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>
40 #include <trace/events/block.h>
45 #define UNSUPPORTED_MDDEV_FLAGS \
46 ((1L << MD_HAS_JOURNAL) | \
47 (1L << MD_JOURNAL_CLEAN))
50 * Number of guaranteed r1bios in case of extreme VM load:
52 #define NR_RAID1_BIOS 256
54 /* when we get a read error on a read-only array, we redirect to another
55 * device without failing the first device, or trying to over-write to
56 * correct the read error. To keep track of bad blocks on a per-bio
57 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
59 #define IO_BLOCKED ((struct bio *)1)
60 /* When we successfully write to a known bad-block, we need to remove the
61 * bad-block marking which must be done from process context. So we record
62 * the success by setting devs[n].bio to IO_MADE_GOOD
64 #define IO_MADE_GOOD ((struct bio *)2)
66 #define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
68 /* When there are this many requests queue to be written by
69 * the raid1 thread, we become 'congested' to provide back-pressure
72 static int max_queued_requests = 1024;
74 static void allow_barrier(struct r1conf *conf, sector_t sector_nr);
75 static void lower_barrier(struct r1conf *conf, sector_t sector_nr);
77 #define raid1_log(md, fmt, args...) \
78 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0)
80 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
82 struct pool_info *pi = data;
83 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
85 /* allocate a r1bio with room for raid_disks entries in the bios array */
86 return kzalloc(size, gfp_flags);
89 static void r1bio_pool_free(void *r1_bio, void *data)
94 #define RESYNC_BLOCK_SIZE (64*1024)
95 #define RESYNC_DEPTH 32
96 #define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
97 #define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
98 #define RESYNC_WINDOW (RESYNC_BLOCK_SIZE * RESYNC_DEPTH)
99 #define RESYNC_WINDOW_SECTORS (RESYNC_WINDOW >> 9)
100 #define CLUSTER_RESYNC_WINDOW (16 * RESYNC_WINDOW)
101 #define CLUSTER_RESYNC_WINDOW_SECTORS (CLUSTER_RESYNC_WINDOW >> 9)
103 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
105 struct pool_info *pi = data;
106 struct r1bio *r1_bio;
111 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
116 * Allocate bios : 1 for reading, n-1 for writing
118 for (j = pi->raid_disks ; j-- ; ) {
119 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
122 r1_bio->bios[j] = bio;
125 * Allocate RESYNC_PAGES data pages and attach them to
127 * If this is a user-requested check/repair, allocate
128 * RESYNC_PAGES for each bio.
130 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
131 need_pages = pi->raid_disks;
134 for (j = 0; j < need_pages; j++) {
135 bio = r1_bio->bios[j];
136 bio->bi_vcnt = RESYNC_PAGES;
138 if (bio_alloc_pages(bio, gfp_flags))
141 /* If not user-requests, copy the page pointers to all bios */
142 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
143 for (i=0; i<RESYNC_PAGES ; i++)
144 for (j=1; j<pi->raid_disks; j++)
145 r1_bio->bios[j]->bi_io_vec[i].bv_page =
146 r1_bio->bios[0]->bi_io_vec[i].bv_page;
149 r1_bio->master_bio = NULL;
155 bio_free_pages(r1_bio->bios[j]);
158 while (++j < pi->raid_disks)
159 bio_put(r1_bio->bios[j]);
160 r1bio_pool_free(r1_bio, data);
164 static void r1buf_pool_free(void *__r1_bio, void *data)
166 struct pool_info *pi = data;
168 struct r1bio *r1bio = __r1_bio;
170 for (i = 0; i < RESYNC_PAGES; i++)
171 for (j = pi->raid_disks; j-- ;) {
173 r1bio->bios[j]->bi_io_vec[i].bv_page !=
174 r1bio->bios[0]->bi_io_vec[i].bv_page)
175 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
177 for (i=0 ; i < pi->raid_disks; i++)
178 bio_put(r1bio->bios[i]);
180 r1bio_pool_free(r1bio, data);
183 static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
187 for (i = 0; i < conf->raid_disks * 2; i++) {
188 struct bio **bio = r1_bio->bios + i;
189 if (!BIO_SPECIAL(*bio))
195 static void free_r1bio(struct r1bio *r1_bio)
197 struct r1conf *conf = r1_bio->mddev->private;
199 put_all_bios(conf, r1_bio);
200 mempool_free(r1_bio, conf->r1bio_pool);
203 static void put_buf(struct r1bio *r1_bio)
205 struct r1conf *conf = r1_bio->mddev->private;
206 sector_t sect = r1_bio->sector;
209 for (i = 0; i < conf->raid_disks * 2; i++) {
210 struct bio *bio = r1_bio->bios[i];
212 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
215 mempool_free(r1_bio, conf->r1buf_pool);
217 lower_barrier(conf, sect);
220 static void reschedule_retry(struct r1bio *r1_bio)
223 struct mddev *mddev = r1_bio->mddev;
224 struct r1conf *conf = mddev->private;
227 idx = sector_to_idx(r1_bio->sector);
228 spin_lock_irqsave(&conf->device_lock, flags);
229 list_add(&r1_bio->retry_list, &conf->retry_list);
230 atomic_inc(&conf->nr_queued[idx]);
231 spin_unlock_irqrestore(&conf->device_lock, flags);
233 wake_up(&conf->wait_barrier);
234 md_wakeup_thread(mddev->thread);
238 * raid_end_bio_io() is called when we have finished servicing a mirrored
239 * operation and are ready to return a success/failure code to the buffer
242 static void call_bio_endio(struct r1bio *r1_bio)
244 struct bio *bio = r1_bio->master_bio;
246 struct r1conf *conf = r1_bio->mddev->private;
247 sector_t bi_sector = bio->bi_iter.bi_sector;
249 if (bio->bi_phys_segments) {
251 spin_lock_irqsave(&conf->device_lock, flags);
252 bio->bi_phys_segments--;
253 done = (bio->bi_phys_segments == 0);
254 spin_unlock_irqrestore(&conf->device_lock, flags);
256 * make_request() might be waiting for
257 * bi_phys_segments to decrease
259 wake_up(&conf->wait_barrier);
263 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
264 bio->bi_error = -EIO;
269 * Wake up any possible resync thread that waits for the device
272 allow_barrier(conf, bi_sector);
276 static void raid_end_bio_io(struct r1bio *r1_bio)
278 struct bio *bio = r1_bio->master_bio;
280 /* if nobody has done the final endio yet, do it now */
281 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
282 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
283 (bio_data_dir(bio) == WRITE) ? "write" : "read",
284 (unsigned long long) bio->bi_iter.bi_sector,
285 (unsigned long long) bio_end_sector(bio) - 1);
287 call_bio_endio(r1_bio);
293 * Update disk head position estimator based on IRQ completion info.
295 static inline void update_head_pos(int disk, struct r1bio *r1_bio)
297 struct r1conf *conf = r1_bio->mddev->private;
299 conf->mirrors[disk].head_position =
300 r1_bio->sector + (r1_bio->sectors);
304 * Find the disk number which triggered given bio
306 static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
309 struct r1conf *conf = r1_bio->mddev->private;
310 int raid_disks = conf->raid_disks;
312 for (mirror = 0; mirror < raid_disks * 2; mirror++)
313 if (r1_bio->bios[mirror] == bio)
316 BUG_ON(mirror == raid_disks * 2);
317 update_head_pos(mirror, r1_bio);
322 static void raid1_end_read_request(struct bio *bio)
324 int uptodate = !bio->bi_error;
325 struct r1bio *r1_bio = bio->bi_private;
326 struct r1conf *conf = r1_bio->mddev->private;
327 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev;
330 * this branch is our 'one mirror IO has finished' event handler:
332 update_head_pos(r1_bio->read_disk, r1_bio);
335 set_bit(R1BIO_Uptodate, &r1_bio->state);
336 else if (test_bit(FailFast, &rdev->flags) &&
337 test_bit(R1BIO_FailFast, &r1_bio->state))
338 /* This was a fail-fast read so we definitely
342 /* If all other devices have failed, we want to return
343 * the error upwards rather than fail the last device.
344 * Here we redefine "uptodate" to mean "Don't want to retry"
347 spin_lock_irqsave(&conf->device_lock, flags);
348 if (r1_bio->mddev->degraded == conf->raid_disks ||
349 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
350 test_bit(In_sync, &rdev->flags)))
352 spin_unlock_irqrestore(&conf->device_lock, flags);
356 raid_end_bio_io(r1_bio);
357 rdev_dec_pending(rdev, conf->mddev);
362 char b[BDEVNAME_SIZE];
363 pr_err_ratelimited("md/raid1:%s: %s: rescheduling sector %llu\n",
365 bdevname(rdev->bdev, b),
366 (unsigned long long)r1_bio->sector);
367 set_bit(R1BIO_ReadError, &r1_bio->state);
368 reschedule_retry(r1_bio);
369 /* don't drop the reference on read_disk yet */
373 static void close_write(struct r1bio *r1_bio)
375 /* it really is the end of this request */
376 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
377 /* free extra copy of the data pages */
378 int i = r1_bio->behind_page_count;
380 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
381 kfree(r1_bio->behind_bvecs);
382 r1_bio->behind_bvecs = NULL;
384 /* clear the bitmap if all writes complete successfully */
385 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
387 !test_bit(R1BIO_Degraded, &r1_bio->state),
388 test_bit(R1BIO_BehindIO, &r1_bio->state));
389 md_write_end(r1_bio->mddev);
392 static void r1_bio_write_done(struct r1bio *r1_bio)
394 if (!atomic_dec_and_test(&r1_bio->remaining))
397 if (test_bit(R1BIO_WriteError, &r1_bio->state))
398 reschedule_retry(r1_bio);
401 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
402 reschedule_retry(r1_bio);
404 raid_end_bio_io(r1_bio);
408 static void raid1_end_write_request(struct bio *bio)
410 struct r1bio *r1_bio = bio->bi_private;
411 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
412 struct r1conf *conf = r1_bio->mddev->private;
413 struct bio *to_put = NULL;
414 int mirror = find_bio_disk(r1_bio, bio);
415 struct md_rdev *rdev = conf->mirrors[mirror].rdev;
418 discard_error = bio->bi_error && bio_op(bio) == REQ_OP_DISCARD;
421 * 'one mirror IO has finished' event handler:
423 if (bio->bi_error && !discard_error) {
424 set_bit(WriteErrorSeen, &rdev->flags);
425 if (!test_and_set_bit(WantReplacement, &rdev->flags))
426 set_bit(MD_RECOVERY_NEEDED, &
427 conf->mddev->recovery);
429 if (test_bit(FailFast, &rdev->flags) &&
430 (bio->bi_opf & MD_FAILFAST) &&
431 /* We never try FailFast to WriteMostly devices */
432 !test_bit(WriteMostly, &rdev->flags)) {
433 md_error(r1_bio->mddev, rdev);
434 if (!test_bit(Faulty, &rdev->flags))
435 /* This is the only remaining device,
436 * We need to retry the write without
439 set_bit(R1BIO_WriteError, &r1_bio->state);
441 /* Finished with this branch */
442 r1_bio->bios[mirror] = NULL;
446 set_bit(R1BIO_WriteError, &r1_bio->state);
449 * Set R1BIO_Uptodate in our master bio, so that we
450 * will return a good error code for to the higher
451 * levels even if IO on some other mirrored buffer
454 * The 'master' represents the composite IO operation
455 * to user-side. So if something waits for IO, then it
456 * will wait for the 'master' bio.
