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 CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
94 #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
95 #define NEXT_NORMALIO_DISTANCE (3 * RESYNC_WINDOW_SECTORS)
97 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
99 struct pool_info *pi = data;
100 struct r1bio *r1_bio;
105 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
110 * Allocate bios : 1 for reading, n-1 for writing
112 for (j = pi->raid_disks ; j-- ; ) {
113 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
116 r1_bio->bios[j] = bio;
119 * Allocate RESYNC_PAGES data pages and attach them to
121 * If this is a user-requested check/repair, allocate
122 * RESYNC_PAGES for each bio.
124 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
125 need_pages = pi->raid_disks;
128 for (j = 0; j < need_pages; j++) {
129 bio = r1_bio->bios[j];
130 bio->bi_vcnt = RESYNC_PAGES;
132 if (bio_alloc_pages(bio, gfp_flags))
135 /* If not user-requests, copy the page pointers to all bios */
136 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
137 for (i=0; i<RESYNC_PAGES ; i++)
138 for (j=1; j<pi->raid_disks; j++)
139 r1_bio->bios[j]->bi_io_vec[i].bv_page =
140 r1_bio->bios[0]->bi_io_vec[i].bv_page;
143 r1_bio->master_bio = NULL;
149 bio_free_pages(r1_bio->bios[j]);
152 while (++j < pi->raid_disks)
153 bio_put(r1_bio->bios[j]);
154 r1bio_pool_free(r1_bio, data);
158 static void r1buf_pool_free(void *__r1_bio, void *data)
160 struct pool_info *pi = data;
162 struct r1bio *r1bio = __r1_bio;
164 for (i = 0; i < RESYNC_PAGES; i++)
165 for (j = pi->raid_disks; j-- ;) {
167 r1bio->bios[j]->bi_io_vec[i].bv_page !=
168 r1bio->bios[0]->bi_io_vec[i].bv_page)
169 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
171 for (i=0 ; i < pi->raid_disks; i++)
172 bio_put(r1bio->bios[i]);
174 r1bio_pool_free(r1bio, data);
177 static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
181 for (i = 0; i < conf->raid_disks * 2; i++) {
182 struct bio **bio = r1_bio->bios + i;
183 if (!BIO_SPECIAL(*bio))
189 static void free_r1bio(struct r1bio *r1_bio)
191 struct r1conf *conf = r1_bio->mddev->private;
193 put_all_bios(conf, r1_bio);
194 mempool_free(r1_bio, conf->r1bio_pool);
197 static void put_buf(struct r1bio *r1_bio)
199 struct r1conf *conf = r1_bio->mddev->private;
202 for (i = 0; i < conf->raid_disks * 2; i++) {
203 struct bio *bio = r1_bio->bios[i];
205 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
208 mempool_free(r1_bio, conf->r1buf_pool);
213 static void reschedule_retry(struct r1bio *r1_bio)
216 struct mddev *mddev = r1_bio->mddev;
217 struct r1conf *conf = mddev->private;
219 spin_lock_irqsave(&conf->device_lock, flags);
220 list_add(&r1_bio->retry_list, &conf->retry_list);
222 spin_unlock_irqrestore(&conf->device_lock, flags);
224 wake_up(&conf->wait_barrier);
225 md_wakeup_thread(mddev->thread);
229 * raid_end_bio_io() is called when we have finished servicing a mirrored
230 * operation and are ready to return a success/failure code to the buffer
233 static void call_bio_endio(struct r1bio *r1_bio)
235 struct bio *bio = r1_bio->master_bio;
237 struct r1conf *conf = r1_bio->mddev->private;
238 sector_t start_next_window = r1_bio->start_next_window;
239 sector_t bi_sector = bio->bi_iter.bi_sector;
241 if (bio->bi_phys_segments) {
243 spin_lock_irqsave(&conf->device_lock, flags);
244 bio->bi_phys_segments--;
245 done = (bio->bi_phys_segments == 0);
246 spin_unlock_irqrestore(&conf->device_lock, flags);
248 * make_request() might be waiting for
249 * bi_phys_segments to decrease
251 wake_up(&conf->wait_barrier);
255 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
256 bio->bi_error = -EIO;
261 * Wake up any possible resync thread that waits for the device
264 allow_barrier(conf, start_next_window, bi_sector);
268 static void raid_end_bio_io(struct r1bio *r1_bio)
270 struct bio *bio = r1_bio->master_bio;
272 /* if nobody has done the final endio yet, do it now */
273 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
274 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
275 (bio_data_dir(bio) == WRITE) ? "write" : "read",
276 (unsigned long long) bio->bi_iter.bi_sector,
277 (unsigned long long) bio_end_sector(bio) - 1);
279 call_bio_endio(r1_bio);
285 * Update disk head position estimator based on IRQ completion info.
287 static inline void update_head_pos(int disk, struct r1bio *r1_bio)
289 struct r1conf *conf = r1_bio->mddev->private;
291 conf->mirrors[disk].head_position =
292 r1_bio->sector + (r1_bio->sectors);
296 * Find the disk number which triggered given bio
298 static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
301 struct r1conf *conf = r1_bio->mddev->private;
302 int raid_disks = conf->raid_disks;
304 for (mirror = 0; mirror < raid_disks * 2; mirror++)
305 if (r1_bio->bios[mirror] == bio)
308 BUG_ON(mirror == raid_disks * 2);
309 update_head_pos(mirror, r1_bio);
314 static void raid1_end_read_request(struct bio *bio)
316 int uptodate = !bio->bi_error;
317 struct r1bio *r1_bio = bio->bi_private;
318 struct r1conf *conf = r1_bio->mddev->private;
319 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;
322 * this branch is our 'one mirror IO has finished' event handler:
324 update_head_pos(r1_bio->read_disk, r1_bio);
327 set_bit(R1BIO_Uptodate, &r1_bio->state);
329 /* If all other devices have failed, we want to return
330 * the error upwards rather than fail the last device.
331 * Here we redefine "uptodate" to mean "Don't want to retry"
334 spin_lock_irqsave(&conf->device_lock, flags);
335 if (r1_bio->mddev->degraded == conf->raid_disks ||
336 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
337 test_bit(In_sync, &rdev->flags)))
339 spin_unlock_irqrestore(&conf->device_lock, flags);
343 raid_end_bio_io(r1_bio);
344 rdev_dec_pending(rdev, conf->mddev);
349 char b[BDEVNAME_SIZE];
351 KERN_ERR "md/raid1:%s: %s: "
352 "rescheduling sector %llu\n",
356 (unsigned long long)r1_bio->sector);
357 set_bit(R1BIO_ReadError, &r1_bio->state);
358 reschedule_retry(r1_bio);
359 /* don't drop the reference on read_disk yet */
363 static void close_write(struct r1bio *r1_bio)
365 /* it really is the end of this request */
366 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
367 /* free extra copy of the data pages */
368 int i = r1_bio->behind_page_count;
370 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
371 kfree(r1_bio->behind_bvecs);
372 r1_bio->behind_bvecs = NULL;
374 /* clear the bitmap if all writes complete successfully */
375 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
377 !test_bit(R1BIO_Degraded, &r1_bio->state),
378 test_bit(R1BIO_BehindIO, &r1_bio->state));
379 md_write_end(r1_bio->mddev);
382 static void r1_bio_write_done(struct r1bio *r1_bio)
384 if (!atomic_dec_and_test(&r1_bio->remaining))
387 if (test_bit(R1BIO_WriteError, &r1_bio->state))
388 reschedule_retry(r1_bio);
391 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
392 reschedule_retry(r1_bio);
394 raid_end_bio_io(r1_bio);
398 static void raid1_end_write_request(struct bio *bio)
400 struct r1bio *r1_bio = bio->bi_private;
401 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
402 struct r1conf *conf = r1_bio->mddev->private;
403 struct bio *to_put = NULL;
404 int mirror = find_bio_disk(r1_bio, bio);
405 struct md_rdev *rdev = conf->mirrors[mirror].rdev;
408 discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD;
411 * 'one mirror IO has finished' event handler:
413 if (bio->bi_error && !discard_error) {
414 set_bit(WriteErrorSeen, &rdev->flags);
415 if (!test_and_set_bit(WantReplacement, &rdev->flags))
416 set_bit(MD_RECOVERY_NEEDED, &
417 conf->mddev->recovery);
419 set_bit(R1BIO_WriteError, &r1_bio->state);
422 * Set R1BIO_Uptodate in our master bio, so that we
423 * will return a good error code for to the higher
424 * levels even if IO on some other mirrored buffer
427 * The 'master' represents the composite IO operation
428 * to user-side. So if something waits for IO, then it
429 * will wait for the 'master' bio.
434 r1_bio->bios[mirror] = NULL;
437 * Do not set R1BIO_Uptodate if the current device is
438 * rebuilding or Faulty. This is because we cannot use
439 * such device for properly reading the data back (we could
440 * potentially use it, if the current write would have felt
441 * before rdev->recovery_offset, but for simplicity we don't
444 if (test_bit(In_sync, &rdev->flags) &&
445 !test_bit(Faulty, &rdev->flags))
446 set_bit(R1BIO_Uptodate, &r1_bio->state);
448 /* Maybe we can clear some bad blocks. */
449 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
450 &first_bad, &bad_sectors) && !discard_error) {
451 r1_bio->bios[mirror] = IO_MADE_GOOD;
452 set_bit(R1BIO_MadeGood, &r1_bio->state);
457 if (test_bit(WriteMostly, &rdev->flags))
458 atomic_dec(&r1_bio->behind_remaining);
461 * In behind mode, we ACK the master bio once the I/O
462 * has safely reached all non-writemostly
463 * disks. Setting the Returned bit ensures that this
464 * gets done only once -- we don't ever want to return
465 * -EIO here, instead we'll wait
467 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
468 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
469 /* Maybe we can return now */
470 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
471 struct bio *mbio = r1_bio->master_bio;
472 pr_debug("raid1: behind end write sectors"
474 (unsigned long long) mbio->bi_iter.bi_sector,
475 (unsigned long long) bio_end_sector(mbio) - 1);
476 call_bio_endio(r1_bio);
480 if (r1_bio->bios[mirror] == NULL)
481 rdev_dec_pending(rdev, conf->mddev);
484 * Let's see if all mirrored write operations have finished
487 r1_bio_write_done(r1_bio);
494 * This routine returns the disk from which the requested read should
495 * be done. There is a per-array 'next expected sequential IO' sector
496 * number - if this matches on the next IO then we use the last disk.
