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()
307 if (blk_queue_io_stat(bio->bi_bdev->bd_disk->queue)) in call_bio_endio()
308 bio_end_io_acct(bio, r1_bio->start_time); in call_bio_endio()
314 struct bio *bio = r1_bio->master_bio; in raid_end_bio_io()
315 struct r1conf *conf = r1_bio->mddev->private; in raid_end_bio_io()
318 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid_end_bio_io()
319 pr_debug("raid1: sync end %s on sectors %llu-%llu\n", in raid_end_bio_io()
321 (unsigned long long) bio->bi_iter.bi_sector, in raid_end_bio_io()
322 (unsigned long long) bio_end_sector(bio) - 1); in raid_end_bio_io()
328 * to go idle. All I/Os, even write-behind writes, are done. in raid_end_bio_io()
330 allow_barrier(conf, r1_bio->sector); in raid_end_bio_io()
340 struct r1conf *conf = r1_bio->mddev->private; in update_head_pos()
342 conf->mirrors[disk].head_position = in update_head_pos()
343 r1_bio->sector + (r1_bio->sectors); in update_head_pos()
351 int mirror; in find_bio_disk() local
352 struct r1conf *conf = r1_bio->mddev->private; in find_bio_disk()
353 int raid_disks = conf->raid_disks; in find_bio_disk()
355 for (mirror = 0; mirror < raid_disks * 2; mirror++) in find_bio_disk()
356 if (r1_bio->bios[mirror] == bio) in find_bio_disk()
359 BUG_ON(mirror == raid_disks * 2); in find_bio_disk()
360 update_head_pos(mirror, r1_bio); in find_bio_disk()
362 return mirror; in find_bio_disk()
367 int uptodate = !bio->bi_status; in raid1_end_read_request()
368 struct r1bio *r1_bio = bio->bi_private; in raid1_end_read_request()
369 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_read_request()
370 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; in raid1_end_read_request()
373 * this branch is our 'one mirror IO has finished' event handler: in raid1_end_read_request()
375 update_head_pos(r1_bio->read_disk, r1_bio); in raid1_end_read_request()
378 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_read_request()
379 else if (test_bit(FailFast, &rdev->flags) && in raid1_end_read_request()
380 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_end_read_request()
381 /* This was a fail-fast read so we definitely in raid1_end_read_request()
390 spin_lock_irqsave(&conf->device_lock, flags); in raid1_end_read_request()
391 if (r1_bio->mddev->degraded == conf->raid_disks || in raid1_end_read_request()
392 (r1_bio->mddev->degraded == conf->raid_disks-1 && in raid1_end_read_request()
393 test_bit(In_sync, &rdev->flags))) in raid1_end_read_request()
395 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_end_read_request()
400 rdev_dec_pending(rdev, conf->mddev); in raid1_end_read_request()
406 mdname(conf->mddev), in raid1_end_read_request()
407 rdev->bdev, in raid1_end_read_request()
408 (unsigned long long)r1_bio->sector); in raid1_end_read_request()
409 set_bit(R1BIO_ReadError, &r1_bio->state); in raid1_end_read_request()
418 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in close_write()
419 bio_free_pages(r1_bio->behind_master_bio); in close_write()
420 bio_put(r1_bio->behind_master_bio); in close_write()
421 r1_bio->behind_master_bio = NULL; in close_write()
424 md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, in close_write()
425 r1_bio->sectors, in close_write()
426 !test_bit(R1BIO_Degraded, &r1_bio->state), in close_write()
427 test_bit(R1BIO_BehindIO, &r1_bio->state)); in close_write()
428 md_write_end(r1_bio->mddev); in close_write()
433 if (!atomic_dec_and_test(&r1_bio->remaining)) in r1_bio_write_done()
436 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in r1_bio_write_done()
440 if (test_bit(R1BIO_MadeGood, &r1_bio->state)) in r1_bio_write_done()
449 struct r1bio *r1_bio = bio->bi_private; in raid1_end_write_request()
450 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); in raid1_end_write_request()
451 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_write_request()
453 int mirror = find_bio_disk(r1_bio, bio); in raid1_end_write_request() local
454 struct md_rdev *rdev = conf->mirrors[mirror].rdev; in raid1_end_write_request()
456 sector_t lo = r1_bio->sector; in raid1_end_write_request()
457 sector_t hi = r1_bio->sector + r1_bio->sectors; in raid1_end_write_request()
459 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; in raid1_end_write_request()
462 * 'one mirror IO has finished' event handler: in raid1_end_write_request()
464 if (bio->bi_status && !discard_error) { in raid1_end_write_request()
465 set_bit(WriteErrorSeen, &rdev->flags); in raid1_end_write_request()
466 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in raid1_end_write_request()
468 conf->mddev->recovery); in raid1_end_write_request()
470 if (test_bit(FailFast, &rdev->flags) && in raid1_end_write_request()
471 (bio->bi_opf & MD_FAILFAST) && in raid1_end_write_request()
473 !test_bit(WriteMostly, &rdev->flags)) { in raid1_end_write_request()
474 md_error(r1_bio->mddev, rdev); in raid1_end_write_request()
481 if (!test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
482 set_bit(R1BIO_WriteError, &r1_bio->state); in raid1_end_write_request()
485 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1_end_write_request()
487 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
498 * to user-side. So if something waits for IO, then it in raid1_end_write_request()
504 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
509 * such device for properly reading the data back (we could in raid1_end_write_request()
511 * before rdev->recovery_offset, but for simplicity we don't in raid1_end_write_request()
514 if (test_bit(In_sync, &rdev->flags) && in raid1_end_write_request()
515 !test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
516 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_write_request()
519 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in raid1_end_write_request()
521 r1_bio->bios[mirror] = IO_MADE_GOOD; in raid1_end_write_request()
522 set_bit(R1BIO_MadeGood, &r1_bio->state); in raid1_end_write_request()
527 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_end_write_request()
529 if (test_bit(WriteMostly, &rdev->flags)) in raid1_end_write_request()
530 atomic_dec(&r1_bio->behind_remaining); in raid1_end_write_request()
534 * has safely reached all non-writemostly in raid1_end_write_request()
536 * gets done only once -- we don't ever want to return in raid1_end_write_request()
537 * -EIO here, instead we'll wait in raid1_end_write_request()
539 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && in raid1_end_write_request()
540 test_bit(R1BIO_Uptodate, &r1_bio->state)) { in raid1_end_write_request()
542 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid1_end_write_request()
543 struct bio *mbio = r1_bio->master_bio; in raid1_end_write_request()
545 " %llu-%llu\n", in raid1_end_write_request()
546 (unsigned long long) mbio->bi_iter.bi_sector, in raid1_end_write_request()
547 (unsigned long long) bio_end_sector(mbio) - 1); in raid1_end_write_request()
551 } else if (rdev->mddev->serialize_policy) in raid1_end_write_request()
553 if (r1_bio->bios[mirror] == NULL) in raid1_end_write_request()
554 rdev_dec_pending(rdev, conf->mddev); in raid1_end_write_request()
576 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - in align_to_barrier_unit_end()
587 * be done. There is a per-array 'next expected sequential IO' sector
588 * number - if this matches on the next IO then we use the last disk.
