Lines Matching +full:data +full:- +full:mirror
1 // SPDX-License-Identifier: GPL-2.0-or-later
9 * RAID-1 management functions.
11 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
19 * - bitmap marked during normal i/o
20 * - bitmap used to skip nondirty blocks during sync
22 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
23 * - persistent bitmap code
38 #include "md-bitmap.h"
50 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0)
52 #include "raid1-10.c"
54 #define START(node) ((node)->start)
55 #define LAST(node) ((node)->last)
64 sector_t lo = r1_bio->sector; in check_and_add_serial()
65 sector_t hi = lo + r1_bio->sectors; in check_and_add_serial()
66 struct serial_in_rdev *serial = &rdev->serial[idx]; in check_and_add_serial()
68 spin_lock_irqsave(&serial->serial_lock, flags); in check_and_add_serial()
70 if (raid1_rb_iter_first(&serial->serial_rb, lo, hi)) in check_and_add_serial()
71 ret = -EBUSY; in check_and_add_serial()
73 si->start = lo; in check_and_add_serial()
74 si->last = hi; in check_and_add_serial()
75 raid1_rb_insert(si, &serial->serial_rb); in check_and_add_serial()
77 spin_unlock_irqrestore(&serial->serial_lock, flags); in check_and_add_serial()
84 struct mddev *mddev = rdev->mddev; in wait_for_serialization()
86 int idx = sector_to_idx(r1_bio->sector); in wait_for_serialization()
87 struct serial_in_rdev *serial = &rdev->serial[idx]; in wait_for_serialization()
89 if (WARN_ON(!mddev->serial_info_pool)) in wait_for_serialization()
91 si = mempool_alloc(mddev->serial_info_pool, GFP_NOIO); in wait_for_serialization()
92 wait_event(serial->serial_io_wait, in wait_for_serialization()
101 struct mddev *mddev = rdev->mddev; in remove_serial()
103 struct serial_in_rdev *serial = &rdev->serial[idx]; in remove_serial()
105 spin_lock_irqsave(&serial->serial_lock, flags); in remove_serial()
106 for (si = raid1_rb_iter_first(&serial->serial_rb, lo, hi); in remove_serial()
108 if (si->start == lo && si->last == hi) { in remove_serial()
109 raid1_rb_remove(si, &serial->serial_rb); in remove_serial()
110 mempool_free(si, mddev->serial_info_pool); in remove_serial()
117 spin_unlock_irqrestore(&serial->serial_lock, flags); in remove_serial()
118 wake_up(&serial->serial_io_wait); in remove_serial()
122 * for resync bio, r1bio pointer can be retrieved from the per-bio
127 return get_resync_pages(bio)->raid_bio; in get_resync_r1bio()
130 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) in r1bio_pool_alloc() argument
132 struct pool_info *pi = data; in r1bio_pool_alloc()
133 int size = offsetof(struct r1bio, bios[pi->raid_disks]); in r1bio_pool_alloc()
146 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) in r1buf_pool_alloc() argument
148 struct pool_info *pi = data; in r1buf_pool_alloc()
159 rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages), in r1buf_pool_alloc()
165 * Allocate bios : 1 for reading, n-1 for writing in r1buf_pool_alloc()
167 for (j = pi->raid_disks ; j-- ; ) { in r1buf_pool_alloc()
171 r1_bio->bios[j] = bio; in r1buf_pool_alloc()
174 * Allocate RESYNC_PAGES data pages and attach them to in r1buf_pool_alloc()
176 * If this is a user-requested check/repair, allocate in r1buf_pool_alloc()
179 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) in r1buf_pool_alloc()
180 need_pages = pi->raid_disks; in r1buf_pool_alloc()
183 for (j = 0; j < pi->raid_disks; j++) { in r1buf_pool_alloc()
186 bio = r1_bio->bios[j]; in r1buf_pool_alloc()
196 rp->raid_bio = r1_bio; in r1buf_pool_alloc()
197 bio->bi_private = rp; in r1buf_pool_alloc()
200 r1_bio->master_bio = NULL; in r1buf_pool_alloc()
205 while (--j >= 0) in r1buf_pool_alloc()
209 while (++j < pi->raid_disks) in r1buf_pool_alloc()
210 bio_put(r1_bio->bios[j]); in r1buf_pool_alloc()
214 rbio_pool_free(r1_bio, data); in r1buf_pool_alloc()
218 static void r1buf_pool_free(void *__r1_bio, void *data) in r1buf_pool_free() argument
220 struct pool_info *pi = data; in r1buf_pool_free()
225 for (i = pi->raid_disks; i--; ) { in r1buf_pool_free()
226 rp = get_resync_pages(r1bio->bios[i]); in r1buf_pool_free()
228 bio_put(r1bio->bios[i]); in r1buf_pool_free()
234 rbio_pool_free(r1bio, data); in r1buf_pool_free()
241 for (i = 0; i < conf->raid_disks * 2; i++) { in put_all_bios()
242 struct bio **bio = r1_bio->bios + i; in put_all_bios()
251 struct r1conf *conf = r1_bio->mddev->private; in free_r1bio()
254 mempool_free(r1_bio, &conf->r1bio_pool); in free_r1bio()
259 struct r1conf *conf = r1_bio->mddev->private; in put_buf()
260 sector_t sect = r1_bio->sector; in put_buf()
263 for (i = 0; i < conf->raid_disks * 2; i++) { in put_buf()
264 struct bio *bio = r1_bio->bios[i]; in put_buf()
265 if (bio->bi_end_io) in put_buf()
266 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); in put_buf()
269 mempool_free(r1_bio, &conf->r1buf_pool); in put_buf()
277 struct mddev *mddev = r1_bio->mddev; in reschedule_retry()
278 struct r1conf *conf = mddev->private; in reschedule_retry()
281 idx = sector_to_idx(r1_bio->sector); in reschedule_retry()
282 spin_lock_irqsave(&conf->device_lock, flags); in reschedule_retry()
283 list_add(&r1_bio->retry_list, &conf->retry_list); in reschedule_retry()
284 atomic_inc(&conf->nr_queued[idx]); in reschedule_retry()
285 spin_unlock_irqrestore(&conf->device_lock, flags); in reschedule_retry()
287 wake_up(&conf->wait_barrier); in reschedule_retry()
288 md_wakeup_thread(mddev->thread); in reschedule_retry()
298 struct bio *bio = r1_bio->master_bio; in call_bio_endio()
300 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in call_bio_endio()
301 bio->bi_status = BLK_STS_IOERR; in call_bio_endio()
308 struct bio *bio = r1_bio->master_bio; in raid_end_bio_io()
309 struct r1conf *conf = r1_bio->mddev->private; in raid_end_bio_io()
312 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid_end_bio_io()
313 pr_debug("raid1: sync end %s on sectors %llu-%llu\n", in raid_end_bio_io()
315 (unsigned long long) bio->bi_iter.bi_sector, in raid_end_bio_io()
316 (unsigned long long) bio_end_sector(bio) - 1); in raid_end_bio_io()
322 * to go idle. All I/Os, even write-behind writes, are done. in raid_end_bio_io()
324 allow_barrier(conf, r1_bio->sector); in raid_end_bio_io()
334 struct r1conf *conf = r1_bio->mddev->private; in update_head_pos()
336 conf->mirrors[disk].head_position = in update_head_pos()
337 r1_bio->sector + (r1_bio->sectors); in update_head_pos()
345 int mirror; in find_bio_disk() local
346 struct r1conf *conf = r1_bio->mddev->private; in find_bio_disk()
347 int raid_disks = conf->raid_disks; in find_bio_disk()
349 for (mirror = 0; mirror < raid_disks * 2; mirror++) in find_bio_disk()
350 if (r1_bio->bios[mirror] == bio) in find_bio_disk()
353 BUG_ON(mirror == raid_disks * 2); in find_bio_disk()
354 update_head_pos(mirror, r1_bio); in find_bio_disk()
356 return mirror; in find_bio_disk()
361 int uptodate = !bio->bi_status; in raid1_end_read_request()
362 struct r1bio *r1_bio = bio->bi_private; in raid1_end_read_request()
363 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_read_request()
364 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; in raid1_end_read_request()
367 * this branch is our 'one mirror IO has finished' event handler: in raid1_end_read_request()
369 update_head_pos(r1_bio->read_disk, r1_bio); in raid1_end_read_request()
372 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_read_request()
373 else if (test_bit(FailFast, &rdev->flags) && in raid1_end_read_request()
374 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_end_read_request()
375 /* This was a fail-fast read so we definitely in raid1_end_read_request()
384 spin_lock_irqsave(&conf->device_lock, flags); in raid1_end_read_request()
385 if (r1_bio->mddev->degraded == conf->raid_disks || in raid1_end_read_request()
386 (r1_bio->mddev->degraded == conf->raid_disks-1 && in raid1_end_read_request()
387 test_bit(In_sync, &rdev->flags))) in raid1_end_read_request()
389 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_end_read_request()
394 rdev_dec_pending(rdev, conf->mddev); in raid1_end_read_request()
401 mdname(conf->mddev), in raid1_end_read_request()
402 bdevname(rdev->bdev, b), in raid1_end_read_request()
403 (unsigned long long)r1_bio->sector); in raid1_end_read_request()
404 set_bit(R1BIO_ReadError, &r1_bio->state); in raid1_end_read_request()
413 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in close_write()
414 bio_free_pages(r1_bio->behind_master_bio); in close_write()
415 bio_put(r1_bio->behind_master_bio); in close_write()
416 r1_bio->behind_master_bio = NULL; in close_write()
419 md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, in close_write()
420 r1_bio->sectors, in close_write()
421 !test_bit(R1BIO_Degraded, &r1_bio->state), in close_write()
422 test_bit(R1BIO_BehindIO, &r1_bio->state)); in close_write()
423 md_write_end(r1_bio->mddev); in close_write()
428 if (!atomic_dec_and_test(&r1_bio->remaining)) in r1_bio_write_done()
431 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in r1_bio_write_done()
435 if (test_bit(R1BIO_MadeGood, &r1_bio->state)) in r1_bio_write_done()
444 struct r1bio *r1_bio = bio->bi_private; in raid1_end_write_request()
445 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); in raid1_end_write_request()
446 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_write_request()
448 int mirror = find_bio_disk(r1_bio, bio); in raid1_end_write_request() local
449 struct md_rdev *rdev = conf->mirrors[mirror].rdev; in raid1_end_write_request()