461 r1_bio->bios[mirror] = NULL;
464 * Do not set R1BIO_Uptodate if the current device is
465 * rebuilding or Faulty. This is because we cannot use
466 * such device for properly reading the data back (we could
467 * potentially use it, if the current write would have felt
468 * before rdev->recovery_offset, but for simplicity we don't
471 if (test_bit(In_sync, &rdev->flags) &&
472 !test_bit(Faulty, &rdev->flags))
473 set_bit(R1BIO_Uptodate, &r1_bio->state);
475 /* Maybe we can clear some bad blocks. */
476 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
477 &first_bad, &bad_sectors) && !discard_error) {
478 r1_bio->bios[mirror] = IO_MADE_GOOD;
479 set_bit(R1BIO_MadeGood, &r1_bio->state);
484 if (test_bit(WriteMostly, &rdev->flags))
485 atomic_dec(&r1_bio->behind_remaining);
488 * In behind mode, we ACK the master bio once the I/O
489 * has safely reached all non-writemostly
490 * disks. Setting the Returned bit ensures that this
491 * gets done only once -- we don't ever want to return
492 * -EIO here, instead we'll wait
494 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
495 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
496 /* Maybe we can return now */
497 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
498 struct bio *mbio = r1_bio->master_bio;
499 pr_debug("raid1: behind end write sectors"
501 (unsigned long long) mbio->bi_iter.bi_sector,
502 (unsigned long long) bio_end_sector(mbio) - 1);
503 call_bio_endio(r1_bio);
507 if (r1_bio->bios[mirror] == NULL)
508 rdev_dec_pending(rdev, conf->mddev);
511 * Let's see if all mirrored write operations have finished
514 r1_bio_write_done(r1_bio);
520 static sector_t align_to_barrier_unit_end(sector_t start_sector,
525 WARN_ON(sectors == 0);
527 * len is the number of sectors from start_sector to end of the
528 * barrier unit which start_sector belongs to.
530 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) -
540 * This routine returns the disk from which the requested read should
541 * be done. There is a per-array 'next expected sequential IO' sector
542 * number - if this matches on the next IO then we use the last disk.
543 * There is also a per-disk 'last know head position' sector that is
544 * maintained from IRQ contexts, both the normal and the resync IO
545 * completion handlers update this position correctly. If there is no
546 * perfect sequential match then we pick the disk whose head is closest.
548 * If there are 2 mirrors in the same 2 devices, performance degrades
549 * because position is mirror, not device based.
551 * The rdev for the device selected will have nr_pending incremented.
553 static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
555 const sector_t this_sector = r1_bio->sector;
557 int best_good_sectors;
558 int best_disk, best_dist_disk, best_pending_disk;
562 unsigned int min_pending;
563 struct md_rdev *rdev;
565 int choose_next_idle;
569 * Check if we can balance. We can balance on the whole
570 * device if no resync is going on, or below the resync window.
571 * We take the first readable disk when above the resync window.
574 sectors = r1_bio->sectors;
577 best_dist = MaxSector;
578 best_pending_disk = -1;
579 min_pending = UINT_MAX;
580 best_good_sectors = 0;
582 choose_next_idle = 0;
583 clear_bit(R1BIO_FailFast, &r1_bio->state);
585 if ((conf->mddev->recovery_cp < this_sector + sectors) ||
586 (mddev_is_clustered(conf->mddev) &&
587 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector,
588 this_sector + sectors)))
593 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
597 unsigned int pending;
600 rdev = rcu_dereference(conf->mirrors[disk].rdev);
601 if (r1_bio->bios[disk] == IO_BLOCKED
603 || test_bit(Faulty, &rdev->flags))
605 if (!test_bit(In_sync, &rdev->flags) &&
606 rdev->recovery_offset < this_sector + sectors)
608 if (test_bit(WriteMostly, &rdev->flags)) {
609 /* Don't balance among write-mostly, just
610 * use the first as a last resort */
611 if (best_dist_disk < 0) {
612 if (is_badblock(rdev, this_sector, sectors,
613 &first_bad, &bad_sectors)) {
614 if (first_bad <= this_sector)
615 /* Cannot use this */
617 best_good_sectors = first_bad - this_sector;
619 best_good_sectors = sectors;
620 best_dist_disk = disk;
621 best_pending_disk = disk;
625 /* This is a reasonable device to use. It might
628 if (is_badblock(rdev, this_sector, sectors,
629 &first_bad, &bad_sectors)) {
630 if (best_dist < MaxSector)
631 /* already have a better device */
633 if (first_bad <= this_sector) {
634 /* cannot read here. If this is the 'primary'
635 * device, then we must not read beyond
636 * bad_sectors from another device..
638 bad_sectors -= (this_sector - first_bad);
639 if (choose_first && sectors > bad_sectors)
640 sectors = bad_sectors;
641 if (best_good_sectors > sectors)
642 best_good_sectors = sectors;
645 sector_t good_sectors = first_bad - this_sector;
646 if (good_sectors > best_good_sectors) {
647 best_good_sectors = good_sectors;
655 best_good_sectors = sectors;
658 /* At least two disks to choose from so failfast is OK */
659 set_bit(R1BIO_FailFast, &r1_bio->state);
661 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
662 has_nonrot_disk |= nonrot;
663 pending = atomic_read(&rdev->nr_pending);
664 dist = abs(this_sector - conf->mirrors[disk].head_position);
669 /* Don't change to another disk for sequential reads */
670 if (conf->mirrors[disk].next_seq_sect == this_sector
672 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
673 struct raid1_info *mirror = &conf->mirrors[disk];
677 * If buffered sequential IO size exceeds optimal
678 * iosize, check if there is idle disk. If yes, choose
679 * the idle disk. read_balance could already choose an
680 * idle disk before noticing it's a sequential IO in
681 * this disk. This doesn't matter because this disk
682 * will idle, next time it will be utilized after the
683 * first disk has IO size exceeds optimal iosize. In
684 * this way, iosize of the first disk will be optimal
685 * iosize at least. iosize of the second disk might be
686 * small, but not a big deal since when the second disk
687 * starts IO, the first disk is likely still busy.
689 if (nonrot && opt_iosize > 0 &&
690 mirror->seq_start != MaxSector &&
691 mirror->next_seq_sect > opt_iosize &&
692 mirror->next_seq_sect - opt_iosize >=
694 choose_next_idle = 1;
700 if (choose_next_idle)
703 if (min_pending > pending) {
704 min_pending = pending;
705 best_pending_disk = disk;
708 if (dist < best_dist) {
710 best_dist_disk = disk;
715 * If all disks are rotational, choose the closest disk. If any disk is
716 * non-rotational, choose the disk with less pending request even the
717 * disk is rotational, which might/might not be optimal for raids with
718 * mixed ratation/non-rotational disks depending on workload.
720 if (best_disk == -1) {
721 if (has_nonrot_disk || min_pending == 0)
722 best_disk = best_pending_disk;
724 best_disk = best_dist_disk;
727 if (best_disk >= 0) {
728 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
731 atomic_inc(&rdev->nr_pending);
732 sectors = best_good_sectors;
734 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
735 conf->mirrors[best_disk].seq_start = this_sector;
737 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
740 *max_sectors = sectors;
745 static int raid1_congested(struct mddev *mddev, int bits)
747 struct r1conf *conf = mddev->private;
750 if ((bits & (1 << WB_async_congested)) &&
751 conf->pending_count >= max_queued_requests)
755 for (i = 0; i < conf->raid_disks * 2; i++) {
756 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
757 if (rdev && !test_bit(Faulty, &rdev->flags)) {
758 struct request_queue *q = bdev_get_queue(rdev->bdev);
762 /* Note the '|| 1' - when read_balance prefers
763 * non-congested targets, it can be removed
765 if ((bits & (1 << WB_async_congested)) || 1)
766 ret |= bdi_congested(&q->backing_dev_info, bits);
768 ret &= bdi_congested(&q->backing_dev_info, bits);
775 static void flush_pending_writes(struct r1conf *conf)
777 /* Any writes that have been queued but are awaiting
778 * bitmap updates get flushed here.
780 spin_lock_irq(&conf->device_lock);
782 if (conf->pending_bio_list.head) {
784 bio = bio_list_get(&conf->pending_bio_list);
785 conf->pending_count = 0;
786 spin_unlock_irq(&conf->device_lock);
787 /* flush any pending bitmap writes to
788 * disk before proceeding w/ I/O */
789 bitmap_unplug(conf->mddev->bitmap);
790 wake_up(&conf->wait_barrier);
792 while (bio) { /* submit pending writes */
793 struct bio *next = bio->bi_next;
794 struct md_rdev *rdev = (void*)bio->bi_bdev;
796 bio->bi_bdev = rdev->bdev;
797 if (test_bit(Faulty, &rdev->flags)) {
798 bio->bi_error = -EIO;
800 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
801 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
805 generic_make_request(bio);
809 spin_unlock_irq(&conf->device_lock);
813 * Sometimes we need to suspend IO while we do something else,
814 * either some resync/recovery, or reconfigure the array.
815 * To do this we raise a 'barrier'.
816 * The 'barrier' is a counter that can be raised multiple times
817 * to count how many activities are happening which preclude
819 * We can only raise the barrier if there is no pending IO.
820 * i.e. if nr_pending == 0.
821 * We choose only to raise the barrier if no-one is waiting for the
822 * barrier to go down. This means that as soon as an IO request
823 * is ready, no other operations which require a barrier will start
824 * until the IO request has had a chance.
826 * So: regular IO calls 'wait_barrier'. When that returns there
827 * is no backgroup IO happening, It must arrange to call
828 * allow_barrier when it has finished its IO.
829 * backgroup IO calls must call raise_barrier. Once that returns
830 * there is no normal IO happeing. It must arrange to call
831 * lower_barrier when the particular background IO completes.
833 static void raise_barrier(struct r1conf *conf, sector_t sector_nr)
835 int idx = sector_to_idx(sector_nr);
837 spin_lock_irq(&conf->resync_lock);
839 /* Wait until no block IO is waiting */
840 wait_event_lock_irq(conf->wait_barrier,
841 !atomic_read(&conf->nr_waiting[idx]),
844 /* block any new IO from starting */
845 atomic_inc(&conf->barrier[idx]);
847 * In raise_barrier() we firstly increase conf->barrier[idx] then
848 * check conf->nr_pending[idx]. In _wait_barrier() we firstly
849 * increase conf->nr_pending[idx] then check conf->barrier[idx].
850 * A memory barrier here to make sure conf->nr_pending[idx] won't
851 * be fetched before conf->barrier[idx] is increased. Otherwise
852 * there will be a race between raise_barrier() and _wait_barrier().
854 smp_mb__after_atomic();
856 /* For these conditions we must wait:
857 * A: while the array is in frozen state
858 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O
859 * existing in corresponding I/O barrier bucket.
860 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches
861 * max resync count which allowed on current I/O barrier bucket.
863 wait_event_lock_irq(conf->wait_barrier,
864 !conf->array_frozen &&
865 !atomic_read(&conf->nr_pending[idx]) &&
866 atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH,
869 atomic_inc(&conf->nr_pending[idx]);
870 spin_unlock_irq(&conf->resync_lock);
873 static void lower_barrier(struct r1conf *conf, sector_t sector_nr)
875 int idx = sector_to_idx(sector_nr);
877 BUG_ON(atomic_read(&conf->barrier[idx]) <= 0);
879 atomic_dec(&conf->barrier[idx]);
880 atomic_dec(&conf->nr_pending[idx]);
881 wake_up(&conf->wait_barrier);
884 static void _wait_barrier(struct r1conf *conf, int idx)
887 * We need to increase conf->nr_pending[idx] very early here,
888 * then raise_barrier() can be blocked when it waits for
889 * conf->nr_pending[idx] to be 0. Then we can avoid holding
890 * conf->resync_lock when there is no barrier raised in same
891 * barrier unit bucket. Also if the array is frozen, I/O
892 * should be blocked until array is unfrozen.
894 atomic_inc(&conf->nr_pending[idx]);
896 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then
897 * check conf->barrier[idx]. In raise_barrier() we firstly increase
898 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory
899 * barrier is necessary here to make sure conf->barrier[idx] won't be
900 * fetched before conf->nr_pending[idx] is increased. Otherwise there
901 * will be a race between _wait_barrier() and raise_barrier().
