors
;
1836 int d
= r1_bio
->read_disk
;
1838 struct md_rdev
*rdev
;
1841 if (s
> (PAGE_SIZE
>>9))
1844 if (r1_bio
->bios
[d
]->bi_end_io
== end_sync_read
) {
1845 /* No rcu protection needed here devices
1846 * can only be removed when no resync is
1847 * active, and resync is currently active
1849 rdev
= conf
->mirrors
[d
].rdev
;
1850 if (sync_page_io(rdev
, sect
, s
<<9,
1851 bio
->bi_io_vec
[idx
].bv_page
,
1858 if (d
== conf
->raid_disks
* 2)
1860 } while (!success
&& d
!= r1_bio
->read_disk
);
1863 char b
[BDEVNAME_SIZE
];
1865 /* Cannot read from anywhere, this block is lost.
1866 * Record a bad block on each device. If that doesn't
1867 * work just disable and interrupt the recovery.
1868 * Don't fail devices as that won't really help.
1870 printk(KERN_ALERT
"md/raid1:%s: %s: unrecoverable I/O read error"
1871 " for block %llu\n",
1873 bdevname(bio
->bi_bdev
, b
),
1874 (unsigned long long)r1_bio
->sector
);
1875 for (d
= 0; d
< conf
->raid_disks
* 2; d
++) {
1876 rdev
= conf
->mirrors
[d
].rdev
;
1877 if (!rdev
|| test_bit(Faulty
, &rdev
->flags
))
1879 if (!rdev_set_badblocks(rdev
, sect
, s
, 0))
1883 conf
->recovery_disabled
=
1884 mddev
->recovery_disabled
;
1885 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
1886 md_done_sync(mddev
, r1_bio
->sectors
, 0);
1898 /* write it back and re-read */
1899 while (d
!= r1_bio
->read_disk
) {
1901 d
= conf
->raid_disks
* 2;
1903 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1905 rdev
= conf
->mirrors
[d
].rdev
;
1906 if (r1_sync_page_io(rdev
, sect
, s
,
1907 bio
->bi_io_vec
[idx
].bv_page
,
1909 r1_bio
->bios
[d
]->bi_end_io
= NULL
;
1910 rdev_dec_pending(rdev
, mddev
);
1914 while (d
!= r1_bio
->read_disk
) {
1916 d
= conf
->raid_disks
* 2;
1918 if (r1_bio
->bios
[d
]->bi_end_io
!= end_sync_read
)
1920 rdev
= conf
->mirrors
[d
].rdev
;
1921 if (r1_sync_page_io(rdev
, sect
, s
,
1922 bio
->bi_io_vec
[idx
].bv_page
,
1924 atomic_add(s
, &rdev
->corrected_errors
);
1930 set_bit(R1BIO_Uptodate
, &r1_bio
->state
);
1931 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1935 static void process_checks(struct r1bio
*r1_bio
)
1937 /* We have read all readable devices. If we haven't
1938 * got the block, then there is no hope left.
1939 * If we have, then we want to do a comparison
1940 * and skip the write if everything is the same.
1941 * If any blocks failed to read, then we need to
1942 * attempt an over-write
1944 struct mddev
*mddev
= r1_bio
->mddev
;
1945 struct r1conf
*conf
= mddev
->private;
1950 /* Fix variable parts of all bios */
1951 vcnt
= (r1_bio
->sectors
+ PAGE_SIZE
/ 512 - 1) >> (PAGE_SHIFT
- 9);
1952 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
1956 struct bio
*b
= r1_bio
->bios
[i
];
1957 if (b
->bi_end_io
!= end_sync_read
)
1959 /* fixup the bio for reuse, but preserve BIO_UPTODATE */
1960 uptodate
= test_bit(BIO_UPTODATE
, &b
->bi_flags
);
1963 clear_bit(BIO_UPTODATE
, &b
->bi_flags
);
1965 b
->bi_iter
.bi_size
= r1_bio
->sectors
<< 9;
1966 b
->bi_iter
.bi_sector
= r1_bio
->sector
+
1967 conf
->mirrors
[i
].rdev
->data_offset
;
1968 b
->bi_bdev
= conf
->mirrors
[i
].rdev
->bdev
;
1969 b
->bi_end_io
= end_sync_read
;
1970 b
->bi_private
= r1_bio
;
1972 size
= b
->bi_iter
.bi_size
;
1973 for (j
= 0; j
< vcnt
; j
++) {
1975 bi
= &b
->bi_io_vec
[j
];
1977 if (size
> PAGE_SIZE
)
1978 bi
->bv_len
= PAGE_SIZE
;
1984 for (primary
= 0; primary
< conf
->raid_disks
* 2; primary
++)
1985 if (r1_bio
->bios
[primary
]->bi_end_io
== end_sync_read
&&
1986 test_bit(BIO_UPTODATE
, &r1_bio
->bios
[primary
]->bi_flags
)) {
1987 r1_bio
->bios
[primary
]->bi_end_io
= NULL
;
1988 rdev_dec_pending(conf
->mirrors
[primary
].rdev
, mddev
);
1991 r1_bio
->read_disk
= primary
;
1992 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
1994 struct bio
*pbio
= r1_bio
->bios
[primary
];
1995 struct bio
*sbio
= r1_bio
->bios
[i
];
1996 int uptodate
= test_bit(BIO_UPTODATE
, &sbio
->bi_flags
);
1998 if (sbio
->bi_end_io
!= end_sync_read
)
2000 /* Now we can 'fixup' the BIO_UPTODATE flag */
2001 set_bit(BIO_UPTODATE
, &sbio
->bi_flags
);
2004 for (j
= vcnt
; j
-- ; ) {
2006 p
= pbio
->bi_io_vec
[j
].bv_page
;
2007 s
= sbio
->bi_io_vec
[j
].bv_page
;
2008 if (memcmp(page_address(p