451 sector_t lo = r1_bio->sector; in raid1_end_write_request()
452 sector_t hi = r1_bio->sector + r1_bio->sectors; in raid1_end_write_request()
454 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; in raid1_end_write_request()
457 * 'one mirror IO has finished' event handler: in raid1_end_write_request()
459 if (bio->bi_status && !discard_error) { in raid1_end_write_request()
460 set_bit(WriteErrorSeen, &rdev->flags); in raid1_end_write_request()
461 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in raid1_end_write_request()
463 conf->mddev->recovery); in raid1_end_write_request()
465 if (test_bit(FailFast, &rdev->flags) && in raid1_end_write_request()
466 (bio->bi_opf & MD_FAILFAST) && in raid1_end_write_request()
468 !test_bit(WriteMostly, &rdev->flags)) { in raid1_end_write_request()
469 md_error(r1_bio->mddev, rdev); in raid1_end_write_request()
478 if (!test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
479 set_bit(R1BIO_WriteError, &r1_bio->state); in raid1_end_write_request()
482 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
493 * to user-side. So if something waits for IO, then it in raid1_end_write_request()
499 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
504 * such device for properly reading the data back (we could in raid1_end_write_request()
506 * before rdev->recovery_offset, but for simplicity we don't in raid1_end_write_request()
509 if (test_bit(In_sync, &rdev->flags) && in raid1_end_write_request()
510 !test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
511 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_write_request()
514 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in raid1_end_write_request()
516 r1_bio->bios[mirror] = IO_MADE_GOOD; in raid1_end_write_request()
517 set_bit(R1BIO_MadeGood, &r1_bio->state); in raid1_end_write_request()
522 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_end_write_request()
524 if (test_bit(WriteMostly, &rdev->flags)) in raid1_end_write_request()
525 atomic_dec(&r1_bio->behind_remaining); in raid1_end_write_request()
529 * has safely reached all non-writemostly in raid1_end_write_request()
531 * gets done only once -- we don't ever want to return in raid1_end_write_request()
532 * -EIO here, instead we'll wait in raid1_end_write_request()
534 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && in raid1_end_write_request()
535 test_bit(R1BIO_Uptodate, &r1_bio->state)) { in raid1_end_write_request()
537 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid1_end_write_request()
538 struct bio *mbio = r1_bio->master_bio; in raid1_end_write_request()
540 " %llu-%llu\n", in raid1_end_write_request()
541 (unsigned long long) mbio->bi_iter.bi_sector, in raid1_end_write_request()
542 (unsigned long long) bio_end_sector(mbio) - 1); in raid1_end_write_request()
546 } else if (rdev->mddev->serialize_policy) in raid1_end_write_request()
548 if (r1_bio->bios[mirror] == NULL) in raid1_end_write_request()
549 rdev_dec_pending(rdev, conf->mddev); in raid1_end_write_request()
571 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - in align_to_barrier_unit_end()
582 * be done. There is a per-array 'next expected sequential IO' sector
583 * number - if this matches on the next IO then we use the last disk.
584 * There is also a per-disk 'last know head position' sector that is
590 * because position is mirror, not device based.
596 const sector_t this_sector = r1_bio->sector; in read_balance()
615 sectors = r1_bio->sectors; in read_balance()
616 best_disk = -1; in read_balance()
617 best_dist_disk = -1; in read_balance()
619 best_pending_disk = -1; in read_balance()
624 clear_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
626 if ((conf->mddev->recovery_cp < this_sector + sectors) || in read_balance()
627 (mddev_is_clustered(conf->mddev) && in read_balance()
628 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, in read_balance()
634 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in read_balance()
641 rdev = rcu_dereference(conf->mirrors[disk].rdev); in read_balance()
642 if (r1_bio->bios[disk] == IO_BLOCKED in read_balance()
644 || test_bit(Faulty, &rdev->flags)) in read_balance()
646 if (!test_bit(In_sync, &rdev->flags) && in read_balance()
647 rdev->recovery_offset < this_sector + sectors) in read_balance()
649 if (test_bit(WriteMostly, &rdev->flags)) { in read_balance()
650 /* Don't balance among write-mostly, just in read_balance()
658 best_good_sectors = first_bad - this_sector; in read_balance()
679 bad_sectors -= (this_sector - first_bad); in read_balance()
686 sector_t good_sectors = first_bad - this_sector; in read_balance()
697 best_disk = -1; in read_balance()
703 set_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
705 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev)); in read_balance()
707 pending = atomic_read(&rdev->nr_pending); in read_balance()
708 dist = abs(this_sector - conf->mirrors[disk].head_position); in read_balance()
714 if (conf->mirrors[disk].next_seq_sect == this_sector in read_balance()
716 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; in read_balance()
717 struct raid1_info *mirror = &conf->mirrors[disk]; in read_balance() local
734 mirror->seq_start != MaxSector && in read_balance()
735 mirror->next_seq_sect > opt_iosize && in read_balance()
736 mirror->next_seq_sect - opt_iosize >= in read_balance()
737 mirror->seq_start) { in read_balance()
760 * non-rotational, choose the disk with less pending request even the in read_balance()
762 * mixed ratation/non-rotational disks depending on workload. in read_balance()
764 if (best_disk == -1) { in read_balance()
772 rdev = rcu_dereference(conf->mirrors[best_disk].rdev); in read_balance()
775 atomic_inc(&rdev->nr_pending); in read_balance()
778 if (conf->mirrors[best_disk].next_seq_sect != this_sector) in read_balance()
779 conf->mirrors[best_disk].seq_start = this_sector; in read_balance()
781 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; in read_balance()
792 md_bitmap_unplug(conf->mddev->bitmap); in flush_bio_list()
793 wake_up(&conf->wait_barrier); in flush_bio_list()
796 struct bio *next = bio->bi_next; in flush_bio_list()
797 struct md_rdev *rdev = (void *)bio->bi_disk; in flush_bio_list()
798 bio->bi_next = NULL; in flush_bio_list()
799 bio_set_dev(bio, rdev->bdev); in flush_bio_list()
800 if (test_bit(Faulty, &rdev->flags)) { in flush_bio_list()
803 !blk_queue_discard(bio->bi_disk->queue))) in flush_bio_list()
818 spin_lock_irq(&conf->device_lock); in flush_pending_writes()
820 if (conf->pending_bio_list.head) { in flush_pending_writes()
824 bio = bio_list_get(&conf->pending_bio_list); in flush_pending_writes()
825 conf->pending_count = 0; in flush_pending_writes()
826 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
830 * current->state might be TASK_UNINTERRUPTIBLE which will in flush_pending_writes()
834 * is a false-positive. Silence the warning by resetting in flush_pending_writes()
842 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
854 * We choose only to raise the barrier if no-one is waiting for the
866 * If resync/recovery is interrupted, returns -EINTR;
873 spin_lock_irq(&conf->resync_lock); in raise_barrier()
876 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
877 !atomic_read(&conf->nr_waiting[idx]), in raise_barrier()
878 conf->resync_lock); in raise_barrier()
881 atomic_inc(&conf->barrier[idx]); in raise_barrier()
883 * In raise_barrier() we firstly increase conf->barrier[idx] then in raise_barrier()
884 * check conf->nr_pending[idx]. In _wait_barrier() we firstly in raise_barrier()
885 * increase conf->nr_pending[idx] then check conf->barrier[idx]. in raise_barrier()
886 * A memory barrier here to make sure conf->nr_pending[idx] won't in raise_barrier()
887 * be fetched before conf->barrier[idx] is increased. Otherwise in raise_barrier()
894 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O in raise_barrier()
896 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches in raise_barrier()
899 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
900 (!conf->array_frozen && in raise_barrier()
901 !atomic_read(&conf->nr_pending[idx]) && in raise_barrier()
902 atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || in raise_barrier()
903 test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), in raise_barrier()
904 conf->resync_lock); in raise_barrier()
906 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { in raise_barrier()
907 atomic_dec(&conf->barrier[idx]); in raise_barrier()
908 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
909 wake_up(&conf->wait_barrier); in raise_barrier()
910 return -EINTR; in raise_barrier()
913 atomic_inc(&conf->nr_sync_pending); in raise_barrier()
914 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
923 BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); in lower_barrier()
925 atomic_dec(&conf->barrier[idx]); in lower_barrier()
926 atomic_dec(&conf->nr_sync_pending); in lower_barrier()
927 wake_up(&conf->wait_barrier); in lower_barrier()
933 * We need to increase conf->nr_pending[idx] very early here, in _wait_barrier()
935 * conf->nr_pending[idx] to be 0. Then we can avoid holding in _wait_barrier()
936 * conf->resync_lock when there is no barrier raised in same in _wait_barrier()
940 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
942 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then in _wait_barrier()