903 smp_mb__after_atomic();
906 * Don't worry about checking two atomic_t variables at same time
907 * here. If during we check conf->barrier[idx], the array is
908 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is
909 * 0, it is safe to return and make the I/O continue. Because the
910 * array is frozen, all I/O returned here will eventually complete
911 * or be queued, no race will happen. See code comment in
914 if (!READ_ONCE(conf->array_frozen) &&
915 !atomic_read(&conf->barrier[idx]))
919 * After holding conf->resync_lock, conf->nr_pending[idx]
920 * should be decreased before waiting for barrier to drop.
921 * Otherwise, we may encounter a race condition because
922 * raise_barrer() might be waiting for conf->nr_pending[idx]
923 * to be 0 at same time.
925 spin_lock_irq(&conf->resync_lock);
926 atomic_inc(&conf->nr_waiting[idx]);
927 atomic_dec(&conf->nr_pending[idx]);
929 * In case freeze_array() is waiting for
930 * get_unqueued_pending() == extra
932 wake_up(&conf->wait_barrier);
933 /* Wait for the barrier in same barrier unit bucket to drop. */
934 wait_event_lock_irq(conf->wait_barrier,
935 !conf->array_frozen &&
936 !atomic_read(&conf->barrier[idx]),
938 atomic_inc(&conf->nr_pending[idx]);
939 atomic_dec(&conf->nr_waiting[idx]);
940 spin_unlock_irq(&conf->resync_lock);
943 static void wait_read_barrier(struct r1conf *conf, sector_t sector_nr)
945 int idx = sector_to_idx(sector_nr);
948 * Very similar to _wait_barrier(). The difference is, for read
949 * I/O we don't need wait for sync I/O, but if the whole array
950 * is frozen, the read I/O still has to wait until the array is
951 * unfrozen. Since there is no ordering requirement with
952 * conf->barrier[idx] here, memory barrier is unnecessary as well.
954 atomic_inc(&conf->nr_pending[idx]);
956 if (!READ_ONCE(conf->array_frozen))
959 spin_lock_irq(&conf->resync_lock);
960 atomic_inc(&conf->nr_waiting[idx]);
961 atomic_dec(&conf->nr_pending[idx]);
963 * In case freeze_array() is waiting for
964 * get_unqueued_pending() == extra
966 wake_up(&conf->wait_barrier);
967 /* Wait for array to be unfrozen */
968 wait_event_lock_irq(conf->wait_barrier,
971 atomic_inc(&conf->nr_pending[idx]);
972 atomic_dec(&conf->nr_waiting[idx]);
973 spin_unlock_irq(&conf->resync_lock);
976 static void wait_barrier(struct r1conf *conf, sector_t sector_nr)
978 int idx = sector_to_idx(sector_nr);
980 _wait_barrier(conf, idx);
983 static void wait_all_barriers(struct r1conf *conf)
987 for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++)
988 _wait_barrier(conf, idx);
991 static void _allow_barrier(struct r1conf *conf, int idx)
993 atomic_dec(&conf->nr_pending[idx]);
994 wake_up(&conf->wait_barrier);
997 static void allow_barrier(struct r1conf *conf, sector_t sector_nr)
999 int idx = sector_to_idx(sector_nr);
1001 _allow_barrier(conf, idx);
1004 static void allow_all_barriers(struct r1conf *conf)
1008 for (idx = 0; idx < BARRIER_BUCKETS_NR; idx++)
1009 _allow_barrier(conf, idx);
1012 /* conf->resync_lock should be held */
1013 static int get_unqueued_pending(struct r1conf *conf)
1017 for (ret = 0, idx = 0; idx < BARRIER_BUCKETS_NR; idx++)
1018 ret += atomic_read(&conf->nr_pending[idx]) -
1019 atomic_read(&conf->nr_queued[idx]);
1024 static void freeze_array(struct r1conf *conf, int extra)
1026 /* Stop sync I/O and normal I/O and wait for everything to
1028 * This is called in two situations:
1029 * 1) management command handlers (reshape, remove disk, quiesce).
1030 * 2) one normal I/O request failed.
1032 * After array_frozen is set to 1, new sync IO will be blocked at
1033 * raise_barrier(), and new normal I/O will blocked at _wait_barrier()
1034 * or wait_read_barrier(). The flying I/Os will either complete or be
1035 * queued. When everything goes quite, there are only queued I/Os left.
1037 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the
1038 * barrier bucket index which this I/O request hits. When all sync and
1039 * normal I/O are queued, sum of all conf->nr_pending[] will match sum
1040 * of all conf->nr_queued[]. But normal I/O failure is an exception,
1041 * in handle_read_error(), we may call freeze_array() before trying to
1042 * fix the read error. In this case, the error read I/O is not queued,
1043 * so get_unqueued_pending() == 1.
1045 * Therefore before this function returns, we need to wait until
1046 * get_unqueued_pendings(conf) gets equal to extra. For
1047 * normal I/O context, extra is 1, in rested situations extra is 0.
1049 spin_lock_irq(&conf->resync_lock);
1050 conf->array_frozen = 1;
1051 raid1_log(conf->mddev, "wait freeze");
1052 wait_event_lock_irq_cmd(
1054 get_unqueued_pending(conf) == extra,
1056 flush_pending_writes(conf));
1057 spin_unlock_irq(&conf->resync_lock);
1059 static void unfreeze_array(struct r1conf *conf)
1061 /* reverse the effect of the freeze */
1062 spin_lock_irq(&conf->resync_lock);
1063 conf->array_frozen = 0;
1064 spin_unlock_irq(&conf->resync_lock);
1065 wake_up(&conf->wait_barrier);
1068 /* duplicate the data pages for behind I/O
1070 static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
1073 struct bio_vec *bvec;
1074 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
1076 if (unlikely(!bvecs))
1079 bio_for_each_segment_all(bvec, bio, i) {
1081 bvecs[i].bv_page = alloc_page(GFP_NOIO);
1082 if (unlikely(!bvecs[i].bv_page))
1084 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
1085 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
1086 kunmap(bvecs[i].bv_page);
1087 kunmap(bvec->bv_page);
1089 r1_bio->behind_bvecs = bvecs;
1090 r1_bio->behind_page_count = bio->bi_vcnt;
1091 set_bit(R1BIO_BehindIO, &r1_bio->state);
1095 for (i = 0; i < bio->bi_vcnt; i++)
1096 if (bvecs[i].bv_page)
1097 put_page(bvecs[i].bv_page);
1099 pr_debug("%dB behind alloc failed, doing sync I/O\n",
1100 bio->bi_iter.bi_size);
1103 struct raid1_plug_cb {
1104 struct blk_plug_cb cb;
1105 struct bio_list pending;
1109 static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
1111 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
1113 struct mddev *mddev = plug->cb.data;
1114 struct r1conf *conf = mddev->private;
1117 if (from_schedule || current->bio_list) {
1118 spin_lock_irq(&conf->device_lock);
1119 bio_list_merge(&conf->pending_bio_list, &plug->pending);
1120 conf->pending_count += plug->pending_cnt;
1121 spin_unlock_irq(&conf->device_lock);
1122 wake_up(&conf->wait_barrier);
1123 md_wakeup_thread(mddev->thread);
1128 /* we aren't scheduling, so we can do the write-out directly. */
1129 bio = bio_list_get(&plug->pending);
1130 bitmap_unplug(mddev->bitmap);
1131 wake_up(&conf->wait_barrier);
1133 while (bio) { /* submit pending writes */
1134 struct bio *next = bio->bi_next;
1135 struct md_rdev *rdev = (void*)bio->bi_bdev;
1136 bio->bi_next = NULL;
1137 bio->bi_bdev = rdev->bdev;
1138 if (test_bit(Faulty, &rdev->flags)) {
1139 bio->bi_error = -EIO;
1141 } else if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
1142 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
1143 /* Just ignore it */
1146 generic_make_request(bio);
1152 static inline struct r1bio *
1153 alloc_r1bio(struct mddev *mddev, struct bio *bio, sector_t sectors_handled)
1155 struct r1conf *conf = mddev->private;
1156 struct r1bio *r1_bio;
1158 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1160 r1_bio->master_bio = bio;
1161 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
1163 r1_bio->mddev = mddev;
1164 r1_bio->sector = bio->bi_iter.bi_sector + sectors_handled;
1169 static void raid1_read_request(struct mddev *mddev, struct bio *bio)
1171 struct r1conf *conf = mddev->private;
1172 struct raid1_info *mirror;
1173 struct r1bio *r1_bio;
1174 struct bio *read_bio;
1175 struct bitmap *bitmap = mddev->bitmap;
1176 const int op = bio_op(bio);
1177 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1178 int sectors_handled;
1183 * Still need barrier for READ in case that whole
1186 wait_read_barrier(conf, bio->bi_iter.bi_sector);
1188 r1_bio = alloc_r1bio(mddev, bio, 0);
1191 * We might need to issue multiple reads to different
1192 * devices if there are bad blocks around, so we keep
1193 * track of the number of reads in bio->bi_phys_segments.
1194 * If this is 0, there is only one r1_bio and no locking
1195 * will be needed when requests complete. If it is
1196 * non-zero, then it is the number of not-completed requests.
1198 bio->bi_phys_segments = 0;
1199 bio_clear_flag(bio, BIO_SEG_VALID);
1202 * make_request() can abort the operation when read-ahead is being
1203 * used and no empty request is available.
1206 rdisk = read_balance(conf, r1_bio, &max_sectors);
1209 /* couldn't find anywhere to read from */
1210 raid_end_bio_io(r1_bio);
1213 mirror = conf->mirrors + rdisk;
1215 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1218 * Reading from a write-mostly device must take care not to
1219 * over-take any writes that are 'behind'
1221 raid1_log(mddev, "wait behind writes");
1222 wait_event(bitmap->behind_wait,
1223 atomic_read(&bitmap->behind_writes) == 0);
1225 r1_bio->read_disk = rdisk;
1227 read_bio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set);
1228 bio_trim(read_bio, r1_bio->sector - bio->bi_iter.bi_sector,
1231 r1_bio->bios[rdisk] = read_bio;
1233 read_bio->bi_iter.bi_sector = r1_bio->sector +
1234 mirror->rdev->data_offset;
1235 read_bio->bi_bdev = mirror->rdev->bdev;
1236 read_bio->bi_end_io = raid1_end_read_request;
1237 bio_set_op_attrs(read_bio, op, do_sync);
1238 if (test_bit(FailFast, &mirror->rdev->flags) &&
1239 test_bit(R1BIO_FailFast, &r1_bio->state))
1240 read_bio->bi_opf |= MD_FAILFAST;
1241 read_bio->bi_private = r1_bio;
1244 trace_block_bio_remap(bdev_get_queue(read_bio->bi_bdev),
1245 read_bio, disk_devt(mddev->gendisk),
1248 if (max_sectors < r1_bio->sectors) {
1250 * could not read all from this device, so we will need another
1253 sectors_handled = (r1_bio->sector + max_sectors
1254 - bio->bi_iter.bi_sector);
1255 r1_bio->sectors = max_sectors;
1256 spin_lock_irq(&conf->device_lock);
1257 if (bio->bi_phys_segments == 0)
1258 bio->bi_phys_segments = 2;
1260 bio->bi_phys_segments++;
1261 spin_unlock_irq(&conf->device_lock);
1264 * Cannot call generic_make_request directly as that will be
1265 * queued in __make_request and subsequent mempool_alloc might
1266 * block waiting for it. So hand bio over to raid1d.
