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 bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); in r1buf_pool_alloc()
172 r1_bio->bios[j] = bio; in r1buf_pool_alloc()
175 * Allocate RESYNC_PAGES data pages and attach them to in r1buf_pool_alloc()
177 * If this is a user-requested check/repair, allocate in r1buf_pool_alloc()
180 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) in r1buf_pool_alloc()
181 need_pages = pi->raid_disks; in r1buf_pool_alloc()
184 for (j = 0; j < pi->raid_disks; j++) { in r1buf_pool_alloc()
187 bio = r1_bio->bios[j]; in r1buf_pool_alloc()
197 rp->raid_bio = r1_bio; in r1buf_pool_alloc()
198 bio->bi_private = rp; in r1buf_pool_alloc()
201 r1_bio->master_bio = NULL; in r1buf_pool_alloc()
206 while (--j >= 0) in r1buf_pool_alloc()
210 while (++j < pi->raid_disks) { in r1buf_pool_alloc()
211 bio_uninit(r1_bio->bios[j]); in r1buf_pool_alloc()
212 kfree(r1_bio->bios[j]); in r1buf_pool_alloc()
217 rbio_pool_free(r1_bio, data); in r1buf_pool_alloc()
221 static void r1buf_pool_free(void *__r1_bio, void *data) in r1buf_pool_free() argument
223 struct pool_info *pi = data; in r1buf_pool_free()
228 for (i = pi->raid_disks; i--; ) { in r1buf_pool_free()
229 rp = get_resync_pages(r1bio->bios[i]); in r1buf_pool_free()
231 bio_uninit(r1bio->bios[i]); in r1buf_pool_free()
232 kfree(r1bio->bios[i]); in r1buf_pool_free()
238 rbio_pool_free(r1bio, data); in r1buf_pool_free()
245 for (i = 0; i < conf->raid_disks * 2; i++) { in put_all_bios()
246 struct bio **bio = r1_bio->bios + i; in put_all_bios()
255 struct r1conf *conf = r1_bio->mddev->private; in free_r1bio()
258 mempool_free(r1_bio, &conf->r1bio_pool); in free_r1bio()
263 struct r1conf *conf = r1_bio->mddev->private; in put_buf()
264 sector_t sect = r1_bio->sector; in put_buf()
267 for (i = 0; i < conf->raid_disks * 2; i++) { in put_buf()
268 struct bio *bio = r1_bio->bios[i]; in put_buf()
269 if (bio->bi_end_io) in put_buf()
270 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); in put_buf()
273 mempool_free(r1_bio, &conf->r1buf_pool); in put_buf()
281 struct mddev *mddev = r1_bio->mddev; in reschedule_retry()
282 struct r1conf *conf = mddev->private; in reschedule_retry()
285 idx = sector_to_idx(r1_bio->sector); in reschedule_retry()
286 spin_lock_irqsave(&conf->device_lock, flags); in reschedule_retry()
287 list_add(&r1_bio->retry_list, &conf->retry_list); in reschedule_retry()
288 atomic_inc(&conf->nr_queued[idx]); in reschedule_retry()
289 spin_unlock_irqrestore(&conf->device_lock, flags); in reschedule_retry()
291 wake_up(&conf->wait_barrier); in reschedule_retry()
292 md_wakeup_thread(mddev->thread); in reschedule_retry()
302 struct bio *bio = r1_bio->master_bio; in call_bio_endio()
304 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in call_bio_endio()
305 bio->bi_status = BLK_STS_IOERR; in call_bio_endio()
312 struct bio *bio = r1_bio->master_bio; in raid_end_bio_io()
313 struct r1conf *conf = r1_bio->mddev->private; in raid_end_bio_io()
314 sector_t sector = r1_bio->sector; in raid_end_bio_io()
317 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid_end_bio_io()
318 pr_debug("raid1: sync end %s on sectors %llu-%llu\n", in raid_end_bio_io()
320 (unsigned long long) bio->bi_iter.bi_sector, in raid_end_bio_io()
321 (unsigned long long) bio_end_sector(bio) - 1); in raid_end_bio_io()
329 * to go idle. All I/Os, even write-behind writes, are done. in raid_end_bio_io()
339 struct r1conf *conf = r1_bio->mddev->private; in update_head_pos()
341 conf->mirrors[disk].head_position = in update_head_pos()
342 r1_bio->sector + (r1_bio->sectors); in update_head_pos()
350 int mirror; in find_bio_disk() local
351 struct r1conf *conf = r1_bio->mddev->private; in find_bio_disk()
352 int raid_disks = conf->raid_disks; in find_bio_disk()
354 for (mirror = 0; mirror < raid_disks * 2; mirror++) in find_bio_disk()
355 if (r1_bio->bios[mirror] == bio) in find_bio_disk()
358 BUG_ON(mirror == raid_disks * 2); in find_bio_disk()
359 update_head_pos(mirror, r1_bio); in find_bio_disk()
361 return mirror; in find_bio_disk()
366 int uptodate = !bio->bi_status; in raid1_end_read_request()
367 struct r1bio *r1_bio = bio->bi_private; in raid1_end_read_request()
368 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_read_request()
369 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; in raid1_end_read_request()
372 * this branch is our 'one mirror IO has finished' event handler: in raid1_end_read_request()
374 update_head_pos(r1_bio->read_disk, r1_bio); in raid1_end_read_request()
377 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_read_request()
378 else if (test_bit(FailFast, &rdev->flags) && in raid1_end_read_request()
379 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_end_read_request()
380 /* This was a fail-fast read so we definitely in raid1_end_read_request()
389 spin_lock_irqsave(&conf->device_lock, flags); in raid1_end_read_request()
390 if (r1_bio->mddev->degraded == conf->raid_disks || in raid1_end_read_request()
391 (r1_bio->mddev->degraded == conf->raid_disks-1 && in raid1_end_read_request()
392 test_bit(In_sync, &rdev->flags))) in raid1_end_read_request()
394 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_end_read_request()
399 rdev_dec_pending(rdev, conf->mddev); in raid1_end_read_request()
405 mdname(conf->mddev), in raid1_end_read_request()
406 rdev->bdev, in raid1_end_read_request()
407 (unsigned long long)r1_bio->sector); in raid1_end_read_request()
408 set_bit(R1BIO_ReadError, &r1_bio->state); in raid1_end_read_request()
417 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in close_write()
418 bio_free_pages(r1_bio->behind_master_bio); in close_write()
419 bio_put(r1_bio->behind_master_bio); in close_write()
420 r1_bio->behind_master_bio = NULL; in close_write()
423 md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, in close_write()
424 r1_bio->sectors, in close_write()
425 !test_bit(R1BIO_Degraded, &r1_bio->state), in close_write()
426 test_bit(R1BIO_BehindIO, &r1_bio->state)); in close_write()
427 md_write_end(r1_bio->mddev); in close_write()
432 if (!atomic_dec_and_test(&r1_bio->remaining)) in r1_bio_write_done()
435 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in r1_bio_write_done()
439 if (test_bit(R1BIO_MadeGood, &r1_bio->state)) in r1_bio_write_done()
448 struct r1bio *r1_bio = bio->bi_private; in raid1_end_write_request()
449 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); in raid1_end_write_request()
450 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_write_request()
452 int mirror = find_bio_disk(r1_bio, bio); in raid1_end_write_request() local
453 struct md_rdev *rdev = conf->mirrors[mirror].rdev; in raid1_end_write_request()
455 sector_t lo = r1_bio->sector; in raid1_end_write_request()
456 sector_t hi = r1_bio->sector + r1_bio->sectors; in raid1_end_write_request()
458 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; in raid1_end_write_request()
461 * 'one mirror IO has finished' event handler: in raid1_end_write_request()
463 if (bio->bi_status && !discard_error) { in raid1_end_write_request()
464 set_bit(WriteErrorSeen, &rdev->flags); in raid1_end_write_request()
465 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in raid1_end_write_request()
467 conf->mddev->recovery); in raid1_end_write_request()
469 if (test_bit(FailFast, &rdev->flags) && in raid1_end_write_request()
470 (bio->bi_opf & MD_FAILFAST) && in raid1_end_write_request()
472 !test_bit(WriteMostly, &rdev->flags)) { in raid1_end_write_request()
473 md_error(r1_bio->mddev, rdev); in raid1_end_write_request()
480 if (!test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
481 set_bit(R1BIO_WriteError, &r1_bio->state); in raid1_end_write_request()
484 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1_end_write_request()
486 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
497 * to user-side. So if something waits for IO, then it in raid1_end_write_request()
503 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
508 * such device for properly reading the data back (we could in raid1_end_write_request()
510 * before rdev->recovery_offset, but for simplicity we don't in raid1_end_write_request()
513 if (test_bit(In_sync, &rdev->flags) && in raid1_end_write_request()
514 !test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
515 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_write_request()
518 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in raid1_end_write_request()
520 r1_bio->bios[mirror] = IO_MADE_GOOD; in raid1_end_write_request()
521 set_bit(R1BIO_MadeGood, &r1_bio->state); in raid1_end_write_request()
526 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_end_write_request()
528 if (test_bit(WriteMostly, &rdev->flags)) in raid1_end_write_request()
529 atomic_dec(&r1_bio->behind_remaining); in raid1_end_write_request()
533 * has safely reached all non-writemostly in raid1_end_write_request()
535 * gets done only once -- we don't ever want to return in raid1_end_write_request()
536 * -EIO here, instead we'll wait in raid1_end_write_request()
538 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && in raid1_end_write_request()
539 test_bit(R1BIO_Uptodate, &r1_bio->state)) { in raid1_end_write_request()
541 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid1_end_write_request()
542 struct bio *mbio = r1_bio->master_bio; in raid1_end_write_request()
544 " %llu-%llu\n", in raid1_end_write_request()
545 (unsigned long long) mbio->bi_iter.bi_sector, in raid1_end_write_request()
546 (unsigned long long) bio_end_sector(mbio) - 1); in raid1_end_write_request()
550 } else if (rdev->mddev->serialize_policy) in raid1_end_write_request()
552 if (r1_bio->bios[mirror] == NULL) in raid1_end_write_request()
553 rdev_dec_pending(rdev, conf->mddev); in raid1_end_write_request()
575 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - in align_to_barrier_unit_end()
586 * be done. There is a per-array 'next expected sequential IO' sector
587 * number - if this matches on the next IO then we use the last disk.