943 * check conf->barrier[idx]. In raise_barrier() we firstly increase in _wait_barrier()
944 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory in _wait_barrier()
945 * barrier is necessary here to make sure conf->barrier[idx] won't be in _wait_barrier()
946 * fetched before conf->nr_pending[idx] is increased. Otherwise there in _wait_barrier()
953 * here. If during we check conf->barrier[idx], the array is in _wait_barrier()
954 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is in _wait_barrier()
960 if (!READ_ONCE(conf->array_frozen) && in _wait_barrier()
961 !atomic_read(&conf->barrier[idx])) in _wait_barrier()
965 * After holding conf->resync_lock, conf->nr_pending[idx] in _wait_barrier()
968 * raise_barrer() might be waiting for conf->nr_pending[idx] in _wait_barrier()
971 spin_lock_irq(&conf->resync_lock); in _wait_barrier()
972 atomic_inc(&conf->nr_waiting[idx]); in _wait_barrier()
973 atomic_dec(&conf->nr_pending[idx]); in _wait_barrier()
978 wake_up(&conf->wait_barrier); in _wait_barrier()
980 wait_event_lock_irq(conf->wait_barrier, in _wait_barrier()
981 !conf->array_frozen && in _wait_barrier()
982 !atomic_read(&conf->barrier[idx]), in _wait_barrier()
983 conf->resync_lock); in _wait_barrier()
984 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
985 atomic_dec(&conf->nr_waiting[idx]); in _wait_barrier()
986 spin_unlock_irq(&conf->resync_lock); in _wait_barrier()
998 * conf->barrier[idx] here, memory barrier is unnecessary as well. in wait_read_barrier()
1000 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1002 if (!READ_ONCE(conf->array_frozen)) in wait_read_barrier()
1005 spin_lock_irq(&conf->resync_lock); in wait_read_barrier()
1006 atomic_inc(&conf->nr_waiting[idx]); in wait_read_barrier()
1007 atomic_dec(&conf->nr_pending[idx]); in wait_read_barrier()
1012 wake_up(&conf->wait_barrier); in wait_read_barrier()
1014 wait_event_lock_irq(conf->wait_barrier, in wait_read_barrier()
1015 !conf->array_frozen, in wait_read_barrier()
1016 conf->resync_lock); in wait_read_barrier()
1017 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1018 atomic_dec(&conf->nr_waiting[idx]); in wait_read_barrier()
1019 spin_unlock_irq(&conf->resync_lock); in wait_read_barrier()
1031 atomic_dec(&conf->nr_pending[idx]); in _allow_barrier()
1032 wake_up(&conf->wait_barrier); in _allow_barrier()
1042 /* conf->resync_lock should be held */
1047 ret = atomic_read(&conf->nr_sync_pending); in get_unqueued_pending()
1049 ret += atomic_read(&conf->nr_pending[idx]) - in get_unqueued_pending()
1050 atomic_read(&conf->nr_queued[idx]); in get_unqueued_pending()
1068 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the in freeze_array()
1070 * normal I/O are queued, sum of all conf->nr_pending[] will match sum in freeze_array()
1071 * of all conf->nr_queued[]. But normal I/O failure is an exception, in freeze_array()
1080 spin_lock_irq(&conf->resync_lock); in freeze_array()
1081 conf->array_frozen = 1; in freeze_array()
1082 raid1_log(conf->mddev, "wait freeze"); in freeze_array()
1084 conf->wait_barrier, in freeze_array()
1086 conf->resync_lock, in freeze_array()
1088 spin_unlock_irq(&conf->resync_lock); in freeze_array()
1093 spin_lock_irq(&conf->resync_lock); in unfreeze_array()
1094 conf->array_frozen = 0; in unfreeze_array()
1095 spin_unlock_irq(&conf->resync_lock); in unfreeze_array()
1096 wake_up(&conf->wait_barrier); in unfreeze_array()
1102 int size = bio->bi_iter.bi_size; in alloc_behind_master_bio()
1103 unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; in alloc_behind_master_bio()
1107 behind_bio = bio_alloc_mddev(GFP_NOIO, vcnt, r1_bio->mddev); in alloc_behind_master_bio()
1113 behind_bio->bi_iter.bi_size = size; in alloc_behind_master_bio()
1117 behind_bio->bi_write_hint = bio->bi_write_hint; in alloc_behind_master_bio()
1129 size -= len; in alloc_behind_master_bio()
1135 r1_bio->behind_master_bio = behind_bio; in alloc_behind_master_bio()
1136 set_bit(R1BIO_BehindIO, &r1_bio->state); in alloc_behind_master_bio()
1142 bio->bi_iter.bi_size); in alloc_behind_master_bio()
1157 struct mddev *mddev = plug->cb.data; in raid1_unplug()
1158 struct r1conf *conf = mddev->private; in raid1_unplug()
1161 if (from_schedule || current->bio_list) { in raid1_unplug()
1162 spin_lock_irq(&conf->device_lock); in raid1_unplug()
1163 bio_list_merge(&conf->pending_bio_list, &plug->pending); in raid1_unplug()
1164 conf->pending_count += plug->pending_cnt; in raid1_unplug()
1165 spin_unlock_irq(&conf->device_lock); in raid1_unplug()
1166 wake_up(&conf->wait_barrier); in raid1_unplug()
1167 md_wakeup_thread(mddev->thread); in raid1_unplug()
1172 /* we aren't scheduling, so we can do the write-out directly. */ in raid1_unplug()
1173 bio = bio_list_get(&plug->pending); in raid1_unplug()
1180 r1_bio->master_bio = bio; in init_r1bio()
1181 r1_bio->sectors = bio_sectors(bio); in init_r1bio()
1182 r1_bio->state = 0; in init_r1bio()
1183 r1_bio->mddev = mddev; in init_r1bio()
1184 r1_bio->sector = bio->bi_iter.bi_sector; in init_r1bio()
1190 struct r1conf *conf = mddev->private; in alloc_r1bio()
1193 r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); in alloc_r1bio()
1195 memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); in alloc_r1bio()
1203 struct r1conf *conf = mddev->private; in raid1_read_request()
1204 struct raid1_info *mirror; in raid1_read_request() local
1206 struct bitmap *bitmap = mddev->bitmap; in raid1_read_request()
1208 const unsigned long do_sync = (bio->bi_opf & REQ_SYNC); in raid1_read_request()
1225 rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); in raid1_read_request()
1227 bdevname(rdev->bdev, b); in raid1_read_request()
1237 wait_read_barrier(conf, bio->bi_iter.bi_sector); in raid1_read_request()
1243 r1_bio->sectors = max_read_sectors; in raid1_read_request()
1246 * make_request() can abort the operation when read-ahead is being in raid1_read_request()
1257 (unsigned long long)r1_bio->sector); in raid1_read_request()
1262 mirror = conf->mirrors + rdisk; in raid1_read_request()
1265 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %s\n", in raid1_read_request()
1267 (unsigned long long)r1_bio->sector, in raid1_read_request()
1268 bdevname(mirror->rdev->bdev, b)); in raid1_read_request()
1270 if (test_bit(WriteMostly, &mirror->rdev->flags) && in raid1_read_request()
1273 * Reading from a write-mostly device must take care not to in raid1_read_request()
1274 * over-take any writes that are 'behind' in raid1_read_request()
1277 wait_event(bitmap->behind_wait, in raid1_read_request()
1278 atomic_read(&bitmap->behind_writes) == 0); in raid1_read_request()
1283 gfp, &conf->bio_split); in raid1_read_request()
1287 r1_bio->master_bio = bio; in raid1_read_request()
1288 r1_bio->sectors = max_sectors; in raid1_read_request()
1291 r1_bio->read_disk = rdisk; in raid1_read_request()
1293 read_bio = bio_clone_fast(bio, gfp, &mddev->bio_set); in raid1_read_request()
1295 r1_bio->bios[rdisk] = read_bio; in raid1_read_request()
1297 read_bio->bi_iter.bi_sector = r1_bio->sector + in raid1_read_request()
1298 mirror->rdev->data_offset; in raid1_read_request()
1299 bio_set_dev(read_bio, mirror->rdev->bdev); in raid1_read_request()
1300 read_bio->bi_end_io = raid1_end_read_request; in raid1_read_request()
1302 if (test_bit(FailFast, &mirror->rdev->flags) && in raid1_read_request()
1303 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_read_request()
1304 read_bio->bi_opf |= MD_FAILFAST; in raid1_read_request()
1305 read_bio->bi_private = r1_bio; in raid1_read_request()
1307 if (mddev->gendisk) in raid1_read_request()
1308 trace_block_bio_remap(read_bio->bi_disk->queue, read_bio, in raid1_read_request()
1309 disk_devt(mddev->gendisk), r1_bio->sector); in raid1_read_request()
1317 struct r1conf *conf = mddev->private; in raid1_write_request()
1320 struct bitmap *bitmap = mddev->bitmap; in raid1_write_request()
1329 md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1330 bio->bi_iter.bi_sector, bio_end_sector(bio))) { in raid1_write_request()
1334 prepare_to_wait(&conf->wait_barrier, in raid1_write_request()
1336 if (!md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1337 bio->bi_iter.bi_sector, in raid1_write_request()
1342 finish_wait(&conf->wait_barrier, &w); in raid1_write_request()
1350 wait_barrier(conf, bio->bi_iter.bi_sector); in raid1_write_request()
1353 r1_bio->sectors = max_write_sectors; in raid1_write_request()
1355 if (conf->pending_count >= max_queued_requests) { in raid1_write_request()
1356 md_wakeup_thread(mddev->thread); in raid1_write_request()
1358 wait_event(conf->wait_barrier, in raid1_write_request()
1359 conf->pending_count < max_queued_requests); in raid1_write_request()
1372 disks = conf->raid_disks * 2; in raid1_write_request()
1376 max_sectors = r1_bio->sectors; in raid1_write_request()
1378 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_write_request()
1379 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { in raid1_write_request()
1380 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1384 r1_bio->bios[i] = NULL; in raid1_write_request()
1385 if (!rdev || test_bit(Faulty, &rdev->flags)) { in raid1_write_request()
1386 if (i < conf->raid_disks) in raid1_write_request()
1387 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1_write_request()
1391 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1392 if (test_bit(WriteErrorSeen, &rdev->flags)) { in raid1_write_request()