497 * There is also a per-disk 'last know head position' sector that is
498 * maintained from IRQ contexts, both the normal and the resync IO
499 * completion handlers update this position correctly. If there is no
500 * perfect sequential match then we pick the disk whose head is closest.
502 * If there are 2 mirrors in the same 2 devices, performance degrades
503 * because position is mirror, not device based.
505 * The rdev for the device selected will have nr_pending incremented.
507 static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
509 const sector_t this_sector = r1_bio->sector;
511 int best_good_sectors;
512 int best_disk, best_dist_disk, best_pending_disk;
516 unsigned int min_pending;
517 struct md_rdev *rdev;
519 int choose_next_idle;
523 * Check if we can balance. We can balance on the whole
524 * device if no resync is going on, or below the resync window.
525 * We take the first readable disk when above the resync window.
528 sectors = r1_bio->sectors;
531 best_dist = MaxSector;
532 best_pending_disk = -1;
533 min_pending = UINT_MAX;
534 best_good_sectors = 0;
536 choose_next_idle = 0;
538 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
539 (mddev_is_clustered(conf->mddev) &&
540 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
541 this_sector + sectors)))
546 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
550 unsigned int pending;
553 rdev = rcu_dereference(conf->mirrors[disk].rdev);
554 if (r1_bio->bios[disk] == IO_BLOCKED
556 || test_bit(Faulty, &rdev->flags))
558 if (!test_bit(In_sync, &rdev->flags) &&
559 rdev->recovery_offset < this_sector + sectors)
561 if (test_bit(WriteMostly, &rdev->flags)) {
562 /* Don't balance among write-mostly, just
563 * use the first as a last resort */
564 if (best_dist_disk < 0) {
565 if (is_badblock(rdev, this_sector, sectors,
566 &first_bad, &bad_sectors)) {
567 if (first_bad <= this_sector)
568 /* Cannot use this */
570 best_good_sectors = first_bad - this_sector;
572 best_good_sectors = sectors;
573 best_dist_disk = disk;
574 best_pending_disk = disk;
578 /* This is a reasonable device to use. It might
581 if (is_badblock(rdev, this_sector, sectors,
582 &first_bad, &bad_sectors)) {
583 if (best_dist < MaxSector)
584 /* already have a better device */
586 if (first_bad <= this_sector) {
587 /* cannot read here. If this is the 'primary'
588 * device, then we must not read beyond
589 * bad_sectors from another device..
591 bad_sectors -= (this_sector - first_bad);
592 if (choose_first && sectors > bad_sectors)
593 sectors = bad_sectors;
594 if (best_good_sectors > sectors)
595 best_good_sectors = sectors;
598 sector_t good_sectors = first_bad - this_sector;
599 if (good_sectors > best_good_sectors) {
600 best_good_sectors = good_sectors;
608 best_good_sectors = sectors;
610 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
611 has_nonrot_disk |= nonrot;
612 pending = atomic_read(&rdev->nr_pending);
613 dist = abs(this_sector - conf->mirrors[disk].head_position);
618 /* Don't change to another disk for sequential reads */
619 if (conf->mirrors[disk].next_seq_sect == this_sector
621 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
622 struct raid1_info *mirror = &conf->mirrors[disk];
626 * If buffered sequential IO size exceeds optimal
627 * iosize, check if there is idle disk. If yes, choose
628 * the idle disk. read_balance could already choose an
629 * idle disk before noticing it's a sequential IO in
630 * this disk. This doesn't matter because this disk
631 * will idle, next time it will be utilized after the
632 * first disk has IO size exceeds optimal iosize. In
633 * this way, iosize of the first disk will be optimal
634 * iosize at least. iosize of the second disk might be
635 * small, but not a big deal since when the second disk
636 * starts IO, the first disk is likely still busy.
638 if (nonrot && opt_iosize > 0 &&
639 mirror->seq_start != MaxSector &&
640 mirror->next_seq_sect > opt_iosize &&
641 mirror->next_seq_sect - opt_iosize >=
643 choose_next_idle = 1;
648 /* If device is idle, use it */
654 if (choose_next_idle)
657 if (min_pending > pending) {
658 min_pending = pending;
659 best_pending_disk = disk;
662 if (dist < best_dist) {
664 best_dist_disk = disk;
669 * If all disks are rotational, choose the closest disk. If any disk is
670 * non-rotational, choose the disk with less pending request even the
671 * disk is rotational, which might/might not be optimal for raids with
672 * mixed ratation/non-rotational disks depending on workload.
674 if (best_disk == -1) {
676 best_disk = best_pending_disk;
678 best_disk = best_dist_disk;
681 if (best_disk >= 0) {
682 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
685 atomic_inc(&rdev->nr_pending);
686 sectors = best_good_sectors;
688 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
689 conf->mirrors[best_disk].seq_start = this_sector;
691 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
694 *max_sectors = sectors;
699 static int raid1_congested(struct mddev *mddev, int bits)
701 struct r1conf *conf = mddev->private;
704 if ((bits & (1 << WB_async_congested)) &&
705 conf->pending_count >= max_queued_requests)
709 for (i = 0; i < conf->raid_disks * 2; i++) {
710 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
711 if (rdev && !test_bit(Faulty, &rdev->flags)) {
712 struct request_queue *q = bdev_get_queue(rdev->bdev);
716 /* Note the '|| 1' - when read_balance prefers
717 * non-congested targets, it can be removed
719 if ((bits & (1 << WB_async_congested)) || 1)
720 ret |= bdi_congested(&q->backing_dev_info, bits);
722 ret &= bdi_congested(&q->backing_dev_info, bits);
729 static void flush_pending_writes(struct r1conf *conf)
731 /* Any writes that have been queued but are awaiting
732 * bitmap updates get flushed here.
734 spin_lock_irq(&conf->device_lock);
736 if (conf->pending_bio_list.head) {
738 bio = bio_list_get(&conf->pending_bio_list);
739 conf->pending_count = 0;
740 spin_unlock_irq(&conf->device_lock);
741 /* flush any pending bitmap writes to
742 * disk before proceeding w/ I/O */
743 bitmap_unplug(conf->mddev->bitmap);
744 wake_up(&conf->wait_barrier);
746 while (bio) { /* submit pending writes */
747 struct bio *next = bio->bi_next;
749 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
750 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
754 generic_make_request(bio);
758 spin_unlock_irq(&conf->device_lock);
762 * Sometimes we need to suspend IO while we do something else,
763 * either some resync/recovery, or reconfigure the array.
764 * To do this we raise a 'barrier'.
765 * The 'barrier' is a counter that can be raised multiple times
766 * to count how many activities are happening which preclude
768 * We can only raise the barrier if there is no pending IO.
769 * i.e. if nr_pending == 0.
770 * We choose only to raise the barrier if no-one is waiting for the
771 * barrier to go down. This means that as soon as an IO request
772 * is ready, no other operations which require a barrier will start
773 * until the IO request has had a chance.
775 * So: regular IO calls 'wait_barrier'. When that returns there
776 * is no backgroup IO happening, It must arrange to call
777 * allow_barrier when it has finished its IO.
778 * backgroup IO calls must call raise_barrier. Once that returns
779 * there is no normal IO happeing. It must arrange to call
780 * lower_barrier when the particular background IO completes.
782 static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
784 spin_lock_irq(&conf->resync_lock);
786 /* Wait until no block IO is waiting */
787 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
790 /* block any new IO from starting */
792 conf->next_resync = sector_nr;
794 /* For these conditions we must wait:
795 * A: while the array is in frozen state
796 * B: while barrier >= RESYNC_DEPTH, meaning resync reach
797 * the max count which allowed.
798 * C: next_resync + RESYNC_SECTORS > start_next_window, meaning
799 * next resync will reach to the window which normal bios are
801 * D: while there are any active requests in the current window.
803 wait_event_lock_irq(conf->wait_barrier,
804 !conf->array_frozen &&
805 conf->barrier < RESYNC_DEPTH &&
806 conf->current_window_requests == 0 &&
807 (conf->start_next_window >=
808 conf->next_resync + RESYNC_SECTORS),
812 spin_unlock_irq(&conf->resync_lock);
815 static void lower_barrier(struct r1conf *conf)
818 BUG_ON(conf->barrier <= 0);
819 spin_lock_irqsave(&conf->resync_lock, flags);
822 spin_unlock_irqrestore(&conf->resync_lock, flags);
823 wake_up(&conf->wait_barrier);
826 static bool need_to_wait_for_sync(struct r1conf *conf, struct bio *bio)
830 if (conf->array_frozen || !bio)
832 else if (conf->barrier && bio_data_dir(bio) == WRITE) {
833 if ((conf->mddev->curr_resync_completed
834 >= bio_end_sector(bio)) ||
835 (conf->next_resync + NEXT_NORMALIO_DISTANCE
836 <= bio->bi_iter.bi_sector))
845 static sector_t wait_barrier(struct r1conf *conf, struct bio *bio)
849 spin_lock_irq(&conf->resync_lock);
850 if (need_to_wait_for_sync(conf, bio)) {
852 /* Wait for the barrier to drop.