1268 reschedule_retry(r1_bio);
1270 r1_bio = alloc_r1bio(mddev, bio, sectors_handled);
1273 generic_make_request(read_bio);
1276 static void raid1_write_request(struct mddev *mddev, struct bio *bio)
1278 struct r1conf *conf = mddev->private;
1279 struct r1bio *r1_bio;
1281 struct bitmap *bitmap = mddev->bitmap;
1282 unsigned long flags;
1283 const int op = bio_op(bio);
1284 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC);
1285 const unsigned long do_fua = (bio->bi_opf & REQ_FUA);
1286 struct md_rdev *blocked_rdev;
1287 struct blk_plug_cb *cb;
1288 struct raid1_plug_cb *plug = NULL;
1290 int sectors_handled;
1294 * Register the new request and wait if the reconstruction
1295 * thread has put up a bar for new requests.
1296 * Continue immediately if no resync is active currently.
1299 md_write_start(mddev, bio); /* wait on superblock update early */
1301 if ((bio_end_sector(bio) > mddev->suspend_lo &&
1302 bio->bi_iter.bi_sector < mddev->suspend_hi) ||
1303 (mddev_is_clustered(mddev) &&
1304 md_cluster_ops->area_resyncing(mddev, WRITE,
1305 bio->bi_iter.bi_sector, bio_end_sector(bio)))) {
1308 * As the suspend_* range is controlled by userspace, we want
1309 * an interruptible wait.
1313 flush_signals(current);
1314 prepare_to_wait(&conf->wait_barrier,
1315 &w, TASK_INTERRUPTIBLE);
1316 if (bio_end_sector(bio) <= mddev->suspend_lo ||
1317 bio->bi_iter.bi_sector >= mddev->suspend_hi ||
1318 (mddev_is_clustered(mddev) &&
1319 !md_cluster_ops->area_resyncing(mddev, WRITE,
1320 bio->bi_iter.bi_sector,
1321 bio_end_sector(bio))))
1325 finish_wait(&conf->wait_barrier, &w);
1327 wait_barrier(conf, bio->bi_iter.bi_sector);
1329 r1_bio = alloc_r1bio(mddev, bio, 0);
1331 /* We might need to issue multiple writes to different
1332 * devices if there are bad blocks around, so we keep
1333 * track of the number of writes in bio->bi_phys_segments.
1334 * If this is 0, there is only one r1_bio and no locking
1335 * will be needed when requests complete. If it is
1336 * non-zero, then it is the number of not-completed requests.
1338 bio->bi_phys_segments = 0;
1339 bio_clear_flag(bio, BIO_SEG_VALID);
1341 if (conf->pending_count >= max_queued_requests) {
1342 md_wakeup_thread(mddev->thread);
1343 raid1_log(mddev, "wait queued");
1344 wait_event(conf->wait_barrier,
1345 conf->pending_count < max_queued_requests);
1347 /* first select target devices under rcu_lock and
1348 * inc refcount on their rdev. Record them by setting
1350 * If there are known/acknowledged bad blocks on any device on
1351 * which we have seen a write error, we want to avoid writing those
1353 * This potentially requires several writes to write around
1354 * the bad blocks. Each set of writes gets it's own r1bio
1355 * with a set of bios attached.
1358 disks = conf->raid_disks * 2;
1360 blocked_rdev = NULL;
1362 max_sectors = r1_bio->sectors;
1363 for (i = 0; i < disks; i++) {
1364 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1365 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1366 atomic_inc(&rdev->nr_pending);
1367 blocked_rdev = rdev;
1370 r1_bio->bios[i] = NULL;
1371 if (!rdev || test_bit(Faulty, &rdev->flags)) {
1372 if (i < conf->raid_disks)
1373 set_bit(R1BIO_Degraded, &r1_bio->state);
1377 atomic_inc(&rdev->nr_pending);
1378 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1383 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors,
1384 &first_bad, &bad_sectors);
1386 /* mustn't write here until the bad block is
1388 set_bit(BlockedBadBlocks, &rdev->flags);
1389 blocked_rdev = rdev;
1392 if (is_bad && first_bad <= r1_bio->sector) {
1393 /* Cannot write here at all */
1394 bad_sectors -= (r1_bio->sector - first_bad);
1395 if (bad_sectors < max_sectors)
1396 /* mustn't write more than bad_sectors
1397 * to other devices yet
1399 max_sectors = bad_sectors;
1400 rdev_dec_pending(rdev, mddev);
1401 /* We don't set R1BIO_Degraded as that
1402 * only applies if the disk is
1403 * missing, so it might be re-added,
1404 * and we want to know to recover this
1406 * In this case the device is here,
1407 * and the fact that this chunk is not
1408 * in-sync is recorded in the bad
1414 int good_sectors = first_bad - r1_bio->sector;
1415 if (good_sectors < max_sectors)
1416 max_sectors = good_sectors;
1419 r1_bio->bios[i] = bio;
1423 if (unlikely(blocked_rdev)) {
1424 /* Wait for this device to become unblocked */
1427 for (j = 0; j < i; j++)
1428 if (r1_bio->bios[j])
1429 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
1431 allow_barrier(conf, bio->bi_iter.bi_sector);
1432 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk);
1433 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1434 wait_barrier(conf, bio->bi_iter.bi_sector);
1438 if (max_sectors < r1_bio->sectors) {
1439 /* We are splitting this write into multiple parts, so
1440 * we need to prepare for allocating another r1_bio.
1442 r1_bio->sectors = max_sectors;
1443 spin_lock_irq(&conf->device_lock);
1444 if (bio->bi_phys_segments == 0)
1445 bio->bi_phys_segments = 2;
1447 bio->bi_phys_segments++;
1448 spin_unlock_irq(&conf->device_lock);
1450 sectors_handled = r1_bio->sector + max_sectors - bio->bi_iter.bi_sector;
1452 atomic_set(&r1_bio->remaining, 1);
1453 atomic_set(&r1_bio->behind_remaining, 0);
1456 for (i = 0; i < disks; i++) {
1457 struct bio *mbio = NULL;
1459 if (!r1_bio->bios[i])
1462 offset = r1_bio->sector - bio->bi_iter.bi_sector;
1466 * Not if there are too many, or cannot
1467 * allocate memory, or a reader on WriteMostly
1468 * is waiting for behind writes to flush */
1470 (atomic_read(&bitmap->behind_writes)
1471 < mddev->bitmap_info.max_write_behind) &&
1472 !waitqueue_active(&bitmap->behind_wait)) {
1473 mbio = bio_clone_bioset_partial(bio, GFP_NOIO,
1477 alloc_behind_pages(mbio, r1_bio);
1480 bitmap_startwrite(bitmap, r1_bio->sector,
1482 test_bit(R1BIO_BehindIO,
1488 if (r1_bio->behind_bvecs)
1489 mbio = bio_clone_bioset_partial(bio, GFP_NOIO,
1494 mbio = bio_clone_fast(bio, GFP_NOIO, mddev->bio_set);
1495 bio_trim(mbio, offset, max_sectors);
1499 if (r1_bio->behind_bvecs) {
1500 struct bio_vec *bvec;
1504 * We trimmed the bio, so _all is legit
1506 bio_for_each_segment_all(bvec, mbio, j)
1507 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
1508 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1509 atomic_inc(&r1_bio->behind_remaining);
1512 r1_bio->bios[i] = mbio;
1514 mbio->bi_iter.bi_sector = (r1_bio->sector +
1515 conf->mirrors[i].rdev->data_offset);
1516 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1517 mbio->bi_end_io = raid1_end_write_request;
1518 bio_set_op_attrs(mbio, op, do_fua | do_sync);
1519 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) &&
1520 !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) &&
1521 conf->raid_disks - mddev->degraded > 1)
1522 mbio->bi_opf |= MD_FAILFAST;
1523 mbio->bi_private = r1_bio;
1525 atomic_inc(&r1_bio->remaining);
1528 trace_block_bio_remap(bdev_get_queue(mbio->bi_bdev),
1529 mbio, disk_devt(mddev->gendisk),
1531 /* flush_pending_writes() needs access to the rdev so...*/
1532 mbio->bi_bdev = (void*)conf->mirrors[i].rdev;
1534 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1536 plug = container_of(cb, struct raid1_plug_cb, cb);
1539 spin_lock_irqsave(&conf->device_lock, flags);
1541 bio_list_add(&plug->pending, mbio);
1542 plug->pending_cnt++;
1544 bio_list_add(&conf->pending_bio_list, mbio);
1545 conf->pending_count++;
1547 spin_unlock_irqrestore(&conf->device_lock, flags);
1549 md_wakeup_thread(mddev->thread);
1551 /* Mustn't call r1_bio_write_done before this next test,
1552 * as it could result in the bio being freed.
1554 if (sectors_handled < bio_sectors(bio)) {
1555 r1_bio_write_done(r1_bio);
1556 /* We need another r1_bio. It has already been counted
1557 * in bio->bi_phys_segments
1559 r1_bio = alloc_r1bio(mddev, bio, sectors_handled);
1563 r1_bio_write_done(r1_bio);
1565 /* In case raid1d snuck in to freeze_array */
1566 wake_up(&conf->wait_barrier);
1569 static void raid1_make_request(struct mddev *mddev, struct bio *bio)
1574 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1575 md_flush_request(mddev, bio);
1579 /* if bio exceeds barrier unit boundary, split it */
1581 sectors = align_to_barrier_unit_end(
1582 bio->bi_iter.bi_sector, bio_sectors(bio));
1583 if (sectors < bio_sectors(bio)) {
1584 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
1585 bio_chain(split, bio);
1590 if (bio_data_dir(split) == READ)
1591 raid1_read_request(mddev, split);
1593 raid1_write_request(mddev, split);
1594 } while (split != bio);
1597 static void raid1_status(struct seq_file *seq, struct mddev *mddev)
1599 struct r1conf *conf = mddev->private;
1602 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
1603 conf->raid_disks - mddev->degraded);
1605 for (i = 0; i < conf->raid_disks; i++) {
1606 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1607 seq_printf(seq, "%s",
1608 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1611 seq_printf(seq, "]");
1614 static void raid1_error(struct mddev *mddev, struct md_rdev *rdev)
1616 char b[BDEVNAME_SIZE];
1617 struct r1conf *conf = mddev->private;
1618 unsigned long flags;
1621 * If it is not operational, then we have already marked it as dead
1622 * else if it is the last working disks, ignore the error, let the
1623 * next level up know.
1624 * else mark the drive as failed
1626 spin_lock_irqsave(&conf->device_lock, flags);
1627 if (test_bit(In_sync, &rdev->flags)
1628 && (conf->raid_disks - mddev->degraded) == 1) {
1630 * Don't fail the drive, act as though we were just a
1631 * normal single drive.
1632 * However don't try a recovery from this drive as
1633 * it is very likely to fail.
1635 conf->recovery_disabled = mddev->recovery_disabled;
1636 spin_unlock_irqrestore(&conf->device_lock, flags);
1639 set_bit(Blocked, &rdev->flags);
1640 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1642 set_bit(Faulty, &rdev->flags);
1644 set_bit(Faulty, &rdev->flags);
1645 spin_unlock_irqrestore(&conf->device_lock, flags);
1647 * if recovery is running, make sure it aborts.
1649 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1650 set_mask_bits(&mddev->sb_flags, 0,
1651 BIT(MD_SB_CHANGE_DEVS) | BIT(MD_SB_CHANGE_PENDING));
1652 pr_crit("md/raid1:%s: Disk failure on %s, disabling device.\n"
1653 "md/raid1:%s: Operation continuing on %d devices.\n",
1654 mdname(mddev), bdevname(rdev->bdev, b),
1655 mdname(mddev), conf->raid_disks - mddev->degraded);
1658 static void print_conf(struct r1conf *conf)
1662 pr_debug("RAID1 conf printout:\n");
1664 pr_debug("(!conf)\n");
1667 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
1671 for (i = 0; i < conf->raid_disks; i++) {
1672 char b[BDEVNAME_SIZE];
1673 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
1675 pr_debug(" disk %d, wo:%d, o:%d, dev:%s\n",
1676 i, !test_bit(In_sync, &rdev->flags),
1677 !test_bit(Faulty, &rdev->flags),
1678 bdevname(rdev->bdev,b));
1683 static void close_sync(struct r1conf *conf)
1685 wait_all_barriers(conf);
1686 allow_all_barriers(conf);
1688 mempool_destroy(conf->r1buf_pool);
1689 conf->r1buf_pool = NULL;
1692 static int raid1_spare_active(struct mddev *mddev)
1695 struct r1conf *conf = mddev->private;
1697 unsigned long flags;
1700 * Find all failed disks within the RAID1 configuration
1701 * and mark them readable.