589 * There is also a per-disk 'last know head position' sector that is
595 * because position is mirror, not device based.
601 const sector_t this_sector = r1_bio->sector; in read_balance()
620 sectors = r1_bio->sectors; in read_balance()
621 best_disk = -1; in read_balance()
622 best_dist_disk = -1; in read_balance()
624 best_pending_disk = -1; in read_balance()
629 clear_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
631 if ((conf->mddev->recovery_cp < this_sector + sectors) || in read_balance()
632 (mddev_is_clustered(conf->mddev) && in read_balance()
633 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, in read_balance()
639 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in read_balance()
646 rdev = rcu_dereference(conf->mirrors[disk].rdev); in read_balance()
647 if (r1_bio->bios[disk] == IO_BLOCKED in read_balance()
649 || test_bit(Faulty, &rdev->flags)) in read_balance()
651 if (!test_bit(In_sync, &rdev->flags) && in read_balance()
652 rdev->recovery_offset < this_sector + sectors) in read_balance()
654 if (test_bit(WriteMostly, &rdev->flags)) { in read_balance()
655 /* Don't balance among write-mostly, just in read_balance()
663 best_good_sectors = first_bad - this_sector; in read_balance()
684 bad_sectors -= (this_sector - first_bad); in read_balance()
691 sector_t good_sectors = first_bad - this_sector; in read_balance()
702 best_disk = -1; in read_balance()
708 set_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
710 nonrot = bdev_nonrot(rdev->bdev); in read_balance()
712 pending = atomic_read(&rdev->nr_pending); in read_balance()
713 dist = abs(this_sector - conf->mirrors[disk].head_position); in read_balance()
719 if (conf->mirrors[disk].next_seq_sect == this_sector in read_balance()
721 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; in read_balance()
722 struct raid1_info *mirror = &conf->mirrors[disk]; in read_balance() local
739 mirror->seq_start != MaxSector && in read_balance()
740 mirror->next_seq_sect > opt_iosize && in read_balance()
741 mirror->next_seq_sect - opt_iosize >= in read_balance()
742 mirror->seq_start) { in read_balance()
765 * non-rotational, choose the disk with less pending request even the in read_balance()
767 * mixed ratation/non-rotational disks depending on workload. in read_balance()
769 if (best_disk == -1) { in read_balance()
777 rdev = rcu_dereference(conf->mirrors[best_disk].rdev); in read_balance()
780 atomic_inc(&rdev->nr_pending); in read_balance()
783 if (conf->mirrors[best_disk].next_seq_sect != this_sector) in read_balance()
784 conf->mirrors[best_disk].seq_start = this_sector; in read_balance()
786 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; in read_balance()
797 md_bitmap_unplug(conf->mddev->bitmap); in flush_bio_list()
798 wake_up(&conf->wait_barrier); in flush_bio_list()
801 struct bio *next = bio->bi_next; in flush_bio_list()
802 struct md_rdev *rdev = (void *)bio->bi_bdev; in flush_bio_list()
803 bio->bi_next = NULL; in flush_bio_list()
804 bio_set_dev(bio, rdev->bdev); in flush_bio_list()
805 if (test_bit(Faulty, &rdev->flags)) { in flush_bio_list()
808 !bdev_max_discard_sectors(bio->bi_bdev))) in flush_bio_list()
823 spin_lock_irq(&conf->device_lock); in flush_pending_writes()
825 if (conf->pending_bio_list.head) { in flush_pending_writes()
829 bio = bio_list_get(&conf->pending_bio_list); in flush_pending_writes()
830 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
834 * current->state might be TASK_UNINTERRUPTIBLE which will in flush_pending_writes()
838 * is a false-positive. Silence the warning by resetting in flush_pending_writes()
846 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
858 * We choose only to raise the barrier if no-one is waiting for the
870 * If resync/recovery is interrupted, returns -EINTR;
877 spin_lock_irq(&conf->resync_lock); in raise_barrier()
880 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
881 !atomic_read(&conf->nr_waiting[idx]), in raise_barrier()
882 conf->resync_lock); in raise_barrier()
885 atomic_inc(&conf->barrier[idx]); in raise_barrier()
887 * In raise_barrier() we firstly increase conf->barrier[idx] then in raise_barrier()
888 * check conf->nr_pending[idx]. In _wait_barrier() we firstly in raise_barrier()
889 * increase conf->nr_pending[idx] then check conf->barrier[idx]. in raise_barrier()
890 * A memory barrier here to make sure conf->nr_pending[idx] won't in raise_barrier()
891 * be fetched before conf->barrier[idx] is increased. Otherwise in raise_barrier()
898 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O in raise_barrier()
900 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches in raise_barrier()
903 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
904 (!conf->array_frozen && in raise_barrier()
905 !atomic_read(&conf->nr_pending[idx]) && in raise_barrier()
906 atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || in raise_barrier()
907 test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), in raise_barrier()
908 conf->resync_lock); in raise_barrier()
910 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { in raise_barrier()
911 atomic_dec(&conf->barrier[idx]); in raise_barrier()
912 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
913 wake_up(&conf->wait_barrier); in raise_barrier()
914 return -EINTR; in raise_barrier()
917 atomic_inc(&conf->nr_sync_pending); in raise_barrier()
918 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
927 BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); in lower_barrier()
929 atomic_dec(&conf->barrier[idx]); in lower_barrier()
930 atomic_dec(&conf->nr_sync_pending); in lower_barrier()
931 wake_up(&conf->wait_barrier); in lower_barrier()
939 * We need to increase conf->nr_pending[idx] very early here, in _wait_barrier()
941 * conf->nr_pending[idx] to be 0. Then we can avoid holding in _wait_barrier()
942 * conf->resync_lock when there is no barrier raised in same in _wait_barrier()
946 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
948 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then in _wait_barrier()
949 * check conf->barrier[idx]. In raise_barrier() we firstly increase in _wait_barrier()
950 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory in _wait_barrier()
951 * barrier is necessary here to make sure conf->barrier[idx] won't be in _wait_barrier()
952 * fetched before conf->nr_pending[idx] is increased. Otherwise there in _wait_barrier()
959 * here. If during we check conf->barrier[idx], the array is in _wait_barrier()
960 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is in _wait_barrier()
966 if (!READ_ONCE(conf->array_frozen) && in _wait_barrier()
967 !atomic_read(&conf->barrier[idx])) in _wait_barrier()
971 * After holding conf->resync_lock, conf->nr_pending[idx] in _wait_barrier()