853 * However if there are already pending
854 * requests (preventing the barrier from
855 * rising completely), and the
856 * per-process bio queue isn't empty,
857 * then don't wait, as we need to empty
858 * that queue to allow conf->start_next_window
861 wait_event_lock_irq(conf->wait_barrier,
862 !conf->array_frozen &&
864 ((conf->start_next_window <
865 conf->next_resync + RESYNC_SECTORS) &&
867 !bio_list_empty(current->bio_list))),
872 if (bio && bio_data_dir(bio) == WRITE) {
873 if (bio->bi_iter.bi_sector >= conf->next_resync) {
874 if (conf->start_next_window == MaxSector)
875 conf->start_next_window =
877 NEXT_NORMALIO_DISTANCE;
879 if ((conf->start_next_window + NEXT_NORMALIO_DISTANCE)
880 <= bio->bi_iter.bi_sector)
881 conf->next_window_requests++;
883 conf->current_window_requests++;
884 sector = conf->start_next_window;
889 spin_unlock_irq(&conf->resync_lock);
893 static void allow_barrier(struct r1conf *conf, sector_t start_next_window,
898 spin_lock_irqsave(&conf->resync_lock, flags);
900 if (start_next_window) {
901 if (start_next_window == conf->start_next_window) {
902 if (conf->start_next_window + NEXT_NORMALIO_DISTANCE
904 conf->next_window_requests--;
906 conf->current_window_requests--;
908 conf->current_window_requests--;
910 if (!conf->current_window_requests) {
911 if (conf->next_window_requests) {
912 conf->current_window_requests =
913 conf->next_window_requests;
914 conf->next_window_requests = 0;
915 conf->start_next_window +=
916 NEXT_NORMALIO_DISTANCE;
918 conf->start_next_window = MaxSector;
921 spin_unlock_irqrestore(&conf->resync_lock, flags);
922 wake_up(&conf->wait_barrier);
925 static void freeze_array(struct r1conf *conf, int extra)
927 /* stop syncio and normal IO and wait for everything to
929 * We wait until nr_pending match nr_queued+extra
930 * This is called in the context of one normal IO request
931 * that has failed. Thus any sync request that might be pending
932 * will be blocked by nr_pending, and we need to wait for
933 * pending IO requests to complete or be queued for re-try.
934 * Thus the number queued (nr_queued) plus this request (extra)
935 * must match the number of pending IOs (nr_pending) before
938 spin_lock_irq(&conf->resync_lock);
939 conf->array_frozen = 1;
940 wait_event_lock_irq_cmd(conf->wait_barrier,
941 conf->nr_pending == conf->nr_queued+extra,
943 flush_pending_writes(conf));
944 spin_unlock_irq(&conf->resync_lock);
946 static void unfreeze_array(struct r1conf *conf)
948 /* reverse the effect of the freeze */
949 spin_lock_irq(&conf->resync_lock);
950 conf->array_frozen = 0;
951 wake_up(&conf->wait_barrier);
952 spin_unlock_irq(&conf->resync_lock);
955 /* duplicate the data pages for behind I/O
957 static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
960 struct bio_vec *bvec;
961 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
963 if (unlikely(!bvecs))
966 bio_for_each_segment_all(bvec, bio, i) {
968 bvecs[i].bv_page = alloc_page(GFP_NOIO);
969 if (unlikely(!bvecs[i].bv_page))
971 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
972 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
973 kunmap(bvecs[i].bv_page);
974 kunmap(bvec->bv_page);
976 r1_bio->behind_bvecs = bvecs;
977 r1_bio->behind_page_count = bio->bi_vcnt;
978 set_bit(R1BIO_BehindIO, &r1_bio->state);
982 for (i = 0; i < bio->bi_vcnt; i++)
983 if (bvecs[i].bv_page)
984 put_page(bvecs[i].bv_page);
986 pr_debug("%dB behind alloc failed, doing sync I/O\n",
987 bio->bi_iter.bi_size);
990 struct raid1_plug_cb {
991 struct blk_plug_cb cb;
992 struct bio_list pending;
996 static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
998 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1000 struct mddev *mddev = plug->cb.data;
1001 struct r1conf *conf = mddev->private;
1004 if (from_schedule || current->bio_list) {
1005 spin_lock_irq(&conf->device_lock);
1006 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1007 conf->pending_count += plug->pending_cnt;
1008 spin_unlock_irq(&conf->device_lock);
1009 wake_up(&conf->wait_barrier);
1010 md_wakeup_thread(mddev->thread);
1015 /* we aren't scheduling, so we can do the write-out directly. */
1016 bio = bio_list_get(&plug->pending);
1017 bitmap_unplug(mddev->bitmap);
1018 wake_up(&conf->wait_barrier);
1020 while (bio) { /* submit pending writes */
1021 struct bio *next = bio->bi_next;
1022 bio->bi_next = NULL;
1023 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
1024 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1025 /* Just ignore it */
1028 generic_make_request(bio);
1034 static void raid1_make_request(struct mddev *mddev, struct bio * bio)
1036 struct r1conf *conf = mddev->private;
1037 struct raid1_info *mirror;
1038 struct r1bio *r1_bio;
1039 struct bio *read_bio;
1041 struct bitmap *bitmap;
1042 unsigned long flags;
1043 const int op = bio_op(bio);
1044 const int rw = bio_data_dir(bio);
1045 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1046 const unsigned long do_flush_fua = (bio->bi_opf &
1047 (REQ_PREFLUSH | REQ_FUA));
1048 struct md_rdev *blocked_rdev;
1049 struct blk_plug_cb *cb;
1050 struct raid1_plug_cb *plug = NULL;
1052 int sectors_handled;
1054 sector_t start_next_window;
1057 * Register the new request and wait if the reconstruction
1058 * thread has put up a bar for new requests.
1059 * Continue immediately if no resync is active currently.
1062 md_write_start(mddev, bio); /* wait on superblock update early */
1064 if (bio_data_dir(bio) == WRITE &&
1065 ((bio_end_sector(bio) > mddev->suspend_lo &&
1066 bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1067 (mddev_is_clustered(mddev) &&
1068 md_cluster_ops->area_resyncing(mddev, WRITE,
1069 bio->bi_iter.bi_sector, bio_end_sector(bio))))) {
1070 /* As the suspend_* range is controlled by
1071 * userspace, we want an interruptible
1077 prepare_to_wait(&conf->wait_barrier,
1078 &w, TASK_INTERRUPTIBLE);
1079 if (bio_end_sector(bio) <= mddev->suspend_lo ||
1080 bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1081 (mddev_is_clustered(mddev) &&
1082 !md_cluster_ops->area_resyncing(mddev, WRITE,
1083 bio->bi_iter.bi_sector, bio_end_sector(bio))))
1086 sigprocmask(SIG_BLOCK, &full, &old);
1088 sigprocmask(SIG_SETMASK, &old, NULL);
1090 finish_wait(&conf->wait_barrier, &w);
1093 start_next_window = wait_barrier(conf, bio);
1095 bitmap = mddev->bitmap;
1098 * make_request() can abort the operation when read-ahead is being
1099 * used and no empty request is available.
1102 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1104 r1_bio->master_bio = bio;
1105 r1_bio->sectors = bio_sectors(bio);
1107 r1_bio->mddev = mddev;
1108 r1_bio->sector = bio->bi_iter.bi_sector;
1110 /* We might need to issue multiple reads to different
1111 * devices if there are bad blocks around, so we keep
1112 * track of the number of reads in bio->bi_phys_segments.
1113 * If this is 0, there is only one r1_bio and no locking
1114 * will be needed when requests complete. If it is
1115 * non-zero, then it is the number of not-completed requests.
1117 bio->bi_phys_segments = 0;
1118 bio_clear_flag(bio, BIO_SEG_VALID);
1122 * read balancing logic:
1127 rdisk = read_balance(conf, r1_bio, &max_sectors);
1130 /* couldn't find anywhere to read from */
1131 raid_end_bio_io(r1_bio);
1134 mirror = conf->mirrors + rdisk;
1136 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1138 /* Reading from a write-mostly device must
1139 * take care not to over-take any writes
1142 wait_event(bitmap->behind_wait,
1143 atomic_read(&bitmap->behind_writes) == 0);
1145 r1_bio->read_disk = rdisk;
1146 r1_bio->start_next_window = 0;
1148 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1149 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
1152 r1_bio->bios[rdisk] = read_bio;
1154 read_bio->bi_iter.bi_sector = r1_bio->sector +
1155 mirror->rdev->data_offset;
1156 read_bio->bi_bdev = mirror->rdev->bdev;
1157 read_bio->bi_end_io = raid1_end_read_request;
1158 bio_set_op_attrs(read_bio, op, do_sync);
1159 read_bio->bi_private = r1_bio;
1161 if (max_sectors < r1_bio->sectors) {
1162 /* could not read all from this device, so we will
1163 * need another r1_bio.
1166 sectors_handled = (r1_bio->sector + max_sectors
1167 - bio->bi_iter.bi_sector);
1168 r1_bio->sectors = max_sectors;
1169 spin_lock_irq(&conf->device_lock);
1170 if (bio->bi_phys_segments == 0)
1171 bio->bi_phys_segments = 2;
1173 bio->bi_phys_segments++;
1174 spin_unlock_irq(&conf->device_lock);
1175 /* Cannot call generic_make_request directly
1176 * as that will be queued in __make_request
1177 * and subsequent mempool_alloc might block waiting
1178 * for it. So hand bio over to raid1d.
1180 reschedule_retry(r1_bio);
1182 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1184 r1_bio->master_bio = bio;
1185 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1187 r1_bio->mddev = mddev;
1188 r1_bio->sector = bio->bi_iter.bi_sector +
1192 generic_make_request(read_bio);
1199 if (conf->pending_count >= max_queued_requests) {
1200 md_wakeup_thread(mddev->thread);
1201 wait_event(conf->wait_barrier,
1202 conf->pending_count < max_queued_requests);
1204 /* first select target devices under rcu_lock and
1205 * inc refcount on their rdev. Record them by setting
1207 * If there are known/acknowledged bad blocks on any device on
1208 * which we have seen a write error, we want to avoid writing those
1210 * This potentially requires several writes to write around
1211 * the bad blocks. Each set of writes gets it's own r1bio
1212 * with a set of bios attached.