1397 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, in raid1_write_request()
1402 set_bit(BlockedBadBlocks, &rdev->flags); in raid1_write_request()
1406 if (is_bad && first_bad <= r1_bio->sector) { in raid1_write_request()
1408 bad_sectors -= (r1_bio->sector - first_bad); in raid1_write_request()
1417 * missing, so it might be re-added, in raid1_write_request()
1422 * in-sync is recorded in the bad in raid1_write_request()
1428 int good_sectors = first_bad - r1_bio->sector; in raid1_write_request()
1433 r1_bio->bios[i] = bio; in raid1_write_request()
1442 if (r1_bio->bios[j]) in raid1_write_request()
1443 rdev_dec_pending(conf->mirrors[j].rdev, mddev); in raid1_write_request()
1444 r1_bio->state = 0; in raid1_write_request()
1445 allow_barrier(conf, bio->bi_iter.bi_sector); in raid1_write_request()
1446 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); in raid1_write_request()
1448 wait_barrier(conf, bio->bi_iter.bi_sector); in raid1_write_request()
1454 GFP_NOIO, &conf->bio_split); in raid1_write_request()
1458 r1_bio->master_bio = bio; in raid1_write_request()
1459 r1_bio->sectors = max_sectors; in raid1_write_request()
1462 atomic_set(&r1_bio->remaining, 1); in raid1_write_request()
1463 atomic_set(&r1_bio->behind_remaining, 0); in raid1_write_request()
1469 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_write_request()
1470 if (!r1_bio->bios[i]) in raid1_write_request()
1479 (atomic_read(&bitmap->behind_writes) in raid1_write_request()
1480 < mddev->bitmap_info.max_write_behind) && in raid1_write_request()
1481 !waitqueue_active(&bitmap->behind_wait)) { in raid1_write_request()
1485 md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, in raid1_write_request()
1486 test_bit(R1BIO_BehindIO, &r1_bio->state)); in raid1_write_request()
1490 if (r1_bio->behind_master_bio) in raid1_write_request()
1491 mbio = bio_clone_fast(r1_bio->behind_master_bio, in raid1_write_request()
1492 GFP_NOIO, &mddev->bio_set); in raid1_write_request()
1494 mbio = bio_clone_fast(bio, GFP_NOIO, &mddev->bio_set); in raid1_write_request()
1496 if (r1_bio->behind_master_bio) { in raid1_write_request()
1497 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_write_request()
1499 if (test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1500 atomic_inc(&r1_bio->behind_remaining); in raid1_write_request()
1501 } else if (mddev->serialize_policy) in raid1_write_request()
1504 r1_bio->bios[i] = mbio; in raid1_write_request()
1506 mbio->bi_iter.bi_sector = (r1_bio->sector + in raid1_write_request()
1507 conf->mirrors[i].rdev->data_offset); in raid1_write_request()
1508 bio_set_dev(mbio, conf->mirrors[i].rdev->bdev); in raid1_write_request()
1509 mbio->bi_end_io = raid1_end_write_request; in raid1_write_request()
1510 mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); in raid1_write_request()
1511 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags) && in raid1_write_request()
1512 !test_bit(WriteMostly, &conf->mirrors[i].rdev->flags) && in raid1_write_request()
1513 conf->raid_disks - mddev->degraded > 1) in raid1_write_request()
1514 mbio->bi_opf |= MD_FAILFAST; in raid1_write_request()
1515 mbio->bi_private = r1_bio; in raid1_write_request()
1517 atomic_inc(&r1_bio->remaining); in raid1_write_request()
1519 if (mddev->gendisk) in raid1_write_request()
1520 trace_block_bio_remap(mbio->bi_disk->queue, in raid1_write_request()
1521 mbio, disk_devt(mddev->gendisk), in raid1_write_request()
1522 r1_bio->sector); in raid1_write_request()
1524 mbio->bi_disk = (void *)conf->mirrors[i].rdev; in raid1_write_request()
1532 bio_list_add(&plug->pending, mbio); in raid1_write_request()
1533 plug->pending_cnt++; in raid1_write_request()
1535 spin_lock_irqsave(&conf->device_lock, flags); in raid1_write_request()
1536 bio_list_add(&conf->pending_bio_list, mbio); in raid1_write_request()
1537 conf->pending_count++; in raid1_write_request()
1538 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_write_request()
1539 md_wakeup_thread(mddev->thread); in raid1_write_request()
1546 wake_up(&conf->wait_barrier); in raid1_write_request()
1553 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid1_make_request()
1565 bio->bi_iter.bi_sector, bio_sectors(bio)); in raid1_make_request()
1579 struct r1conf *conf = mddev->private; in raid1_status()
1582 seq_printf(seq, " [%d/%d] [", conf->raid_disks, in raid1_status()
1583 conf->raid_disks - mddev->degraded); in raid1_status()
1585 for (i = 0; i < conf->raid_disks; i++) { in raid1_status()
1586 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_status()
1588 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid1_status()
1597 struct r1conf *conf = mddev->private; in raid1_error()
1606 spin_lock_irqsave(&conf->device_lock, flags); in raid1_error()
1607 if (test_bit(In_sync, &rdev->flags) && !mddev->fail_last_dev in raid1_error()
1608 && (conf->raid_disks - mddev->degraded) == 1) { in raid1_error()
1615 conf->recovery_disabled = mddev->recovery_disabled; in raid1_error()
1616 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1619 set_bit(Blocked, &rdev->flags); in raid1_error()
1620 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid1_error()
1621 mddev->degraded++; in raid1_error()
1622 set_bit(Faulty, &rdev->flags); in raid1_error()
1623 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1627 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_error()
1628 set_mask_bits(&mddev->sb_flags, 0, in raid1_error()
1632 mdname(mddev), bdevname(rdev->bdev, b), in raid1_error()
1633 mdname(mddev), conf->raid_disks - mddev->degraded); in raid1_error()
1645 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, in print_conf()
1646 conf->raid_disks); in print_conf()
1649 for (i = 0; i < conf->raid_disks; i++) { in print_conf()
1651 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in print_conf()
1654 i, !test_bit(In_sync, &rdev->flags), in print_conf()
1655 !test_bit(Faulty, &rdev->flags), in print_conf()
1656 bdevname(rdev->bdev,b)); in print_conf()
1670 mempool_exit(&conf->r1buf_pool); in close_sync()
1676 struct r1conf *conf = mddev->private; in raid1_spare_active()
1685 * which expects 'In_sync' flags and ->degraded to be consistent. in raid1_spare_active()
1687 spin_lock_irqsave(&conf->device_lock, flags); in raid1_spare_active()
1688 for (i = 0; i < conf->raid_disks; i++) { in raid1_spare_active()
1689 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_spare_active()
1690 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; in raid1_spare_active()
1692 && !test_bit(Candidate, &repl->flags) in raid1_spare_active()
1693 && repl->recovery_offset == MaxSector in raid1_spare_active()
1694 && !test_bit(Faulty, &repl->flags) in raid1_spare_active()
1695 && !test_and_set_bit(In_sync, &repl->flags)) { in raid1_spare_active()
1698 !test_and_clear_bit(In_sync, &rdev->flags)) in raid1_spare_active()
1703 * it gets removed and never re-added in raid1_spare_active()
1705 set_bit(Faulty, &rdev->flags); in raid1_spare_active()
1707 rdev->sysfs_state); in raid1_spare_active()
1711 && rdev->recovery_offset == MaxSector in raid1_spare_active()
1712 && !test_bit(Faulty, &rdev->flags) in raid1_spare_active()
1713 && !test_and_set_bit(In_sync, &rdev->flags)) { in raid1_spare_active()
1715 sysfs_notify_dirent_safe(rdev->sysfs_state); in raid1_spare_active()
1718 mddev->degraded -= count; in raid1_spare_active()
1719 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_spare_active()
1727 struct r1conf *conf = mddev->private; in raid1_add_disk()
1728 int err = -EEXIST; in raid1_add_disk()
1729 int mirror = 0; in raid1_add_disk() local
1732 int last = conf->raid_disks - 1; in raid1_add_disk()
1734 if (mddev->recovery_disabled == conf->recovery_disabled) in raid1_add_disk()
1735 return -EBUSY; in raid1_add_disk()
1738 return -ENXIO; in raid1_add_disk()
1740 if (rdev->raid_disk >= 0) in raid1_add_disk()
1741 first = last = rdev->raid_disk; in raid1_add_disk()
1744 * find the disk ... but prefer rdev->saved_raid_disk in raid1_add_disk()
1747 if (rdev->saved_raid_disk >= 0 && in raid1_add_disk()
1748 rdev->saved_raid_disk >= first && in raid1_add_disk()
1749 rdev->saved_raid_disk < conf->raid_disks && in raid1_add_disk()
1750 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid1_add_disk()
1751 first = last = rdev->saved_raid_disk; in raid1_add_disk()
1753 for (mirror = first; mirror <= last; mirror++) { in raid1_add_disk()
1754 p = conf->mirrors + mirror; in raid1_add_disk()
1755 if (!p->rdev) { in raid1_add_disk()
1756 if (mddev->gendisk) in raid1_add_disk()
1757 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_add_disk()
1758 rdev->data_offset << 9); in raid1_add_disk()
1760 p->head_position = 0; in raid1_add_disk()
1761 rdev->raid_disk = mirror; in raid1_add_disk()
1766 if (rdev->saved_raid_disk < 0) in raid1_add_disk()
1767 conf->fullsync = 1; in raid1_add_disk()
1768 rcu_assign_pointer(p->rdev, rdev); in raid1_add_disk()
1771 if (test_bit(WantReplacement, &p->rdev->flags) && in raid1_add_disk()
1772 p[conf->raid_disks].rdev == NULL) { in raid1_add_disk()
1774 clear_bit(In_sync, &rdev->flags); in raid1_add_disk()
1775 set_bit(Replacement, &rdev->flags); in raid1_add_disk()
1776 rdev->raid_disk = mirror; in raid1_add_disk()
1778 conf->fullsync = 1; in raid1_add_disk()
1779 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); in raid1_add_disk()
1783 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev))) in raid1_add_disk()