974 * raise_barrer() might be waiting for conf->nr_pending[idx] in _wait_barrier()
977 spin_lock_irq(&conf->resync_lock); in _wait_barrier()
978 atomic_inc(&conf->nr_waiting[idx]); in _wait_barrier()
979 atomic_dec(&conf->nr_pending[idx]); in _wait_barrier()
984 wake_up(&conf->wait_barrier); in _wait_barrier()
991 wait_event_lock_irq(conf->wait_barrier, in _wait_barrier()
992 !conf->array_frozen && in _wait_barrier()
993 !atomic_read(&conf->barrier[idx]), in _wait_barrier()
994 conf->resync_lock); in _wait_barrier()
995 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
998 atomic_dec(&conf->nr_waiting[idx]); in _wait_barrier()
999 spin_unlock_irq(&conf->resync_lock); in _wait_barrier()
1013 * conf->barrier[idx] here, memory barrier is unnecessary as well. in wait_read_barrier()
1015 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1017 if (!READ_ONCE(conf->array_frozen)) in wait_read_barrier()
1020 spin_lock_irq(&conf->resync_lock); in wait_read_barrier()
1021 atomic_inc(&conf->nr_waiting[idx]); in wait_read_barrier()
1022 atomic_dec(&conf->nr_pending[idx]); in wait_read_barrier()
1027 wake_up(&conf->wait_barrier); in wait_read_barrier()
1035 wait_event_lock_irq(conf->wait_barrier, in wait_read_barrier()
1036 !conf->array_frozen, in wait_read_barrier()
1037 conf->resync_lock); in wait_read_barrier()
1038 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1041 atomic_dec(&conf->nr_waiting[idx]); in wait_read_barrier()
1042 spin_unlock_irq(&conf->resync_lock); in wait_read_barrier()
1055 atomic_dec(&conf->nr_pending[idx]); in _allow_barrier()
1056 wake_up(&conf->wait_barrier); in _allow_barrier()
1066 /* conf->resync_lock should be held */
1071 ret = atomic_read(&conf->nr_sync_pending); in get_unqueued_pending()
1073 ret += atomic_read(&conf->nr_pending[idx]) - in get_unqueued_pending()
1074 atomic_read(&conf->nr_queued[idx]); in get_unqueued_pending()
1092 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the in freeze_array()
1094 * normal I/O are queued, sum of all conf->nr_pending[] will match sum in freeze_array()
1095 * of all conf->nr_queued[]. But normal I/O failure is an exception, in freeze_array()
1104 spin_lock_irq(&conf->resync_lock); in freeze_array()
1105 conf->array_frozen = 1; in freeze_array()
1106 raid1_log(conf->mddev, "wait freeze"); in freeze_array()
1108 conf->wait_barrier, in freeze_array()
1110 conf->resync_lock, in freeze_array()
1112 spin_unlock_irq(&conf->resync_lock); in freeze_array()
1117 spin_lock_irq(&conf->resync_lock); in unfreeze_array()
1118 conf->array_frozen = 0; in unfreeze_array()
1119 spin_unlock_irq(&conf->resync_lock); in unfreeze_array()
1120 wake_up(&conf->wait_barrier); in unfreeze_array()
1126 int size = bio->bi_iter.bi_size; in alloc_behind_master_bio()
1127 unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; in alloc_behind_master_bio()
1132 &r1_bio->mddev->bio_set); in alloc_behind_master_bio()
1138 behind_bio->bi_iter.bi_size = size; in alloc_behind_master_bio()
1152 size -= len; in alloc_behind_master_bio()
1158 r1_bio->behind_master_bio = behind_bio; in alloc_behind_master_bio()
1159 set_bit(R1BIO_BehindIO, &r1_bio->state); in alloc_behind_master_bio()
1165 bio->bi_iter.bi_size); in alloc_behind_master_bio()
1174 struct mddev *mddev = plug->cb.data; in raid1_unplug()
1175 struct r1conf *conf = mddev->private; in raid1_unplug()
1178 if (from_schedule || current->bio_list) { in raid1_unplug()
1179 spin_lock_irq(&conf->device_lock); in raid1_unplug()
1180 bio_list_merge(&conf->pending_bio_list, &plug->pending); in raid1_unplug()
1181 spin_unlock_irq(&conf->device_lock); in raid1_unplug()
1182 wake_up(&conf->wait_barrier); in raid1_unplug()
1183 md_wakeup_thread(mddev->thread); in raid1_unplug()
1188 /* we aren't scheduling, so we can do the write-out directly. */ in raid1_unplug()
1189 bio = bio_list_get(&plug->pending); in raid1_unplug()
1196 r1_bio->master_bio = bio; in init_r1bio()
1197 r1_bio->sectors = bio_sectors(bio); in init_r1bio()
1198 r1_bio->state = 0; in init_r1bio()
1199 r1_bio->mddev = mddev; in init_r1bio()
1200 r1_bio->sector = bio->bi_iter.bi_sector; in init_r1bio()
1206 struct r1conf *conf = mddev->private; in alloc_r1bio()
1209 r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); in alloc_r1bio()
1211 memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); in alloc_r1bio()
1219 struct r1conf *conf = mddev->private; in raid1_read_request()
1220 struct raid1_info *mirror; in raid1_read_request() local
1222 struct bitmap *bitmap = mddev->bitmap; in raid1_read_request()
1224 const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; in raid1_read_request()
1241 rdev = rcu_dereference(conf->mirrors[r1_bio->read_disk].rdev); in raid1_read_request()
1243 snprintf(b, sizeof(b), "%pg", rdev->bdev); in raid1_read_request()
1253 if (!wait_read_barrier(conf, bio->bi_iter.bi_sector, in raid1_read_request()
1254 bio->bi_opf & REQ_NOWAIT)) { in raid1_read_request()
1263 r1_bio->sectors = max_read_sectors; in raid1_read_request()
1266 * make_request() can abort the operation when read-ahead is being in raid1_read_request()
1277 (unsigned long long)r1_bio->sector); in raid1_read_request()
1282 mirror = conf->mirrors + rdisk; in raid1_read_request()
1285 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n", in raid1_read_request()
1287 (unsigned long long)r1_bio->sector, in raid1_read_request()
1288 mirror->rdev->bdev); in raid1_read_request()
1290 if (test_bit(WriteMostly, &mirror->rdev->flags) && in raid1_read_request()
1293 * Reading from a write-mostly device must take care not to in raid1_read_request()
1294 * over-take any writes that are 'behind' in raid1_read_request()
1297 wait_event(bitmap->behind_wait, in raid1_read_request()
1298 atomic_read(&bitmap->behind_writes) == 0); in raid1_read_request()
1303 gfp, &conf->bio_split); in raid1_read_request()
1307 r1_bio->master_bio = bio; in raid1_read_request()
1308 r1_bio->sectors = max_sectors; in raid1_read_request()
1311 r1_bio->read_disk = rdisk; in raid1_read_request()
1313 if (!r1bio_existed && blk_queue_io_stat(bio->bi_bdev->bd_disk->queue)) in raid1_read_request()
1314 r1_bio->start_time = bio_start_io_acct(bio); in raid1_read_request()
1316 read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp, in raid1_read_request()
1317 &mddev->bio_set); in raid1_read_request()
1319 r1_bio->bios[rdisk] = read_bio; in raid1_read_request()
1321 read_bio->bi_iter.bi_sector = r1_bio->sector + in raid1_read_request()
1322 mirror->rdev->data_offset; in raid1_read_request()
1323 read_bio->bi_end_io = raid1_end_read_request; in raid1_read_request()
1325 if (test_bit(FailFast, &mirror->rdev->flags) && in raid1_read_request()
1326 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_read_request()
1327 read_bio->bi_opf |= MD_FAILFAST; in raid1_read_request()