1215 disks = conf->raid_disks * 2;
1217 r1_bio->start_next_window = start_next_window;
1218 blocked_rdev = NULL;
1220 max_sectors = r1_bio->sectors;
1221 for (i = 0; i < disks; i++) {
1222 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1223 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1224 atomic_inc(&rdev->nr_pending);
1225 blocked_rdev = rdev;
1228 r1_bio->bios[i] = NULL;
1229 if (!rdev || test_bit(Faulty, &rdev->flags)) {
1230 if (i < conf->raid_disks)
1231 set_bit(R1BIO_Degraded, &r1_bio->state);
1235 atomic_inc(&rdev->nr_pending);
1236 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1241 is_bad = is_badblock(rdev, r1_bio->sector,
1243 &first_bad, &bad_sectors);
1245 /* mustn't write here until the bad block is
1247 set_bit(BlockedBadBlocks, &rdev->flags);
1248 blocked_rdev = rdev;
1251 if (is_bad && first_bad <= r1_bio->sector) {
1252 /* Cannot write here at all */
1253 bad_sectors -= (r1_bio->sector - first_bad);
1254 if (bad_sectors < max_sectors)
1255 /* mustn't write more than bad_sectors
1256 * to other devices yet
1258 max_sectors = bad_sectors;
1259 rdev_dec_pending(rdev, mddev);
1260 /* We don't set R1BIO_Degraded as that
1261 * only applies if the disk is
1262 * missing, so it might be re-added,
1263 * and we want to know to recover this
1265 * In this case the device is here,
1266 * and the fact that this chunk is not
1267 * in-sync is recorded in the bad
1273 int good_sectors = first_bad - r1_bio->sector;
1274 if (good_sectors < max_sectors)
1275 max_sectors = good_sectors;
1278 r1_bio->bios[i] = bio;
1282 if (unlikely(blocked_rdev)) {
1283 /* Wait for this device to become unblocked */
1285 sector_t old = start_next_window;
1287 for (j = 0; j < i; j++)
1288 if (r1_bio->bios[j])
1289 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
1291 allow_barrier(conf, start_next_window, bio->bi_iter.bi_sector);
1292 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1293 start_next_window = wait_barrier(conf, bio);
1295 * We must make sure the multi r1bios of bio have
1296 * the same value of bi_phys_segments
1298 if (bio->bi_phys_segments && old &&
1299 old != start_next_window)
1300 /* Wait for the former r1bio(s) to complete */
1301 wait_event(conf->wait_barrier,
1302 bio->bi_phys_segments == 1);
1306 if (max_sectors < r1_bio->sectors) {
1307 /* We are splitting this write into multiple parts, so
1308 * we need to prepare for allocating another r1_bio.
1310 r1_bio->sectors = max_sectors;
1311 spin_lock_irq(&conf->device_lock);
1312 if (bio->bi_phys_segments == 0)
1313 bio->bi_phys_segments = 2;
1315 bio->bi_phys_segments++;
1316 spin_unlock_irq(&conf->device_lock);
1318 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
1320 atomic_set(&r1_bio->remaining, 1);
1321 atomic_set(&r1_bio->behind_remaining, 0);
1324 for (i = 0; i < disks; i++) {
1326 if (!r1_bio->bios[i])
1329 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
1330 bio_trim(mbio, r1_bio->sector - bio->bi_iter.bi_sector, max_sectors);
1334 * Not if there are too many, or cannot
1335 * allocate memory, or a reader on WriteMostly
1336 * is waiting for behind writes to flush */
1338 (atomic_read(&bitmap->behind_writes)
1339 < mddev->bitmap_info.max_write_behind) &&
1340 !waitqueue_active(&bitmap->behind_wait))
1341 alloc_behind_pages(mbio, r1_bio);
1343 bitmap_startwrite(bitmap, r1_bio->sector,
1345 test_bit(R1BIO_BehindIO,
1349 if (r1_bio->behind_bvecs) {
1350 struct bio_vec *bvec;
1354 * We trimmed the bio, so _all is legit
1356 bio_for_each_segment_all(bvec, mbio, j)
1357 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
1358 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1359 atomic_inc(&r1_bio->behind_remaining);
1362 r1_bio->bios[i] = mbio;
1364 mbio->bi_iter.bi_sector = (r1_bio->sector +
1365 conf->mirrors[i].rdev->data_offset);
1366 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1367 mbio->bi_end_io = raid1_end_write_request;
1368 bio_set_op_attrs(mbio, op, do_flush_fua | do_sync);
1369 mbio->bi_private = r1_bio;
1371 atomic_inc(&r1_bio->remaining);
1373 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1375 plug = container_of(cb, struct raid1_plug_cb, cb);
1378 spin_lock_irqsave(&conf->device_lock, flags);
1380 bio_list_add(&plug->pending, mbio);
1381 plug->pending_cnt++;
1383 bio_list_add(&conf->pending_bio_list, mbio);
1384 conf->pending_count++;
1386 spin_unlock_irqrestore(&conf->device_lock, flags);
1388 md_wakeup_thread(mddev->thread);
1390 /* Mustn't call r1_bio_write_done before this next test,
1391 * as it could result in the bio being freed.
1393 if (sectors_handled < bio_sectors(bio)) {
1394 r1_bio_write_done(r1_bio);
1395 /* We need another r1_bio. It has already been counted
1396 * in bio->bi_phys_segments
1398 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1399 r1_bio->master_bio = bio;
1400 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1402 r1_bio->mddev = mddev;
1403 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
1407 r1_bio_write_done(r1_bio);
1409 /* In case raid1d snuck in to freeze_array */
1410 wake_up(&conf->wait_barrier);
1413 static void raid1_status(struct seq_file *seq, struct mddev *mddev)
1415 struct r1conf *conf = mddev->private;
1418 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
1419 conf->raid_disks - mddev->degraded);
1421 for (i = 0; i < conf->raid_disks; i++) {
1422 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1423 seq_printf(seq, "%s",
1424 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1427 seq_printf(seq, "]");
1430 static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
1432 char b[BDEVNAME_SIZE];
1433 struct r1conf *conf = mddev->private;
1434 unsigned long flags;
1437 * If it is not operational, then we have already marked it as dead
1438 * else if it is the last working disks, ignore the error, let the
1439 * next level up know.
1440 * else mark the drive as failed
1442 if (test_bit(In_sync, &rdev->flags)
1443 && (conf->raid_disks - mddev->degraded) == 1) {
1445 * Don't fail the drive, act as though we were just a
1446 * normal single drive.
1447 * However don't try a recovery from this drive as
1448 * it is very likely to fail.
1450 conf->recovery_disabled = mddev->recovery_disabled;
1453 set_bit(Blocked, &rdev->flags);
1454 spin_lock_irqsave(&conf->device_lock, flags);
1455 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1457 set_bit(Faulty, &rdev->flags);
1459 set_bit(Faulty, &rdev->flags);
1460 spin_unlock_irqrestore(&conf->device_lock, flags);
1462 * if recovery is running, make sure it aborts.
1464 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1465 set_mask_bits(&mddev->flags, 0,
1466 BIT(MD_CHANGE_DEVS) | BIT(MD_CHANGE_PENDING));
1468 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1469 "md/raid1:%s: Operation continuing on %d devices.\n",
1470 mdname(mddev), bdevname(rdev->bdev, b),
1471 mdname(mddev), conf->raid_disks - mddev->degraded);
1474 static void print_conf(struct r1conf *conf)
1478 printk(KERN_DEBUG "RAID1 conf printout:\n");
1480 printk(KERN_DEBUG "(!conf)\n");
1483 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1487 for (i = 0; i < conf->raid_disks; i++) {
1488 char b[BDEVNAME_SIZE];
1489 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1491 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
1492 i, !test_bit(In_sync, &rdev->flags),
1493 !test_bit(Faulty, &rdev->flags),
1494 bdevname(rdev->bdev,b));
1499 static void close_sync(struct r1conf *conf)
1501 wait_barrier(conf, NULL);
1502 allow_barrier(conf, 0, 0);
1504 mempool_destroy(conf->r1buf_pool);
1505 conf->r1buf_pool = NULL;
1507 spin_lock_irq(&conf->resync_lock);
1508 conf->next_resync = MaxSector - 2 * NEXT_NORMALIO_DISTANCE;
1509 conf->start_next_window = MaxSector;
1510 conf->current_window_requests +=
1511 conf->next_window_requests;
1512 conf->next_window_requests = 0;
1513 spin_unlock_irq(&conf->resync_lock);
1516 static int raid1_spare_active(struct mddev *mddev)
1519 struct r1conf *conf = mddev->private;
1521 unsigned long flags;
1524 * Find all failed disks within the RAID1 configuration
1525 * and mark them readable.
1526 * Called under mddev lock, so rcu protection not needed.
1527 * device_lock used to avoid races with raid1_end_read_request
1528 * which expects 'In_sync' flags and ->degraded to be consistent.
1530 spin_lock_irqsave(&conf->device_lock, flags);
1531 for (i = 0; i < conf->raid_disks; i++) {
1532 struct md_rdev *rdev = conf->mirrors[i].rdev;
1533 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1535 && !test_bit(Candidate, &repl->flags)
1536 && repl->recovery_offset == MaxSector
1537 && !test_bit(Faulty, &repl->flags)
1538 && !test_and_set_bit(In_sync, &repl->flags)) {
1539 /* replacement has just become active */
1541 !test_and_clear_bit(In_sync, &rdev->flags))
1544 /* Replaced device not technically
1545 * faulty, but we need to be sure
1546 * it gets removed and never re-added
1548 set_bit(Faulty, &rdev->flags);
1549 sysfs_notify_dirent_safe(
1554 && rdev->recovery_offset == MaxSector
1555 && !test_bit(Faulty, &rdev->flags)
1556 && !test_and_set_bit(In_sync, &rdev->flags)) {
1558 sysfs_notify_dirent_safe(rdev->sysfs_state);
1561 mddev->degraded -= count;
1562 spin_unlock_irqrestore(&conf->device_lock, flags);
1568 static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1570 struct r1conf *conf = mddev->private;
1573 struct raid1_info *p;
1575 int last = conf->raid_disks - 1;
1577 if (mddev->recovery_disabled == conf->recovery_disabled)
1580 if (md_integrity_add_rdev(rdev, mddev))
1583 if (rdev->raid_disk >= 0)
1584 first = last = rdev->raid_disk;
1587 * find the disk ... but prefer rdev->saved_raid_disk
1590 if (rdev->saved_raid_disk >= 0 &&
1591 rdev->saved_raid_disk >= first &&
1592 rdev->saved_raid_disk < conf->raid_disks &&
1593 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1594 first = last = rdev->saved_raid_disk;
1596 for (mirror = first; mirror <= last; mirror++) {
1597 p = conf->mirrors+mirror;
1601 disk_stack_limits(mddev->gendisk, rdev->bdev,
1602 rdev->data_offset << 9);
1604 p->head_position = 0;
1605 rdev->raid_disk = mirror;
1607 /* As all devices are equivalent, we don't need a full recovery
1608 * if this was recently any drive of the array
1610 if (rdev->saved_raid_disk < 0)
1612 rcu_assign_pointer(p->rdev, rdev);
1615 if (test_bit(WantReplacement, &p->rdev->flags) &&
1616 p[conf->raid_disks].rdev == NULL) {
1617 /* Add this device as a replacement */
1618 clear_bit(In_sync, &rdev->flags);
1619 set_bit(Replacement, &rdev->flags);
1620 rdev->raid_disk = mirror;
1623 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1627 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
1628 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1633 static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1635 struct r1conf *conf = mddev->private;
1637 int number = rdev->raid_disk;
1638 struct raid1_info *p = conf->mirrors + number;
1640 if (rdev != p->rdev)
1641 p = conf->mirrors + conf->raid_disks + number;
1644 if (rdev == p->rdev) {
1645 if (test_bit(In_sync, &rdev->flags) ||
1646 atomic_read(&rdev->nr_pending)) {
1650 /* Only remove non-faulty devices if recovery
1653 if (!test_bit(Faulty, &rdev->flags) &&
1654 mddev->recovery_disabled != conf->recovery_disabled &&
1655 mddev->degraded < conf->raid_disks) {
1660 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1662 if (atomic_read(&rdev->nr_pending)) {
1663 /* lost the race, try later */
1669 if (conf->mirrors[conf->raid_disks + number].rdev) {
1670 /* We just removed a device that is being replaced.