1702 * Called under mddev lock, so rcu protection not needed.
1703 * device_lock used to avoid races with raid1_end_read_request
1704 * which expects 'In_sync' flags and ->degraded to be consistent.
1706 spin_lock_irqsave(&conf->device_lock, flags);
1707 for (i = 0; i < conf->raid_disks; i++) {
1708 struct md_rdev *rdev = conf->mirrors[i].rdev;
1709 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1711 && !test_bit(Candidate, &repl->flags)
1712 && repl->recovery_offset == MaxSector
1713 && !test_bit(Faulty, &repl->flags)
1714 && !test_and_set_bit(In_sync, &repl->flags)) {
1715 /* replacement has just become active */
1717 !test_and_clear_bit(In_sync, &rdev->flags))
1720 /* Replaced device not technically
1721 * faulty, but we need to be sure
1722 * it gets removed and never re-added
1724 set_bit(Faulty, &rdev->flags);
1725 sysfs_notify_dirent_safe(
1730 && rdev->recovery_offset == MaxSector
1731 && !test_bit(Faulty, &rdev->flags)
1732 && !test_and_set_bit(In_sync, &rdev->flags)) {
1734 sysfs_notify_dirent_safe(rdev->sysfs_state);
1737 mddev->degraded -= count;
1738 spin_unlock_irqrestore(&conf->device_lock, flags);
1744 static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
1746 struct r1conf *conf = mddev->private;
1749 struct raid1_info *p;
1751 int last = conf->raid_disks - 1;
1753 if (mddev->recovery_disabled == conf->recovery_disabled)
1756 if (md_integrity_add_rdev(rdev, mddev))
1759 if (rdev->raid_disk >= 0)
1760 first = last = rdev->raid_disk;
1763 * find the disk ... but prefer rdev->saved_raid_disk
1766 if (rdev->saved_raid_disk >= 0 &&
1767 rdev->saved_raid_disk >= first &&
1768 conf->mirrors[rdev->saved_raid_disk].rdev == NULL)
1769 first = last = rdev->saved_raid_disk;
1771 for (mirror = first; mirror <= last; mirror++) {
1772 p = conf->mirrors+mirror;
1776 disk_stack_limits(mddev->gendisk, rdev->bdev,
1777 rdev->data_offset << 9);
1779 p->head_position = 0;
1780 rdev->raid_disk = mirror;
1782 /* As all devices are equivalent, we don't need a full recovery
1783 * if this was recently any drive of the array
1785 if (rdev->saved_raid_disk < 0)
1787 rcu_assign_pointer(p->rdev, rdev);
1790 if (test_bit(WantReplacement, &p->rdev->flags) &&
1791 p[conf->raid_disks].rdev == NULL) {
1792 /* Add this device as a replacement */
1793 clear_bit(In_sync, &rdev->flags);
1794 set_bit(Replacement, &rdev->flags);
1795 rdev->raid_disk = mirror;
1798 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1802 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
1803 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
1808 static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
1810 struct r1conf *conf = mddev->private;
1812 int number = rdev->raid_disk;
1813 struct raid1_info *p = conf->mirrors + number;
1815 if (rdev != p->rdev)
1816 p = conf->mirrors + conf->raid_disks + number;
1819 if (rdev == p->rdev) {
1820 if (test_bit(In_sync, &rdev->flags) ||
1821 atomic_read(&rdev->nr_pending)) {
1825 /* Only remove non-faulty devices if recovery
1828 if (!test_bit(Faulty, &rdev->flags) &&
1829 mddev->recovery_disabled != conf->recovery_disabled &&
1830 mddev->degraded < conf->raid_disks) {
1835 if (!test_bit(RemoveSynchronized, &rdev->flags)) {
1837 if (atomic_read(&rdev->nr_pending)) {
1838 /* lost the race, try later */
1844 if (conf->mirrors[conf->raid_disks + number].rdev) {
1845 /* We just removed a device that is being replaced.
1846 * Move down the replacement. We drain all IO before
1847 * doing this to avoid confusion.
1849 struct md_rdev *repl =
1850 conf->mirrors[conf->raid_disks + number].rdev;
1851 freeze_array(conf, 0);
1852 clear_bit(Replacement, &repl->flags);
1854 conf->mirrors[conf->raid_disks + number].rdev = NULL;
1855 unfreeze_array(conf);
1856 clear_bit(WantReplacement, &rdev->flags);
1858 clear_bit(WantReplacement, &rdev->flags);
1859 err = md_integrity_register(mddev);
1867 static void end_sync_read(struct bio *bio)
1869 struct r1bio *r1_bio = bio->bi_private;
1871 update_head_pos(r1_bio->read_disk, r1_bio);
1874 * we have read a block, now it needs to be re-written,
1875 * or re-read if the read failed.
1876 * We don't do much here, just schedule handling by raid1d
1879 set_bit(R1BIO_Uptodate, &r1_bio->state);
1881 if (atomic_dec_and_test(&r1_bio->remaining))
1882 reschedule_retry(r1_bio);
1885 static void end_sync_write(struct bio *bio)
1887 int uptodate = !bio->bi_error;
1888 struct r1bio *r1_bio = bio->bi_private;
1889 struct mddev *mddev = r1_bio->mddev;
1890 struct r1conf *conf = mddev->private;
1893 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev;
1896 sector_t sync_blocks = 0;
1897 sector_t s = r1_bio->sector;
1898 long sectors_to_go = r1_bio->sectors;
1899 /* make sure these bits doesn't get cleared. */
1901 bitmap_end_sync(mddev->bitmap, s,
1904 sectors_to_go -= sync_blocks;
1905 } while (sectors_to_go > 0);
1906 set_bit(WriteErrorSeen, &rdev->flags);
1907 if (!test_and_set_bit(WantReplacement, &rdev->flags))
1908 set_bit(MD_RECOVERY_NEEDED, &
1910 set_bit(R1BIO_WriteError, &r1_bio->state);
1911 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors,
1912 &first_bad, &bad_sectors) &&
1913 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1916 &first_bad, &bad_sectors)
1918 set_bit(R1BIO_MadeGood, &r1_bio->state);
1920 if (atomic_dec_and_test(&r1_bio->remaining)) {
1921 int s = r1_bio->sectors;
1922 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1923 test_bit(R1BIO_WriteError, &r1_bio->state))
1924 reschedule_retry(r1_bio);
1927 md_done_sync(mddev, s, uptodate);
1932 static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
1933 int sectors, struct page *page, int rw)
1935 if (sync_page_io(rdev, sector, sectors << 9, page, rw, 0, false))
1939 set_bit(WriteErrorSeen, &rdev->flags);
1940 if (!test_and_set_bit(WantReplacement,
1942 set_bit(MD_RECOVERY_NEEDED, &
1943 rdev->mddev->recovery);
1945 /* need to record an error - either for the block or the device */
1946 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1947 md_error(rdev->mddev, rdev);
1951 static int fix_sync_read_error(struct r1bio *r1_bio)
1953 /* Try some synchronous reads of other devices to get
1954 * good data, much like with normal read errors. Only
1955 * read into the pages we already have so we don't
1956 * need to re-issue the read request.
1957 * We don't need to freeze the array, because being in an
1958 * active sync request, there is no normal IO, and
1959 * no overlapping syncs.
1960 * We don't need to check is_badblock() again as we
1961 * made sure that anything with a bad block in range
1962 * will have bi_end_io clear.
1964 struct mddev *mddev = r1_bio->mddev;
1965 struct r1conf *conf = mddev->private;
1966 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1967 sector_t sect = r1_bio->sector;
1968 int sectors = r1_bio->sectors;
1970 struct md_rdev *rdev;
1972 rdev = conf->mirrors[r1_bio->read_disk].rdev;
1973 if (test_bit(FailFast, &rdev->flags)) {
1974 /* Don't try recovering from here - just fail it
1975 * ... unless it is the last working device of course */
1976 md_error(mddev, rdev);
1977 if (test_bit(Faulty, &rdev->flags))
1978 /* Don't try to read from here, but make sure
1979 * put_buf does it's thing
1981 bio->bi_end_io = end_sync_write;
1986 int d = r1_bio->read_disk;
1990 if (s > (PAGE_SIZE>>9))
1993 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1994 /* No rcu protection needed here devices
1995 * can only be removed when no resync is
1996 * active, and resync is currently active
1998 rdev = conf->mirrors[d].rdev;
1999 if (sync_page_io(rdev, sect, s<<9,
2000 bio->bi_io_vec[idx].bv_page,
2001 REQ_OP_READ, 0, false)) {
2007 if (d == conf->raid_disks * 2)
2009 } while (!success && d != r1_bio->read_disk);
2012 char b[BDEVNAME_SIZE];
2014 /* Cannot read from anywhere, this block is lost.
2015 * Record a bad block on each device. If that doesn't
2016 * work just disable and interrupt the recovery.
2017 * Don't fail devices as that won't really help.
2019 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
2021 bdevname(bio->bi_bdev, b),
2022 (unsigned long long)r1_bio->sector);
2023 for (d = 0; d < conf->raid_disks * 2; d++) {
2024 rdev = conf->mirrors[d].rdev;
2025 if (!rdev || test_bit(Faulty, &rdev->flags))
2027 if (!rdev_set_badblocks(rdev, sect, s, 0))
2031 conf->recovery_disabled =
2032 mddev->recovery_disabled;
2033 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2034 md_done_sync(mddev, r1_bio->sectors, 0);
2046 /* write it back and re-read */
2047 while (d != r1_bio->read_disk) {
2049 d = conf->raid_disks * 2;
2051 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
2053 rdev = conf->mirrors[d].rdev;
2054 if (r1_sync_page_io(rdev, sect, s,
2055 bio->bi_io_vec[idx].bv_page,
2057 r1_bio->bios[d]->bi_end_io = NULL;
2058 rdev_dec_pending(rdev, mddev);
2062 while (d != r1_bio->read_disk) {
2064 d = conf->raid_disks * 2;
2066 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
2068 rdev = conf->mirrors[d].rdev;
2069 if (r1_sync_page_io(rdev, sect, s,
2070 bio->bi_io_vec[idx].bv_page,
2072 atomic_add(s, &rdev->corrected_errors);
2078 set_bit(R1BIO_Uptodate, &r1_bio->state);
2083 static void process_checks(struct r1bio *r1_bio)
2085 /* We have read all readable devices. If we haven't
2086 * got the block, then there is no hope left.
2087 * If we have, then we want to do a comparison
2088 * and skip the write if everything is the same.