1784 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue); in raid1_add_disk()
1791 struct r1conf *conf = mddev->private; in raid1_remove_disk()
1793 int number = rdev->raid_disk; in raid1_remove_disk()
1794 struct raid1_info *p = conf->mirrors + number; in raid1_remove_disk()
1796 if (rdev != p->rdev) in raid1_remove_disk()
1797 p = conf->mirrors + conf->raid_disks + number; in raid1_remove_disk()
1800 if (rdev == p->rdev) { in raid1_remove_disk()
1801 if (test_bit(In_sync, &rdev->flags) || in raid1_remove_disk()
1802 atomic_read(&rdev->nr_pending)) { in raid1_remove_disk()
1803 err = -EBUSY; in raid1_remove_disk()
1806 /* Only remove non-faulty devices if recovery in raid1_remove_disk()
1809 if (!test_bit(Faulty, &rdev->flags) && in raid1_remove_disk()
1810 mddev->recovery_disabled != conf->recovery_disabled && in raid1_remove_disk()
1811 mddev->degraded < conf->raid_disks) { in raid1_remove_disk()
1812 err = -EBUSY; in raid1_remove_disk()
1815 p->rdev = NULL; in raid1_remove_disk()
1816 if (!test_bit(RemoveSynchronized, &rdev->flags)) { in raid1_remove_disk()
1818 if (atomic_read(&rdev->nr_pending)) { in raid1_remove_disk()
1820 err = -EBUSY; in raid1_remove_disk()
1821 p->rdev = rdev; in raid1_remove_disk()
1825 if (conf->mirrors[conf->raid_disks + number].rdev) { in raid1_remove_disk()
1831 conf->mirrors[conf->raid_disks + number].rdev; in raid1_remove_disk()
1833 if (atomic_read(&repl->nr_pending)) { in raid1_remove_disk()
1840 err = -EBUSY; in raid1_remove_disk()
1844 clear_bit(Replacement, &repl->flags); in raid1_remove_disk()
1845 p->rdev = repl; in raid1_remove_disk()
1846 conf->mirrors[conf->raid_disks + number].rdev = NULL; in raid1_remove_disk()
1850 clear_bit(WantReplacement, &rdev->flags); in raid1_remove_disk()
1863 update_head_pos(r1_bio->read_disk, r1_bio); in end_sync_read()
1866 * we have read a block, now it needs to be re-written, in end_sync_read()
1867 * or re-read if the read failed. in end_sync_read()
1870 if (!bio->bi_status) in end_sync_read()
1871 set_bit(R1BIO_Uptodate, &r1_bio->state); in end_sync_read()
1873 if (atomic_dec_and_test(&r1_bio->remaining)) in end_sync_read()
1880 sector_t s = r1_bio->sector; in abort_sync_write()
1881 long sectors_to_go = r1_bio->sectors; in abort_sync_write()
1885 md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); in abort_sync_write()
1887 sectors_to_go -= sync_blocks; in abort_sync_write()
1893 if (atomic_dec_and_test(&r1_bio->remaining)) { in put_sync_write_buf()
1894 struct mddev *mddev = r1_bio->mddev; in put_sync_write_buf()
1895 int s = r1_bio->sectors; in put_sync_write_buf()
1897 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in put_sync_write_buf()
1898 test_bit(R1BIO_WriteError, &r1_bio->state)) in put_sync_write_buf()
1909 int uptodate = !bio->bi_status; in end_sync_write()
1911 struct mddev *mddev = r1_bio->mddev; in end_sync_write()
1912 struct r1conf *conf = mddev->private; in end_sync_write()
1915 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; in end_sync_write()
1919 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
1920 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
1922 mddev->recovery); in end_sync_write()
1923 set_bit(R1BIO_WriteError, &r1_bio->state); in end_sync_write()
1924 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in end_sync_write()
1926 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, in end_sync_write()
1927 r1_bio->sector, in end_sync_write()
1928 r1_bio->sectors, in end_sync_write()
1931 set_bit(R1BIO_MadeGood, &r1_bio->state); in end_sync_write()
1943 set_bit(WriteErrorSeen, &rdev->flags); in r1_sync_page_io()
1945 &rdev->flags)) in r1_sync_page_io()
1947 rdev->mddev->recovery); in r1_sync_page_io()
1949 /* need to record an error - either for the block or the device */ in r1_sync_page_io()
1951 md_error(rdev->mddev, rdev); in r1_sync_page_io()
1958 * good data, much like with normal read errors. Only in fix_sync_read_error()
1960 * need to re-issue the read request. in fix_sync_read_error()
1968 struct mddev *mddev = r1_bio->mddev; in fix_sync_read_error()
1969 struct r1conf *conf = mddev->private; in fix_sync_read_error()
1970 struct bio *bio = r1_bio->bios[r1_bio->read_disk]; in fix_sync_read_error()
1971 struct page **pages = get_resync_pages(bio)->pages; in fix_sync_read_error()
1972 sector_t sect = r1_bio->sector; in fix_sync_read_error()
1973 int sectors = r1_bio->sectors; in fix_sync_read_error()
1977 rdev = conf->mirrors[r1_bio->read_disk].rdev; in fix_sync_read_error()
1978 if (test_bit(FailFast, &rdev->flags)) { in fix_sync_read_error()
1979 /* Don't try recovering from here - just fail it in fix_sync_read_error()
1982 if (test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
1986 bio->bi_end_io = end_sync_write; in fix_sync_read_error()
1991 int d = r1_bio->read_disk; in fix_sync_read_error()
1998 if (r1_bio->bios[d]->bi_end_io == end_sync_read) { in fix_sync_read_error()
2003 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2012 if (d == conf->raid_disks * 2) in fix_sync_read_error()
2014 } while (!success && d != r1_bio->read_disk); in fix_sync_read_error()
2026 (unsigned long long)r1_bio->sector); in fix_sync_read_error()
2027 for (d = 0; d < conf->raid_disks * 2; d++) { in fix_sync_read_error()
2028 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2029 if (!rdev || test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2035 conf->recovery_disabled = in fix_sync_read_error()
2036 mddev->recovery_disabled; in fix_sync_read_error()
2037 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in fix_sync_read_error()
2038 md_done_sync(mddev, r1_bio->sectors, 0); in fix_sync_read_error()
2043 sectors -= s; in fix_sync_read_error()
2050 /* write it back and re-read */ in fix_sync_read_error()
2051 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2053 d = conf->raid_disks * 2; in fix_sync_read_error()
2054 d--; in fix_sync_read_error()
2055 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2057 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2061 r1_bio->bios[d]->bi_end_io = NULL; in fix_sync_read_error()
2066 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2068 d = conf->raid_disks * 2; in fix_sync_read_error()
2069 d--; in fix_sync_read_error()
2070 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2072 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2076 atomic_add(s, &rdev->corrected_errors); in fix_sync_read_error()
2078 sectors -= s; in fix_sync_read_error()
2082 set_bit(R1BIO_Uptodate, &r1_bio->state); in fix_sync_read_error()
2083 bio->bi_status = 0; in fix_sync_read_error()
2094 * attempt an over-write in process_checks()
2096 struct mddev *mddev = r1_bio->mddev; in process_checks()
2097 struct r1conf *conf = mddev->private; in process_checks()
2103 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); in process_checks()
2104 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2106 struct bio *b = r1_bio->bios[i]; in process_checks()
2108 if (b->bi_end_io != end_sync_read) in process_checks()
2111 status = b->bi_status; in process_checks()
2113 b->bi_status = status; in process_checks()
2114 b->bi_iter.bi_sector = r1_bio->sector + in process_checks()
2115 conf->mirrors[i].rdev->data_offset; in process_checks()
2116 bio_set_dev(b, conf->mirrors[i].rdev->bdev); in process_checks()
2117 b->bi_end_io = end_sync_read; in process_checks()
2118 rp->raid_bio = r1_bio; in process_checks()
2119 b->bi_private = rp; in process_checks()
2122 md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); in process_checks()
2124 for (primary = 0; primary < conf->raid_disks * 2; primary++) in process_checks()
2125 if (r1_bio->bios[primary]->bi_end_io == end_sync_read && in process_checks()
2126 !r1_bio->bios[primary]->bi_status) { in process_checks()
2127 r1_bio->bios[primary]->bi_end_io = NULL; in process_checks()
2128 rdev_dec_pending(conf->mirrors[primary].rdev, mddev); in process_checks()
2131 r1_bio->read_disk = primary; in process_checks()
2132 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2134 struct bio *pbio = r1_bio->bios[primary]; in process_checks()
2135 struct bio *sbio = r1_bio->bios[i]; in process_checks()
2136 blk_status_t status = sbio->bi_status; in process_checks()
2137 struct page **ppages = get_resync_pages(pbio)->pages; in process_checks()
2138 struct page **spages = get_resync_pages(sbio)->pages; in process_checks()
2143 if (sbio->bi_end_io != end_sync_read) in process_checks()
2146 sbio->bi_status = 0; in process_checks()
2149 page_len[j++] = bi->bv_len; in process_checks()
2152 for (j = vcnt; j-- ; ) { in process_checks()
2161 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); in process_checks()
2162 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) in process_checks()
2165 sbio->bi_end_io = NULL; in process_checks()
2166 rdev_dec_pending(conf->mirrors[i].rdev, mddev); in process_checks()
2176 struct r1conf *conf = mddev->private; in sync_request_write()
2178 int disks = conf->raid_disks * 2; in sync_request_write()
2181 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in sync_request_write()
2182 /* ouch - failed to read all of that. */ in sync_request_write()
2186 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in sync_request_write()
2192 atomic_set(&r1_bio->remaining, 1); in sync_request_write()
2194 wbio = r1_bio->bios[i]; in sync_request_write()
2195 if (wbio->bi_end_io == NULL || in sync_request_write()
2196 (wbio->bi_end_io == end_sync_read && in sync_request_write()