1671 * Move down the replacement. We drain all IO before
1672 * doing this to avoid confusion.
1674 struct md_rdev *repl =
1675 conf->mirrors[conf->raid_disks + number].rdev;
1676 freeze_array(conf, 0);
1677 if (atomic_read(&repl->nr_pending)) {
1678 /* It means that some queued IO of retry_list
1679 * hold repl. Thus, we cannot set replacement
1680 * as NULL, avoiding rdev NULL pointer
1681 * dereference in sync_request_write and
1682 * handle_write_finished.
1685 unfreeze_array(conf);
1688 clear_bit(Replacement, &repl->flags);
1690 conf->mirrors[conf->raid_disks + number].rdev = NULL;
1691 unfreeze_array(conf);
1692 clear_bit(WantReplacement, &rdev->flags);
1694 clear_bit(WantReplacement, &rdev->flags);
1695 err = md_integrity_register(mddev);
1703 static void end_sync_read(struct bio *bio)
1705 struct r1bio *r1_bio = bio->bi_private;
1707 update_head_pos(r1_bio->read_disk, r1_bio);
1710 * we have read a block, now it needs to be re-written,
1711 * or re-read if the read failed.
1712 * We don't do much here, just schedule handling by raid1d
1715 set_bit(R1BIO_Uptodate, &r1_bio->state);
1717 if (atomic_dec_and_test(&r1_bio->remaining))
1718 reschedule_retry(r1_bio);
1721 static void end_sync_write(struct bio *bio)
1723 int uptodate = !bio->bi_error;
1724 struct r1bio *r1_bio = bio->bi_private;
1725 struct mddev *mddev = r1_bio->mddev;
1726 struct r1conf *conf = mddev->private;
1729 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
1732 sector_t sync_blocks = 0;
1733 sector_t s = r1_bio->sector;
1734 long sectors_to_go = r1_bio->sectors;
1735 /* make sure these bits doesn't get cleared. */
1737 bitmap_end_sync(mddev->bitmap, s,
1740 sectors_to_go -= sync_blocks;
1741 } while (sectors_to_go > 0);
1742 set_bit(WriteErrorSeen, &rdev->flags);
1743 if (!test_and_set_bit(WantReplacement, &rdev->flags))
1744 set_bit(MD_RECOVERY_NEEDED, &
1746 set_bit(R1BIO_WriteError, &r1_bio->state);
1747 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
1748 &first_bad, &bad_sectors) &&
1749 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1752 &first_bad, &bad_sectors)
1754 set_bit(R1BIO_MadeGood, &r1_bio->state);
1756 if (atomic_dec_and_test(&r1_bio->remaining)) {
1757 int s = r1_bio->sectors;
1758 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1759 test_bit(R1BIO_WriteError, &r1_bio->state))
1760 reschedule_retry(r1_bio);
1763 md_done_sync(mddev, s, uptodate);
1768 static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
1769 int sectors, struct page *page, int rw)
1771 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
1775 set_bit(WriteErrorSeen, &rdev->flags);
1776 if (!test_and_set_bit(WantReplacement,
1778 set_bit(MD_RECOVERY_NEEDED, &
1779 rdev->mddev->recovery);
1781 /* need to record an error - either for the block or the device */
1782 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1783 md_error(rdev->mddev, rdev);
1787 static int fix_sync_read_error(struct r1bio *r1_bio)
1789 /* Try some synchronous reads of other devices to get
1790 * good data, much like with normal read errors. Only
1791 * read into the pages we already have so we don't
1792 * need to re-issue the read request.
1793 * We don't need to freeze the array, because being in an
1794 * active sync request, there is no normal IO, and
1795 * no overlapping syncs.
1796 * We don't need to check is_badblock() again as we
1797 * made sure that anything with a bad block in range
1798 * will have bi_end_io clear.
1800 struct mddev *mddev = r1_bio->mddev;
1801 struct r1conf *conf = mddev->private;
1802 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1803 sector_t sect = r1_bio->sector;
1804 int sectors = r1_bio->sectors;
1809 int d = r1_bio->read_disk;
1811 struct md_rdev *rdev;
1814 if (s > (PAGE_SIZE>>9))
1817 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1818 /* No rcu protection needed here devices
1819 * can only be removed when no resync is
1820 * active, and resync is currently active
1822 rdev = conf->mirrors[d].rdev;
1823 if (sync_page_io(rdev, sect, s<<9,
1824 bio->bi_io_vec[idx].bv_page,
1825 REQ_OP_READ, 0, false)) {
1831 if (d == conf->raid_disks * 2)
1833 } while (!success && d != r1_bio->read_disk);
1836 char b[BDEVNAME_SIZE];
1838 /* Cannot read from anywhere, this block is lost.
1839 * Record a bad block on each device. If that doesn't
1840 * work just disable and interrupt the recovery.
1841 * Don't fail devices as that won't really help.
1843 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1844 " for block %llu\n",
1846 bdevname(bio->bi_bdev, b),
1847 (unsigned long long)r1_bio->sector);
1848 for (d = 0; d < conf->raid_disks * 2; d++) {
1849 rdev = conf->mirrors[d].rdev;
1850 if (!rdev || test_bit(Faulty, &rdev->flags))
1852 if (!rdev_set_badblocks(rdev, sect, s, 0))
1856 conf->recovery_disabled =
1857 mddev->recovery_disabled;
1858 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1859 md_done_sync(mddev, r1_bio->sectors, 0);
1871 /* write it back and re-read */
1872 while (d != r1_bio->read_disk) {
1874 d = conf->raid_disks * 2;
1876 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1878 rdev = conf->mirrors[d].rdev;
1879 if (r1_sync_page_io(rdev, sect, s,
1880 bio->bi_io_vec[idx].bv_page,
1882 r1_bio->bios[d]->bi_end_io = NULL;
1883 rdev_dec_pending(rdev, mddev);
1887 while (d != r1_bio->read_disk) {
1889 d = conf->raid_disks * 2;
1891 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1893 rdev = conf->mirrors[d].rdev;
1894 if (r1_sync_page_io(rdev, sect, s,
1895 bio->bi_io_vec[idx].bv_page,
1897 atomic_add(s, &rdev->corrected_errors);
1903 set_bit(R1BIO_Uptodate, &r1_bio->state);
1908 static void process_checks(struct r1bio *r1_bio)
1910 /* We have read all readable devices. If we haven't
1911 * got the block, then there is no hope left.
1912 * If we have, then we want to do a comparison
1913 * and skip the write if everything is the same.
1914 * If any blocks failed to read, then we need to
1915 * attempt an over-write
1917 struct mddev *mddev = r1_bio->mddev;
1918 struct r1conf *conf = mddev->private;
1923 /* Fix variable parts of all bios */
1924 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1925 for (i = 0; i < conf->raid_disks * 2; i++) {
1929 struct bio *b = r1_bio->bios[i];
1930 if (b->bi_end_io != end_sync_read)
1932 /* fixup the bio for reuse, but preserve errno */
1933 error = b->bi_error;
1935 b->bi_error = error;
1937 b->bi_iter.bi_size = r1_bio->sectors << 9;
1938 b->bi_iter.bi_sector = r1_bio->sector +
1939 conf->mirrors[i].rdev->data_offset;
1940 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1941 b->bi_end_io = end_sync_read;
1942 b->bi_private = r1_bio;
1944 size = b->bi_iter.bi_size;
1945 for (j = 0; j < vcnt ; j++) {
1947 bi = &b->bi_io_vec[j];
1949 if (size > PAGE_SIZE)
1950 bi->bv_len = PAGE_SIZE;
1956 for (primary = 0; primary < conf->raid_disks * 2; primary++)
1957 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1958 !r1_bio->bios[primary]->bi_error) {
1959 r1_bio->bios[primary]->bi_end_io = NULL;
1960 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1963 r1_bio->read_disk = primary;
1964 for (i = 0; i < conf->raid_disks * 2; i++) {
1966 struct bio *pbio = r1_bio->bios[primary];
1967 struct bio *sbio = r1_bio->bios[i];
1968 int error = sbio->bi_error;
1970 if (sbio->bi_end_io != end_sync_read)
1972 /* Now we can 'fixup' the error value */
1976 for (j = vcnt; j-- ; ) {
1978 p = pbio->bi_io_vec[j].bv_page;
1979 s = sbio->bi_io_vec[j].bv_page;
1980 if (memcmp(page_address(p),
1982 sbio->bi_io_vec[j].bv_len))
1988 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
1989 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1991 /* No need to write to this device. */
1992 sbio->bi_end_io = NULL;
1993 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1997 bio_copy_data(sbio, pbio);
2001 static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
2003 struct r1conf *conf = mddev->private;
2005 int disks = conf->raid_disks * 2;
2006 struct bio *bio, *wbio;
2008 bio = r1_bio->bios[r1_bio->read_disk];
2010 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2011 /* ouch - failed to read all of that. */
2012 if (!fix_sync_read_error(r1_bio))
2015 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2016 process_checks(r1_bio);
2021 atomic_set(&r1_bio->remaining, 1);
2022 for (i = 0; i < disks ; i++) {
2023 wbio = r1_bio->bios[i];
2024 if (wbio->bi_end_io == NULL ||
2025 (wbio->bi_end_io == end_sync_read &&
2026 (i == r1_bio->read_disk ||
2027 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
2030 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2031 wbio->bi_end_io = end_sync_write;
2032 atomic_inc(&r1_bio->remaining);
2033 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
2035 generic_make_request(wbio);
2038 if (atomic_dec_and_test(&r1_bio->remaining)) {
2039 /* if we're here, all write(s) have completed, so clean up */
2040 int s = r1_bio->sectors;
2041 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2042 test_bit(R1BIO_WriteError, &r1_bio->state))
2043 reschedule_retry(r1_bio);
2046 md_done_sync(mddev, s, 1);
2052 * This is a kernel thread which:
2054 * 1. Retries failed read operations on working mirrors.