1328 read_bio->bi_private = r1_bio; in raid1_read_request()
1330 if (mddev->gendisk) in raid1_read_request()
1331 trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), in raid1_read_request()
1332 r1_bio->sector); in raid1_read_request()
1340 struct r1conf *conf = mddev->private; in raid1_write_request()
1343 struct bitmap *bitmap = mddev->bitmap; in raid1_write_request()
1353 md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1354 bio->bi_iter.bi_sector, bio_end_sector(bio))) { in raid1_write_request()
1357 if (bio->bi_opf & REQ_NOWAIT) { in raid1_write_request()
1362 prepare_to_wait(&conf->wait_barrier, in raid1_write_request()
1364 if (!md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1365 bio->bi_iter.bi_sector, in raid1_write_request()
1370 finish_wait(&conf->wait_barrier, &w); in raid1_write_request()
1378 if (!wait_barrier(conf, bio->bi_iter.bi_sector, in raid1_write_request()
1379 bio->bi_opf & REQ_NOWAIT)) { in raid1_write_request()
1385 r1_bio->sectors = max_write_sectors; in raid1_write_request()
1398 disks = conf->raid_disks * 2; in raid1_write_request()
1402 max_sectors = r1_bio->sectors; in raid1_write_request()
1404 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_write_request()
1407 * The write-behind io is only attempted on drives marked as in raid1_write_request()
1408 * write-mostly, which means we could allocate write behind in raid1_write_request()
1411 if (rdev && test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1414 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { in raid1_write_request()
1415 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1419 r1_bio->bios[i] = NULL; in raid1_write_request()
1420 if (!rdev || test_bit(Faulty, &rdev->flags)) { in raid1_write_request()
1421 if (i < conf->raid_disks) in raid1_write_request()
1422 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1_write_request()
1426 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1427 if (test_bit(WriteErrorSeen, &rdev->flags)) { in raid1_write_request()
1432 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, in raid1_write_request()
1437 set_bit(BlockedBadBlocks, &rdev->flags); in raid1_write_request()
1441 if (is_bad && first_bad <= r1_bio->sector) { in raid1_write_request()
1443 bad_sectors -= (r1_bio->sector - first_bad); in raid1_write_request()
1452 * missing, so it might be re-added, in raid1_write_request()
1457 * in-sync is recorded in the bad in raid1_write_request()
1463 int good_sectors = first_bad - r1_bio->sector; in raid1_write_request()
1468 r1_bio->bios[i] = bio; in raid1_write_request()
1477 if (r1_bio->bios[j]) in raid1_write_request()
1478 rdev_dec_pending(conf->mirrors[j].rdev, mddev); in raid1_write_request()
1479 r1_bio->state = 0; in raid1_write_request()
1480 allow_barrier(conf, bio->bi_iter.bi_sector); in raid1_write_request()
1482 if (bio->bi_opf & REQ_NOWAIT) { in raid1_write_request()
1486 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); in raid1_write_request()
1488 wait_barrier(conf, bio->bi_iter.bi_sector, false); in raid1_write_request()
1494 * alloc_behind_master_bio allocates a copy of the data payload a page in raid1_write_request()
1503 GFP_NOIO, &conf->bio_split); in raid1_write_request()
1507 r1_bio->master_bio = bio; in raid1_write_request()
1508 r1_bio->sectors = max_sectors; in raid1_write_request()
1511 if (blk_queue_io_stat(bio->bi_bdev->bd_disk->queue)) in raid1_write_request()
1512 r1_bio->start_time = bio_start_io_acct(bio); in raid1_write_request()
1513 atomic_set(&r1_bio->remaining, 1); in raid1_write_request()
1514 atomic_set(&r1_bio->behind_remaining, 0); in raid1_write_request()
1520 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_write_request()
1521 if (!r1_bio->bios[i]) in raid1_write_request()
1530 test_bit(WriteMostly, &rdev->flags) && in raid1_write_request()
1531 (atomic_read(&bitmap->behind_writes) in raid1_write_request()
1532 < mddev->bitmap_info.max_write_behind) && in raid1_write_request()
1533 !waitqueue_active(&bitmap->behind_wait)) { in raid1_write_request()
1537 md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, in raid1_write_request()
1538 test_bit(R1BIO_BehindIO, &r1_bio->state)); in raid1_write_request()
1542 if (r1_bio->behind_master_bio) { in raid1_write_request()
1543 mbio = bio_alloc_clone(rdev->bdev, in raid1_write_request()
1544 r1_bio->behind_master_bio, in raid1_write_request()
1545 GFP_NOIO, &mddev->bio_set); in raid1_write_request()
1546 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_write_request()
1548 if (test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1549 atomic_inc(&r1_bio->behind_remaining); in raid1_write_request()
1551 mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, in raid1_write_request()
1552 &mddev->bio_set); in raid1_write_request()
1554 if (mddev->serialize_policy) in raid1_write_request()
1558 r1_bio->bios[i] = mbio; in raid1_write_request()
1560 mbio->bi_iter.bi_sector = (r1_bio->sector + rdev->data_offset); in raid1_write_request()
1561 mbio->bi_end_io = raid1_end_write_request; in raid1_write_request()
1562 mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); in raid1_write_request()
1563 if (test_bit(FailFast, &rdev->flags) && in raid1_write_request()
1564 !test_bit(WriteMostly, &rdev->flags) && in raid1_write_request()
1565 conf->raid_disks - mddev->degraded > 1) in raid1_write_request()
1566 mbio->bi_opf |= MD_FAILFAST; in raid1_write_request()
1567 mbio->bi_private = r1_bio; in raid1_write_request()
1569 atomic_inc(&r1_bio->remaining); in raid1_write_request()
1571 if (mddev->gendisk) in raid1_write_request()
1572 trace_block_bio_remap(mbio, disk_devt(mddev->gendisk), in raid1_write_request()
1573 r1_bio->sector); in raid1_write_request()
1575 mbio->bi_bdev = (void *)rdev; in raid1_write_request()
1583 bio_list_add(&plug->pending, mbio); in raid1_write_request()
1585 spin_lock_irqsave(&conf->device_lock, flags); in raid1_write_request()
1586 bio_list_add(&conf->pending_bio_list, mbio); in raid1_write_request()
1587 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_write_request()
1588 md_wakeup_thread(mddev->thread); in raid1_write_request()
1595 wake_up(&conf->wait_barrier); in raid1_write_request()
1602 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid1_make_request()
1614 bio->bi_iter.bi_sector, bio_sectors(bio)); in raid1_make_request()
1628 struct r1conf *conf = mddev->private; in raid1_status()
1631 seq_printf(seq, " [%d/%d] [", conf->raid_disks, in raid1_status()
1632 conf->raid_disks - mddev->degraded); in raid1_status()
1634 for (i = 0; i < conf->raid_disks; i++) { in raid1_status()
1635 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_status()
1637 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid1_status()
1644 * raid1_error() - RAID1 error handler.