2197 (i == r1_bio->read_disk || in sync_request_write()
2198 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) in sync_request_write()
2200 if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { in sync_request_write()
2206 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) in sync_request_write()
2207 wbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2209 wbio->bi_end_io = end_sync_write; in sync_request_write()
2210 atomic_inc(&r1_bio->remaining); in sync_request_write()
2211 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); in sync_request_write()
2230 struct mddev *mddev = conf->mddev; in fix_read_error()
2246 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2248 (test_bit(In_sync, &rdev->flags) || in fix_read_error()
2249 (!test_bit(Faulty, &rdev->flags) && in fix_read_error()
2250 rdev->recovery_offset >= sect + s)) && in fix_read_error()
2253 atomic_inc(&rdev->nr_pending); in fix_read_error()
2256 conf->tmppage, REQ_OP_READ, 0, false)) in fix_read_error()
2264 if (d == conf->raid_disks * 2) in fix_read_error()
2269 /* Cannot read from anywhere - mark it bad */ in fix_read_error()
2270 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; in fix_read_error()
2275 /* write it back and re-read */ in fix_read_error()
2279 d = conf->raid_disks * 2; in fix_read_error()
2280 d--; in fix_read_error()
2282 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2284 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2285 atomic_inc(&rdev->nr_pending); in fix_read_error()
2288 conf->tmppage, WRITE); in fix_read_error()
2297 d = conf->raid_disks * 2; in fix_read_error()
2298 d--; in fix_read_error()
2300 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2302 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2303 atomic_inc(&rdev->nr_pending); in fix_read_error()
2306 conf->tmppage, READ)) { in fix_read_error()
2307 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2311 rdev->data_offset), in fix_read_error()
2312 bdevname(rdev->bdev, b)); in fix_read_error()
2318 sectors -= s; in fix_read_error()
2325 struct mddev *mddev = r1_bio->mddev; in narrow_write_error()
2326 struct r1conf *conf = mddev->private; in narrow_write_error()
2327 struct md_rdev *rdev = conf->mirrors[i].rdev; in narrow_write_error()
2329 /* bio has the data to be written to device 'i' where in narrow_write_error()
2343 int sect_to_write = r1_bio->sectors; in narrow_write_error()
2346 if (rdev->badblocks.shift < 0) in narrow_write_error()
2349 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2350 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2351 sector = r1_bio->sector; in narrow_write_error()
2353 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2354 - sector; in narrow_write_error()
2362 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in narrow_write_error()
2363 wbio = bio_clone_fast(r1_bio->behind_master_bio, in narrow_write_error()
2365 &mddev->bio_set); in narrow_write_error()
2367 wbio = bio_clone_fast(r1_bio->master_bio, GFP_NOIO, in narrow_write_error()
2368 &mddev->bio_set); in narrow_write_error()
2372 wbio->bi_iter.bi_sector = r1_bio->sector; in narrow_write_error()
2373 wbio->bi_iter.bi_size = r1_bio->sectors << 9; in narrow_write_error()
2375 bio_trim(wbio, sector - r1_bio->sector, sectors); in narrow_write_error()
2376 wbio->bi_iter.bi_sector += rdev->data_offset; in narrow_write_error()
2377 bio_set_dev(wbio, rdev->bdev); in narrow_write_error()
2386 sect_to_write -= sectors; in narrow_write_error()
2396 int s = r1_bio->sectors; in handle_sync_write_finished()
2397 for (m = 0; m < conf->raid_disks * 2 ; m++) { in handle_sync_write_finished()
2398 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_sync_write_finished()
2399 struct bio *bio = r1_bio->bios[m]; in handle_sync_write_finished()
2400 if (bio->bi_end_io == NULL) in handle_sync_write_finished()
2402 if (!bio->bi_status && in handle_sync_write_finished()
2403 test_bit(R1BIO_MadeGood, &r1_bio->state)) { in handle_sync_write_finished()
2404 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); in handle_sync_write_finished()
2406 if (bio->bi_status && in handle_sync_write_finished()
2407 test_bit(R1BIO_WriteError, &r1_bio->state)) { in handle_sync_write_finished()
2408 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) in handle_sync_write_finished()
2409 md_error(conf->mddev, rdev); in handle_sync_write_finished()
2413 md_done_sync(conf->mddev, s, 1); in handle_sync_write_finished()
2421 for (m = 0; m < conf->raid_disks * 2 ; m++) in handle_write_finished()
2422 if (r1_bio->bios[m] == IO_MADE_GOOD) { in handle_write_finished()
2423 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_write_finished()
2425 r1_bio->sector, in handle_write_finished()
2426 r1_bio->sectors, 0); in handle_write_finished()
2427 rdev_dec_pending(rdev, conf->mddev); in handle_write_finished()
2428 } else if (r1_bio->bios[m] != NULL) { in handle_write_finished()
2435 md_error(conf->mddev, in handle_write_finished()
2436 conf->mirrors[m].rdev); in handle_write_finished()
2438 set_bit(R1BIO_Degraded, &r1_bio->state); in handle_write_finished()
2440 rdev_dec_pending(conf->mirrors[m].rdev, in handle_write_finished()
2441 conf->mddev); in handle_write_finished()
2444 spin_lock_irq(&conf->device_lock); in handle_write_finished()
2445 list_add(&r1_bio->retry_list, &conf->bio_end_io_list); in handle_write_finished()
2446 idx = sector_to_idx(r1_bio->sector); in handle_write_finished()
2447 atomic_inc(&conf->nr_queued[idx]); in handle_write_finished()
2448 spin_unlock_irq(&conf->device_lock); in handle_write_finished()
2453 wake_up(&conf->wait_barrier); in handle_write_finished()
2454 md_wakeup_thread(conf->mddev->thread); in handle_write_finished()
2456 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in handle_write_finished()
2464 struct mddev *mddev = conf->mddev; in handle_read_error()
2468 clear_bit(R1BIO_ReadError, &r1_bio->state); in handle_read_error()
2472 * other devices. When we find one, we re-write in handle_read_error()
2478 bio = r1_bio->bios[r1_bio->read_disk]; in handle_read_error()
2480 r1_bio->bios[r1_bio->read_disk] = NULL; in handle_read_error()
2482 rdev = conf->mirrors[r1_bio->read_disk].rdev; in handle_read_error()
2483 if (mddev->ro == 0 in handle_read_error()
2484 && !test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2486 fix_read_error(conf, r1_bio->read_disk, in handle_read_error()
2487 r1_bio->sector, r1_bio->sectors); in handle_read_error()
2489 } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2492 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; in handle_read_error()
2495 rdev_dec_pending(rdev, conf->mddev); in handle_read_error()
2496 allow_barrier(conf, r1_bio->sector); in handle_read_error()
2497 bio = r1_bio->master_bio; in handle_read_error()
2500 r1_bio->state = 0; in handle_read_error()
2501 raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); in handle_read_error()
2506 struct mddev *mddev = thread->mddev; in raid1d()
2509 struct r1conf *conf = mddev->private; in raid1d()
2510 struct list_head *head = &conf->retry_list; in raid1d()
2516 if (!list_empty_careful(&conf->bio_end_io_list) && in raid1d()
2517 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid1d()
2519 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2520 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) in raid1d()
2521 list_splice_init(&conf->bio_end_io_list, &tmp); in raid1d()
2522 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2526 list_del(&r1_bio->retry_list); in raid1d()
2527 idx = sector_to_idx(r1_bio->sector); in raid1d()
2528 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2529 if (mddev->degraded) in raid1d()
2530 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1d()
2531 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2542 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2544 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2547 r1_bio = list_entry(head->prev, struct r1bio, retry_list); in raid1d()
2548 list_del(head->prev); in raid1d()
2549 idx = sector_to_idx(r1_bio->sector); in raid1d()
2550 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2551 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2553 mddev = r1_bio->mddev; in raid1d()
2554 conf = mddev->private; in raid1d()
2555 if (test_bit(R1BIO_IsSync, &r1_bio->state)) { in raid1d()
2556 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2557 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2561 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2562 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2564 else if (test_bit(R1BIO_ReadError, &r1_bio->state)) in raid1d()
2570 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid1d()
2581 BUG_ON(mempool_initialized(&conf->r1buf_pool)); in init_resync()
2583 return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, in init_resync()
2584 r1buf_pool_free, conf->poolinfo); in init_resync()
2589 struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); in raid1_alloc_init_r1buf()
2594 for (i = conf->poolinfo->raid_disks; i--; ) { in raid1_alloc_init_r1buf()
2595 bio = r1bio->bios[i]; in raid1_alloc_init_r1buf()
2596 rps = bio->bi_private; in raid1_alloc_init_r1buf()
2598 bio->bi_private = rps; in raid1_alloc_init_r1buf()
2600 r1bio->master_bio = NULL; in raid1_alloc_init_r1buf()
2607 * We need to make sure that no normal I/O request - particularly write
2608 * requests - conflict with active sync requests.