2089 * If any blocks failed to read, then we need to
2090 * attempt an over-write
2092 struct mddev *mddev = r1_bio->mddev;
2093 struct r1conf *conf = mddev->private;
2098 /* Fix variable parts of all bios */
2099 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
2100 for (i = 0; i < conf->raid_disks * 2; i++) {
2104 struct bio *b = r1_bio->bios[i];
2105 if (b->bi_end_io != end_sync_read)
2107 /* fixup the bio for reuse, but preserve errno */
2108 error = b->bi_error;
2110 b->bi_error = error;
2112 b->bi_iter.bi_size = r1_bio->sectors << 9;
2113 b->bi_iter.bi_sector = r1_bio->sector +
2114 conf->mirrors[i].rdev->data_offset;
2115 b->bi_bdev = conf->mirrors[i].rdev->bdev;
2116 b->bi_end_io = end_sync_read;
2117 b->bi_private = r1_bio;
2119 size = b->bi_iter.bi_size;
2120 for (j = 0; j < vcnt ; j++) {
2122 bi = &b->bi_io_vec[j];
2124 if (size > PAGE_SIZE)
2125 bi->bv_len = PAGE_SIZE;
2131 for (primary = 0; primary < conf->raid_disks * 2; primary++)
2132 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
2133 !r1_bio->bios[primary]->bi_error) {
2134 r1_bio->bios[primary]->bi_end_io = NULL;
2135 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
2138 r1_bio->read_disk = primary;
2139 for (i = 0; i < conf->raid_disks * 2; i++) {
2141 struct bio *pbio = r1_bio->bios[primary];
2142 struct bio *sbio = r1_bio->bios[i];
2143 int error = sbio->bi_error;
2145 if (sbio->bi_end_io != end_sync_read)
2147 /* Now we can 'fixup' the error value */
2151 for (j = vcnt; j-- ; ) {
2153 p = pbio->bi_io_vec[j].bv_page;
2154 s = sbio->bi_io_vec[j].bv_page;
2155 if (memcmp(page_address(p),
2157 sbio->bi_io_vec[j].bv_len))
2163 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
2164 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
2166 /* No need to write to this device. */
2167 sbio->bi_end_io = NULL;
2168 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
2172 bio_copy_data(sbio, pbio);
2176 static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
2178 struct r1conf *conf = mddev->private;
2180 int disks = conf->raid_disks * 2;
2181 struct bio *bio, *wbio;
2183 bio = r1_bio->bios[r1_bio->read_disk];
2185 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
2186 /* ouch - failed to read all of that. */
2187 if (!fix_sync_read_error(r1_bio))
2190 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2191 process_checks(r1_bio);
2196 atomic_set(&r1_bio->remaining, 1);
2197 for (i = 0; i < disks ; i++) {
2198 wbio = r1_bio->bios[i];
2199 if (wbio->bi_end_io == NULL ||
2200 (wbio->bi_end_io == end_sync_read &&
2201 (i == r1_bio->read_disk ||
2202 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
2205 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2206 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags))
2207 wbio->bi_opf |= MD_FAILFAST;
2209 wbio->bi_end_io = end_sync_write;
2210 atomic_inc(&r1_bio->remaining);
2211 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
2213 generic_make_request(wbio);
2216 if (atomic_dec_and_test(&r1_bio->remaining)) {
2217 /* if we're here, all write(s) have completed, so clean up */
2218 int s = r1_bio->sectors;
2219 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2220 test_bit(R1BIO_WriteError, &r1_bio->state))
2221 reschedule_retry(r1_bio);
2224 md_done_sync(mddev, s, 1);
2230 * This is a kernel thread which:
2232 * 1. Retries failed read operations on working mirrors.
2233 * 2. Updates the raid superblock when problems encounter.
2234 * 3. Performs writes following reads for array synchronising.
2237 static void fix_read_error(struct r1conf *conf, int read_disk,
2238 sector_t sect, int sectors)
2240 struct mddev *mddev = conf->mddev;
2246 struct md_rdev *rdev;
2248 if (s > (PAGE_SIZE>>9))
2256 rdev = rcu_dereference(conf->mirrors[d].rdev);
2258 (test_bit(In_sync, &rdev->flags) ||
2259 (!test_bit(Faulty, &rdev->flags) &&
2260 rdev->recovery_offset >= sect + s)) &&
2261 is_badblock(rdev, sect, s,
2262 &first_bad, &bad_sectors) == 0) {
2263 atomic_inc(&rdev->nr_pending);
2265 if (sync_page_io(rdev, sect, s<<9,
2266 conf->tmppage, REQ_OP_READ, 0, false))
2268 rdev_dec_pending(rdev, mddev);
2274 if (d == conf->raid_disks * 2)
2276 } while (!success && d != read_disk);
2279 /* Cannot read from anywhere - mark it bad */
2280 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
2281 if (!rdev_set_badblocks(rdev, sect, s, 0))
2282 md_error(mddev, rdev);
2285 /* write it back and re-read */
2287 while (d != read_disk) {
2289 d = conf->raid_disks * 2;
2292 rdev = rcu_dereference(conf->mirrors[d].rdev);
2294 !test_bit(Faulty, &rdev->flags)) {
2295 atomic_inc(&rdev->nr_pending);
2297 r1_sync_page_io(rdev, sect, s,
2298 conf->tmppage, WRITE);
2299 rdev_dec_pending(rdev, mddev);
2304 while (d != read_disk) {
2305 char b[BDEVNAME_SIZE];
2307 d = conf->raid_disks * 2;
2310 rdev = rcu_dereference(conf->mirrors[d].rdev);
2312 !test_bit(Faulty, &rdev->flags)) {
2313 atomic_inc(&rdev->nr_pending);
2315 if (r1_sync_page_io(rdev, sect, s,
2316 conf->tmppage, READ)) {
2317 atomic_add(s, &rdev->corrected_errors);
2318 pr_info("md/raid1:%s: read error corrected (%d sectors at %llu on %s)\n",
2320 (unsigned long long)(sect +
2322 bdevname(rdev->bdev, b));
2324 rdev_dec_pending(rdev, mddev);
2333 static int narrow_write_error(struct r1bio *r1_bio, int i)
2335 struct mddev *mddev = r1_bio->mddev;
2336 struct r1conf *conf = mddev->private;
2337 struct md_rdev *rdev = conf->mirrors[i].rdev;
2339 /* bio has the data to be written to device 'i' where
2340 * we just recently had a write error.
2341 * We repeatedly clone the bio and trim down to one block,
2342 * then try the write. Where the write fails we record
2344 * It is conceivable that the bio doesn't exactly align with
2345 * blocks. We must handle this somehow.
2347 * We currently own a reference on the rdev.
2353 int sect_to_write = r1_bio->sectors;
2356 if (rdev->badblocks.shift < 0)
2359 block_sectors = roundup(1 << rdev->badblocks.shift,
2360 bdev_logical_block_size(rdev->bdev) >> 9);
2361 sector = r1_bio->sector;
2362 sectors = ((sector + block_sectors)
2363 & ~(sector_t)(block_sectors - 1))
2366 while (sect_to_write) {
2368 if (sectors > sect_to_write)
2369 sectors = sect_to_write;
2370 /* Write at 'sector' for 'sectors'*/
2372 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2373 unsigned vcnt = r1_bio->behind_page_count;
2374 struct bio_vec *vec = r1_bio->behind_bvecs;
2376 while (!vec->bv_page) {
2381 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2382 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2384 wbio->bi_vcnt = vcnt;
2386 wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO,
2390 bio_set_op_attrs(wbio, REQ_OP_WRITE, 0);
2391 wbio->bi_iter.bi_sector = r1_bio->sector;
2392 wbio->bi_iter.bi_size = r1_bio->sectors << 9;
2394 bio_trim(wbio, sector - r1_bio->sector, sectors);
2395 wbio->bi_iter.bi_sector += rdev->data_offset;
2396 wbio->bi_bdev = rdev->bdev;
2398 if (submit_bio_wait(wbio) < 0)
2400 ok = rdev_set_badblocks(rdev, sector,
2405 sect_to_write -= sectors;
2407 sectors = block_sectors;
2412 static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2415 int s = r1_bio->sectors;
2416 for (m = 0; m < conf->raid_disks * 2 ; m++) {
2417 struct md_rdev *rdev = conf->mirrors[m].rdev;
2418 struct bio *bio = r1_bio->bios[m];
2419 if (bio->bi_end_io == NULL)
2421 if (!bio->bi_error &&
2422 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
2423 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
2425 if (bio->bi_error &&
2426 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2427 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2428 md_error(conf->mddev, rdev);
2432 md_done_sync(conf->mddev, s, 1);
2435 static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
2440 for (m = 0; m < conf->raid_disks * 2 ; m++)
2441 if (r1_bio->bios[m] == IO_MADE_GOOD) {
2442 struct md_rdev *rdev = conf->mirrors[m].rdev;
2443 rdev_clear_badblocks(rdev,
2445 r1_bio->sectors, 0);
2446 rdev_dec_pending(rdev, conf->mddev);
2447 } else if (r1_bio->bios[m] != NULL) {
2448 /* This drive got a write error. We need to
2449 * narrow down and record precise write
2453 if (!narrow_write_error(r1_bio, m)) {
2454 md_error(conf->mddev,
2455 conf->mirrors[m].rdev);
2456 /* an I/O failed, we can't clear the bitmap */
2457 set_bit(R1BIO_Degraded, &r1_bio->state);
2459 rdev_dec_pending(conf->mirrors[m].rdev,
2463 spin_lock_irq(&conf->device_lock);
2464 list_add(&r1_bio->retry_list, &conf->bio_end_io_list);
2465 idx = sector_to_idx(r1_bio->sector);
2466 atomic_inc(&conf->nr_queued[idx]);
2467 spin_unlock_irq(&conf->device_lock);
2469 * In case freeze_array() is waiting for condition
2470 * get_unqueued_pending() == extra to be true.
2472 wake_up(&conf->wait_barrier);
2473 md_wakeup_thread(conf->mddev->thread);
2475 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2476 close_write(r1_bio);
2477 raid_end_bio_io(r1_bio);
2481 static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
2485 struct mddev *mddev = conf->mddev;
2487 char b[BDEVNAME_SIZE];
2488 struct md_rdev *rdev;
2490 sector_t bio_sector;
2492 clear_bit(R1BIO_ReadError, &r1_bio->state);
2493 /* we got a read error. Maybe the drive is bad. Maybe just
2494 * the block and we can fix it.
2495 * We freeze all other IO, and try reading the block from
2496 * other devices. When we find one, we re-write
2497 * and check it that fixes the read error.