588 * There is also a per-disk 'last know head position' sector that is
594 * because position is mirror, not device based.
600 const sector_t this_sector = r1_bio->sector; in read_balance()
619 sectors = r1_bio->sectors; in read_balance()
620 best_disk = -1; in read_balance()
621 best_dist_disk = -1; in read_balance()
623 best_pending_disk = -1; in read_balance()
628 clear_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
630 if ((conf->mddev->recovery_cp < this_sector + sectors) || in read_balance()
631 (mddev_is_clustered(conf->mddev) && in read_balance()
632 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, in read_balance()
638 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in read_balance()
645 rdev = rcu_dereference(conf->mirrors[disk].rdev); in read_balance()
646 if (r1_bio->bios[disk] == IO_BLOCKED in read_balance()
648 || test_bit(Faulty, &rdev->flags)) in read_balance()
650 if (!test_bit(In_sync, &rdev->flags) && in read_balance()
651 rdev->recovery_offset < this_sector + sectors) in read_balance()
653 if (test_bit(WriteMostly, &rdev->flags)) { in read_balance()
654 /* Don't balance among write-mostly, just in read_balance()
662 best_good_sectors = first_bad - this_sector; in read_balance()
683 bad_sectors -= (this_sector - first_bad); in read_balance()
690 sector_t good_sectors = first_bad - this_sector; in read_balance()
701 best_disk = -1; in read_balance()
707 set_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
709 nonrot = bdev_nonrot(rdev->bdev); in read_balance()
711 pending = atomic_read(&rdev->nr_pending); in read_balance()
712 dist = abs(this_sector - conf->mirrors[disk].head_position); in read_balance()
718 if (conf->mirrors[disk].next_seq_sect == this_sector in read_balance()
720 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; in read_balance()
721 struct raid1_info *mirror = &conf->mirrors[disk]; in read_balance() local
738 mirror->seq_start != MaxSector && in read_balance()
739 mirror->next_seq_sect > opt_iosize && in read_balance()
740 mirror->next_seq_sect - opt_iosize >= in read_balance()
741 mirror->seq_start) { in read_balance()
764 * non-rotational, choose the disk with less pending request even the in read_balance()
766 * mixed ratation/non-rotational disks depending on workload. in read_balance()
768 if (best_disk == -1) { in read_balance()
776 rdev = rcu_dereference(conf->mirrors[best_disk].rdev); in read_balance()
779 atomic_inc(&rdev->nr_pending); in read_balance()
782 if (conf->mirrors[best_disk].next_seq_sect != this_sector) in read_balance()
783 conf->mirrors[best_disk].seq_start = this_sector; in read_balance()
785 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; in read_balance()
795 if (wq_has_sleeper(&conf->wait_barrier)) in wake_up_barrier()
796 wake_up(&conf->wait_barrier); in wake_up_barrier()
802 raid1_prepare_flush_writes(conf->mddev->bitmap); in flush_bio_list()
806 struct bio *next = bio->bi_next; in flush_bio_list()
819 spin_lock_irq(&conf->device_lock); in flush_pending_writes()
821 if (conf->pending_bio_list.head) { in flush_pending_writes()
825 bio = bio_list_get(&conf->pending_bio_list); 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()
935 * We need to increase conf->nr_pending[idx] very early here, in _wait_barrier()
937 * conf->nr_pending[idx] to be 0. Then we can avoid holding in _wait_barrier()
938 * conf->resync_lock when there is no barrier raised in same in _wait_barrier()
942 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
944 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then in _wait_barrier()
945 * check conf->barrier[idx]. In raise_barrier() we firstly increase in _wait_barrier()
946 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory in _wait_barrier()
947 * barrier is necessary here to make sure conf->barrier[idx] won't be in _wait_barrier()
948 * fetched before conf->nr_pending[idx] is increased. Otherwise there in _wait_barrier()
955 * here. If during we check conf->barrier[idx], the array is in _wait_barrier()
956 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is in _wait_barrier()
962 if (!READ_ONCE(conf->array_frozen) && in _wait_barrier()
963 !atomic_read(&conf->barrier[idx])) in _wait_barrier()
967 * After holding conf->resync_lock, conf->nr_pending[idx] in _wait_barrier()
970 * raise_barrer() might be waiting for conf->nr_pending[idx] in _wait_barrier()
973 spin_lock_irq(&conf->resync_lock); in _wait_barrier()
974 atomic_inc(&conf->nr_waiting[idx]); in _wait_barrier()
975 atomic_dec(&conf->nr_pending[idx]); in _wait_barrier()
987 wait_event_lock_irq(conf->wait_barrier, in _wait_barrier()
988 !conf->array_frozen && in _wait_barrier()
989 !atomic_read(&conf->barrier[idx]), in _wait_barrier()
990 conf->resync_lock); in _wait_barrier()
991 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
994 atomic_dec(&conf->nr_waiting[idx]); in _wait_barrier()
995 spin_unlock_irq(&conf->resync_lock); in _wait_barrier()
1009 * conf->barrier[idx] here, memory barrier is unnecessary as well. in wait_read_barrier()
1011 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1013 if (!READ_ONCE(conf->array_frozen)) in wait_read_barrier()
1016 spin_lock_irq(&conf->resync_lock); in wait_read_barrier()
1017 atomic_inc(&conf->nr_waiting[idx]); in wait_read_barrier()
1018 atomic_dec(&conf->nr_pending[idx]); in wait_read_barrier()
1031 wait_event_lock_irq(conf->wait_barrier, in wait_read_barrier()
1032 !conf->array_frozen, in wait_read_barrier()
1033 conf->resync_lock); in wait_read_barrier()
1034 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1037 atomic_dec(&conf->nr_waiting[idx]); in wait_read_barrier()
1038 spin_unlock_irq(&conf->resync_lock); in wait_read_barrier()
1051 atomic_dec(&conf->nr_pending[idx]); in _allow_barrier()
1062 /* conf->resync_lock should be held */
1067 ret = atomic_read(&conf->nr_sync_pending); in get_unqueued_pending()
1069 ret += atomic_read(&conf->nr_pending[idx]) - in get_unqueued_pending()
1070 atomic_read(&conf->nr_queued[idx]); in get_unqueued_pending()
1088 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the in freeze_array()
1090 * normal I/O are queued, sum of all conf->nr_pending[] will match sum in freeze_array()
1091 * of all conf->nr_queued[]. But normal I/O failure is an exception, in freeze_array()
1100 spin_lock_irq(&conf->resync_lock); in freeze_array()
1101 conf->array_frozen = 1; in freeze_array()
1102 raid1_log(conf->mddev, "wait freeze"); in freeze_array()
1104 conf->wait_barrier, in freeze_array()
1106 conf->resync_lock, in freeze_array()
1108 spin_unlock_irq(&conf->resync_lock); in freeze_array()
1113 spin_lock_irq(&conf->resync_lock); in unfreeze_array()
1114 conf->array_frozen = 0; in unfreeze_array()
1115 spin_unlock_irq(&conf->resync_lock); in unfreeze_array()
1116 wake_up(&conf->wait_barrier); in unfreeze_array()
1122 int size = bio->bi_iter.bi_size; in alloc_behind_master_bio()
1123 unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; in alloc_behind_master_bio()
1128 &r1_bio->mddev->bio_set); in alloc_behind_master_bio()
1134 behind_bio->bi_iter.bi_size = size; in alloc_behind_master_bio()
1151 size -= len; in alloc_behind_master_bio()
1157 r1_bio->behind_master_bio = behind_bio; in alloc_behind_master_bio()
1158 set_bit(R1BIO_BehindIO, &r1_bio->state); in alloc_behind_master_bio()
1164 bio->bi_iter.bi_size); in alloc_behind_master_bio()
1173 struct mddev *mddev = plug->cb.data; in raid1_unplug()
1174 struct r1conf *conf = mddev->private; in raid1_unplug()
1178 spin_lock_irq(&conf->device_lock); in raid1_unplug()
1179 bio_list_merge(&conf->pending_bio_list, &plug->pending); in raid1_unplug()
1180 spin_unlock_irq(&conf->device_lock); in raid1_unplug()
1182 md_wakeup_thread(mddev->thread); in raid1_unplug()
1187 /* we aren't scheduling, so we can do the write-out directly. */ in raid1_unplug()
1188 bio = bio_list_get(&plug->pending); in raid1_unplug()
1195 r1_bio->master_bio = bio; in init_r1bio()
1196 r1_bio->sectors = bio_sectors(bio); in init_r1bio()
1197 r1_bio->state = 0; in init_r1bio()
1198 r1_bio->mddev = mddev; in init_r1bio()
1199 r1_bio->sector = bio->bi_iter.bi_sector; in init_r1bio()
1205 struct r1conf *conf = mddev->private; in alloc_r1bio()
1208 r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); in alloc_r1bio()
1210 memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); in alloc_r1bio()
1218 struct r1conf *conf = mddev->private; in raid1_read_request()
1219 struct raid1_info *mirror; in raid1_read_request() local
1221 struct bitmap *bitmap = mddev->bitmap; in raid1_read_request()
1223 const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; in raid1_read_request()