2617 struct r1conf *conf = mddev->private; in raid1_sync_request()
2621 int disk = -1; in raid1_sync_request()
2623 int wonly = -1; in raid1_sync_request()
2632 if (!mempool_initialized(&conf->r1buf_pool)) in raid1_sync_request()
2636 max_sector = mddev->dev_sectors; in raid1_sync_request()
2641 * We can find the current addess in mddev->curr_resync in raid1_sync_request()
2643 if (mddev->curr_resync < max_sector) /* aborted */ in raid1_sync_request()
2644 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, in raid1_sync_request()
2647 conf->fullsync = 0; in raid1_sync_request()
2649 md_bitmap_close_sync(mddev->bitmap); in raid1_sync_request()
2653 conf->cluster_sync_low = 0; in raid1_sync_request()
2654 conf->cluster_sync_high = 0; in raid1_sync_request()
2659 if (mddev->bitmap == NULL && in raid1_sync_request()
2660 mddev->recovery_cp == MaxSector && in raid1_sync_request()
2661 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid1_sync_request()
2662 conf->fullsync == 0) { in raid1_sync_request()
2664 return max_sector - sector_nr; in raid1_sync_request()
2669 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && in raid1_sync_request()
2670 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2677 * If there is non-resync activity waiting for a turn, then let it in raid1_sync_request()
2680 if (atomic_read(&conf->nr_waiting[idx])) in raid1_sync_request()
2687 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2688 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid1_sync_request()
2701 * and any others (which will be non-In_sync devices) for WRITE. in raid1_sync_request()
2706 r1_bio->mddev = mddev; in raid1_sync_request()
2707 r1_bio->sector = sector_nr; in raid1_sync_request()
2708 r1_bio->state = 0; in raid1_sync_request()
2709 set_bit(R1BIO_IsSync, &r1_bio->state); in raid1_sync_request()
2713 for (i = 0; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2715 bio = r1_bio->bios[i]; in raid1_sync_request()
2717 rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_sync_request()
2719 test_bit(Faulty, &rdev->flags)) { in raid1_sync_request()
2720 if (i < conf->raid_disks) in raid1_sync_request()
2722 } else if (!test_bit(In_sync, &rdev->flags)) { in raid1_sync_request()
2724 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2734 good_sectors = first_bad - sector_nr; in raid1_sync_request()
2736 bad_sectors -= (sector_nr - first_bad); in raid1_sync_request()
2743 if (test_bit(WriteMostly, &rdev->flags)) { in raid1_sync_request()
2751 bio->bi_end_io = end_sync_read; in raid1_sync_request()
2753 } else if (!test_bit(WriteErrorSeen, &rdev->flags) && in raid1_sync_request()
2754 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid1_sync_request()
2755 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { in raid1_sync_request()
2763 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2767 if (rdev && bio->bi_end_io) { in raid1_sync_request()
2768 atomic_inc(&rdev->nr_pending); in raid1_sync_request()
2769 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; in raid1_sync_request()
2770 bio_set_dev(bio, rdev->bdev); in raid1_sync_request()
2771 if (test_bit(FailFast, &rdev->flags)) in raid1_sync_request()
2772 bio->bi_opf |= MD_FAILFAST; in raid1_sync_request()
2778 r1_bio->read_disk = disk; in raid1_sync_request()
2785 for (i = 0 ; i < conf->raid_disks * 2 ; i++) in raid1_sync_request()
2786 if (r1_bio->bios[i]->bi_end_io == end_sync_write) { in raid1_sync_request()
2787 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_sync_request()
2792 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid1_sync_request()
2802 conf->recovery_disabled = mddev->recovery_disabled; in raid1_sync_request()
2803 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_sync_request()
2810 /* only resync enough to reach the next bad->good in raid1_sync_request()
2815 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) in raid1_sync_request()
2817 write_targets += read_targets-1; in raid1_sync_request()
2820 /* There is nowhere to write, so all non-sync in raid1_sync_request()
2821 * drives must be failed - so we are finished in raid1_sync_request()
2826 rv = max_sector - sector_nr; in raid1_sync_request()
2832 if (max_sector > mddev->resync_max) in raid1_sync_request()
2833 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid1_sync_request()
2842 len = (max_sector - sector_nr) << 9; in raid1_sync_request()
2846 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2848 !conf->fullsync && in raid1_sync_request()
2849 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in raid1_sync_request()
2855 for (i = 0 ; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2858 bio = r1_bio->bios[i]; in raid1_sync_request()
2860 if (bio->bi_end_io) { in raid1_sync_request()
2872 sync_blocks -= (len>>9); in raid1_sync_request()
2875 r1_bio->sectors = nr_sectors; in raid1_sync_request()
2878 conf->cluster_sync_high < sector_nr + nr_sectors) { in raid1_sync_request()
2879 conf->cluster_sync_low = mddev->curr_resync_completed; in raid1_sync_request()
2880 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; in raid1_sync_request()
2882 md_cluster_ops->resync_info_update(mddev, in raid1_sync_request()
2883 conf->cluster_sync_low, in raid1_sync_request()
2884 conf->cluster_sync_high); in raid1_sync_request()
2887 /* For a user-requested sync, we read all readable devices and do a in raid1_sync_request()
2890 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2891 atomic_set(&r1_bio->remaining, read_targets); in raid1_sync_request()
2892 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { in raid1_sync_request()
2893 bio = r1_bio->bios[i]; in raid1_sync_request()
2894 if (bio->bi_end_io == end_sync_read) { in raid1_sync_request()
2895 read_targets--; in raid1_sync_request()
2898 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2903 atomic_set(&r1_bio->remaining, 1); in raid1_sync_request()
2904 bio = r1_bio->bios[r1_bio->read_disk]; in raid1_sync_request()
2907 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2918 return mddev->dev_sectors; in raid1_size()
2927 int err = -ENOMEM; in setup_conf()
2933 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2935 if (!conf->nr_pending) in setup_conf()
2938 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2940 if (!conf->nr_waiting) in setup_conf()
2943 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2945 if (!conf->nr_queued) in setup_conf()
2948 conf->barrier = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2950 if (!conf->barrier) in setup_conf()
2953 conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), in setup_conf()
2954 mddev->raid_disks, 2), in setup_conf()
2956 if (!conf->mirrors) in setup_conf()
2959 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
2960 if (!conf->tmppage) in setup_conf()
2963 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); in setup_conf()
2964 if (!conf->poolinfo) in setup_conf()
2966 conf->poolinfo->raid_disks = mddev->raid_disks * 2; in setup_conf()
2967 err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, in setup_conf()
2968 rbio_pool_free, conf->poolinfo); in setup_conf()
2972 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
2976 conf->poolinfo->mddev = mddev; in setup_conf()
2978 err = -EINVAL; in setup_conf()
2979 spin_lock_init(&conf->device_lock); in setup_conf()
2981 int disk_idx = rdev->raid_disk; in setup_conf()
2982 if (disk_idx >= mddev->raid_disks in setup_conf()
2985 if (test_bit(Replacement, &rdev->flags)) in setup_conf()
2986 disk = conf->mirrors + mddev->raid_disks + disk_idx; in setup_conf()
2988 disk = conf->mirrors + disk_idx; in setup_conf()
2990 if (disk->rdev) in setup_conf()
2992 disk->rdev = rdev; in setup_conf()
2993 disk->head_position = 0; in setup_conf()