1650 * - &MD_BROKEN flag is set in &mddev->flags.
1651 * - recovery is disabled.
1653 * - recovery is interrupted.
1654 * - &mddev->degraded is bumped.
1657 * &mddev->fail_last_dev is off.
1661 struct r1conf *conf = mddev->private; in raid1_error()
1664 spin_lock_irqsave(&conf->device_lock, flags); in raid1_error()
1666 if (test_bit(In_sync, &rdev->flags) && in raid1_error()
1667 (conf->raid_disks - mddev->degraded) == 1) { in raid1_error()
1668 set_bit(MD_BROKEN, &mddev->flags); in raid1_error()
1670 if (!mddev->fail_last_dev) { in raid1_error()
1671 conf->recovery_disabled = mddev->recovery_disabled; in raid1_error()
1672 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1676 set_bit(Blocked, &rdev->flags); in raid1_error()
1677 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid1_error()
1678 mddev->degraded++; in raid1_error()
1679 set_bit(Faulty, &rdev->flags); in raid1_error()
1680 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1684 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_error()
1685 set_mask_bits(&mddev->sb_flags, 0, in raid1_error()
1689 mdname(mddev), rdev->bdev, in raid1_error()
1690 mdname(mddev), conf->raid_disks - mddev->degraded); in raid1_error()
1702 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, in print_conf()
1703 conf->raid_disks); in print_conf()
1706 for (i = 0; i < conf->raid_disks; i++) { in print_conf()
1707 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in print_conf()
1710 i, !test_bit(In_sync, &rdev->flags), in print_conf()
1711 !test_bit(Faulty, &rdev->flags), in print_conf()
1712 rdev->bdev); in print_conf()
1726 mempool_exit(&conf->r1buf_pool); in close_sync()
1732 struct r1conf *conf = mddev->private; in raid1_spare_active()
1741 * which expects 'In_sync' flags and ->degraded to be consistent. in raid1_spare_active()
1743 spin_lock_irqsave(&conf->device_lock, flags); in raid1_spare_active()
1744 for (i = 0; i < conf->raid_disks; i++) { in raid1_spare_active()
1745 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_spare_active()
1746 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; in raid1_spare_active()
1748 && !test_bit(Candidate, &repl->flags) in raid1_spare_active()
1749 && repl->recovery_offset == MaxSector in raid1_spare_active()
1750 && !test_bit(Faulty, &repl->flags) in raid1_spare_active()
1751 && !test_and_set_bit(In_sync, &repl->flags)) { in raid1_spare_active()
1754 !test_and_clear_bit(In_sync, &rdev->flags)) in raid1_spare_active()
1759 * it gets removed and never re-added in raid1_spare_active()
1761 set_bit(Faulty, &rdev->flags); in raid1_spare_active()
1763 rdev->sysfs_state); in raid1_spare_active()
1767 && rdev->recovery_offset == MaxSector in raid1_spare_active()
1768 && !test_bit(Faulty, &rdev->flags) in raid1_spare_active()
1769 && !test_and_set_bit(In_sync, &rdev->flags)) { in raid1_spare_active()
1771 sysfs_notify_dirent_safe(rdev->sysfs_state); in raid1_spare_active()
1774 mddev->degraded -= count; in raid1_spare_active()
1775 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_spare_active()
1783 struct r1conf *conf = mddev->private; in raid1_add_disk()
1784 int err = -EEXIST; in raid1_add_disk()
1785 int mirror = 0; in raid1_add_disk() local
1788 int last = conf->raid_disks - 1; in raid1_add_disk()
1790 if (mddev->recovery_disabled == conf->recovery_disabled) in raid1_add_disk()
1791 return -EBUSY; in raid1_add_disk()
1794 return -ENXIO; in raid1_add_disk()
1796 if (rdev->raid_disk >= 0) in raid1_add_disk()
1797 first = last = rdev->raid_disk; in raid1_add_disk()
1800 * find the disk ... but prefer rdev->saved_raid_disk in raid1_add_disk()
1803 if (rdev->saved_raid_disk >= 0 && in raid1_add_disk()
1804 rdev->saved_raid_disk >= first && in raid1_add_disk()
1805 rdev->saved_raid_disk < conf->raid_disks && in raid1_add_disk()
1806 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid1_add_disk()
1807 first = last = rdev->saved_raid_disk; in raid1_add_disk()
1809 for (mirror = first; mirror <= last; mirror++) { in raid1_add_disk()
1810 p = conf->mirrors + mirror; in raid1_add_disk()
1811 if (!p->rdev) { in raid1_add_disk()
1812 if (mddev->gendisk) in raid1_add_disk()
1813 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_add_disk()
1814 rdev->data_offset << 9); in raid1_add_disk()
1816 p->head_position = 0; in raid1_add_disk()
1817 rdev->raid_disk = mirror; in raid1_add_disk()
1822 if (rdev->saved_raid_disk < 0) in raid1_add_disk()
1823 conf->fullsync = 1; in raid1_add_disk()
1824 rcu_assign_pointer(p->rdev, rdev); in raid1_add_disk()
1827 if (test_bit(WantReplacement, &p->rdev->flags) && in raid1_add_disk()
1828 p[conf->raid_disks].rdev == NULL) { in raid1_add_disk()
1830 clear_bit(In_sync, &rdev->flags); in raid1_add_disk()
1831 set_bit(Replacement, &rdev->flags); in raid1_add_disk()
1832 rdev->raid_disk = mirror; in raid1_add_disk()
1834 conf->fullsync = 1; in raid1_add_disk()
1835 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev); in raid1_add_disk()
1845 struct r1conf *conf = mddev->private; in raid1_remove_disk()
1847 int number = rdev->raid_disk; in raid1_remove_disk()
1848 struct raid1_info *p = conf->mirrors + number; in raid1_remove_disk()
1850 if (rdev != p->rdev) in raid1_remove_disk()
1851 p = conf->mirrors + conf->raid_disks + number; in raid1_remove_disk()
1854 if (rdev == p->rdev) { in raid1_remove_disk()
1855 if (test_bit(In_sync, &rdev->flags) || in raid1_remove_disk()
1856 atomic_read(&rdev->nr_pending)) { in raid1_remove_disk()
1857 err = -EBUSY; in raid1_remove_disk()
1860 /* Only remove non-faulty devices if recovery in raid1_remove_disk()
1863 if (!test_bit(Faulty, &rdev->flags) && in raid1_remove_disk()
1864 mddev->recovery_disabled != conf->recovery_disabled && in raid1_remove_disk()
1865 mddev->degraded < conf->raid_disks) { in raid1_remove_disk()
1866 err = -EBUSY; in raid1_remove_disk()
1869 p->rdev = NULL; in raid1_remove_disk()
1870 if (!test_bit(RemoveSynchronized, &rdev->flags)) { in raid1_remove_disk()
1872 if (atomic_read(&rdev->nr_pending)) { in raid1_remove_disk()
1874 err = -EBUSY; in raid1_remove_disk()