2498 * This is all done synchronously while the array is
2502 bio = r1_bio->bios[r1_bio->read_disk];
2503 bdevname(bio->bi_bdev, b);
2504 bio_dev = bio->bi_bdev->bd_dev;
2505 bio_sector = conf->mirrors[r1_bio->read_disk].rdev->data_offset + r1_bio->sector;
2507 r1_bio->bios[r1_bio->read_disk] = NULL;
2509 rdev = conf->mirrors[r1_bio->read_disk].rdev;
2511 && !test_bit(FailFast, &rdev->flags)) {
2512 freeze_array(conf, 1);
2513 fix_read_error(conf, r1_bio->read_disk,
2514 r1_bio->sector, r1_bio->sectors);
2515 unfreeze_array(conf);
2517 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED;
2520 rdev_dec_pending(rdev, conf->mddev);
2523 disk = read_balance(conf, r1_bio, &max_sectors);
2525 pr_crit_ratelimited("md/raid1:%s: %s: unrecoverable I/O read error for block %llu\n",
2526 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2527 raid_end_bio_io(r1_bio);
2529 const unsigned long do_sync
2530 = r1_bio->master_bio->bi_opf & REQ_SYNC;
2531 r1_bio->read_disk = disk;
2532 bio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO,
2534 bio_trim(bio, r1_bio->sector - bio->bi_iter.bi_sector,
2536 r1_bio->bios[r1_bio->read_disk] = bio;
2537 rdev = conf->mirrors[disk].rdev;
2538 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n",
2540 (unsigned long long)r1_bio->sector,
2541 bdevname(rdev->bdev, b));
2542 bio->bi_iter.bi_sector = r1_bio->sector + rdev->data_offset;
2543 bio->bi_bdev = rdev->bdev;
2544 bio->bi_end_io = raid1_end_read_request;
2545 bio_set_op_attrs(bio, REQ_OP_READ, do_sync);
2546 if (test_bit(FailFast, &rdev->flags) &&
2547 test_bit(R1BIO_FailFast, &r1_bio->state))
2548 bio->bi_opf |= MD_FAILFAST;
2549 bio->bi_private = r1_bio;
2550 if (max_sectors < r1_bio->sectors) {
2551 /* Drat - have to split this up more */
2552 struct bio *mbio = r1_bio->master_bio;
2553 int sectors_handled = (r1_bio->sector + max_sectors
2554 - mbio->bi_iter.bi_sector);
2555 r1_bio->sectors = max_sectors;
2556 spin_lock_irq(&conf->device_lock);
2557 if (mbio->bi_phys_segments == 0)
2558 mbio->bi_phys_segments = 2;
2560 mbio->bi_phys_segments++;
2561 spin_unlock_irq(&conf->device_lock);
2562 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev),
2563 bio, bio_dev, bio_sector);
2564 generic_make_request(bio);
2567 r1_bio = alloc_r1bio(mddev, mbio, sectors_handled);
2568 set_bit(R1BIO_ReadError, &r1_bio->state);
2572 trace_block_bio_remap(bdev_get_queue(bio->bi_bdev),
2573 bio, bio_dev, bio_sector);
2574 generic_make_request(bio);
2579 static void raid1d(struct md_thread *thread)
2581 struct mddev *mddev = thread->mddev;
2582 struct r1bio *r1_bio;
2583 unsigned long flags;
2584 struct r1conf *conf = mddev->private;
2585 struct list_head *head = &conf->retry_list;
2586 struct blk_plug plug;
2589 md_check_recovery(mddev);
2591 if (!list_empty_careful(&conf->bio_end_io_list) &&
2592 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) {
2594 spin_lock_irqsave(&conf->device_lock, flags);
2595 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags))
2596 list_splice_init(&conf->bio_end_io_list, &tmp);
2597 spin_unlock_irqrestore(&conf->device_lock, flags);
2598 while (!list_empty(&tmp)) {
2599 r1_bio = list_first_entry(&tmp, struct r1bio,
2601 list_del(&r1_bio->retry_list);
2602 idx = sector_to_idx(r1_bio->sector);
2603 atomic_dec(&conf->nr_queued[idx]);
2604 if (mddev->degraded)
2605 set_bit(R1BIO_Degraded, &r1_bio->state);
2606 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2607 close_write(r1_bio);
2608 raid_end_bio_io(r1_bio);
2612 blk_start_plug(&plug);
2615 flush_pending_writes(conf);
2617 spin_lock_irqsave(&conf->device_lock, flags);
2618 if (list_empty(head)) {
2619 spin_unlock_irqrestore(&conf->device_lock, flags);
2622 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
2623 list_del(head->prev);
2624 idx = sector_to_idx(r1_bio->sector);
2625 atomic_dec(&conf->nr_queued[idx]);
2626 spin_unlock_irqrestore(&conf->device_lock, flags);
2628 mddev = r1_bio->mddev;
2629 conf = mddev->private;
2630 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
2631 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2632 test_bit(R1BIO_WriteError, &r1_bio->state))
2633 handle_sync_write_finished(conf, r1_bio);
2635 sync_request_write(mddev, r1_bio);
2636 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
2637 test_bit(R1BIO_WriteError, &r1_bio->state))
2638 handle_write_finished(conf, r1_bio);
2639 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2640 handle_read_error(conf, r1_bio);
2642 /* just a partial read to be scheduled from separate
2645 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
2648 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING))
2649 md_check_recovery(mddev);
2651 blk_finish_plug(&plug);
2654 static int init_resync(struct r1conf *conf)
2658 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
2659 BUG_ON(conf->r1buf_pool);
2660 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2662 if (!conf->r1buf_pool)
2668 * perform a "sync" on one "block"
2670 * We need to make sure that no normal I/O request - particularly write
2671 * requests - conflict with active sync requests.
2673 * This is achieved by tracking pending requests and a 'barrier' concept
2674 * that can be installed to exclude normal IO requests.
2677 static sector_t raid1_sync_request(struct mddev *mddev, sector_t sector_nr,
2680 struct r1conf *conf = mddev->private;
2681 struct r1bio *r1_bio;
2683 sector_t max_sector, nr_sectors;
2687 int write_targets = 0, read_targets = 0;
2688 sector_t sync_blocks;
2689 int still_degraded = 0;
2690 int good_sectors = RESYNC_SECTORS;
2691 int min_bad = 0; /* number of sectors that are bad in all devices */
2692 int idx = sector_to_idx(sector_nr);
2694 if (!conf->r1buf_pool)
2695 if (init_resync(conf))
2698 max_sector = mddev->dev_sectors;
2699 if (sector_nr >= max_sector) {
2700 /* If we aborted, we need to abort the
2701 * sync on the 'current' bitmap chunk (there will
2702 * only be one in raid1 resync.
2703 * We can find the current addess in mddev->curr_resync
2705 if (mddev->curr_resync < max_sector) /* aborted */
2706 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
2708 else /* completed sync */
2711 bitmap_close_sync(mddev->bitmap);
2714 if (mddev_is_clustered(mddev)) {
2715 conf->cluster_sync_low = 0;
2716 conf->cluster_sync_high = 0;
2721 if (mddev->bitmap == NULL &&
2722 mddev->recovery_cp == MaxSector &&
2723 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
2724 conf->fullsync == 0) {
2726 return max_sector - sector_nr;
2728 /* before building a request, check if we can skip these blocks..
2729 * This call the bitmap_start_sync doesn't actually record anything
2731 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
2732 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2733 /* We can skip this block, and probably several more */
2739 * If there is non-resync activity waiting for a turn, then let it
2740 * though before starting on this new sync request.
2742 if (atomic_read(&conf->nr_waiting[idx]))
2743 schedule_timeout_uninterruptible(1);
2745 /* we are incrementing sector_nr below. To be safe, we check against
2746 * sector_nr + two times RESYNC_SECTORS
2749 bitmap_cond_end_sync(mddev->bitmap, sector_nr,
2750 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high));
2751 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
2753 raise_barrier(conf, sector_nr);
2757 * If we get a correctably read error during resync or recovery,
2758 * we might want to read from a different device. So we
2759 * flag all drives that could conceivably be read from for READ,
2760 * and any others (which will be non-In_sync devices) for WRITE.
2761 * If a read fails, we try reading from something else for which READ
2765 r1_bio->mddev = mddev;
2766 r1_bio->sector = sector_nr;
2768 set_bit(R1BIO_IsSync, &r1_bio->state);
2769 /* make sure good_sectors won't go across barrier unit boundary */
2770 good_sectors = align_to_barrier_unit_end(sector_nr, good_sectors);
2772 for (i = 0; i < conf->raid_disks * 2; i++) {
2773 struct md_rdev *rdev;
2774 bio = r1_bio->bios[i];
2777 rdev = rcu_dereference(conf->mirrors[i].rdev);
2779 test_bit(Faulty, &rdev->flags)) {
2780 if (i < conf->raid_disks)
2782 } else if (!test_bit(In_sync, &rdev->flags)) {
2783 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2784 bio->bi_end_io = end_sync_write;
2787 /* may need to read from here */
2788 sector_t first_bad = MaxSector;
2791 if (is_badblock(rdev, sector_nr, good_sectors,
2792 &first_bad, &bad_sectors)) {
2793 if (first_bad > sector_nr)
2794 good_sectors = first_bad - sector_nr;
2796 bad_sectors -= (sector_nr - first_bad);
2798 min_bad > bad_sectors)
2799 min_bad = bad_sectors;
2802 if (sector_nr < first_bad) {
2803 if (test_bit(WriteMostly, &rdev->flags)) {
2810 bio_set_op_attrs(bio, REQ_OP_READ, 0);
2811 bio->bi_end_io = end_sync_read;
2813 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2814 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2815 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2817 * The device is suitable for reading (InSync),
2818 * but has bad block(s) here. Let's try to correct them,
2819 * if we are doing resync or repair. Otherwise, leave
2820 * this device alone for this sync request.
2822 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
2823 bio->bi_end_io = end_sync_write;
2827 if (bio->bi_end_io) {
2828 atomic_inc(&rdev->nr_pending);
2829 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset;
2830 bio->bi_bdev = rdev->bdev;
2831 bio->bi_private = r1_bio;
2832 if (test_bit(FailFast, &rdev->flags))
2833 bio->bi_opf |= MD_FAILFAST;
2839 r1_bio->read_disk = disk;
2841 if (read_targets == 0 && min_bad > 0) {
2842 /* These sectors are bad on all InSync devices, so we
2843 * need to mark them bad on all write targets
2846 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
2847 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
2848 struct md_rdev *rdev = conf->mirrors[i].rdev;
2849 ok = rdev_set_badblocks(rdev, sector_nr,
2853 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags);
2858 /* Cannot record the badblocks, so need to
2860 * If there are multiple read targets, could just
2861 * fail the really bad ones ???