2055 * 2. Updates the raid superblock when problems encounter.
2056 * 3. Performs writes following reads for array synchronising.
2059 static void fix_read_error(struct r1conf *conf, int read_disk,
2060 sector_t sect, int sectors)
2062 struct mddev *mddev = conf->mddev;
2068 struct md_rdev *rdev;
2070 if (s > (PAGE_SIZE>>9))
2078 rdev = rcu_dereference(conf->mirrors[d].rdev);
2080 (test_bit(In_sync, &rdev->flags) ||
2081 (!test_bit(Faulty, &rdev->flags) &&
2082 rdev->recovery_offset >= sect + s)) &&
2083 is_badblock(rdev, sect, s,
2084 &first_bad, &bad_sectors) == 0) {
2085 atomic_inc(&rdev->nr_pending);
2087 if (sync_page_io(rdev, sect, s<<9,
2088 conf->tmppage, REQ_OP_READ, 0, false))
2090 rdev_dec_pending(rdev, mddev);
2096 if (d == conf->raid_disks * 2)
2098 } while (!success && d != read_disk);
2101 /* Cannot read from anywhere - mark it bad */
2102 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2103 if (!rdev_set_badblocks(rdev, sect, s, 0))
2104 md_error(mddev, rdev);
2107 /* write it back and re-read */
2109 while (d != read_disk) {
2111 d = conf->raid_disks * 2;
2114 rdev = rcu_dereference(conf->mirrors[d].rdev);
2116 !test_bit(Faulty, &rdev->flags)) {
2117 atomic_inc(&rdev->nr_pending);
2119 r1_sync_page_io(rdev, sect, s,
2120 conf->tmppage, WRITE);
2121 rdev_dec_pending(rdev, mddev);
2126 while (d != read_disk) {
2127 char b[BDEVNAME_SIZE];
2129 d = conf->raid_disks * 2;
2132 rdev = rcu_dereference(conf->mirrors[d].rdev);
2134 !test_bit(Faulty, &rdev->flags)) {
2135 atomic_inc(&rdev->nr_pending);
2137 if (r1_sync_page_io(rdev, sect, s,
2138 conf->tmppage, READ)) {
2139 atomic_add(s, &rdev->corrected_errors);
2141 "md/raid1:%s: read error corrected "
2142 "(%d sectors at %llu on %s)\n",
2144 (unsigned long long)(sect +
2146 bdevname(rdev->bdev, b));
2148 rdev_dec_pending(rdev, mddev);
2157 static int narrow_write_error(struct r1bio *r1_bio, int i)
2159 struct mddev *mddev = r1_bio->mddev;
2160 struct r1conf *conf = mddev->private;
2161 struct md_rdev *rdev = conf->mirrors[i].rdev;
2163 /* bio has the data to be written to device 'i' where
2164 * we just recently had a write error.
2165 * We repeatedly clone the bio and trim down to one block,
2166 * then try the write. Where the write fails we record
2168 * It is conceivable that the bio doesn't exactly align with
2169 * blocks. We must handle this somehow.
2171 * We currently own a reference on the rdev.
2177 int sect_to_write = r1_bio->sectors;
2180 if (rdev->badblocks.shift < 0)
2183 block_sectors = roundup(1 << rdev->badblocks.shift,
2184 bdev_logical_block_size(rdev->bdev) >> 9);
2185 sector = r1_bio->sector;
2186 sectors = ((sector + block_sectors)
2187 & ~(sector_t)(block_sectors - 1))
2190 while (sect_to_write) {
2192 if (sectors > sect_to_write)
2193 sectors = sect_to_write;
2194 /* Write at 'sector' for 'sectors'*/
2196 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2197 unsigned vcnt = r1_bio->behind_page_count;
2198 struct bio_vec *vec = r1_bio->behind_bvecs;
2200 while (!vec->bv_page) {
2205 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2206 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2208 wbio->bi_vcnt = vcnt;
2210 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2213 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2214 wbio->bi_iter.bi_sector = r1_bio->sector;
2215 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
2217 bio_trim(wbio, sector - r1_bio->sector, sectors);
2218 wbio->bi_iter.bi_sector += rdev->data_offset;
2219 wbio->bi_bdev = rdev->bdev;
2221 if (submit_bio_wait(wbio) < 0)
2223 ok = rdev_set_badblocks(rdev, sector,
2228 sect_to_write -= sectors;
2230 sectors = block_sectors;
2235 static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2238 int s = r1_bio->sectors;
2239 for (m = 0; m < conf->raid_disks * 2 ; m++) {
2240 struct md_rdev *rdev = conf->mirrors[m].rdev;
2241 struct bio *bio = r1_bio->bios[m];
2242 if (bio->bi_end_io == NULL)
2244 if (!bio->bi_error &&
2245 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2246 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
2248 if (bio->bi_error &&
2249 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2250 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2251 md_error(conf->mddev, rdev);
2255 md_done_sync(conf->mddev, s, 1);
2258 static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2262 for (m = 0; m < conf->raid_disks * 2 ; m++)
2263 if (r1_bio->bios[m] == IO_MADE_GOOD) {
2264 struct md_rdev *rdev = conf->mirrors[m].rdev;
2265 rdev_clear_badblocks(rdev,
2267 r1_bio->sectors, 0);
2268 rdev_dec_pending(rdev, conf->mddev);
2269 } else if (r1_bio->bios[m] != NULL) {
2270 /* This drive got a write error. We need to
2271 * narrow down and record precise write
2275 if (!narrow_write_error(r1_bio, m)) {
2276 md_error(conf->mddev,
2277 conf->mirrors[m].rdev);
2278 /* an I/O failed, we can't clear the bitmap */
2279 set_bit(R1BIO_Degraded, &r1_bio->state);
2281 rdev_dec_pending(conf->mirrors[m].rdev,
2285 spin_lock_irq(&conf->device_lock);
2286 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2288 spin_unlock_irq(&conf->device_lock);
2289 md_wakeup_thread(conf->mddev->thread);
2291 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2292 close_write(r1_bio);
2293 raid_end_bio_io(r1_bio);
2297 static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
2301 struct mddev *mddev = conf->mddev;
2303 char b[BDEVNAME_SIZE];
2304 struct md_rdev *rdev;
2306 clear_bit(R1BIO_ReadError, &r1_bio->state);
2307 /* we got a read error. Maybe the drive is bad. Maybe just
2308 * the block and we can fix it.
2309 * We freeze all other IO, and try reading the block from
2310 * other devices. When we find one, we re-write
2311 * and check it that fixes the read error.
2312 * This is all done synchronously while the array is
2316 bio = r1_bio->bios[r1_bio->read_disk];
2317 bdevname(bio->bi_bdev, b);
2319 r1_bio->bios[r1_bio->read_disk] = NULL;
2321 if (mddev->ro == 0) {
2322 freeze_array(conf, 1);
2323 fix_read_error(conf, r1_bio->read_disk,
2324 r1_bio->sector, r1_bio->sectors);
2325 unfreeze_array(conf);
2327 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2330 rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
2333 disk = read_balance(conf, r1_bio, &max_sectors);
2335 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2336 " read error for block %llu\n",
2337 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2338 raid_end_bio_io(r1_bio);
2340 const unsigned long do_sync
2341 = r1_bio->master_bio->bi_opf & REQ_SYNC;
2342 r1_bio->read_disk = disk;
2343 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2344 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2346 r1_bio->bios[r1_bio->read_disk] = bio;
2347 rdev = conf->mirrors[disk].rdev;
2348 printk_ratelimited(KERN_ERR
2349 "md/raid1:%s: redirecting sector %llu"
2350 " to other mirror: %s\n",
2352 (unsigned long long)r1_bio->sector,
2353 bdevname(rdev->bdev, b));
2354 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
2355 bio->bi_bdev = rdev->bdev;
2356 bio->bi_end_io = raid1_end_read_request;
2357 bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
2358 bio->bi_private = r1_bio;
2359 if (max_sectors < r1_bio->sectors) {
2360 /* Drat - have to split this up more */
2361 struct bio *mbio = r1_bio->master_bio;
2362 int sectors_handled = (r1_bio->sector + max_sectors
2363 - mbio->bi_iter.bi_sector);
2364 r1_bio->sectors = max_sectors;
2365 spin_lock_irq(&conf->device_lock);
2366 if (mbio->bi_phys_segments == 0)
2367 mbio->bi_phys_segments = 2;
2369 mbio->bi_phys_segments++;
2370 spin_unlock_irq(&conf->device_lock);
2371 generic_make_request(bio);
2374 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2376 r1_bio->master_bio = mbio;
2377 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
2379 set_bit(R1BIO_ReadError, &r1_bio->state);
2380 r1_bio->mddev = mddev;
2381 r1_bio->sector = mbio->bi_iter.bi_sector +
2386 generic_make_request(bio);
2390 static void raid1d(struct md_thread *thread)
2392 struct mddev *mddev = thread->mddev;
2393 struct r1bio *r1_bio;
2394 unsigned long flags;
2395 struct r1conf *conf = mddev->private;
2396 struct list_head *head = &conf->retry_list;
2397 struct blk_plug plug;
2399 md_check_recovery(mddev);
2401 if (!list_empty_careful(&conf->bio_end_io_list) &&
2402 !test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2404 spin_lock_irqsave(&conf->device_lock, flags);
2405 if (!test_bit(MD_CHANGE_PENDING, &mddev->flags)) {
2406 while (!list_empty(&conf->bio_end_io_list)) {
2407 list_move(conf->bio_end_io_list.prev, &tmp);
2411 spin_unlock_irqrestore(&conf->device_lock, flags);
2412 while (!list_empty(&tmp)) {
2413 r1_bio = list_first_entry(&tmp, struct r1bio,
2415 list_del(&r1_bio->retry_list);
2416 if (mddev->degraded)
2417 set_bit(R1BIO_Degraded, &r1_bio->state);
2418 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2419 close_write(r1_bio);
2420 raid_end_bio_io(r1_bio);
2424 blk_start_plug(&plug);
2427 flush_pending_writes(conf);
2429 spin_lock_irqsave(&conf->device_lock, flags);
2430 if (list_empty(head)) {
2431 spin_unlock_irqrestore(&conf->device_lock, flags);
2434 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
2435 list_del(head->prev);
2437 spin_unlock_irqrestore(&conf->device_lock, flags);
2439 mddev = r1_bio->mddev;
2440 conf = mddev->private;
2441 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
2442 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2443 test_bit(R1BIO_WriteError, &r1_bio->state))
2444 handle_sync_write_finished(conf, r1_bio);
2446 sync_request_write(mddev, r1_bio);
2447 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2448 test_bit(R1BIO_WriteError, &r1_bio->state))
2449 handle_write_finished(conf, r1_bio);
2450 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2451 handle_read_error(conf, r1_bio);
2453 /* just a partial read to be scheduled from separate
2456 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
2459 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2460 md_check_recovery(mddev);
2462 blk_finish_plug(&plug);
2465 static int init_resync(struct r1conf *conf)
2469 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
2470 BUG_ON(conf->r1buf_pool);
2471 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2473 if (!conf->r1buf_pool)
2475 conf->next_resync = 0;
2480 * perform a "sync" on one "block"
2482 * We need to make sure that no normal I/O request - particularly write
2483 * requests - conflict with active sync requests.