1875 p->rdev = rdev; in raid1_remove_disk()
1879 if (conf->mirrors[conf->raid_disks + number].rdev) { in raid1_remove_disk()
1885 conf->mirrors[conf->raid_disks + number].rdev; in raid1_remove_disk()
1887 if (atomic_read(&repl->nr_pending)) { in raid1_remove_disk()
1894 err = -EBUSY; in raid1_remove_disk()
1898 clear_bit(Replacement, &repl->flags); in raid1_remove_disk()
1899 p->rdev = repl; in raid1_remove_disk()
1900 conf->mirrors[conf->raid_disks + number].rdev = NULL; in raid1_remove_disk()
1904 clear_bit(WantReplacement, &rdev->flags); in raid1_remove_disk()
1917 update_head_pos(r1_bio->read_disk, r1_bio); in end_sync_read()
1920 * we have read a block, now it needs to be re-written, in end_sync_read()
1921 * or re-read if the read failed. in end_sync_read()
1924 if (!bio->bi_status) in end_sync_read()
1925 set_bit(R1BIO_Uptodate, &r1_bio->state); in end_sync_read()
1927 if (atomic_dec_and_test(&r1_bio->remaining)) in end_sync_read()
1934 sector_t s = r1_bio->sector; in abort_sync_write()
1935 long sectors_to_go = r1_bio->sectors; in abort_sync_write()
1939 md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); in abort_sync_write()
1941 sectors_to_go -= sync_blocks; in abort_sync_write()
1947 if (atomic_dec_and_test(&r1_bio->remaining)) { in put_sync_write_buf()
1948 struct mddev *mddev = r1_bio->mddev; in put_sync_write_buf()
1949 int s = r1_bio->sectors; in put_sync_write_buf()
1951 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in put_sync_write_buf()
1952 test_bit(R1BIO_WriteError, &r1_bio->state)) in put_sync_write_buf()
1963 int uptodate = !bio->bi_status; in end_sync_write()
1965 struct mddev *mddev = r1_bio->mddev; in end_sync_write()
1966 struct r1conf *conf = mddev->private; in end_sync_write()
1969 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; in end_sync_write()
1973 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
1974 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
1976 mddev->recovery); in end_sync_write()
1977 set_bit(R1BIO_WriteError, &r1_bio->state); in end_sync_write()
1978 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in end_sync_write()
1980 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, in end_sync_write()
1981 r1_bio->sector, in end_sync_write()
1982 r1_bio->sectors, in end_sync_write()
1985 set_bit(R1BIO_MadeGood, &r1_bio->state); in end_sync_write()
1997 set_bit(WriteErrorSeen, &rdev->flags); in r1_sync_page_io()
1999 &rdev->flags)) in r1_sync_page_io()
2001 rdev->mddev->recovery); in r1_sync_page_io()
2003 /* need to record an error - either for the block or the device */ in r1_sync_page_io()
2005 md_error(rdev->mddev, rdev); in r1_sync_page_io()
2012 * good data, much like with normal read errors. Only in fix_sync_read_error()
2014 * need to re-issue the read request. in fix_sync_read_error()
2022 struct mddev *mddev = r1_bio->mddev; in fix_sync_read_error()
2023 struct r1conf *conf = mddev->private; in fix_sync_read_error()
2024 struct bio *bio = r1_bio->bios[r1_bio->read_disk]; in fix_sync_read_error()
2025 struct page **pages = get_resync_pages(bio)->pages; in fix_sync_read_error()
2026 sector_t sect = r1_bio->sector; in fix_sync_read_error()
2027 int sectors = r1_bio->sectors; in fix_sync_read_error()
2031 rdev = conf->mirrors[r1_bio->read_disk].rdev; in fix_sync_read_error()
2032 if (test_bit(FailFast, &rdev->flags)) { in fix_sync_read_error()
2033 /* Don't try recovering from here - just fail it in fix_sync_read_error()
2036 if (test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2040 bio->bi_end_io = end_sync_write; in fix_sync_read_error()
2045 int d = r1_bio->read_disk; in fix_sync_read_error()
2052 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()
2066 if (d == conf->raid_disks * 2) in fix_sync_read_error()
2068 } while (!success && d != r1_bio->read_disk); in fix_sync_read_error()
2078 mdname(mddev), bio->bi_bdev, in fix_sync_read_error()
2079 (unsigned long long)r1_bio->sector); in fix_sync_read_error()
2080 for (d = 0; d < conf->raid_disks * 2; d++) { in fix_sync_read_error()
2081 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2082 if (!rdev || test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2088 conf->recovery_disabled = in fix_sync_read_error()
2089 mddev->recovery_disabled; in fix_sync_read_error()
2090 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in fix_sync_read_error()
2091 md_done_sync(mddev, r1_bio->sectors, 0); in fix_sync_read_error()
2096 sectors -= s; in fix_sync_read_error()
2103 /* write it back and re-read */ in fix_sync_read_error()
2104 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2106 d = conf->raid_disks * 2; in fix_sync_read_error()
2107 d--; in fix_sync_read_error()
2108 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2110 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2114 r1_bio->bios[d]->bi_end_io = NULL; in fix_sync_read_error()
2119 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2121 d = conf->raid_disks * 2; in fix_sync_read_error()
2122 d--; in fix_sync_read_error()
2123 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2125 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2129 atomic_add(s, &rdev->corrected_errors); in fix_sync_read_error()
2131 sectors -= s; in fix_sync_read_error()
2135 set_bit(R1BIO_Uptodate, &r1_bio->state); in fix_sync_read_error()
2136 bio->bi_status = 0; in fix_sync_read_error()
2147 * attempt an over-write in process_checks()
2149 struct mddev *mddev = r1_bio->mddev; in process_checks()
2150 struct r1conf *conf = mddev->private; in process_checks()
2156 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); in process_checks()
2157 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2159 struct bio *b = r1_bio->bios[i]; in process_checks()
2161 if (b->bi_end_io != end_sync_read) in process_checks()
2164 status = b->bi_status; in process_checks()
2165 bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ); in process_checks()
2166 b->bi_status = status; in process_checks()
2167 b->bi_iter.bi_sector = r1_bio->sector + in process_checks()
2168 conf->mirrors[i].rdev->data_offset; in process_checks()