2994 disk->seq_start = MaxSector; in setup_conf()
2996 conf->raid_disks = mddev->raid_disks; in setup_conf()
2997 conf->mddev = mddev; in setup_conf()
2998 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
2999 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
3001 spin_lock_init(&conf->resync_lock); in setup_conf()
3002 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
3004 bio_list_init(&conf->pending_bio_list); in setup_conf()
3005 conf->pending_count = 0; in setup_conf()
3006 conf->recovery_disabled = mddev->recovery_disabled - 1; in setup_conf()
3008 err = -EIO; in setup_conf()
3009 for (i = 0; i < conf->raid_disks * 2; i++) { in setup_conf()
3011 disk = conf->mirrors + i; in setup_conf()
3013 if (i < conf->raid_disks && in setup_conf()
3014 disk[conf->raid_disks].rdev) { in setup_conf()
3016 if (!disk->rdev) { in setup_conf()
3020 disk->rdev = in setup_conf()
3021 disk[conf->raid_disks].rdev; in setup_conf()
3022 disk[conf->raid_disks].rdev = NULL; in setup_conf()
3023 } else if (!test_bit(In_sync, &disk->rdev->flags)) in setup_conf()
3024 /* Original is not in_sync - bad */ in setup_conf()
3028 if (!disk->rdev || in setup_conf()
3029 !test_bit(In_sync, &disk->rdev->flags)) { in setup_conf()
3030 disk->head_position = 0; in setup_conf()
3031 if (disk->rdev && in setup_conf()
3032 (disk->rdev->saved_raid_disk < 0)) in setup_conf()
3033 conf->fullsync = 1; in setup_conf()
3037 err = -ENOMEM; in setup_conf()
3038 conf->thread = md_register_thread(raid1d, mddev, "raid1"); in setup_conf()
3039 if (!conf->thread) in setup_conf()
3046 mempool_exit(&conf->r1bio_pool); in setup_conf()
3047 kfree(conf->mirrors); in setup_conf()
3048 safe_put_page(conf->tmppage); in setup_conf()
3049 kfree(conf->poolinfo); in setup_conf()
3050 kfree(conf->nr_pending); in setup_conf()
3051 kfree(conf->nr_waiting); in setup_conf()
3052 kfree(conf->nr_queued); in setup_conf()
3053 kfree(conf->barrier); in setup_conf()
3054 bioset_exit(&conf->bio_split); in setup_conf()
3069 if (mddev->level != 1) { in raid1_run()
3071 mdname(mddev), mddev->level); in raid1_run()
3072 return -EIO; in raid1_run()
3074 if (mddev->reshape_position != MaxSector) { in raid1_run()
3077 return -EIO; in raid1_run()
3080 return -ENOMEM; in raid1_run()
3086 if (mddev->private == NULL) in raid1_run()
3089 conf = mddev->private; in raid1_run()
3094 if (mddev->queue) { in raid1_run()
3095 blk_queue_max_write_same_sectors(mddev->queue, 0); in raid1_run()
3096 blk_queue_max_write_zeroes_sectors(mddev->queue, 0); in raid1_run()
3100 if (!mddev->gendisk) in raid1_run()
3102 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_run()
3103 rdev->data_offset << 9); in raid1_run()
3104 if (blk_queue_discard(bdev_get_queue(rdev->bdev))) in raid1_run()
3108 mddev->degraded = 0; in raid1_run()
3109 for (i = 0; i < conf->raid_disks; i++) in raid1_run()
3110 if (conf->mirrors[i].rdev == NULL || in raid1_run()
3111 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || in raid1_run()
3112 test_bit(Faulty, &conf->mirrors[i].rdev->flags)) in raid1_run()
3113 mddev->degraded++; in raid1_run()
3117 if (conf->raid_disks - mddev->degraded < 1) { in raid1_run()
3118 ret = -EINVAL; in raid1_run()
3122 if (conf->raid_disks - mddev->degraded == 1) in raid1_run()
3123 mddev->recovery_cp = MaxSector; in raid1_run()
3125 if (mddev->recovery_cp != MaxSector) in raid1_run()
3126 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", in raid1_run()
3129 mdname(mddev), mddev->raid_disks - mddev->degraded, in raid1_run()
3130 mddev->raid_disks); in raid1_run()
3135 mddev->thread = conf->thread; in raid1_run()
3136 conf->thread = NULL; in raid1_run()
3137 mddev->private = conf; in raid1_run()
3138 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid1_run()
3142 if (mddev->queue) { in raid1_run()
3145 mddev->queue); in raid1_run()
3148 mddev->queue); in raid1_run()
3153 md_unregister_thread(&mddev->thread); in raid1_run()
3167 mempool_exit(&conf->r1bio_pool); in raid1_free()
3168 kfree(conf->mirrors); in raid1_free()
3169 safe_put_page(conf->tmppage); in raid1_free()
3170 kfree(conf->poolinfo); in raid1_free()
3171 kfree(conf->nr_pending); in raid1_free()
3172 kfree(conf->nr_waiting); in raid1_free()
3173 kfree(conf->nr_queued); in raid1_free()
3174 kfree(conf->barrier); in raid1_free()
3175 bioset_exit(&conf->bio_split); in raid1_free()
3189 if (mddev->external_size && in raid1_resize()
3190 mddev->array_sectors > newsize) in raid1_resize()
3191 return -EINVAL; in raid1_resize()
3192 if (mddev->bitmap) { in raid1_resize()
3193 int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid1_resize()
3198 if (sectors > mddev->dev_sectors && in raid1_resize()
3199 mddev->recovery_cp > mddev->dev_sectors) { in raid1_resize()
3200 mddev->recovery_cp = mddev->dev_sectors; in raid1_resize()
3201 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_resize()
3203 mddev->dev_sectors = sectors; in raid1_resize()
3204 mddev->resync_max_sectors = sectors; in raid1_resize()
3212 * 2/ resize conf->mirrors in raid1_reshape()
3216 * Then resize conf->mirrors and swap in the new r1bio pool. in raid1_reshape()
3224 struct r1conf *conf = mddev->private; in raid1_reshape()
3234 if (mddev->chunk_sectors != mddev->new_chunk_sectors || in raid1_reshape()
3235 mddev->layout != mddev->new_layout || in raid1_reshape()
3236 mddev->level != mddev->new_level) { in raid1_reshape()
3237 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid1_reshape()
3238 mddev->new_layout = mddev->layout; in raid1_reshape()
3239 mddev->new_level = mddev->level; in raid1_reshape()
3240 return -EINVAL; in raid1_reshape()
3246 raid_disks = mddev->raid_disks + mddev->delta_disks; in raid1_reshape()
3248 if (raid_disks < conf->raid_disks) { in raid1_reshape()
3250 for (d= 0; d < conf->raid_disks; d++) in raid1_reshape()
3251 if (conf->mirrors[d].rdev) in raid1_reshape()
3254 return -EBUSY; in raid1_reshape()
3259 return -ENOMEM; in raid1_reshape()
3260 newpoolinfo->mddev = mddev; in raid1_reshape()
3261 newpoolinfo->raid_disks = raid_disks * 2; in raid1_reshape()
3275 return -ENOMEM; in raid1_reshape()
3281 oldpool = conf->r1bio_pool; in raid1_reshape()
3282 conf->r1bio_pool = newpool; in raid1_reshape()
3284 for (d = d2 = 0; d < conf->raid_disks; d++) { in raid1_reshape()
3285 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid1_reshape()
3286 if (rdev && rdev->raid_disk != d2) { in raid1_reshape()
3288 rdev->raid_disk = d2; in raid1_reshape()
3292 mdname(mddev), rdev->raid_disk); in raid1_reshape()
3297 kfree(conf->mirrors); in raid1_reshape()
3298 conf->mirrors = newmirrors; in raid1_reshape()
3299 kfree(conf->poolinfo); in raid1_reshape()
3300 conf->poolinfo = newpoolinfo; in raid1_reshape()
3302 spin_lock_irqsave(&conf->device_lock, flags); in raid1_reshape()
3303 mddev->degraded += (raid_disks - conf->raid_disks); in raid1_reshape()
3304 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_reshape()
3305 conf->raid_disks = mddev->raid_disks = raid_disks; in raid1_reshape()
3306 mddev->delta_disks = 0; in raid1_reshape()
3310 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); in raid1_reshape()
3311 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_reshape()
3312 md_wakeup_thread(mddev->thread); in raid1_reshape()
3320 struct r1conf *conf = mddev->private; in raid1_quiesce()
3333 if (mddev->level == 5 && mddev->raid_disks == 2) { in raid1_takeover()
3335 mddev->new_level = 1; in raid1_takeover()
3336 mddev->new_layout = 0; in raid1_takeover()
3337 mddev->new_chunk_sectors = 0; in raid1_takeover()
3341 conf->array_frozen = 1; in raid1_takeover()
3347 return ERR_PTR(-EINVAL); in raid1_takeover()
3385 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3386 MODULE_ALIAS("md-raid1");
3387 MODULE_ALIAS("md-level-1");