1240 rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); in raid1_read_request()
1242 snprintf(b, sizeof(b), "%pg", rdev->bdev); in raid1_read_request()
1252 if (!wait_read_barrier(conf, bio->bi_iter.bi_sector, in raid1_read_request()
1253 bio->bi_opf & REQ_NOWAIT)) { in raid1_read_request()
1262 r1_bio->sectors = max_read_sectors; in raid1_read_request()
1265 * make_request() can abort the operation when read-ahead is being in raid1_read_request()
1276 (unsigned long long)r1_bio->sector); in raid1_read_request()
1281 mirror = conf->mirrors + rdisk; in raid1_read_request()
1284 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n", in raid1_read_request()
1286 (unsigned long long)r1_bio->sector, in raid1_read_request()
1287 mirror->rdev->bdev); in raid1_read_request()
1289 if (test_bit(WriteMostly, &mirror->rdev->flags) && in raid1_read_request()
1292 * Reading from a write-mostly device must take care not to in raid1_read_request()
1293 * over-take any writes that are 'behind' in raid1_read_request()
1296 wait_event(bitmap->behind_wait, in raid1_read_request()
1297 atomic_read(&bitmap->behind_writes) == 0); in raid1_read_request()
1302 gfp, &conf->bio_split); in raid1_read_request()
1306 r1_bio->master_bio = bio; in raid1_read_request()
1307 r1_bio->sectors = max_sectors; in raid1_read_request()
1310 r1_bio->read_disk = rdisk; in raid1_read_request()
1313 r1_bio->master_bio = bio; in raid1_read_request()
1315 read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp, in raid1_read_request()
1316 &mddev->bio_set); in raid1_read_request()
1318 r1_bio->bios[rdisk] = read_bio; in raid1_read_request()
1320 read_bio->bi_iter.bi_sector = r1_bio->sector + in raid1_read_request()
1321 mirror->rdev->data_offset; in raid1_read_request()
1322 read_bio->bi_end_io = raid1_end_read_request; in raid1_read_request()
1323 read_bio->bi_opf = op | do_sync; in raid1_read_request()
1324 if (test_bit(FailFast, &mirror->rdev->flags) && in raid1_read_request()
1325 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_read_request()
1326 read_bio->bi_opf |= MD_FAILFAST; in raid1_read_request()
1327 read_bio->bi_private = r1_bio; in raid1_read_request()
1329 if (mddev->gendisk) in raid1_read_request()
1330 trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), in raid1_read_request()
1331 r1_bio->sector); in raid1_read_request()
1339 struct r1conf *conf = mddev->private; in raid1_write_request()
1342 struct bitmap *bitmap = mddev->bitmap; in raid1_write_request()
1350 md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1351 bio->bi_iter.bi_sector, bio_end_sector(bio))) { in raid1_write_request()
1354 if (bio->bi_opf & REQ_NOWAIT) { in raid1_write_request()
1359 prepare_to_wait(&conf->wait_barrier, in raid1_write_request()
1361 if (!md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1362 bio->bi_iter.bi_sector, in raid1_write_request()
1367 finish_wait(&conf->wait_barrier, &w); in raid1_write_request()
1375 if (!wait_barrier(conf, bio->bi_iter.bi_sector, in raid1_write_request()
1376 bio->bi_opf & REQ_NOWAIT)) { in raid1_write_request()
1383 r1_bio->sectors = max_write_sectors; in raid1_write_request()
1396 disks = conf->raid_disks * 2; in raid1_write_request()
1399 max_sectors = r1_bio->sectors; in raid1_write_request()
1401 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_write_request()
1404 * The write-behind io is only attempted on drives marked as in raid1_write_request()
1405 * write-mostly, which means we could allocate write behind in raid1_write_request()
1408 if (rdev && test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1411 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { in raid1_write_request()
1412 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1416 r1_bio->bios[i] = NULL; in raid1_write_request()
1417 if (!rdev || test_bit(Faulty, &rdev->flags)) { in raid1_write_request()
1418 if (i < conf->raid_disks) in raid1_write_request()
1419 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1_write_request()
1423 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1424 if (test_bit(WriteErrorSeen, &rdev->flags)) { in raid1_write_request()
1429 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, in raid1_write_request()
1434 set_bit(BlockedBadBlocks, &rdev->flags); in raid1_write_request()
1438 if (is_bad && first_bad <= r1_bio->sector) { in raid1_write_request()
1440 bad_sectors -= (r1_bio->sector - first_bad); in raid1_write_request()
1449 * missing, so it might be re-added, in raid1_write_request()
1454 * in-sync is recorded in the bad in raid1_write_request()
1460 int good_sectors = first_bad - r1_bio->sector; in raid1_write_request()
1465 r1_bio->bios[i] = bio; in raid1_write_request()
1474 if (r1_bio->bios[j]) in raid1_write_request()
1475 rdev_dec_pending(conf->mirrors[j].rdev, mddev); in raid1_write_request()
1477 allow_barrier(conf, bio->bi_iter.bi_sector); in raid1_write_request()
1479 if (bio->bi_opf & REQ_NOWAIT) { in raid1_write_request()
1483 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); in raid1_write_request()
1485 wait_barrier(conf, bio->bi_iter.bi_sector, false); in raid1_write_request()
1491 * alloc_behind_master_bio allocates a copy of the data payload a page in raid1_write_request()
1500 GFP_NOIO, &conf->bio_split); in raid1_write_request()
1504 r1_bio->master_bio = bio; in raid1_write_request()
1505 r1_bio->sectors = max_sectors; in raid1_write_request()
1509 r1_bio->master_bio = bio; in raid1_write_request()
1510 atomic_set(&r1_bio->remaining, 1); in raid1_write_request()
1511 atomic_set(&r1_bio->behind_remaining, 0); in raid1_write_request()
1517 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_write_request()
1518 if (!r1_bio->bios[i]) in raid1_write_request()
1527 (atomic_read(&bitmap->behind_writes) in raid1_write_request()
1528 < mddev->bitmap_info.max_write_behind) && in raid1_write_request()
1529 !waitqueue_active(&bitmap->behind_wait)) { in raid1_write_request()
1533 md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, in raid1_write_request()
1534 test_bit(R1BIO_BehindIO, &r1_bio->state)); in raid1_write_request()
1538 if (r1_bio->behind_master_bio) { in raid1_write_request()
1539 mbio = bio_alloc_clone(rdev->bdev, in raid1_write_request()
1540 r1_bio->behind_master_bio, in raid1_write_request()
1541 GFP_NOIO, &mddev->bio_set); in raid1_write_request()
1542 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_write_request()
1544 if (test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1545 atomic_inc(&r1_bio->behind_remaining); in raid1_write_request()
1547 mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, in raid1_write_request()
1548 &mddev->bio_set); in raid1_write_request()
1550 if (mddev->serialize_policy) in raid1_write_request()
1554 r1_bio->bios[i] = mbio; in raid1_write_request()
1556 mbio->bi_iter.bi_sector = (r1_bio->sector + rdev->data_offset); in raid1_write_request()
1557 mbio->bi_end_io = raid1_end_write_request; in raid1_write_request()
1558 mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); in raid1_write_request()
1559 if (test_bit(FailFast, &rdev->flags) && in raid1_write_request()
1560 !test_bit(WriteMostly, &rdev->flags) && in raid1_write_request()
1561 conf->raid_disks - mddev->degraded > 1) in raid1_write_request()
1562 mbio->bi_opf |= MD_FAILFAST; in raid1_write_request()
1563 mbio->bi_private = r1_bio; in raid1_write_request()
1565 atomic_inc(&r1_bio->remaining); in raid1_write_request()
1567 if (mddev->gendisk) in raid1_write_request()
1568 trace_block_bio_remap(mbio, disk_devt(mddev->gendisk), in raid1_write_request()
1569 r1_bio->sector); in raid1_write_request()
1571 mbio->bi_bdev = (void *)rdev; in raid1_write_request()
1573 spin_lock_irqsave(&conf->device_lock, flags); in raid1_write_request()
1574 bio_list_add(&conf->pending_bio_list, mbio); in raid1_write_request()
1575 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_write_request()
1576 md_wakeup_thread(mddev->thread); in raid1_write_request()
1590 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid1_make_request()
1602 bio->bi_iter.bi_sector, bio_sectors(bio)); in raid1_make_request()
1616 struct r1conf *conf = mddev->private; in raid1_status()
1619 seq_printf(seq, " [%d/%d] [", conf->raid_disks, in raid1_status()
1620 conf->raid_disks - mddev->degraded); in raid1_status()
1622 for (i = 0; i < conf->raid_disks; i++) { in raid1_status()
1623 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_status()
1625 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid1_status()
1632 * raid1_error() - RAID1 error handler.