2169 b->bi_end_io = end_sync_read; in process_checks()
2170 rp->raid_bio = r1_bio; in process_checks()
2171 b->bi_private = rp; in process_checks()
2174 md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); in process_checks()
2176 for (primary = 0; primary < conf->raid_disks * 2; primary++) in process_checks()
2177 if (r1_bio->bios[primary]->bi_end_io == end_sync_read && in process_checks()
2178 !r1_bio->bios[primary]->bi_status) { in process_checks()
2179 r1_bio->bios[primary]->bi_end_io = NULL; in process_checks()
2180 rdev_dec_pending(conf->mirrors[primary].rdev, mddev); in process_checks()
2183 r1_bio->read_disk = primary; in process_checks()
2184 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2186 struct bio *pbio = r1_bio->bios[primary]; in process_checks()
2187 struct bio *sbio = r1_bio->bios[i]; in process_checks()
2188 blk_status_t status = sbio->bi_status; in process_checks()
2189 struct page **ppages = get_resync_pages(pbio)->pages; in process_checks()
2190 struct page **spages = get_resync_pages(sbio)->pages; in process_checks()
2195 if (sbio->bi_end_io != end_sync_read) in process_checks()
2198 sbio->bi_status = 0; in process_checks()
2201 page_len[j++] = bi->bv_len; in process_checks()
2204 for (j = vcnt; j-- ; ) { in process_checks()
2213 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); in process_checks()
2214 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) in process_checks()
2217 sbio->bi_end_io = NULL; in process_checks()
2218 rdev_dec_pending(conf->mirrors[i].rdev, mddev); in process_checks()
2228 struct r1conf *conf = mddev->private; in sync_request_write()
2230 int disks = conf->raid_disks * 2; in sync_request_write()
2233 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in sync_request_write()
2234 /* ouch - failed to read all of that. */ in sync_request_write()
2238 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in sync_request_write()
2244 atomic_set(&r1_bio->remaining, 1); in sync_request_write()
2246 wbio = r1_bio->bios[i]; in sync_request_write()
2247 if (wbio->bi_end_io == NULL || in sync_request_write()
2248 (wbio->bi_end_io == end_sync_read && in sync_request_write()
2249 (i == r1_bio->read_disk || in sync_request_write()
2250 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) in sync_request_write()
2252 if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { in sync_request_write()
2258 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) in sync_request_write()
2259 wbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2261 wbio->bi_end_io = end_sync_write; in sync_request_write()
2262 atomic_inc(&r1_bio->remaining); in sync_request_write()
2263 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); in sync_request_write()
2282 struct mddev *mddev = conf->mddev; in fix_read_error()
2298 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2300 (test_bit(In_sync, &rdev->flags) || in fix_read_error()
2301 (!test_bit(Faulty, &rdev->flags) && in fix_read_error()
2302 rdev->recovery_offset >= sect + s)) && in fix_read_error()
2305 atomic_inc(&rdev->nr_pending); in fix_read_error()
2308 conf->tmppage, REQ_OP_READ, false)) in fix_read_error()
2316 if (d == conf->raid_disks * 2) in fix_read_error()
2321 /* Cannot read from anywhere - mark it bad */ in fix_read_error()
2322 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; in fix_read_error()
2327 /* write it back and re-read */ in fix_read_error()
2331 d = conf->raid_disks * 2; in fix_read_error()
2332 d--; in fix_read_error()
2334 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2336 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2337 atomic_inc(&rdev->nr_pending); in fix_read_error()
2340 conf->tmppage, WRITE); in fix_read_error()
2348 d = conf->raid_disks * 2; in fix_read_error()
2349 d--; in fix_read_error()
2351 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2353 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2354 atomic_inc(&rdev->nr_pending); in fix_read_error()
2357 conf->tmppage, READ)) { in fix_read_error()
2358 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2362 rdev->data_offset), in fix_read_error()
2363 rdev->bdev); in fix_read_error()
2369 sectors -= s; in fix_read_error()
2376 struct mddev *mddev = r1_bio->mddev; in narrow_write_error()
2377 struct r1conf *conf = mddev->private; in narrow_write_error()
2378 struct md_rdev *rdev = conf->mirrors[i].rdev; in narrow_write_error()
2380 /* bio has the data to be written to device 'i' where in narrow_write_error()
2394 int sect_to_write = r1_bio->sectors; in narrow_write_error()
2397 if (rdev->badblocks.shift < 0) in narrow_write_error()
2400 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2401 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2402 sector = r1_bio->sector; in narrow_write_error()
2404 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2405 - sector; in narrow_write_error()
2413 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in narrow_write_error()
2414 wbio = bio_alloc_clone(rdev->bdev, in narrow_write_error()
2415 r1_bio->behind_master_bio, in narrow_write_error()
2416 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2418 wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio, in narrow_write_error()
2419 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2423 wbio->bi_iter.bi_sector = r1_bio->sector; in narrow_write_error()
2424 wbio->bi_iter.bi_size = r1_bio->sectors << 9; in narrow_write_error()
2426 bio_trim(wbio, sector - r1_bio->sector, sectors); in narrow_write_error()
2427 wbio->bi_iter.bi_sector += rdev->data_offset; in narrow_write_error()
2436 sect_to_write -= sectors; in narrow_write_error()
2446 int s = r1_bio->sectors; in handle_sync_write_finished()
2447 for (m = 0; m < conf->raid_disks * 2 ; m++) { in handle_sync_write_finished()
2448 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_sync_write_finished()
2449 struct bio *bio = r1_bio->bios[m]; in handle_sync_write_finished()
2450 if (bio->bi_end_io == NULL) in handle_sync_write_finished()
2452 if (!bio->bi_status && in handle_sync_write_finished()