),
2010 sbio
->bi_io_vec
[j
].bv_len
))
2016 atomic64_add(r1_bio
->sectors
, &mddev
->resync_mismatches
);
2017 if (j
< 0 || (test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)
2019 /* No need to write to this device. */
2020 sbio
->bi_end_io
= NULL
;
2021 rdev_dec_pending(conf
->mirrors
[i
].rdev
, mddev
);
2025 bio_copy_data(sbio
, pbio
);
2029 static void sync_request_write(struct mddev
*mddev
, struct r1bio
*r1_bio
)
2031 struct r1conf
*conf
= mddev
->private;
2033 int disks
= conf
->raid_disks
* 2;
2034 struct bio
*bio
, *wbio
;
2036 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
2038 if (!test_bit(R1BIO_Uptodate
, &r1_bio
->state
))
2039 /* ouch - failed to read all of that. */
2040 if (!fix_sync_read_error(r1_bio
))
2043 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2044 process_checks(r1_bio
);
2049 atomic_set(&r1_bio
->remaining
, 1);
2050 for (i
= 0; i
< disks
; i
++) {
2051 wbio
= r1_bio
->bios
[i
];
2052 if (wbio
->bi_end_io
== NULL
||
2053 (wbio
->bi_end_io
== end_sync_read
&&
2054 (i
== r1_bio
->read_disk
||
2055 !test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
))))
2058 wbio
->bi_rw
= WRITE
;
2059 wbio
->bi_end_io
= end_sync_write
;
2060 atomic_inc(&r1_bio
->remaining
);
2061 md_sync_acct(conf
->mirrors
[i
].rdev
->bdev
, bio_sectors(wbio
));
2063 generic_make_request(wbio
);
2066 if (atomic_dec_and_test(&r1_bio
->remaining
)) {
2067 /* if we're here, all write(s) have completed, so clean up */
2068 int s
= r1_bio
->sectors
;
2069 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
2070 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2071 reschedule_retry(r1_bio
);
2074 md_done_sync(mddev
, s
, 1);
2080 * This is a kernel thread which:
2082 * 1. Retries failed read operations on working mirrors.
2083 * 2. Updates the raid superblock when problems encounter.
2084 * 3. Performs writes following reads for array synchronising.
2087 static void fix_read_error(struct r1conf
*conf
, int read_disk
,
2088 sector_t sect
, int sectors
)
2090 struct mddev
*mddev
= conf
->mddev
;
2096 struct md_rdev
*rdev
;
2098 if (s
> (PAGE_SIZE
>>9))
2102 /* Note: no rcu protection needed here
2103 * as this is synchronous in the raid1d thread
2104 * which is the thread that might remove
2105 * a device. If raid1d ever becomes multi-threaded....
2110 rdev
= conf
->mirrors
[d
].rdev
;
2112 (test_bit(In_sync
, &rdev
->flags
) ||
2113 (!test_bit(Faulty
, &rdev
->flags
) &&
2114 rdev
->recovery_offset
>= sect
+ s
)) &&
2115 is_badblock(rdev
, sect
, s
,
2116 &first_bad
, &bad_sectors
) == 0 &&
2117 sync_page_io(rdev
, sect
, s
<<9,
2118 conf
->tmppage
, READ
, false))
2122 if (d
== conf
->raid_disks
* 2)
2125 } while (!success
&& d
!= read_disk
);
2128 /* Cannot read from anywhere - mark it bad */
2129 struct md_rdev
*rdev
= conf
->mirrors
[read_disk
].rdev
;
2130 if (!rdev_set_badblocks(rdev
, sect
, s
, 0))
2131 md_error(mddev
, rdev
);
2134 /* write it back and re-read */
2136 while (d
!= read_disk
) {
2138 d
= conf
->raid_disks
* 2;
2140 rdev
= conf
->mirrors
[d
].rdev
;
2142 !test_bit(Faulty
, &rdev
->flags
))
2143 r1_sync_page_io(rdev
, sect
, s
,
2144 conf
->tmppage
, WRITE
);
2147 while (d
!= read_disk
) {
2148 char b
[BDEVNAME_SIZE
];
2150 d
= conf
->raid_disks
* 2;
2152 rdev
= conf
->mirrors
[d
].rdev
;
2154 !test_bit(Faulty
, &rdev
->flags
)) {
2155 if (r1_sync_page_io(rdev
, sect
, s
,
2156 conf
->tmppage
, READ
)) {
2157 atomic_add(s
, &rdev
->corrected_errors
);
2159 "md/raid1:%s: read error corrected "
2160 "(%d sectors at %llu on %s)\n",
2162 (unsigned long long)(sect
+
2164 bdevname(rdev
->bdev
, b
));
2173 static int narrow_write_error(struct r1bio
*r1_bio
, int i
)
2175 struct mddev
*mddev
= r1_bio
->mddev
;
2176 struct r1conf
*conf
= mddev
->private;
2177 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
2179 /* bio has the data to be written to device 'i' where
2180 * we just recently had a write error.
2181 * We repeatedly clone the bio and trim down to one block,
2182 * then try the write. Where the write fails we record
2184 * It is conceivable that the bio doesn't exactly align with
2185 * blocks. We must handle this somehow.
2187 * We currently own a reference on the rdev.