1638 * - &MD_BROKEN flag is set in &mddev->flags.
1639 * - recovery is disabled.
1641 * - recovery is interrupted.
1642 * - &mddev->degraded is bumped.
1645 * &mddev->fail_last_dev is off.
1649 struct r1conf *conf = mddev->private; in raid1_error()
1652 spin_lock_irqsave(&conf->device_lock, flags); in raid1_error()
1654 if (test_bit(In_sync, &rdev->flags) && in raid1_error()
1655 (conf->raid_disks - mddev->degraded) == 1) { in raid1_error()
1656 set_bit(MD_BROKEN, &mddev->flags); in raid1_error()
1658 if (!mddev->fail_last_dev) { in raid1_error()
1659 conf->recovery_disabled = mddev->recovery_disabled; in raid1_error()
1660 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1664 set_bit(Blocked, &rdev->flags); in raid1_error()
1665 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid1_error()
1666 mddev->degraded++; in raid1_error()
1667 set_bit(Faulty, &rdev->flags); in raid1_error()
1668 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1672 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_error()
1673 set_mask_bits(&mddev->sb_flags, 0, in raid1_error()
1677 mdname(mddev), rdev->bdev, in raid1_error()
1678 mdname(mddev), conf->raid_disks - mddev->degraded); in raid1_error()
1690 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, in print_conf()
1691 conf->raid_disks); in print_conf()
1694 for (i = 0; i < conf->raid_disks; i++) { in print_conf()
1695 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in print_conf()
1698 i, !test_bit(In_sync, &rdev->flags), in print_conf()
1699 !test_bit(Faulty, &rdev->flags), in print_conf()
1700 rdev->bdev); in print_conf()
1714 mempool_exit(&conf->r1buf_pool); in close_sync()
1720 struct r1conf *conf = mddev->private; in raid1_spare_active()
1729 * which expects 'In_sync' flags and ->degraded to be consistent. in raid1_spare_active()
1731 spin_lock_irqsave(&conf->device_lock, flags); in raid1_spare_active()
1732 for (i = 0; i < conf->raid_disks; i++) { in raid1_spare_active()
1733 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_spare_active()
1734 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; in raid1_spare_active()
1736 && !test_bit(Candidate, &repl->flags) in raid1_spare_active()
1737 && repl->recovery_offset == MaxSector in raid1_spare_active()
1738 && !test_bit(Faulty, &repl->flags) in raid1_spare_active()
1739 && !test_and_set_bit(In_sync, &repl->flags)) { in raid1_spare_active()
1742 !test_and_clear_bit(In_sync, &rdev->flags)) in raid1_spare_active()
1747 * it gets removed and never re-added in raid1_spare_active()
1749 set_bit(Faulty, &rdev->flags); in raid1_spare_active()
1751 rdev->sysfs_state); in raid1_spare_active()
1755 && rdev->recovery_offset == MaxSector in raid1_spare_active()
1756 && !test_bit(Faulty, &rdev->flags) in raid1_spare_active()
1757 && !test_and_set_bit(In_sync, &rdev->flags)) { in raid1_spare_active()
1759 sysfs_notify_dirent_safe(rdev->sysfs_state); in raid1_spare_active()
1762 mddev->degraded -= count; in raid1_spare_active()
1763 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_spare_active()
1771 struct r1conf *conf = mddev->private; in raid1_add_disk()
1772 int err = -EEXIST; in raid1_add_disk()
1773 int mirror = 0, repl_slot = -1; in raid1_add_disk() local
1776 int last = conf->raid_disks - 1; in raid1_add_disk()
1778 if (mddev->recovery_disabled == conf->recovery_disabled) in raid1_add_disk()
1779 return -EBUSY; in raid1_add_disk()
1782 return -ENXIO; in raid1_add_disk()
1784 if (rdev->raid_disk >= 0) in raid1_add_disk()
1785 first = last = rdev->raid_disk; in raid1_add_disk()
1788 * find the disk ... but prefer rdev->saved_raid_disk in raid1_add_disk()
1791 if (rdev->saved_raid_disk >= 0 && in raid1_add_disk()
1792 rdev->saved_raid_disk >= first && in raid1_add_disk()
1793 rdev->saved_raid_disk < conf->raid_disks && in raid1_add_disk()
1794 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid1_add_disk()
1795 first = last = rdev->saved_raid_disk; in raid1_add_disk()
1797 for (mirror = first; mirror <= last; mirror++) { in raid1_add_disk()
1798 p = conf->mirrors + mirror; in raid1_add_disk()
1799 if (!p->rdev) { in raid1_add_disk()
1800 if (mddev->gendisk) in raid1_add_disk()
1801 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_add_disk()
1802 rdev->data_offset << 9); in raid1_add_disk()
1804 p->head_position = 0; in raid1_add_disk()
1805 rdev->raid_disk = mirror; in raid1_add_disk()
1810 if (rdev->saved_raid_disk < 0) in raid1_add_disk()
1811 conf->fullsync = 1; in raid1_add_disk()
1812 rcu_assign_pointer(p->rdev, rdev); in raid1_add_disk()
1815 if (test_bit(WantReplacement, &p->rdev->flags) && in raid1_add_disk()
1816 p[conf->raid_disks].rdev == NULL && repl_slot < 0) in raid1_add_disk()
1817 repl_slot = mirror; in raid1_add_disk()
1822 p = conf->mirrors + repl_slot; in raid1_add_disk()
1823 clear_bit(In_sync, &rdev->flags); in raid1_add_disk()
1824 set_bit(Replacement, &rdev->flags); in raid1_add_disk()
1825 rdev->raid_disk = repl_slot; in raid1_add_disk()
1827 conf->fullsync = 1; in raid1_add_disk()
1828 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); in raid1_add_disk()
1837 struct r1conf *conf = mddev->private; in raid1_remove_disk()
1839 int number = rdev->raid_disk; in raid1_remove_disk()
1840 struct raid1_info *p = conf->mirrors + number; in raid1_remove_disk()
1842 if (unlikely(number >= conf->raid_disks)) in raid1_remove_disk()
1845 if (rdev != p->rdev) in raid1_remove_disk()
1846 p = conf->mirrors + conf->raid_disks + number; in raid1_remove_disk()
1849 if (rdev == p->rdev) { in raid1_remove_disk()
1850 if (test_bit(In_sync, &rdev->flags) || in raid1_remove_disk()
1851 atomic_read(&rdev->nr_pending)) { in raid1_remove_disk()
1852 err = -EBUSY; in raid1_remove_disk()
1855 /* Only remove non-faulty devices if recovery in raid1_remove_disk()
1858 if (!test_bit(Faulty, &rdev->flags) && in raid1_remove_disk()
1859 mddev->recovery_disabled != conf->recovery_disabled && in raid1_remove_disk()
1860 mddev->degraded < conf->raid_disks) { in raid1_remove_disk()
1861 err = -EBUSY; in raid1_remove_disk()
1864 p->rdev = NULL; in raid1_remove_disk()
1865 if (!test_bit(RemoveSynchronized, &rdev->flags)) { in raid1_remove_disk()
1867 if (atomic_read(&rdev->nr_pending)) { in raid1_remove_disk()
1869 err = -EBUSY; in raid1_remove_disk()
1870 p->rdev = rdev; in raid1_remove_disk()
1874 if (conf->mirrors[conf->raid_disks + number].rdev) { in raid1_remove_disk()
1880 conf->mirrors[conf->raid_disks + number].rdev; in raid1_remove_disk()
1882 if (atomic_read(&repl->nr_pending)) { in raid1_remove_disk()
1889 err = -EBUSY; in raid1_remove_disk()
1893 clear_bit(Replacement, &repl->flags); in raid1_remove_disk()
1894 p->rdev = repl; in raid1_remove_disk()
1895 conf->mirrors[conf->raid_disks + number].rdev = NULL; in raid1_remove_disk()
1899 clear_bit(WantReplacement, &rdev->flags); in raid1_remove_disk()
1912 update_head_pos(r1_bio->read_disk, r1_bio); in end_sync_read()
1915 * we have read a block, now it needs to be re-written, in end_sync_read()
1916 * or re-read if the read failed. in end_sync_read()
1919 if (!bio->bi_status) in end_sync_read()
1920 set_bit(R1BIO_Uptodate, &r1_bio->state); in end_sync_read()
1922 if (atomic_dec_and_test(&r1_bio->remaining)) in end_sync_read()
1929 sector_t s = r1_bio->sector; in abort_sync_write()
1930 long sectors_to_go = r1_bio->sectors; in abort_sync_write()
1934 md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); in abort_sync_write()
1936 sectors_to_go -= sync_blocks; in abort_sync_write()
1942 if (atomic_dec_and_test(&r1_bio->remaining)) { in put_sync_write_buf()
1943 struct mddev *mddev = r1_bio->mddev; in put_sync_write_buf()
1944 int s = r1_bio->sectors; in put_sync_write_buf()
1946 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in put_sync_write_buf()
1947 test_bit(R1BIO_WriteError, &r1_bio->state)) in put_sync_write_buf()
1958 int uptodate = !bio->bi_status; in end_sync_write()
1960 struct mddev *mddev = r1_bio->mddev; in end_sync_write()
1961 struct r1conf *conf = mddev->private; in end_sync_write()