2485 * This is achieved by tracking pending requests and a 'barrier' concept
2486 * that can be installed to exclude normal IO requests.
2489 static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2492 struct r1conf *conf = mddev->private;
2493 struct r1bio *r1_bio;
2495 sector_t max_sector, nr_sectors;
2499 int write_targets = 0, read_targets = 0;
2500 sector_t sync_blocks;
2501 int still_degraded = 0;
2502 int good_sectors = RESYNC_SECTORS;
2503 int min_bad = 0; /* number of sectors that are bad in all devices */
2505 if (!conf->r1buf_pool)
2506 if (init_resync(conf))
2509 max_sector = mddev->dev_sectors;
2510 if (sector_nr >= max_sector) {
2511 /* If we aborted, we need to abort the
2512 * sync on the 'current' bitmap chunk (there will
2513 * only be one in raid1 resync.
2514 * We can find the current addess in mddev->curr_resync
2516 if (mddev->curr_resync < max_sector) /* aborted */
2517 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2519 else /* completed sync */
2522 bitmap_close_sync(mddev->bitmap);
2525 if (mddev_is_clustered(mddev)) {
2526 conf->cluster_sync_low = 0;
2527 conf->cluster_sync_high = 0;
2532 if (mddev->bitmap == NULL &&
2533 mddev->recovery_cp == MaxSector &&
2534 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
2535 conf->fullsync == 0) {
2537 return max_sector - sector_nr;
2539 /* before building a request, check if we can skip these blocks..
2540 * This call the bitmap_start_sync doesn't actually record anything
2542 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
2543 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2544 /* We can skip this block, and probably several more */
2550 * If there is non-resync activity waiting for a turn, then let it
2551 * though before starting on this new sync request.
2553 if (conf->nr_waiting)
2554 schedule_timeout_uninterruptible(1);
2556 /* we are incrementing sector_nr below. To be safe, we check against
2557 * sector_nr + two times RESYNC_SECTORS
2560 bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2561 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
2562 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
2564 raise_barrier(conf, sector_nr);
2568 * If we get a correctably read error during resync or recovery,
2569 * we might want to read from a different device. So we
2570 * flag all drives that could conceivably be read from for READ,
2571 * and any others (which will be non-In_sync devices) for WRITE.
2572 * If a read fails, we try reading from something else for which READ
2576 r1_bio->mddev = mddev;
2577 r1_bio->sector = sector_nr;
2579 set_bit(R1BIO_IsSync, &r1_bio->state);
2581 for (i = 0; i < conf->raid_disks * 2; i++) {
2582 struct md_rdev *rdev;
2583 bio = r1_bio->bios[i];
2586 rdev = rcu_dereference(conf->mirrors[i].rdev);
2588 test_bit(Faulty, &rdev->flags)) {
2589 if (i < conf->raid_disks)
2591 } else if (!test_bit(In_sync, &rdev->flags)) {
2592 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2593 bio->bi_end_io = end_sync_write;
2596 /* may need to read from here */
2597 sector_t first_bad = MaxSector;
2600 if (is_badblock(rdev, sector_nr, good_sectors,
2601 &first_bad, &bad_sectors)) {
2602 if (first_bad > sector_nr)
2603 good_sectors = first_bad - sector_nr;
2605 bad_sectors -= (sector_nr - first_bad);
2607 min_bad > bad_sectors)
2608 min_bad = bad_sectors;
2611 if (sector_nr < first_bad) {
2612 if (test_bit(WriteMostly, &rdev->flags)) {
2619 bio_set_op_attrs(bio, REQ_OP_READ, 0);
2620 bio->bi_end_io = end_sync_read;
2622 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2623 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2624 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2626 * The device is suitable for reading (InSync),
2627 * but has bad block(s) here. Let's try to correct them,
2628 * if we are doing resync or repair. Otherwise, leave
2629 * this device alone for this sync request.
2631 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2632 bio->bi_end_io = end_sync_write;
2636 if (rdev && bio->bi_end_io) {
2637 atomic_inc(&rdev->nr_pending);
2638 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
2639 bio->bi_bdev = rdev->bdev;
2640 bio->bi_private = r1_bio;
2646 r1_bio->read_disk = disk;
2648 if (read_targets == 0 && min_bad > 0) {
2649 /* These sectors are bad on all InSync devices, so we
2650 * need to mark them bad on all write targets
2653 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
2654 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
2655 struct md_rdev *rdev = conf->mirrors[i].rdev;
2656 ok = rdev_set_badblocks(rdev, sector_nr,
2660 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2665 /* Cannot record the badblocks, so need to
2667 * If there are multiple read targets, could just
2668 * fail the really bad ones ???
2670 conf->recovery_disabled = mddev->recovery_disabled;
2671 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2677 if (min_bad > 0 && min_bad < good_sectors) {
2678 /* only resync enough to reach the next bad->good
2680 good_sectors = min_bad;
2683 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2684 /* extra read targets are also write targets */
2685 write_targets += read_targets-1;
2687 if (write_targets == 0 || read_targets == 0) {
2688 /* There is nowhere to write, so all non-sync
2689 * drives must be failed - so we are finished
2693 max_sector = sector_nr + min_bad;
2694 rv = max_sector - sector_nr;
2700 if (max_sector > mddev->resync_max)
2701 max_sector = mddev->resync_max; /* Don't do IO beyond here */
2702 if (max_sector > sector_nr + good_sectors)
2703 max_sector = sector_nr + good_sectors;
2708 int len = PAGE_SIZE;
2709 if (sector_nr + (len>>9) > max_sector)
2710 len = (max_sector - sector_nr) << 9;
2713 if (sync_blocks == 0) {
2714 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
2715 &sync_blocks, still_degraded) &&
2717 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2719 if ((len >> 9) > sync_blocks)
2720 len = sync_blocks<<9;
2723 for (i = 0 ; i < conf->raid_disks * 2; i++) {
2724 bio = r1_bio->bios[i];
2725 if (bio->bi_end_io) {
2726 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
2727 if (bio_add_page(bio, page, len, 0) == 0) {
2729 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
2732 bio = r1_bio->bios[i];
2733 if (bio->bi_end_io==NULL)
2735 /* remove last page from this bio */
2737 bio->bi_iter.bi_size -= len;
2738 bio_clear_flag(bio, BIO_SEG_VALID);
2744 nr_sectors += len>>9;
2745 sector_nr += len>>9;
2746 sync_blocks -= (len>>9);
2747 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2749 r1_bio->sectors = nr_sectors;
2751 if (mddev_is_clustered(mddev) &&
2752 conf->cluster_sync_high < sector_nr + nr_sectors) {
2753 conf->cluster_sync_low = mddev->curr_resync_completed;
2754 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2755 /* Send resync message */
2756 md_cluster_ops->resync_info_update(mddev,
2757 conf->cluster_sync_low,
2758 conf->cluster_sync_high);
2761 /* For a user-requested sync, we read all readable devices and do a
2764 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2765 atomic_set(&r1_bio->remaining, read_targets);
2766 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
2767 bio = r1_bio->bios[i];
2768 if (bio->bi_end_io == end_sync_read) {
2770 md_sync_acct(bio->bi_bdev, nr_sectors);
2771 generic_make_request(bio);
2775 atomic_set(&r1_bio->remaining, 1);
2776 bio = r1_bio->bios[r1_bio->read_disk];
2777 md_sync_acct(bio->bi_bdev, nr_sectors);
2778 generic_make_request(bio);
2784 static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
2789 return mddev->dev_sectors;
2792 static struct r1conf *setup_conf(struct mddev *mddev)
2794 struct r1conf *conf;
2796 struct raid1_info *disk;
2797 struct md_rdev *rdev;
2800 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
2804 conf->mirrors = kzalloc(sizeof(struct raid1_info)
2805 * mddev->raid_disks * 2,
2810 conf->tmppage = alloc_page(GFP_KERNEL);
2814 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
2815 if (!conf->poolinfo)
2817 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
2818 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2821 if (!conf->r1bio_pool)
2824 conf->poolinfo->mddev = mddev;
2827 spin_lock_init(&conf->device_lock);
2828 rdev_for_each(rdev, mddev) {
2829 struct request_queue *q;
2830 int disk_idx = rdev->raid_disk;
2831 if (disk_idx >= mddev->raid_disks
2834 if (test_bit(Replacement, &rdev->flags))
2835 disk = conf->mirrors + mddev->raid_disks + disk_idx;
2837 disk = conf->mirrors + disk_idx;
2842 q = bdev_get_queue(rdev->bdev);
2844 disk->head_position = 0;
2845 disk->seq_start = MaxSector;
2847 conf->raid_disks = mddev->raid_disks;
2848 conf->mddev = mddev;
2849 INIT_LIST_HEAD(&conf->retry_list);
2850 INIT_LIST_HEAD(&conf->bio_end_io_list);
2852 spin_lock_init(&conf->resync_lock);