2193 int sect_to_write
= r1_bio
->sectors
;
2196 if (rdev
->badblocks
.shift
< 0)
2199 block_sectors
= roundup(1 << rdev
->badblocks
.shift
,
2200 bdev_logical_block_size(rdev
->bdev
) >> 9);
2201 sector
= r1_bio
->sector
;
2202 sectors
= ((sector
+ block_sectors
)
2203 & ~(sector_t
)(block_sectors
- 1))
2206 while (sect_to_write
) {
2208 if (sectors
> sect_to_write
)
2209 sectors
= sect_to_write
;
2210 /* Write at 'sector' for 'sectors'*/
2212 if (test_bit(R1BIO_BehindIO
, &r1_bio
->state
)) {
2213 unsigned vcnt
= r1_bio
->behind_page_count
;
2214 struct bio_vec
*vec
= r1_bio
->behind_bvecs
;
2216 while (!vec
->bv_page
) {
2221 wbio
= bio_alloc_mddev(GFP_NOIO
, vcnt
, mddev
);
2222 memcpy(wbio
->bi_io_vec
, vec
, vcnt
* sizeof(struct bio_vec
));
2224 wbio
->bi_vcnt
= vcnt
;
2226 wbio
= bio_clone_mddev(r1_bio
->master_bio
, GFP_NOIO
, mddev
);
2229 wbio
->bi_rw
= WRITE
;
2230 wbio
->bi_iter
.bi_sector
= r1_bio
->sector
;
2231 wbio
->bi_iter
.bi_size
= r1_bio
->sectors
<< 9;
2233 bio_trim(wbio
, sector
- r1_bio
->sector
, sectors
);
2234 wbio
->bi_iter
.bi_sector
+= rdev
->data_offset
;
2235 wbio
->bi_bdev
= rdev
->bdev
;
2236 if (submit_bio_wait(WRITE
, wbio
) == 0)
2238 ok
= rdev_set_badblocks(rdev
, sector
,
2243 sect_to_write
-= sectors
;
2245 sectors
= block_sectors
;
2250 static void handle_sync_write_finished(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2253 int s
= r1_bio
->sectors
;
2254 for (m
= 0; m
< conf
->raid_disks
* 2 ; m
++) {
2255 struct md_rdev
*rdev
= conf
->mirrors
[m
].rdev
;
2256 struct bio
*bio
= r1_bio
->bios
[m
];
2257 if (bio
->bi_end_io
== NULL
)
2259 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
2260 test_bit(R1BIO_MadeGood
, &r1_bio
->state
)) {
2261 rdev_clear_badblocks(rdev
, r1_bio
->sector
, s
, 0);
2263 if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
) &&
2264 test_bit(R1BIO_WriteError
, &r1_bio
->state
)) {
2265 if (!rdev_set_badblocks(rdev
, r1_bio
->sector
, s
, 0))
2266 md_error(conf
->mddev
, rdev
);
2270 md_done_sync(conf
->mddev
, s
, 1);
2273 static void handle_write_finished(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2276 for (m
= 0; m
< conf
->raid_disks
* 2 ; m
++)
2277 if (r1_bio
->bios
[m
] == IO_MADE_GOOD
) {
2278 struct md_rdev
*rdev
= conf
->mirrors
[m
].rdev
;
2279 rdev_clear_badblocks(rdev
,
2281 r1_bio
->sectors
, 0);
2282 rdev_dec_pending(rdev
, conf
->mddev
);
2283 } else if (r1_bio
->bios
[m
] != NULL
) {
2284 /* This drive got a write error. We need to
2285 * narrow down and record precise write
2288 if (!narrow_write_error(r1_bio
, m
)) {
2289 md_error(conf
->mddev
,
2290 conf
->mirrors
[m
].rdev
);
2291 /* an I/O failed, we can't clear the bitmap */
2292 set_bit(R1BIO_Degraded
, &r1_bio
->state
);
2294 rdev_dec_pending(conf
->mirrors
[m
].rdev
,
2297 if (test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2298 close_write(r1_bio
);
2299 raid_end_bio_io(r1_bio
);
2302 static void handle_read_error(struct r1conf
*conf
, struct r1bio
*r1_bio
)
2306 struct mddev
*mddev
= conf
->mddev
;
2308 char b
[BDEVNAME_SIZE
];
2309 struct md_rdev
*rdev
;
2311 clear_bit(R1BIO_ReadError
, &r1_bio
->state
);
2312 /* we got a read error. Maybe the drive is bad. Maybe just
2313 * the block and we can fix it.
2314 * We freeze all other IO, and try reading the block from
2315 * other devices. When we find one, we re-write
2316 * and check it that fixes the read error.