1964 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; in end_sync_write()
1968 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
1969 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
1971 mddev->recovery); in end_sync_write()
1972 set_bit(R1BIO_WriteError, &r1_bio->state); in end_sync_write()
1973 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in end_sync_write()
1975 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, in end_sync_write()
1976 r1_bio->sector, in end_sync_write()
1977 r1_bio->sectors, in end_sync_write()
1980 set_bit(R1BIO_MadeGood, &r1_bio->state); in end_sync_write()
1992 set_bit(WriteErrorSeen, &rdev->flags); in r1_sync_page_io()
1994 &rdev->flags)) in r1_sync_page_io()
1996 rdev->mddev->recovery); in r1_sync_page_io()
1998 /* need to record an error - either for the block or the device */ in r1_sync_page_io()
2000 md_error(rdev->mddev, rdev); in r1_sync_page_io()
2007 * good data, much like with normal read errors. Only in fix_sync_read_error()
2009 * need to re-issue the read request. in fix_sync_read_error()
2017 struct mddev *mddev = r1_bio->mddev; in fix_sync_read_error()
2018 struct r1conf *conf = mddev->private; in fix_sync_read_error()
2019 struct bio *bio = r1_bio->bios[r1_bio->read_disk]; in fix_sync_read_error()
2020 struct page **pages = get_resync_pages(bio)->pages; in fix_sync_read_error()
2021 sector_t sect = r1_bio->sector; in fix_sync_read_error()
2022 int sectors = r1_bio->sectors; in fix_sync_read_error()
2026 rdev = conf->mirrors[r1_bio->read_disk].rdev; in fix_sync_read_error()
2027 if (test_bit(FailFast, &rdev->flags)) { in fix_sync_read_error()
2028 /* Don't try recovering from here - just fail it in fix_sync_read_error()
2031 if (test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2035 bio->bi_end_io = end_sync_write; in fix_sync_read_error()
2040 int d = r1_bio->read_disk; in fix_sync_read_error()
2047 if (r1_bio->bios[d]->bi_end_io == end_sync_read) { in fix_sync_read_error()
2052 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2061 if (d == conf->raid_disks * 2) in fix_sync_read_error()
2063 } while (!success && d != r1_bio->read_disk); in fix_sync_read_error()
2073 mdname(mddev), bio->bi_bdev, in fix_sync_read_error()
2074 (unsigned long long)r1_bio->sector); in fix_sync_read_error()
2075 for (d = 0; d < conf->raid_disks * 2; d++) { in fix_sync_read_error()
2076 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2077 if (!rdev || test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2083 conf->recovery_disabled = in fix_sync_read_error()
2084 mddev->recovery_disabled; in fix_sync_read_error()
2085 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in fix_sync_read_error()
2086 md_done_sync(mddev, r1_bio->sectors, 0); in fix_sync_read_error()
2091 sectors -= s; in fix_sync_read_error()
2098 /* write it back and re-read */ in fix_sync_read_error()
2099 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2101 d = conf->raid_disks * 2; in fix_sync_read_error()
2102 d--; in fix_sync_read_error()
2103 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2105 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2109 r1_bio->bios[d]->bi_end_io = NULL; in fix_sync_read_error()
2114 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2116 d = conf->raid_disks * 2; in fix_sync_read_error()
2117 d--; in fix_sync_read_error()
2118 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2120 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2124 atomic_add(s, &rdev->corrected_errors); in fix_sync_read_error()
2126 sectors -= s; in fix_sync_read_error()
2130 set_bit(R1BIO_Uptodate, &r1_bio->state); in fix_sync_read_error()
2131 bio->bi_status = 0; in fix_sync_read_error()
2142 * attempt an over-write in process_checks()
2144 struct mddev *mddev = r1_bio->mddev; in process_checks()
2145 struct r1conf *conf = mddev->private; in process_checks()
2151 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); in process_checks()
2152 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2154 struct bio *b = r1_bio->bios[i]; in process_checks()
2156 if (b->bi_end_io != end_sync_read) in process_checks()
2159 status = b->bi_status; in process_checks()
2160 bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ); in process_checks()
2161 b->bi_status = status; in process_checks()
2162 b->bi_iter.bi_sector = r1_bio->sector + in process_checks()
2163 conf->mirrors[i].rdev->data_offset; in process_checks()
2164 b->bi_end_io = end_sync_read; in process_checks()
2165 rp->raid_bio = r1_bio; in process_checks()
2166 b->bi_private = rp; in process_checks()
2169 md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); in process_checks()
2171 for (primary = 0; primary < conf->raid_disks * 2; primary++) in process_checks()
2172 if (r1_bio->bios[primary]->bi_end_io == end_sync_read && in process_checks()
2173 !r1_bio->bios[primary]->bi_status) { in process_checks()
2174 r1_bio->bios[primary]->bi_end_io = NULL; in process_checks()
2175 rdev_dec_pending(conf->mirrors[primary].rdev, mddev); in process_checks()
2178 r1_bio->read_disk = primary; in process_checks()
2179 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2181 struct bio *pbio = r1_bio->bios[primary]; in process_checks()
2182 struct bio *sbio = r1_bio->bios[i]; in process_checks()
2183 blk_status_t status = sbio->bi_status; in process_checks()
2184 struct page **ppages = get_resync_pages(pbio)->pages; in process_checks()
2185 struct page **spages = get_resync_pages(sbio)->pages; in process_checks()
2190 if (sbio->bi_end_io != end_sync_read) in process_checks()
2193 sbio->bi_status = 0; in process_checks()
2196 page_len[j++] = bi->bv_len; in process_checks()
2199 for (j = vcnt; j-- ; ) { in process_checks()
2208 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); in process_checks()
2209 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) in process_checks()
2212 sbio->bi_end_io = NULL; in process_checks()
2213 rdev_dec_pending(conf->mirrors[i].rdev, mddev); in process_checks()
2223 struct r1conf *conf = mddev->private; in sync_request_write()
2225 int disks = conf->raid_disks * 2; in sync_request_write()
2228 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in sync_request_write()
2229 /* ouch - failed to read all of that. */ in sync_request_write()
2233 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in sync_request_write()
2239 atomic_set(&r1_bio->remaining, 1); in sync_request_write()
2241 wbio = r1_bio->bios[i]; in sync_request_write()
2242 if (wbio->bi_end_io == NULL || in sync_request_write()
2243 (wbio->bi_end_io == end_sync_read && in sync_request_write()
2244 (i == r1_bio->read_disk || in sync_request_write()
2245 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) in sync_request_write()
2247 if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { in sync_request_write()
2252 wbio->bi_opf = REQ_OP_WRITE; in sync_request_write()
2253 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) in sync_request_write()
2254 wbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2256 wbio->bi_end_io = end_sync_write; in sync_request_write()
2257 atomic_inc(&r1_bio->remaining); in sync_request_write()
2258 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); in sync_request_write()
2277 struct mddev *mddev = conf->mddev; in fix_read_error()
2293 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2295 (test_bit(In_sync, &rdev->flags) || in fix_read_error()
2296 (!test_bit(Faulty, &rdev->flags) && in fix_read_error()
2297 rdev->recovery_offset >= sect + s)) && in fix_read_error()
2300 atomic_inc(&rdev->nr_pending); in fix_read_error()
2303 conf->tmppage, REQ_OP_READ, false)) in fix_read_error()
2311 if (d == conf->raid_disks * 2) in fix_read_error()
2316 /* Cannot read from anywhere - mark it bad */ in fix_read_error()
2317 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; in fix_read_error()
2322 /* write it back and re-read */ in fix_read_error()
2326 d = conf->raid_disks * 2; in fix_read_error()
2327 d--; in fix_read_error()