2853 init_waitqueue_head(&conf->wait_barrier);
2855 bio_list_init(&conf->pending_bio_list);
2856 conf->pending_count = 0;
2857 conf->recovery_disabled = mddev->recovery_disabled - 1;
2859 conf->start_next_window = MaxSector;
2860 conf->current_window_requests = conf->next_window_requests = 0;
2863 for (i = 0; i < conf->raid_disks * 2; i++) {
2865 disk = conf->mirrors + i;
2867 if (i < conf->raid_disks &&
2868 disk[conf->raid_disks].rdev) {
2869 /* This slot has a replacement. */
2871 /* No original, just make the replacement
2872 * a recovering spare
2875 disk[conf->raid_disks].rdev;
2876 disk[conf->raid_disks].rdev = NULL;
2877 } else if (!test_bit(In_sync, &disk->rdev->flags))
2878 /* Original is not in_sync - bad */
2883 !test_bit(In_sync, &disk->rdev->flags)) {
2884 disk->head_position = 0;
2886 (disk->rdev->saved_raid_disk < 0))
2892 conf->thread = md_register_thread(raid1d, mddev, "raid1");
2893 if (!conf->thread) {
2895 "md/raid1:%s: couldn't allocate thread\n",
2904 mempool_destroy(conf->r1bio_pool);
2905 kfree(conf->mirrors);
2906 safe_put_page(conf->tmppage);
2907 kfree(conf->poolinfo);
2910 return ERR_PTR(err);
2913 static void raid1_free(struct mddev *mddev, void *priv);
2914 static int raid1_run(struct mddev *mddev)
2916 struct r1conf *conf;
2918 struct md_rdev *rdev;
2920 bool discard_supported = false;
2922 if (mddev->level != 1) {
2923 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
2924 mdname(mddev), mddev->level);
2927 if (mddev->reshape_position != MaxSector) {
2928 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
2933 * copy the already verified devices into our private RAID1
2934 * bookkeeping area. [whatever we allocate in run(),
2935 * should be freed in raid1_free()]
2937 if (mddev->private == NULL)
2938 conf = setup_conf(mddev);
2940 conf = mddev->private;
2943 return PTR_ERR(conf);
2946 blk_queue_max_write_same_sectors(mddev->queue, 0);
2948 rdev_for_each(rdev, mddev) {
2949 if (!mddev->gendisk)
2951 disk_stack_limits(mddev->gendisk, rdev->bdev,
2952 rdev->data_offset << 9);
2953 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2954 discard_supported = true;
2957 mddev->degraded = 0;
2958 for (i=0; i < conf->raid_disks; i++)
2959 if (conf->mirrors[i].rdev == NULL ||
2960 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2961 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2964 * RAID1 needs at least one disk in active
2966 if (conf->raid_disks - mddev->degraded < 1) {
2971 if (conf->raid_disks - mddev->degraded == 1)
2972 mddev->recovery_cp = MaxSector;
2974 if (mddev->recovery_cp != MaxSector)
2975 printk(KERN_NOTICE "md/raid1:%s: not clean"
2976 " -- starting background reconstruction\n",
2979 "md/raid1:%s: active with %d out of %d mirrors\n",
2980 mdname(mddev), mddev->raid_disks - mddev->degraded,
2984 * Ok, everything is just fine now
2986 mddev->thread = conf->thread;
2987 conf->thread = NULL;
2988 mddev->private = conf;
2990 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
2993 if (discard_supported)
2994 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
2997 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
3001 ret = md_integrity_register(mddev);
3003 md_unregister_thread(&mddev->thread);
3009 raid1_free(mddev, conf);
3013 static void raid1_free(struct mddev *mddev, void *priv)
3015 struct r1conf *conf = priv;
3017 mempool_destroy(conf->r1bio_pool);
3018 kfree(conf->mirrors);
3019 safe_put_page(conf->tmppage);
3020 kfree(conf->poolinfo);
3024 static int raid1_resize(struct mddev *mddev, sector_t sectors)
3026 /* no resync is happening, and there is enough space
3027 * on all devices, so we can resize.
3028 * We need to make sure resync covers any new space.
3029 * If the array is shrinking we should possibly wait until
3030 * any io in the removed space completes, but it hardly seems
3033 sector_t newsize = raid1_size(mddev, sectors, 0);
3034 if (mddev->external_size &&
3035 mddev->array_sectors > newsize)
3037 if (mddev->bitmap) {
3038 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
3042 md_set_array_sectors(mddev, newsize);
3043 set_capacity(mddev->gendisk, mddev->array_sectors);
3044 revalidate_disk(mddev->gendisk);
3045 if (sectors > mddev->dev_sectors &&
3046 mddev->recovery_cp > mddev->dev_sectors) {
3047 mddev->recovery_cp = mddev->dev_sectors;
3048 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3050 mddev->dev_sectors = sectors;
3051 mddev->resync_max_sectors = sectors;
3055 static int raid1_reshape(struct mddev *mddev)
3058 * 1/ resize the r1bio_pool
3059 * 2/ resize conf->mirrors
3061 * We allocate a new r1bio_pool if we can.
3062 * Then raise a device barrier and wait until all IO stops.
3063 * Then resize conf->mirrors and swap in the new r1bio pool.
3065 * At the same time, we "pack" the devices so that all the missing
3066 * devices have the higher raid_disk numbers.
3068 mempool_t *newpool, *oldpool;
3069 struct pool_info *newpoolinfo;
3070 struct raid1_info *newmirrors;
3071 struct r1conf *conf = mddev->private;
3072 int cnt, raid_disks;
3073 unsigned long flags;
3076 /* Cannot change chunk_size, layout, or level */
3077 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
3078 mddev->layout != mddev->new_layout ||
3079 mddev->level != mddev->new_level) {
3080 mddev->new_chunk_sectors = mddev->chunk_sectors;
3081 mddev->new_layout = mddev->layout;
3082 mddev->new_level = mddev->level;
3086 if (!mddev_is_clustered(mddev)) {
3087 err = md_allow_write(mddev);
3092 raid_disks = mddev->raid_disks + mddev->delta_disks;
3094 if (raid_disks < conf->raid_disks) {
3096 for (d= 0; d < conf->raid_disks; d++)
3097 if (conf->mirrors[d].rdev)
3099 if (cnt > raid_disks)
3103 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3106 newpoolinfo->mddev = mddev;
3107 newpoolinfo->raid_disks = raid_disks * 2;
3109 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3110 r1bio_pool_free, newpoolinfo);
3115 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
3119 mempool_destroy(newpool);
3123 freeze_array(conf, 0);
3125 /* ok, everything is stopped */
3126 oldpool = conf->r1bio_pool;
3127 conf->r1bio_pool = newpool;
3129 for (d = d2 = 0; d < conf->raid_disks; d++) {
3130 struct md_rdev *rdev = conf->mirrors[d].rdev;
3131 if (rdev && rdev->raid_disk != d2) {
3132 sysfs_unlink_rdev(mddev, rdev);
3133 rdev->raid_disk = d2;
3134 sysfs_unlink_rdev(mddev, rdev);
3135 if (sysfs_link_rdev(mddev, rdev))
3137 "md/raid1:%s: cannot register rd%d\n",
3138 mdname(mddev), rdev->raid_disk);
3141 newmirrors[d2++].rdev = rdev;
3143 kfree(conf->mirrors);
3144 conf->mirrors = newmirrors;
3145 kfree(conf->poolinfo);
3146 conf->poolinfo = newpoolinfo;
3148 spin_lock_irqsave(&conf->device_lock, flags);
3149 mddev->degraded += (raid_disks - conf->raid_disks);
3150 spin_unlock_irqrestore(&conf->device_lock, flags);
3151 conf->raid_disks = mddev->raid_disks = raid_disks;
3152 mddev->delta_disks = 0;
3154 unfreeze_array(conf);
3156 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3157 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3158 md_wakeup_thread(mddev->thread);
3160 mempool_destroy(oldpool);
3164 static void raid1_quiesce(struct mddev *mddev, int state)
3166 struct r1conf *conf = mddev->private;
3169 case 2: /* wake for suspend */
3170 wake_up(&conf->wait_barrier);
3173 freeze_array(conf, 0);
3176 unfreeze_array(conf);
3181 static void *raid1_takeover(struct mddev *mddev)
3183 /* raid1 can take over:
3184 * raid5 with 2 devices, any layout or chunk size
3186 if (mddev->level == 5 && mddev->raid_disks == 2) {
3187 struct r1conf *conf;
3188 mddev->new_level = 1;
3189 mddev->new_layout = 0;
3190 mddev->new_chunk_sectors = 0;
3191 conf = setup_conf(mddev);
3193 /* Array must appear to be quiesced */
3194 conf->array_frozen = 1;
3197 return ERR_PTR(-EINVAL);
3200 static struct md_personality raid1_personality =
3204 .owner = THIS_MODULE,
3205 .make_request = raid1_make_request,
3208 .status = raid1_status,
3209 .error_handler = raid1_error,
3210 .hot_add_disk = raid1_add_disk,
3211 .hot_remove_disk= raid1_remove_disk,
3212 .spare_active = raid1_spare_active,
3213 .sync_request = raid1_sync_request,
3214 .resize = raid1_resize,
3216 .check_reshape = raid1_reshape,
3217 .quiesce = raid1_quiesce,
3218 .takeover = raid1_takeover,
3219 .congested = raid1_congested,
3222 static int __init raid_init(void)
3224 return register_md_personality(&raid1_personality);
3227 static void raid_exit(void)
3229 unregister_md_personality(&raid1_personality);
3232 module_init(raid_init);
3233 module_exit(raid_exit);
3234 MODULE_LICENSE("GPL");
3235 MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3236 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3237 MODULE_ALIAS("md-raid1");
3238 MODULE_ALIAS("md-level-1");
3240 module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);