2317 * This is all done synchronously while the array is
2320 if (mddev
->ro
== 0) {
2321 freeze_array(conf
, 1);
2322 fix_read_error(conf
, r1_bio
->read_disk
,
2323 r1_bio
->sector
, r1_bio
->sectors
);
2324 unfreeze_array(conf
);
2326 md_error(mddev
, conf
->mirrors
[r1_bio
->read_disk
].rdev
);
2327 rdev_dec_pending(conf
->mirrors
[r1_bio
->read_disk
].rdev
, conf
->mddev
);
2329 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
2330 bdevname(bio
->bi_bdev
, b
);
2332 disk
= read_balance(conf
, r1_bio
, &max_sectors
);
2334 printk(KERN_ALERT
"md/raid1:%s: %s: unrecoverable I/O"
2335 " read error for block %llu\n",
2336 mdname(mddev
), b
, (unsigned long long)r1_bio
->sector
);
2337 raid_end_bio_io(r1_bio
);
2339 const unsigned long do_sync
2340 = r1_bio
->master_bio
->bi_rw
& REQ_SYNC
;
2342 r1_bio
->bios
[r1_bio
->read_disk
] =
2343 mddev
->ro
? IO_BLOCKED
: NULL
;
2346 r1_bio
->read_disk
= disk
;
2347 bio
= bio_clone_mddev(r1_bio
->master_bio
, GFP_NOIO
, mddev
);
2348 bio_trim(bio
, r1_bio
->sector
- bio
->bi_iter
.bi_sector
,
2350 r1_bio
->bios
[r1_bio
->read_disk
] = bio
;
2351 rdev
= conf
->mirrors
[disk
].rdev
;
2352 printk_ratelimited(KERN_ERR
2353 "md/raid1:%s: redirecting sector %llu"
2354 " to other mirror: %s\n",
2356 (unsigned long long)r1_bio
->sector
,
2357 bdevname(rdev
->bdev
, b
));
2358 bio
->bi_iter
.bi_sector
= r1_bio
->sector
+ rdev
->data_offset
;
2359 bio
->bi_bdev
= rdev
->bdev
;
2360 bio
->bi_end_io
= raid1_end_read_request
;
2361 bio
->bi_rw
= READ
| do_sync
;
2362 bio
->bi_private
= r1_bio
;
2363 if (max_sectors
< r1_bio
->sectors
) {
2364 /* Drat - have to split this up more */
2365 struct bio
*mbio
= r1_bio
->master_bio
;
2366 int sectors_handled
= (r1_bio
->sector
+ max_sectors
2367 - mbio
->bi_iter
.bi_sector
);
2368 r1_bio
->sectors
= max_sectors
;
2369 spin_lock_irq(&conf
->device_lock
);
2370 if (mbio
->bi_phys_segments
== 0)
2371 mbio
->bi_phys_segments
= 2;
2373 mbio
->bi_phys_segments
++;
2374 spin_unlock_irq(&conf
->device_lock
);
2375 generic_make_request(bio
);
2378 r1_bio
= mempool_alloc(conf
->r1bio_pool
, GFP_NOIO
);
2380 r1_bio
->master_bio
= mbio
;
2381 r1_bio
->sectors
= bio_sectors(mbio
) - sectors_handled
;
2383 set_bit(R1BIO_ReadError
, &r1_bio
->state
);
2384 r1_bio
->mddev
= mddev
;
2385 r1_bio
->sector
= mbio
->bi_iter
.bi_sector
+
2390 generic_make_request(bio
);
2394 static void raid1d(struct md_thread
*thread
)
2396 struct mddev
*mddev
= thread
->mddev
;
2397 struct r1bio
*r1_bio
;
2398 unsigned long flags
;
2399 struct r1conf
*conf
= mddev
->private;
2400 struct list_head
*head
= &conf
->retry_list
;
2401 struct blk_plug plug
;
2403 md_check_recovery(mddev
);
2405 blk_start_plug(&plug
);
2408 flush_pending_writes(conf
);
2410 spin_lock_irqsave(&conf
->device_lock
, flags
);
2411 if (list_empty(head
)) {
2412 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2415 r1_bio
= list_entry(head
->prev
, struct r1bio
, retry_list
);
2416 list_del(head
->prev
);
2418 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
2420 mddev
= r1_bio
->mddev
;
2421 conf
= mddev
->private;
2422 if (test_bit(R1BIO_IsSync
, &r1_bio
->state
)) {
2423 if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
2424 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2425 handle_sync_write_finished(conf
, r1_bio
);
2427 sync_request_write(mddev
, r1_bio
);
2428 } else if (test_bit(R1BIO_MadeGood
, &r1_bio
->state
) ||
2429 test_bit(R1BIO_WriteError
, &r1_bio
->state
))
2430 handle_write_finished(conf
, r1_bio
);
2431 else if (test_bit(R1BIO_ReadError
, &r1_bio
->state
))
2432 handle_read_error(conf
, r1_bio
);
2434 /* just a partial read to be scheduled from separate
2437 generic_make_request(r1_bio
->bios
[r1_bio
->read_disk
]);
2440 if (mddev
->flags
& ~(1<<MD_CHANGE_PENDING
))
2441 md_check_recovery(mddev
);
2443 blk_finish_plug(&plug
);
2446 static int init_resync(struct r1conf
*conf
)
2450 buffs
= RESYNC_WINDOW
/ RESYNC_BLOCK_SIZE
;
2451 BUG_ON(conf
->r1buf_pool
);
2452 conf
->r1buf_pool
= mempool_create(buffs
, r1buf_pool_alloc
, r1buf_pool_free
,
2454 if (!conf
->r1buf_pool
)
2456 conf
->next_resync
= 0;
2461 * perform a "sync" on one "block"
2463 * We need to make sure that no normal I/O request - particularly write
2464 * requests - conflict with active sync requests.
2466 * This is achieved by tracking pending requests and a 'barrier' concept
2467 * that can be installed to exclude normal IO requests.
2470 static sector_t
sync_request(struct mddev
*mddev
, sector_t sector_nr
, int *skipped
, int go_faster
)
2472 struct r1conf
*conf
= mddev
->private;
2473 struct r1bio
*r1_bio
;
2475 sector_t max_sector
, nr_sectors
;
2479 int write_targets
= 0, read_targets
= 0;
2480 sector_t sync_blocks
;
2481 int still_degraded
= 0;
2482 int good_sectors
= RESYNC_SECTORS
;
2483 int min_bad
= 0; /* number of sectors that are bad in all devices */
2485 if (!conf
->r1buf_pool
)
2486 if (init_resync(conf
))
2489 max_sector
= mddev
->dev_sectors
;
2490 if (sector_nr
>= max_sector
) {
2491 /* If we aborted, we need to abort the
2492 * sync on the 'current' bitmap chunk (there will
2493 * only be one in raid1 resync.