2329 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2331 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2332 atomic_inc(&rdev->nr_pending); in fix_read_error()
2335 conf->tmppage, WRITE); in fix_read_error()
2343 d = conf->raid_disks * 2; in fix_read_error()
2344 d--; in fix_read_error()
2346 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2348 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2349 atomic_inc(&rdev->nr_pending); in fix_read_error()
2352 conf->tmppage, READ)) { in fix_read_error()
2353 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2357 rdev->data_offset), in fix_read_error()
2358 rdev->bdev); in fix_read_error()
2364 sectors -= s; in fix_read_error()
2371 struct mddev *mddev = r1_bio->mddev; in narrow_write_error()
2372 struct r1conf *conf = mddev->private; in narrow_write_error()
2373 struct md_rdev *rdev = conf->mirrors[i].rdev; in narrow_write_error()
2375 /* bio has the data to be written to device 'i' where in narrow_write_error()
2389 int sect_to_write = r1_bio->sectors; in narrow_write_error()
2392 if (rdev->badblocks.shift < 0) in narrow_write_error()
2395 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2396 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2397 sector = r1_bio->sector; in narrow_write_error()
2399 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2400 - sector; in narrow_write_error()
2408 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in narrow_write_error()
2409 wbio = bio_alloc_clone(rdev->bdev, in narrow_write_error()
2410 r1_bio->behind_master_bio, in narrow_write_error()
2411 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2413 wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio, in narrow_write_error()
2414 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2417 wbio->bi_opf = REQ_OP_WRITE; in narrow_write_error()
2418 wbio->bi_iter.bi_sector = r1_bio->sector; in narrow_write_error()
2419 wbio->bi_iter.bi_size = r1_bio->sectors << 9; in narrow_write_error()
2421 bio_trim(wbio, sector - r1_bio->sector, sectors); in narrow_write_error()
2422 wbio->bi_iter.bi_sector += rdev->data_offset; in narrow_write_error()
2431 sect_to_write -= sectors; in narrow_write_error()
2441 int s = r1_bio->sectors; in handle_sync_write_finished()
2442 for (m = 0; m < conf->raid_disks * 2 ; m++) { in handle_sync_write_finished()
2443 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_sync_write_finished()
2444 struct bio *bio = r1_bio->bios[m]; in handle_sync_write_finished()
2445 if (bio->bi_end_io == NULL) in handle_sync_write_finished()
2447 if (!bio->bi_status && in handle_sync_write_finished()
2448 test_bit(R1BIO_MadeGood, &r1_bio->state)) { in handle_sync_write_finished()
2449 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); in handle_sync_write_finished()
2451 if (bio->bi_status && in handle_sync_write_finished()
2452 test_bit(R1BIO_WriteError, &r1_bio->state)) { in handle_sync_write_finished()
2453 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) in handle_sync_write_finished()
2454 md_error(conf->mddev, rdev); in handle_sync_write_finished()
2458 md_done_sync(conf->mddev, s, 1); in handle_sync_write_finished()
2466 for (m = 0; m < conf->raid_disks * 2 ; m++) in handle_write_finished()
2467 if (r1_bio->bios[m] == IO_MADE_GOOD) { in handle_write_finished()
2468 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_write_finished()
2470 r1_bio->sector, in handle_write_finished()
2471 r1_bio->sectors, 0); in handle_write_finished()
2472 rdev_dec_pending(rdev, conf->mddev); in handle_write_finished()
2473 } else if (r1_bio->bios[m] != NULL) { in handle_write_finished()
2480 md_error(conf->mddev, in handle_write_finished()
2481 conf->mirrors[m].rdev); in handle_write_finished()
2483 set_bit(R1BIO_Degraded, &r1_bio->state); in handle_write_finished()
2485 rdev_dec_pending(conf->mirrors[m].rdev, in handle_write_finished()
2486 conf->mddev); in handle_write_finished()
2489 spin_lock_irq(&conf->device_lock); in handle_write_finished()
2490 list_add(&r1_bio->retry_list, &conf->bio_end_io_list); in handle_write_finished()
2491 idx = sector_to_idx(r1_bio->sector); in handle_write_finished()
2492 atomic_inc(&conf->nr_queued[idx]); in handle_write_finished()
2493 spin_unlock_irq(&conf->device_lock); in handle_write_finished()
2498 wake_up(&conf->wait_barrier); in handle_write_finished()
2499 md_wakeup_thread(conf->mddev->thread); in handle_write_finished()
2501 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in handle_write_finished()
2509 struct mddev *mddev = conf->mddev; in handle_read_error()
2514 clear_bit(R1BIO_ReadError, &r1_bio->state); in handle_read_error()
2518 * other devices. When we find one, we re-write in handle_read_error()
2524 bio = r1_bio->bios[r1_bio->read_disk]; in handle_read_error()
2526 r1_bio->bios[r1_bio->read_disk] = NULL; in handle_read_error()
2528 rdev = conf->mirrors[r1_bio->read_disk].rdev; in handle_read_error()
2529 if (mddev->ro == 0 in handle_read_error()
2530 && !test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2532 fix_read_error(conf, r1_bio->read_disk, in handle_read_error()
2533 r1_bio->sector, r1_bio->sectors); in handle_read_error()
2535 } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2538 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; in handle_read_error()
2541 rdev_dec_pending(rdev, conf->mddev); in handle_read_error()
2542 sector = r1_bio->sector; in handle_read_error()
2543 bio = r1_bio->master_bio; in handle_read_error()
2546 r1_bio->state = 0; in handle_read_error()
2547 raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); in handle_read_error()
2553 struct mddev *mddev = thread->mddev; in raid1d()
2556 struct r1conf *conf = mddev->private; in raid1d()
2557 struct list_head *head = &conf->retry_list; in raid1d()
2563 if (!list_empty_careful(&conf->bio_end_io_list) && in raid1d()
2564 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid1d()
2566 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2567 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) in raid1d()
2568 list_splice_init(&conf->bio_end_io_list, &tmp); in raid1d()
2569 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2573 list_del(&r1_bio->retry_list); in raid1d()
2574 idx = sector_to_idx(r1_bio->sector); in raid1d()
2575 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2576 if (mddev->degraded) in raid1d()
2577 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1d()
2578 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2589 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2591 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2594 r1_bio = list_entry(head->prev, struct r1bio, retry_list); in raid1d()
2595 list_del(head->prev); in raid1d()
2596 idx = sector_to_idx(r1_bio->sector); in raid1d()
2597 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2598 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2600 mddev = r1_bio->mddev; in raid1d()
2601 conf = mddev->private; in raid1d()
2602 if (test_bit(R1BIO_IsSync, &r1_bio->state)) { in raid1d()
2603 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2604 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2608 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2609 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2611 else if (test_bit(R1BIO_ReadError, &r1_bio->state)) in raid1d()
2617 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid1d()
2628 BUG_ON(mempool_initialized(&conf->r1buf_pool)); in init_resync()
2630 return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, in init_resync()
2631 r1buf_pool_free, conf->poolinfo); in init_resync()
2636 struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); in raid1_alloc_init_r1buf()
2641 for (i = conf->poolinfo->raid_disks; i--; ) { in raid1_alloc_init_r1buf()
2642 bio = r1bio->bios[i]; in raid1_alloc_init_r1buf()
2643 rps = bio->bi_private; in raid1_alloc_init_r1buf()
2645 bio->bi_private = rps; in raid1_alloc_init_r1buf()
2647 r1bio->master_bio = NULL; in raid1_alloc_init_r1buf()
2654 * We need to make sure that no normal I/O request - particularly write
2655 * requests - conflict with active sync requests.