2863 conf->recovery_disabled = mddev->recovery_disabled;
2864 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2870 if (min_bad > 0 && min_bad < good_sectors) {
2871 /* only resync enough to reach the next bad->good
2873 good_sectors = min_bad;
2876 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2877 /* extra read targets are also write targets */
2878 write_targets += read_targets-1;
2880 if (write_targets == 0 || read_targets == 0) {
2881 /* There is nowhere to write, so all non-sync
2882 * drives must be failed - so we are finished
2886 max_sector = sector_nr + min_bad;
2887 rv = max_sector - sector_nr;
2893 if (max_sector > mddev->resync_max)
2894 max_sector = mddev->resync_max; /* Don't do IO beyond here */
2895 if (max_sector > sector_nr + good_sectors)
2896 max_sector = sector_nr + good_sectors;
2901 int len = PAGE_SIZE;
2902 if (sector_nr + (len>>9) > max_sector)
2903 len = (max_sector - sector_nr) << 9;
2906 if (sync_blocks == 0) {
2907 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
2908 &sync_blocks, still_degraded) &&
2910 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2912 if ((len >> 9) > sync_blocks)
2913 len = sync_blocks<<9;
2916 for (i = 0 ; i < conf->raid_disks * 2; i++) {
2917 bio = r1_bio->bios[i];
2918 if (bio->bi_end_io) {
2919 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
2920 if (bio_add_page(bio, page, len, 0) == 0) {
2922 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
2925 bio = r1_bio->bios[i];
2926 if (bio->bi_end_io==NULL)
2928 /* remove last page from this bio */
2930 bio->bi_iter.bi_size -= len;
2931 bio_clear_flag(bio, BIO_SEG_VALID);
2937 nr_sectors += len>>9;
2938 sector_nr += len>>9;
2939 sync_blocks -= (len>>9);
2940 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2942 r1_bio->sectors = nr_sectors;
2944 if (mddev_is_clustered(mddev) &&
2945 conf->cluster_sync_high < sector_nr + nr_sectors) {
2946 conf->cluster_sync_low = mddev->curr_resync_completed;
2947 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS;
2948 /* Send resync message */
2949 md_cluster_ops->resync_info_update(mddev,
2950 conf->cluster_sync_low,
2951 conf->cluster_sync_high);
2954 /* For a user-requested sync, we read all readable devices and do a
2957 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2958 atomic_set(&r1_bio->remaining, read_targets);
2959 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
2960 bio = r1_bio->bios[i];
2961 if (bio->bi_end_io == end_sync_read) {
2963 md_sync_acct(bio->bi_bdev, nr_sectors);
2964 if (read_targets == 1)
2965 bio->bi_opf &= ~MD_FAILFAST;
2966 generic_make_request(bio);
2970 atomic_set(&r1_bio->remaining, 1);
2971 bio = r1_bio->bios[r1_bio->read_disk];
2972 md_sync_acct(bio->bi_bdev, nr_sectors);
2973 if (read_targets == 1)
2974 bio->bi_opf &= ~MD_FAILFAST;
2975 generic_make_request(bio);
2981 static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
2986 return mddev->dev_sectors;
2989 static struct r1conf *setup_conf(struct mddev *mddev)
2991 struct r1conf *conf;
2993 struct raid1_info *disk;
2994 struct md_rdev *rdev;
2997 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
3001 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR,
3002 sizeof(atomic_t), GFP_KERNEL);
3003 if (!conf->nr_pending)
3006 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR,
3007 sizeof(atomic_t), GFP_KERNEL);
3008 if (!conf->nr_waiting)
3011 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR,
3012 sizeof(atomic_t), GFP_KERNEL);
3013 if (!conf->nr_queued)
3016 conf->barrier = kcalloc(BARRIER_BUCKETS_NR,
3017 sizeof(atomic_t), GFP_KERNEL);
3021 conf->mirrors = kzalloc(sizeof(struct raid1_info)
3022 * mddev->raid_disks * 2,
3027 conf->tmppage = alloc_page(GFP_KERNEL);
3031 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
3032 if (!conf->poolinfo)
3034 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
3035 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3038 if (!conf->r1bio_pool)
3041 conf->poolinfo->mddev = mddev;
3044 spin_lock_init(&conf->device_lock);
3045 rdev_for_each(rdev, mddev) {
3046 struct request_queue *q;
3047 int disk_idx = rdev->raid_disk;
3048 if (disk_idx >= mddev->raid_disks
3051 if (test_bit(Replacement, &rdev->flags))
3052 disk = conf->mirrors + mddev->raid_disks + disk_idx;
3054 disk = conf->mirrors + disk_idx;
3059 q = bdev_get_queue(rdev->bdev);
3061 disk->head_position = 0;
3062 disk->seq_start = MaxSector;
3064 conf->raid_disks = mddev->raid_disks;
3065 conf->mddev = mddev;
3066 INIT_LIST_HEAD(&conf->retry_list);
3067 INIT_LIST_HEAD(&conf->bio_end_io_list);
3069 spin_lock_init(&conf->resync_lock);
3070 init_waitqueue_head(&conf->wait_barrier);
3072 bio_list_init(&conf->pending_bio_list);
3073 conf->pending_count = 0;
3074 conf->recovery_disabled = mddev->recovery_disabled - 1;
3077 for (i = 0; i < conf->raid_disks * 2; i++) {
3079 disk = conf->mirrors + i;
3081 if (i < conf->raid_disks &&
3082 disk[conf->raid_disks].rdev) {
3083 /* This slot has a replacement. */
3085 /* No original, just make the replacement
3086 * a recovering spare
3089 disk[conf->raid_disks].rdev;
3090 disk[conf->raid_disks].rdev = NULL;
3091 } else if (!test_bit(In_sync, &disk->rdev->flags))
3092 /* Original is not in_sync - bad */
3097 !test_bit(In_sync, &disk->rdev->flags)) {
3098 disk->head_position = 0;
3100 (disk->rdev->saved_raid_disk < 0))
3106 conf->thread = md_register_thread(raid1d, mddev, "raid1");
3114 mempool_destroy(conf->r1bio_pool);
3115 kfree(conf->mirrors);
3116 safe_put_page(conf->tmppage);
3117 kfree(conf->poolinfo);
3118 kfree(conf->nr_pending);
3119 kfree(conf->nr_waiting);
3120 kfree(conf->nr_queued);
3121 kfree(conf->barrier);
3124 return ERR_PTR(err);
3127 static void raid1_free(struct mddev *mddev, void *priv);
3128 static int raid1_run(struct mddev *mddev)
3130 struct r1conf *conf;
3132 struct md_rdev *rdev;
3134 bool discard_supported = false;
3136 if (mddev->level != 1) {
3137 pr_warn("md/raid1:%s: raid level not set to mirroring (%d)\n",
3138 mdname(mddev), mddev->level);
3141 if (mddev->reshape_position != MaxSector) {
3142 pr_warn("md/raid1:%s: reshape_position set but not supported\n",
3147 * copy the already verified devices into our private RAID1
3148 * bookkeeping area. [whatever we allocate in run(),
3149 * should be freed in raid1_free()]
3151 if (mddev->private == NULL)
3152 conf = setup_conf(mddev);
3154 conf = mddev->private;
3157 return PTR_ERR(conf);
3160 blk_queue_max_write_same_sectors(mddev->queue, 0);
3162 rdev_for_each(rdev, mddev) {
3163 if (!mddev->gendisk)
3165 disk_stack_limits(mddev->gendisk, rdev->bdev,
3166 rdev->data_offset << 9);
3167 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
3168 discard_supported = true;
3171 mddev->degraded = 0;
3172 for (i=0; i < conf->raid_disks; i++)
3173 if (conf->mirrors[i].rdev == NULL ||
3174 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
3175 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
3178 if (conf->raid_disks - mddev->degraded == 1)
3179 mddev->recovery_cp = MaxSector;
3181 if (mddev->recovery_cp != MaxSector)
3182 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n",
3184 pr_info("md/raid1:%s: active with %d out of %d mirrors\n",
3185 mdname(mddev), mddev->raid_disks - mddev->degraded,
3189 * Ok, everything is just fine now
3191 mddev->thread = conf->thread;
3192 conf->thread = NULL;
3193 mddev->private = conf;
3194 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags);
3196 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
3199 if (discard_supported)
3200 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
3203 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
3207 ret = md_integrity_register(mddev);
3209 md_unregister_thread(&mddev->thread);
3210 raid1_free(mddev, conf);
3215 static void raid1_free(struct mddev *mddev, void *priv)
3217 struct r1conf *conf = priv;
3219 mempool_destroy(conf->r1bio_pool);
3220 kfree(conf->mirrors);
3221 safe_put_page(conf->tmppage);
3222 kfree(conf->poolinfo);
3223 kfree(conf->nr_pending);
3224 kfree(conf->nr_waiting);
3225 kfree(conf->nr_queued);
3226 kfree(conf->barrier);
3230 static int raid1_resize(struct mddev *mddev, sector_t sectors)
3232 /* no resync is happening, and there is enough space
3233 * on all devices, so we can resize.
3234 * We need to make sure resync covers any new space.
3235 * If the array is shrinking we should possibly wait until
3236 * any io in the removed space completes, but it hardly seems
3239 sector_t newsize = raid1_size(mddev, sectors, 0);
3240 if (mddev->external_size &&
3241 mddev->array_sectors > newsize)
3243 if (mddev->bitmap) {
3244 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
3248 md_set_array_sectors(mddev, newsize);
3249 set_capacity(mddev->gendisk, mddev->array_sectors);
3250 revalidate_disk(mddev->gendisk);
3251 if (sectors > mddev->dev_sectors &&
3252 mddev->recovery_cp > mddev->dev_sectors) {
3253 mddev->recovery_cp = mddev->dev_sectors;
3254 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3256 mddev->dev_sectors = sectors;
3257 mddev->resync_max_sectors = sectors;
3261 static int raid1_reshape(struct mddev *mddev)
3264 * 1/ resize the r1bio_pool
3265 * 2/ resize conf->mirrors
3267 * We allocate a new r1bio_pool if we can.
3268 * Then raise a device barrier and wait until all IO stops.
3269 * Then resize conf->mirrors and swap in the new r1bio pool.
3271 * At the same time, we "pack" the devices so that all the missing
3272 * devices have the higher raid_disk numbers.
3274 mempool_t *newpool, *oldpool;
3275 struct pool_info *newpoolinfo;
3276 struct raid1_info *newmirrors;
3277 struct r1conf *conf = mddev->private;
3278 int cnt, raid_disks;
3279 unsigned long flags;
3282 /* Cannot change chunk_size, layout, or level */
3283 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
3284 mddev->layout != mddev->new_layout ||
3285 mddev->level != mddev->new_level) {
3286 mddev->new_chunk_sectors = mddev->chunk_sectors;
3287 mddev->new_layout = mddev->layout;
3288 mddev->new_level = mddev->level;
3292 if (!mddev_is_clustered(mddev)) {
3293 err = md_allow_write(mddev);
3298 raid_disks = mddev->raid_disks + mddev->delta_disks;
3300 if (raid_disks < conf->raid_disks) {
3302 for (d= 0; d < conf->raid_disks; d++)
3303 if (conf->mirrors[d].rdev)
3305 if (cnt > raid_disks)
3309 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
3312 newpoolinfo->mddev = mddev;
3313 newpoolinfo->raid_disks = raid_disks * 2;
3315 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
3316 r1bio_pool_free, newpoolinfo);
3321 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
3325 mempool_destroy(newpool);
3329 freeze_array(conf, 0);
3331 /* ok, everything is stopped */
3332 oldpool = conf->r1bio_pool;
3333 conf->r1bio_pool = newpool;
3335 for (d = d2 = 0; d < conf->raid_disks; d++) {
3336 struct md_rdev *rdev = conf->mirrors[d].rdev;
3337 if (rdev && rdev->raid_disk != d2) {
3338 sysfs_unlink_rdev(mddev, rdev);
3339 rdev->raid_disk = d2;
3340 sysfs_unlink_rdev(mddev, rdev);
3341 if (sysfs_link_rdev(mddev, rdev))
3342 pr_warn("md/raid1:%s: cannot register rd%d\n",
3343 mdname(mddev), rdev->raid_disk);
3346 newmirrors[d2++].rdev = rdev;
3348 kfree(conf->mirrors);
3349 conf->mirrors = newmirrors;
3350 kfree(conf->poolinfo);
3351 conf->poolinfo = newpoolinfo;
3353 spin_lock_irqsave(&conf->device_lock, flags);
3354 mddev->degraded += (raid_disks - conf->raid_disks);
3355 spin_unlock_irqrestore(&conf->device_lock, flags);
3356 conf->raid_disks = mddev->raid_disks = raid_disks;
3357 mddev->delta_disks = 0;
3359 unfreeze_array(conf);
3361 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3362 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3363 md_wakeup_thread(mddev->thread);
3365 mempool_destroy(oldpool);
3369 static void raid1_quiesce(struct mddev *mddev, int state)
3371 struct r1conf *conf = mddev->private;
3374 case 2: /* wake for suspend */
3375 wake_up(&conf->wait_barrier);
3378 freeze_array(conf, 0);
3381 unfreeze_array(conf);
3386 static void *raid1_takeover(struct mddev *mddev)
3388 /* raid1 can take over:
3389 * raid5 with 2 devices, any layout or chunk size
3391 if (mddev->level == 5 && mddev->raid_disks == 2) {
3392 struct r1conf *conf;
3393 mddev->new_level = 1;
3394 mddev->new_layout = 0;
3395 mddev->new_chunk_sectors = 0;
3396 conf = setup_conf(mddev);
3397 if (!IS_ERR(conf)) {
3398 /* Array must appear to be quiesced */
3399 conf->array_frozen = 1;
3400 mddev_clear_unsupported_flags(mddev,
3401 UNSUPPORTED_MDDEV_FLAGS);
3405 return ERR_PTR(-EINVAL);
3408 static struct md_personality raid1_personality =
3412 .owner = THIS_MODULE,
3413 .make_request = raid1_make_request,
3416 .status = raid1_status,
3417 .error_handler = raid1_error,
3418 .hot_add_disk = raid1_add_disk,
3419 .hot_remove_disk= raid1_remove_disk,
3420 .spare_active = raid1_spare_active,
3421 .sync_request = raid1_sync_request,
3422 .resize = raid1_resize,
3424 .check_reshape = raid1_reshape,
3425 .quiesce = raid1_quiesce,
3426 .takeover = raid1_takeover,
3427 .congested = raid1_congested,
3430 static int __init raid_init(void)
3432 return register_md_personality(&raid1_personality);
3435 static void raid_exit(void)
3437 unregister_md_personality(&raid1_personality);
3440 module_init(raid_init);
3441 module_exit(raid_exit);
3442 MODULE_LICENSE("GPL");
3443 MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
3444 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3445 MODULE_ALIAS("md-raid1");
3446 MODULE_ALIAS("md-level-1");
3448 module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);