2494 * We can find the current addess in mddev->curr_resync
2496 if (mddev
->curr_resync
< max_sector
) /* aborted */
2497 bitmap_end_sync(mddev
->bitmap
, mddev
->curr_resync
,
2499 else /* completed sync */
2502 bitmap_close_sync(mddev
->bitmap
);
2507 if (mddev
->bitmap
== NULL
&&
2508 mddev
->recovery_cp
== MaxSector
&&
2509 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
) &&
2510 conf
->fullsync
== 0) {
2512 return max_sector
- sector_nr
;
2514 /* before building a request, check if we can skip these blocks..
2515 * This call the bitmap_start_sync doesn't actually record anything
2517 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
, &sync_blocks
, 1) &&
2518 !conf
->fullsync
&& !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
2519 /* We can skip this block, and probably several more */
2524 * If there is non-resync activity waiting for a turn,
2525 * and resync is going fast enough,
2526 * then let it though before starting on this new sync request.
2528 if (!go_faster
&& conf
->nr_waiting
)
2529 msleep_interruptible(1000);
2531 bitmap_cond_end_sync(mddev
->bitmap
, sector_nr
);
2532 r1_bio
= mempool_alloc(conf
->r1buf_pool
, GFP_NOIO
);
2534 raise_barrier(conf
, sector_nr
);
2538 * If we get a correctably read error during resync or recovery,
2539 * we might want to read from a different device. So we
2540 * flag all drives that could conceivably be read from for READ,
2541 * and any others (which will be non-In_sync devices) for WRITE.
2542 * If a read fails, we try reading from something else for which READ
2546 r1_bio
->mddev
= mddev
;
2547 r1_bio
->sector
= sector_nr
;
2549 set_bit(R1BIO_IsSync
, &r1_bio
->state
);
2551 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
2552 struct md_rdev
*rdev
;
2553 bio
= r1_bio
->bios
[i
];
2556 rdev
= rcu_dereference(conf
->mirrors
[i
].rdev
);
2558 test_bit(Faulty
, &rdev
->flags
)) {
2559 if (i
< conf
->raid_disks
)
2561 } else if (!test_bit(In_sync
, &rdev
->flags
)) {
2563 bio
->bi_end_io
= end_sync_write
;
2566 /* may need to read from here */
2567 sector_t first_bad
= MaxSector
;
2570 if (is_badblock(rdev
, sector_nr
, good_sectors
,
2571 &first_bad
, &bad_sectors
)) {
2572 if (first_bad
> sector_nr
)
2573 good_sectors
= first_bad
- sector_nr
;
2575 bad_sectors
-= (sector_nr
- first_bad
);
2577 min_bad
> bad_sectors
)
2578 min_bad
= bad_sectors
;
2581 if (sector_nr
< first_bad
) {
2582 if (test_bit(WriteMostly
, &rdev
->flags
)) {
2590 bio
->bi_end_io
= end_sync_read
;
2592 } else if (!test_bit(WriteErrorSeen
, &rdev
->flags
) &&
2593 test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) &&
2594 !test_bit(MD_RECOVERY_CHECK
, &mddev
->recovery
)) {
2596 * The device is suitable for reading (InSync),
2597 * but has bad block(s) here. Let's try to correct them,
2598 * if we are doing resync or repair. Otherwise, leave
2599 * this device alone for this sync request.
2602 bio
->bi_end_io
= end_sync_write
;
2606 if (bio
->bi_end_io
) {
2607 atomic_inc(&rdev
->nr_pending
);
2608 bio
->bi_iter
.bi_sector
= sector_nr
+ rdev
->data_offset
;
2609 bio
->bi_bdev
= rdev
->bdev
;
2610 bio
->bi_private
= r1_bio
;
2616 r1_bio
->read_disk
= disk
;
2618 if (read_targets
== 0 && min_bad
> 0) {
2619 /* These sectors are bad on all InSync devices, so we
2620 * need to mark them bad on all write targets
2623 for (i
= 0 ; i
< conf
->raid_disks
* 2 ; i
++)
2624 if (r1_bio
->bios
[i
]->bi_end_io
== end_sync_write
) {
2625 struct md_rdev
*rdev
= conf
->mirrors
[i
].rdev
;
2626 ok
= rdev_set_badblocks(rdev
, sector_nr
,
2630 set_bit(MD_CHANGE_DEVS
, &mddev
->flags
);
2635 /* Cannot record the badblocks, so need to
2637 * If there are multiple read targets, could just
2638 * fail the really bad ones ???