2664 struct r1conf *conf = mddev->private; in raid1_sync_request()
2668 int disk = -1; in raid1_sync_request()
2670 int wonly = -1; in raid1_sync_request()
2679 if (!mempool_initialized(&conf->r1buf_pool)) in raid1_sync_request()
2683 max_sector = mddev->dev_sectors; in raid1_sync_request()
2688 * We can find the current addess in mddev->curr_resync in raid1_sync_request()
2690 if (mddev->curr_resync < max_sector) /* aborted */ in raid1_sync_request()
2691 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, in raid1_sync_request()
2694 conf->fullsync = 0; in raid1_sync_request()
2696 md_bitmap_close_sync(mddev->bitmap); in raid1_sync_request()
2700 conf->cluster_sync_low = 0; in raid1_sync_request()
2701 conf->cluster_sync_high = 0; in raid1_sync_request()
2706 if (mddev->bitmap == NULL && in raid1_sync_request()
2707 mddev->recovery_cp == MaxSector && in raid1_sync_request()
2708 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid1_sync_request()
2709 conf->fullsync == 0) { in raid1_sync_request()
2711 return max_sector - sector_nr; in raid1_sync_request()
2716 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && in raid1_sync_request()
2717 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2724 * If there is non-resync activity waiting for a turn, then let it in raid1_sync_request()
2727 if (atomic_read(&conf->nr_waiting[idx])) in raid1_sync_request()
2734 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2735 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid1_sync_request()
2748 * and any others (which will be non-In_sync devices) for WRITE. in raid1_sync_request()
2753 r1_bio->mddev = mddev; in raid1_sync_request()
2754 r1_bio->sector = sector_nr; in raid1_sync_request()
2755 r1_bio->state = 0; in raid1_sync_request()
2756 set_bit(R1BIO_IsSync, &r1_bio->state); in raid1_sync_request()
2760 for (i = 0; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2762 bio = r1_bio->bios[i]; in raid1_sync_request()
2764 rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_sync_request()
2766 test_bit(Faulty, &rdev->flags)) { in raid1_sync_request()
2767 if (i < conf->raid_disks) in raid1_sync_request()
2769 } else if (!test_bit(In_sync, &rdev->flags)) { in raid1_sync_request()
2770 bio->bi_opf = REQ_OP_WRITE; in raid1_sync_request()
2771 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2781 good_sectors = first_bad - sector_nr; in raid1_sync_request()
2783 bad_sectors -= (sector_nr - first_bad); in raid1_sync_request()
2790 if (test_bit(WriteMostly, &rdev->flags)) { in raid1_sync_request()
2797 bio->bi_opf = REQ_OP_READ; in raid1_sync_request()
2798 bio->bi_end_io = end_sync_read; in raid1_sync_request()
2800 } else if (!test_bit(WriteErrorSeen, &rdev->flags) && in raid1_sync_request()
2801 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid1_sync_request()
2802 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { in raid1_sync_request()
2809 bio->bi_opf = REQ_OP_WRITE; in raid1_sync_request()
2810 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2814 if (rdev && bio->bi_end_io) { in raid1_sync_request()
2815 atomic_inc(&rdev->nr_pending); in raid1_sync_request()
2816 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; in raid1_sync_request()
2817 bio_set_dev(bio, rdev->bdev); in raid1_sync_request()
2818 if (test_bit(FailFast, &rdev->flags)) in raid1_sync_request()
2819 bio->bi_opf |= MD_FAILFAST; in raid1_sync_request()
2825 r1_bio->read_disk = disk; in raid1_sync_request()
2832 for (i = 0 ; i < conf->raid_disks * 2 ; i++) in raid1_sync_request()
2833 if (r1_bio->bios[i]->bi_end_io == end_sync_write) { in raid1_sync_request()
2834 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_sync_request()
2839 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid1_sync_request()
2849 conf->recovery_disabled = mddev->recovery_disabled; in raid1_sync_request()
2850 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_sync_request()
2857 /* only resync enough to reach the next bad->good in raid1_sync_request()
2862 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) in raid1_sync_request()
2864 write_targets += read_targets-1; in raid1_sync_request()
2867 /* There is nowhere to write, so all non-sync in raid1_sync_request()
2868 * drives must be failed - so we are finished in raid1_sync_request()
2873 rv = max_sector - sector_nr; in raid1_sync_request()
2879 if (max_sector > mddev->resync_max) in raid1_sync_request()
2880 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid1_sync_request()
2889 len = (max_sector - sector_nr) << 9; in raid1_sync_request()
2893 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2895 !conf->fullsync && in raid1_sync_request()
2896 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in raid1_sync_request()
2902 for (i = 0 ; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2905 bio = r1_bio->bios[i]; in raid1_sync_request()
2907 if (bio->bi_end_io) { in raid1_sync_request()
2919 sync_blocks -= (len>>9); in raid1_sync_request()
2922 r1_bio->sectors = nr_sectors; in raid1_sync_request()
2925 conf->cluster_sync_high < sector_nr + nr_sectors) { in raid1_sync_request()
2926 conf->cluster_sync_low = mddev->curr_resync_completed; in raid1_sync_request()
2927 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; in raid1_sync_request()
2929 md_cluster_ops->resync_info_update(mddev, in raid1_sync_request()
2930 conf->cluster_sync_low, in raid1_sync_request()
2931 conf->cluster_sync_high); in raid1_sync_request()
2934 /* For a user-requested sync, we read all readable devices and do a in raid1_sync_request()
2937 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2938 atomic_set(&r1_bio->remaining, read_targets); in raid1_sync_request()
2939 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { in raid1_sync_request()
2940 bio = r1_bio->bios[i]; in raid1_sync_request()
2941 if (bio->bi_end_io == end_sync_read) { in raid1_sync_request()
2942 read_targets--; in raid1_sync_request()
2945 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2950 atomic_set(&r1_bio->remaining, 1); in raid1_sync_request()
2951 bio = r1_bio->bios[r1_bio->read_disk]; in raid1_sync_request()
2954 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2965 return mddev->dev_sectors; in raid1_size()
2974 int err = -ENOMEM; in setup_conf()
2980 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2982 if (!conf->nr_pending) in setup_conf()
2985 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2987 if (!conf->nr_waiting) in setup_conf()
2990 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2992 if (!conf->nr_queued) in setup_conf()
2995 conf->barrier = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2997 if (!conf->barrier) in setup_conf()
3000 conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), in setup_conf()
3001 mddev->raid_disks, 2), in setup_conf()
3003 if (!conf->mirrors) in setup_conf()
3006 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
3007 if (!conf->tmppage) in setup_conf()
3010 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); in setup_conf()
3011 if (!conf->poolinfo) in setup_conf()
3013 conf->poolinfo->raid_disks = mddev->raid_disks * 2; in setup_conf()
3014 err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, in setup_conf()
3015 rbio_pool_free, conf->poolinfo); in setup_conf()
3019 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
3023 conf->poolinfo->mddev = mddev; in setup_conf()
3025 err = -EINVAL; in setup_conf()
3026 spin_lock_init(&conf->device_lock); in setup_conf()
3028 int disk_idx = rdev->raid_disk; in setup_conf()
3029 if (disk_idx >= mddev->raid_disks in setup_conf()
3032 if (test_bit(Replacement, &rdev->flags)) in setup_conf()
3033 disk = conf->mirrors + mddev->raid_disks + disk_idx; in setup_conf()