2640 conf
->recovery_disabled
= mddev
->recovery_disabled
;
2641 set_bit(MD_RECOVERY_INTR
, &mddev
->recovery
);
2647 if (min_bad
> 0 && min_bad
< good_sectors
) {
2648 /* only resync enough to reach the next bad->good
2650 good_sectors
= min_bad
;
2653 if (test_bit(MD_RECOVERY_SYNC
, &mddev
->recovery
) && read_targets
> 0)
2654 /* extra read targets are also write targets */
2655 write_targets
+= read_targets
-1;
2657 if (write_targets
== 0 || read_targets
== 0) {
2658 /* There is nowhere to write, so all non-sync
2659 * drives must be failed - so we are finished
2663 max_sector
= sector_nr
+ min_bad
;
2664 rv
= max_sector
- sector_nr
;
2670 if (max_sector
> mddev
->resync_max
)
2671 max_sector
= mddev
->resync_max
; /* Don't do IO beyond here */
2672 if (max_sector
> sector_nr
+ good_sectors
)
2673 max_sector
= sector_nr
+ good_sectors
;
2678 int len
= PAGE_SIZE
;
2679 if (sector_nr
+ (len
>>9) > max_sector
)
2680 len
= (max_sector
- sector_nr
) << 9;
2683 if (sync_blocks
== 0) {
2684 if (!bitmap_start_sync(mddev
->bitmap
, sector_nr
,
2685 &sync_blocks
, still_degraded
) &&
2687 !test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
))
2689 BUG_ON(sync_blocks
< (PAGE_SIZE
>>9));
2690 if ((len
>> 9) > sync_blocks
)
2691 len
= sync_blocks
<<9;
2694 for (i
= 0 ; i
< conf
->raid_disks
* 2; i
++) {
2695 bio
= r1_bio
->bios
[i
];
2696 if (bio
->bi_end_io
) {
2697 page
= bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
;
2698 if (bio_add_page(bio
, page
, len
, 0) == 0) {
2700 bio
->bi_io_vec
[bio
->bi_vcnt
].bv_page
= page
;
2703 bio
= r1_bio
->bios
[i
];
2704 if (bio
->bi_end_io
==NULL
)
2706 /* remove last page from this bio */
2708 bio
->bi_iter
.bi_size
-= len
;
2709 __clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
2715 nr_sectors
+= len
>>9;
2716 sector_nr
+= len
>>9;
2717 sync_blocks
-= (len
>>9);
2718 } while (r1_bio
->bios
[disk
]->bi_vcnt
< RESYNC_PAGES
);
2720 r1_bio
->sectors
= nr_sectors
;
2722 /* For a user-requested sync, we read all readable devices and do a
2725 if (test_bit(MD_RECOVERY_REQUESTED
, &mddev
->recovery
)) {
2726 atomic_set(&r1_bio
->remaining
, read_targets
);
2727 for (i
= 0; i
< conf
->raid_disks
* 2 && read_targets
; i
++) {
2728 bio
= r1_bio
->bios
[i
];
2729 if (bio
->bi_end_io
== end_sync_read
) {
2731 md_sync_acct(bio
->bi_bdev
, nr_sectors
);
2732 generic_make_request(bio
);
2736 atomic_set(&r1_bio
->remaining
, 1);
2737 bio
= r1_bio
->bios
[r1_bio
->read_disk
];
2738 md_sync_acct(bio
->bi_bdev
, nr_sectors
);
2739 generic_make_request(bio
);
2745 static sector_t
raid1_size(struct mddev
*mddev
, sector_t sectors
, int raid_disks
)
2750 return mddev
->dev_sectors
;
2753 static struct r1conf
*setup_conf(struct mddev
*mddev
)
2755 struct r1conf
*conf
;
2757 struct raid1_info
*disk
;
2758 struct md_rdev
*rdev
;
2761 conf
= kzalloc(sizeof(struct r1conf
), GFP_KERNEL
);
2765 conf
->mirrors
= kzalloc(sizeof(struct raid1_info
)
2766 * mddev
->raid_disks
* 2,
2771 conf
->tmppage
= alloc_page(GFP_KERNEL
);
2775 conf
->poolinfo
= kzalloc(sizeof(*conf
->poolinfo
), GFP_KERNEL
);
2776 if (!conf
->poolinfo
)
2778 conf
->poolinfo
->raid_disks
= mddev
->raid_disks
* 2;
2779 conf
->r1bio_pool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
2782 if (!conf
->r1bio_pool
)
2785 conf
->poolinfo
->mddev
= mddev
;
2788 spin_lock_init(&conf
->device_lock
);
2789 rdev_for_each(rdev
, mddev
) {
2790 struct request_queue
*q
;
2791 int disk_idx
= rdev
->raid_disk
;
2792 if (disk_idx
>= mddev
->raid_disks
2795 if (test_bit(Replacement
, &rdev
->flags
))
2796 disk
= conf
->mirrors
+ mddev
->raid_disks
+ disk_idx
;
2798 disk
= conf
->mirrors
+ disk_idx
;
2803 q
= bdev_get_queue(rdev
->bdev
);
2804 if (q
->merge_bvec_fn
)
2805 mddev
->merge_check_needed
= 1;
2807 disk
->head_position
= 0;
2808 disk
->seq_start
= MaxSector
;
2810 conf
->raid_disks
= mddev
->raid_disks
;
2811 conf
->mddev
= mddev
;
2812 INIT_LIST_HEAD(&conf
->retry_list
);
2814 spin_lock_init(&conf
->resync_lock
);
2815 init_waitqueue_head(&conf
->wait_barrier
);
2817 bio_list_init(&conf
->pending_bio_list
);
2818 conf
->pending_count
= 0;
2819 conf
->recovery_disabled
= mddev
->recovery_disabled
- 1;
2821 conf
->start_next_window
= MaxSector
;
2822 conf
->current_window_requests
= conf
->next_window_requests
= 0;
2825 for (i
= 0; i
< conf
->raid_disks
* 2; i
++) {
2827 disk
= conf
->mirrors
+ i
;
2829 if (i
< conf
->raid_disks
&&
2830 disk
[conf
->raid_disks
].rdev
) {
2831 /* This slot has a replacement. */
2833 /* No original, just make the replacement
2834 * a recovering spare
2837 disk
[conf
->raid_disks
].rdev
;
2838 disk
[conf
->raid_disks
].rdev
= NULL
;
2839 } else if (!test_bit(In_sync
, &disk
->rdev
->flags
))
2840 /* Original is not in_sync - bad */
2845 !test_bit(In_sync
, &disk
->rdev
->flags
)) {
2846 disk
->head_position
= 0;
2848 (disk
->rdev
->saved_raid_disk
< 0))
2854 conf
->thread
= md_register_thread(raid1d
, mddev
, "raid1");
2855 if (!conf
->thread
) {
2857 "md/raid1:%s: couldn't allocate thread\n",
2866 if (conf
->r1bio_pool
)
2867 mempool_destroy(conf
->r1bio_pool
);
2868 kfree(conf
->mirrors
);
2869 safe_put_page(conf
->tmppage
);
2870 kfree(conf
->poolinfo
);
2873 return ERR_PTR(err
);
2876 static void raid1_free(struct mddev
*mddev
, void *priv
);
2877 static int run(struct mddev
*mddev
)
2879 struct r1conf
*conf
;
2881 struct md_rdev
*rdev
;
2883 bool discard_supported
= false;
2885 if (mddev
->level
!= 1) {
2886 printk(KERN_ERR
"md/raid1:%s: raid level not set to mirroring (%d)\n",
2887 mdname(mddev
), mddev
->level
);
2890 if (mddev
->reshape_position
!= MaxSector
) {
2891 printk(KERN_ERR
"md/raid1:%s: reshape_position set but not supported\n",
2896 * copy the already verified devices into our private RAID1
2897 * bookkeeping area. [whatever we allocate in run(),
2898 * should be freed in raid1_free()]
2900 if (mddev
->private == NULL
)
2901 conf
= setup_conf(mddev
);
2903 conf
= mddev
->private;
2906 return PTR_ERR(conf
);
2909 blk_queue_max_write_same_sectors(mddev
->queue
, 0);
2911 rdev_for_each(rdev
, mddev
) {
2912 if (!mddev
->gendisk
)
2914 disk_stack_limits(mddev
->gendisk
, rdev
->bdev
,
2915 rdev
->data_offset
<< 9);
2916 if (blk_queue_discard(bdev_get_queue(rdev
->bdev
)))
2917 discard_supported
= true;
2920 mddev
->degraded
= 0;
2921 for (i
=0; i
< conf
->raid_disks
; i
++)
2922 if (conf
->mirrors
[i
].rdev
== NULL
||
2923 !test_bit(In_sync
, &conf
->mirrors
[i
].rdev
->flags
) ||
2924 test_bit(Faulty
, &conf
->mirrors
[i
].rdev
->flags
))
2927 if (conf
->raid_disks
- mddev
->degraded
== 1)
2928 mddev
->recovery_cp
= MaxSector
;
2930 if (mddev
->recovery_cp
!= MaxSector
)
2931 printk(KERN_NOTICE
"md/raid1:%s: not clean"
2932 " -- starting background reconstruction\n",
2935 "md/raid1:%s: active with %d out of %d mirrors\n",
2936 mdname(mddev
), mddev
->raid_disks
- mddev
->degraded
,
2940 * Ok, everything is just fine now
2942 mddev
->thread
= conf
->thread
;
2943 conf
->thread
= NULL
;
2944 mddev
->private = conf
;
2946 md_set_array_sectors(mddev
, raid1_size(mddev
, 0, 0));
2949 if (discard_supported
)
2950 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
,
2953 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD
,
2957 ret
= md_integrity_register(mddev
);
2959 md_unregister_thread(&mddev
->thread
);
2960 raid1_free(mddev
, conf
);
2965 static void raid1_free(struct mddev
*mddev
, void *priv
)
2967 struct r1conf
*conf
= priv
;
2969 if (conf
->r1bio_pool
)
2970 mempool_destroy(conf
->r1bio_pool
);
2971 kfree(conf
->mirrors
);
2972 safe_put_page(conf
->tmppage
);
2973 kfree(conf
->poolinfo
);
2977 static int raid1_resize(struct mddev
*mddev
, sector_t sectors
)
2979 /* no resync is happening, and there is enough space
2980 * on all devices, so we can resize.
2981 * We need to make sure resync covers any new space.
2982 * If the array is shrinking we should possibly wait until
2983 * any io in the removed space completes, but it hardly seems
2986 sector_t newsize
= raid1_size(mddev
, sectors
, 0);
2987 if (mddev
->external_size
&&
2988 mddev
->array_sectors
> newsize
)
2990 if (mddev
->bitmap
) {
2991 int ret
= bitmap_resize(mddev
->bitmap
, newsize
, 0, 0);
2995 md_set_array_sectors(mddev
, newsize
);
2996 set_capacity(mddev
->gendisk
, mddev
->array_sectors
);
2997 revalidate_disk(mddev
->gendisk
);
2998 if (sectors
> mddev
->dev_sectors
&&
2999 mddev
->recovery_cp
> mddev
->dev_sectors
) {
3000 mddev
->recovery_cp
= mddev
->dev_sectors
;
3001 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3003 mddev
->dev_sectors
= sectors
;
3004 mddev
->resync_max_sectors
= sectors
;
3008 static int raid1_reshape(struct mddev
*mddev
)
3011 * 1/ resize the r1bio_pool
3012 * 2/ resize conf->mirrors
3014 * We allocate a new r1bio_pool if we can.
3015 * Then raise a device barrier and wait until all IO stops.
3016 * Then resize conf->mirrors and swap in the new r1bio pool.
3018 * At the same time, we "pack" the devices so that all the missing
3019 * devices have the higher raid_disk numbers.