3035 disk = conf->mirrors + disk_idx; in setup_conf()
3037 if (disk->rdev) in setup_conf()
3039 disk->rdev = rdev; in setup_conf()
3040 disk->head_position = 0; in setup_conf()
3041 disk->seq_start = MaxSector; in setup_conf()
3043 conf->raid_disks = mddev->raid_disks; in setup_conf()
3044 conf->mddev = mddev; in setup_conf()
3045 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
3046 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
3048 spin_lock_init(&conf->resync_lock); in setup_conf()
3049 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
3051 bio_list_init(&conf->pending_bio_list); in setup_conf()
3052 conf->recovery_disabled = mddev->recovery_disabled - 1; in setup_conf()
3054 err = -EIO; in setup_conf()
3055 for (i = 0; i < conf->raid_disks * 2; i++) { in setup_conf()
3057 disk = conf->mirrors + i; in setup_conf()
3059 if (i < conf->raid_disks && in setup_conf()
3060 disk[conf->raid_disks].rdev) { in setup_conf()
3062 if (!disk->rdev) { in setup_conf()
3066 disk->rdev = in setup_conf()
3067 disk[conf->raid_disks].rdev; in setup_conf()
3068 disk[conf->raid_disks].rdev = NULL; in setup_conf()
3069 } else if (!test_bit(In_sync, &disk->rdev->flags)) in setup_conf()
3070 /* Original is not in_sync - bad */ in setup_conf()
3074 if (!disk->rdev || in setup_conf()
3075 !test_bit(In_sync, &disk->rdev->flags)) { in setup_conf()
3076 disk->head_position = 0; in setup_conf()
3077 if (disk->rdev && in setup_conf()
3078 (disk->rdev->saved_raid_disk < 0)) in setup_conf()
3079 conf->fullsync = 1; in setup_conf()
3083 err = -ENOMEM; in setup_conf()
3084 rcu_assign_pointer(conf->thread, in setup_conf()
3086 if (!conf->thread) in setup_conf()
3093 mempool_exit(&conf->r1bio_pool); in setup_conf()
3094 kfree(conf->mirrors); in setup_conf()
3095 safe_put_page(conf->tmppage); in setup_conf()
3096 kfree(conf->poolinfo); in setup_conf()
3097 kfree(conf->nr_pending); in setup_conf()
3098 kfree(conf->nr_waiting); in setup_conf()
3099 kfree(conf->nr_queued); in setup_conf()
3100 kfree(conf->barrier); in setup_conf()
3101 bioset_exit(&conf->bio_split); in setup_conf()
3115 if (mddev->level != 1) { in raid1_run()
3117 mdname(mddev), mddev->level); in raid1_run()
3118 return -EIO; in raid1_run()
3120 if (mddev->reshape_position != MaxSector) { in raid1_run()
3123 return -EIO; in raid1_run()
3126 return -ENOMEM; in raid1_run()
3132 if (mddev->private == NULL) in raid1_run()
3135 conf = mddev->private; in raid1_run()
3140 if (mddev->queue) in raid1_run()
3141 blk_queue_max_write_zeroes_sectors(mddev->queue, 0); in raid1_run()
3144 if (!mddev->gendisk) in raid1_run()
3146 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_run()
3147 rdev->data_offset << 9); in raid1_run()
3150 mddev->degraded = 0; in raid1_run()
3151 for (i = 0; i < conf->raid_disks; i++) in raid1_run()
3152 if (conf->mirrors[i].rdev == NULL || in raid1_run()
3153 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || in raid1_run()
3154 test_bit(Faulty, &conf->mirrors[i].rdev->flags)) in raid1_run()
3155 mddev->degraded++; in raid1_run()
3159 if (conf->raid_disks - mddev->degraded < 1) { in raid1_run()
3160 md_unregister_thread(mddev, &conf->thread); in raid1_run()
3161 ret = -EINVAL; in raid1_run()
3165 if (conf->raid_disks - mddev->degraded == 1) in raid1_run()
3166 mddev->recovery_cp = MaxSector; in raid1_run()
3168 if (mddev->recovery_cp != MaxSector) in raid1_run()
3169 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", in raid1_run()
3172 mdname(mddev), mddev->raid_disks - mddev->degraded, in raid1_run()
3173 mddev->raid_disks); in raid1_run()
3178 rcu_assign_pointer(mddev->thread, conf->thread); in raid1_run()
3179 rcu_assign_pointer(conf->thread, NULL); in raid1_run()
3180 mddev->private = conf; in raid1_run()
3181 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid1_run()
3187 md_unregister_thread(mddev, &mddev->thread); in raid1_run()
3201 mempool_exit(&conf->r1bio_pool); in raid1_free()
3202 kfree(conf->mirrors); in raid1_free()
3203 safe_put_page(conf->tmppage); in raid1_free()
3204 kfree(conf->poolinfo); in raid1_free()
3205 kfree(conf->nr_pending); in raid1_free()
3206 kfree(conf->nr_waiting); in raid1_free()
3207 kfree(conf->nr_queued); in raid1_free()
3208 kfree(conf->barrier); in raid1_free()
3209 bioset_exit(&conf->bio_split); in raid1_free()
3223 if (mddev->external_size && in raid1_resize()
3224 mddev->array_sectors > newsize) in raid1_resize()
3225 return -EINVAL; in raid1_resize()
3226 if (mddev->bitmap) { in raid1_resize()
3227 int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid1_resize()
3232 if (sectors > mddev->dev_sectors && in raid1_resize()
3233 mddev->recovery_cp > mddev->dev_sectors) { in raid1_resize()
3234 mddev->recovery_cp = mddev->dev_sectors; in raid1_resize()
3235 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_resize()
3237 mddev->dev_sectors = sectors; in raid1_resize()
3238 mddev->resync_max_sectors = sectors; in raid1_resize()
3246 * 2/ resize conf->mirrors in raid1_reshape()
3250 * Then resize conf->mirrors and swap in the new r1bio pool. in raid1_reshape()
3258 struct r1conf *conf = mddev->private; in raid1_reshape()
3268 if (mddev->chunk_sectors != mddev->new_chunk_sectors || in raid1_reshape()
3269 mddev->layout != mddev->new_layout || in raid1_reshape()
3270 mddev->level != mddev->new_level) { in raid1_reshape()
3271 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid1_reshape()
3272 mddev->new_layout = mddev->layout; in raid1_reshape()
3273 mddev->new_level = mddev->level; in raid1_reshape()
3274 return -EINVAL; in raid1_reshape()
3280 raid_disks = mddev->raid_disks + mddev->delta_disks; in raid1_reshape()
3282 if (raid_disks < conf->raid_disks) { in raid1_reshape()
3284 for (d= 0; d < conf->raid_disks; d++) in raid1_reshape()
3285 if (conf->mirrors[d].rdev) in raid1_reshape()
3288 return -EBUSY; in raid1_reshape()
3293 return -ENOMEM; in raid1_reshape()
3294 newpoolinfo->mddev = mddev; in raid1_reshape()
3295 newpoolinfo->raid_disks = raid_disks * 2; in raid1_reshape()
3309 return -ENOMEM; in raid1_reshape()
3315 oldpool = conf->r1bio_pool; in raid1_reshape()
3316 conf->r1bio_pool = newpool; in raid1_reshape()
3318 for (d = d2 = 0; d < conf->raid_disks; d++) { in raid1_reshape()
3319 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid1_reshape()
3320 if (rdev && rdev->raid_disk != d2) { in raid1_reshape()
3322 rdev->raid_disk = d2; in raid1_reshape()
3326 mdname(mddev), rdev->raid_disk); in raid1_reshape()
3331 kfree(conf->mirrors); in raid1_reshape()
3332 conf->mirrors = newmirrors; in raid1_reshape()
3333 kfree(conf->poolinfo); in raid1_reshape()
3334 conf->poolinfo = newpoolinfo; in raid1_reshape()
3336 spin_lock_irqsave(&conf->device_lock, flags); in raid1_reshape()
3337 mddev->degraded += (raid_disks - conf->raid_disks); in raid1_reshape()
3338 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_reshape()
3339 conf->raid_disks = mddev->raid_disks = raid_disks; in raid1_reshape()
3340 mddev->delta_disks = 0; in raid1_reshape()
3344 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); in raid1_reshape()
3345 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_reshape()
3346 md_wakeup_thread(mddev->thread); in raid1_reshape()
3354 struct r1conf *conf = mddev->private; in raid1_quiesce()
3367 if (mddev->level == 5 && mddev->raid_disks == 2) { in raid1_takeover()
3369 mddev->new_level = 1; in raid1_takeover()
3370 mddev->new_layout = 0; in raid1_takeover()
3371 mddev->new_chunk_sectors = 0; in raid1_takeover()
3375 conf->array_frozen = 1; in raid1_takeover()
3381 return ERR_PTR(-EINVAL); in raid1_takeover()
3419 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3420 MODULE_ALIAS("md-raid1");
3421 MODULE_ALIAS("md-level-1");