3021 mempool_t
*newpool
, *oldpool
;
3022 struct pool_info
*newpoolinfo
;
3023 struct raid1_info
*newmirrors
;
3024 struct r1conf
*conf
= mddev
->private;
3025 int cnt
, raid_disks
;
3026 unsigned long flags
;
3029 /* Cannot change chunk_size, layout, or level */
3030 if (mddev
->chunk_sectors
!= mddev
->new_chunk_sectors
||
3031 mddev
->layout
!= mddev
->new_layout
||
3032 mddev
->level
!= mddev
->new_level
) {
3033 mddev
->new_chunk_sectors
= mddev
->chunk_sectors
;
3034 mddev
->new_layout
= mddev
->layout
;
3035 mddev
->new_level
= mddev
->level
;
3039 err
= md_allow_write(mddev
);
3043 raid_disks
= mddev
->raid_disks
+ mddev
->delta_disks
;
3045 if (raid_disks
< conf
->raid_disks
) {
3047 for (d
= 0; d
< conf
->raid_disks
; d
++)
3048 if (conf
->mirrors
[d
].rdev
)
3050 if (cnt
> raid_disks
)
3054 newpoolinfo
= kmalloc(sizeof(*newpoolinfo
), GFP_KERNEL
);
3057 newpoolinfo
->mddev
= mddev
;
3058 newpoolinfo
->raid_disks
= raid_disks
* 2;
3060 newpool
= mempool_create(NR_RAID1_BIOS
, r1bio_pool_alloc
,
3061 r1bio_pool_free
, newpoolinfo
);
3066 newmirrors
= kzalloc(sizeof(struct raid1_info
) * raid_disks
* 2,
3070 mempool_destroy(newpool
);
3074 freeze_array(conf
, 0);
3076 /* ok, everything is stopped */
3077 oldpool
= conf
->r1bio_pool
;
3078 conf
->r1bio_pool
= newpool
;
3080 for (d
= d2
= 0; d
< conf
->raid_disks
; d
++) {
3081 struct md_rdev
*rdev
= conf
->mirrors
[d
].rdev
;
3082 if (rdev
&& rdev
->raid_disk
!= d2
) {
3083 sysfs_unlink_rdev(mddev
, rdev
);
3084 rdev
->raid_disk
= d2
;
3085 sysfs_unlink_rdev(mddev
, rdev
);
3086 if (sysfs_link_rdev(mddev
, rdev
))
3088 "md/raid1:%s: cannot register rd%d\n",
3089 mdname(mddev
), rdev
->raid_disk
);
3092 newmirrors
[d2
++].rdev
= rdev
;
3094 kfree(conf
->mirrors
);
3095 conf
->mirrors
= newmirrors
;
3096 kfree(conf
->poolinfo
);
3097 conf
->poolinfo
= newpoolinfo
;
3099 spin_lock_irqsave(&conf
->device_lock
, flags
);
3100 mddev
->degraded
+= (raid_disks
- conf
->raid_disks
);
3101 spin_unlock_irqrestore(&conf
->device_lock
, flags
);
3102 conf
->raid_disks
= mddev
->raid_disks
= raid_disks
;
3103 mddev
->delta_disks
= 0;
3105 unfreeze_array(conf
);
3107 set_bit(MD_RECOVERY_NEEDED
, &mddev
->recovery
);
3108 md_wakeup_thread(mddev
->thread
);
3110 mempool_destroy(oldpool
);
3114 static void raid1_quiesce(struct mddev
*mddev
, int state
)
3116 struct r1conf
*conf
= mddev
->private;
3119 case 2: /* wake for suspend */
3120 wake_up(&conf
->wait_barrier
);
3123 freeze_array(conf
, 0);
3126 unfreeze_array(conf
);
3131 static void *raid1_takeover(struct mddev
*mddev
)
3133 /* raid1 can take over:
3134 * raid5 with 2 devices, any layout or chunk size
3136 if (mddev
->level
== 5 && mddev
->raid_disks
== 2) {
3137 struct r1conf
*conf
;
3138 mddev
->new_level
= 1;
3139 mddev
->new_layout
= 0;
3140 mddev
->new_chunk_sectors
= 0;
3141 conf
= setup_conf(mddev
);
3143 /* Array must appear to be quiesced */
3144 conf
->array_frozen
= 1;
3147 return ERR_PTR(-EINVAL
);
3150 static struct md_personality raid1_personality
=
3154 .owner
= THIS_MODULE
,
3155 .make_request
= make_request
,
3159 .error_handler
= error
,
3160 .hot_add_disk
= raid1_add_disk
,
3161 .hot_remove_disk
= raid1_remove_disk
,
3162 .spare_active
= raid1_spare_active
,
3163 .sync_request
= sync_request
,
3164 .resize
= raid1_resize
,
3166 .check_reshape
= raid1_reshape
,
3167 .quiesce
= raid1_quiesce
,
3168 .takeover
= raid1_takeover
,
3169 .congested
= raid1_congested
,
3170 .mergeable_bvec
= raid1_mergeable_bvec
,
3173 static int __init
raid_init(void)
3175 return register_md_personality(&raid1_personality
);
3178 static void raid_exit(void)
3180 unregister_md_personality(&raid1_personality
);
3183 module_init(raid_init
);
3184 module_exit(raid_exit
);
3185 MODULE_LICENSE("GPL");
3186 MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3187 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3188 MODULE_ALIAS("md-raid1");
3189 MODULE_ALIAS("md-level-1");
3191 module_param(max_queued_requests
, int, S_IRUGO
|S_IWUSR
);