2453 test_bit(R1BIO_MadeGood, &r1_bio->state)) { in handle_sync_write_finished()
2454 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); in handle_sync_write_finished()
2456 if (bio->bi_status && in handle_sync_write_finished()
2457 test_bit(R1BIO_WriteError, &r1_bio->state)) { in handle_sync_write_finished()
2458 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) in handle_sync_write_finished()
2459 md_error(conf->mddev, rdev); in handle_sync_write_finished()
2463 md_done_sync(conf->mddev, s, 1); in handle_sync_write_finished()
2471 for (m = 0; m < conf->raid_disks * 2 ; m++) in handle_write_finished()
2472 if (r1_bio->bios[m] == IO_MADE_GOOD) { in handle_write_finished()
2473 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_write_finished()
2475 r1_bio->sector, in handle_write_finished()
2476 r1_bio->sectors, 0); in handle_write_finished()
2477 rdev_dec_pending(rdev, conf->mddev); in handle_write_finished()
2478 } else if (r1_bio->bios[m] != NULL) { in handle_write_finished()
2485 md_error(conf->mddev, in handle_write_finished()
2486 conf->mirrors[m].rdev); in handle_write_finished()
2488 set_bit(R1BIO_Degraded, &r1_bio->state); in handle_write_finished()
2490 rdev_dec_pending(conf->mirrors[m].rdev, in handle_write_finished()
2491 conf->mddev); in handle_write_finished()
2494 spin_lock_irq(&conf->device_lock); in handle_write_finished()
2495 list_add(&r1_bio->retry_list, &conf->bio_end_io_list); in handle_write_finished()
2496 idx = sector_to_idx(r1_bio->sector); in handle_write_finished()
2497 atomic_inc(&conf->nr_queued[idx]); in handle_write_finished()
2498 spin_unlock_irq(&conf->device_lock); in handle_write_finished()
2503 wake_up(&conf->wait_barrier); in handle_write_finished()
2504 md_wakeup_thread(conf->mddev->thread); in handle_write_finished()
2506 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in handle_write_finished()
2514 struct mddev *mddev = conf->mddev; in handle_read_error()
2518 clear_bit(R1BIO_ReadError, &r1_bio->state); in handle_read_error()
2522 * other devices. When we find one, we re-write in handle_read_error()
2528 bio = r1_bio->bios[r1_bio->read_disk]; in handle_read_error()
2530 r1_bio->bios[r1_bio->read_disk] = NULL; in handle_read_error()
2532 rdev = conf->mirrors[r1_bio->read_disk].rdev; in handle_read_error()
2533 if (mddev->ro == 0 in handle_read_error()
2534 && !test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2536 fix_read_error(conf, r1_bio->read_disk, in handle_read_error()
2537 r1_bio->sector, r1_bio->sectors); in handle_read_error()
2539 } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2542 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; in handle_read_error()
2545 rdev_dec_pending(rdev, conf->mddev); in handle_read_error()
2546 allow_barrier(conf, r1_bio->sector); in handle_read_error()
2547 bio = r1_bio->master_bio; in handle_read_error()
2550 r1_bio->state = 0; in handle_read_error()
2551 raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); in handle_read_error()
2556 struct mddev *mddev = thread->mddev; in raid1d()
2559 struct r1conf *conf = mddev->private; in raid1d()
2560 struct list_head *head = &conf->retry_list; in raid1d()
2566 if (!list_empty_careful(&conf->bio_end_io_list) && in raid1d()
2567 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid1d()
2569 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2570 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) in raid1d()
2571 list_splice_init(&conf->bio_end_io_list, &tmp); in raid1d()
2572 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2576 list_del(&r1_bio->retry_list); in raid1d()
2577 idx = sector_to_idx(r1_bio->sector); in raid1d()
2578 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2579 if (mddev->degraded) in raid1d()
2580 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1d()
2581 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2592 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2594 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2597 r1_bio = list_entry(head->prev, struct r1bio, retry_list); in raid1d()
2598 list_del(head->prev); in raid1d()
2599 idx = sector_to_idx(r1_bio->sector); in raid1d()
2600 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2601 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2603 mddev = r1_bio->mddev; in raid1d()
2604 conf = mddev->private; in raid1d()
2605 if (test_bit(R1BIO_IsSync, &r1_bio->state)) { in raid1d()
2606 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2607 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2611 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2612 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2614 else if (test_bit(R1BIO_ReadError, &r1_bio->state)) in raid1d()
2620 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid1d()
2631 BUG_ON(mempool_initialized(&conf->r1buf_pool)); in init_resync()
2633 return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, in init_resync()
2634 r1buf_pool_free, conf->poolinfo); in init_resync()
2639 struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); in raid1_alloc_init_r1buf()
2644 for (i = conf->poolinfo->raid_disks; i--; ) { in raid1_alloc_init_r1buf()
2645 bio = r1bio->bios[i]; in raid1_alloc_init_r1buf()
2646 rps = bio->bi_private; in raid1_alloc_init_r1buf()
2648 bio->bi_private = rps; in raid1_alloc_init_r1buf()
2650 r1bio->master_bio = NULL; in raid1_alloc_init_r1buf()
2657 * We need to make sure that no normal I/O request - particularly write
2658 * requests - conflict with active sync requests.
2667 struct r1conf *conf = mddev->private; in raid1_sync_request()
2671 int disk = -1; in raid1_sync_request()
2673 int wonly = -1; in raid1_sync_request()
2682 if (!mempool_initialized(&conf->r1buf_pool)) in raid1_sync_request()
2686 max_sector = mddev->dev_sectors; in raid1_sync_request()
2691 * We can find the current addess in mddev->curr_resync in raid1_sync_request()
2693 if (mddev->curr_resync < max_sector) /* aborted */ in raid1_sync_request()
2694 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, in raid1_sync_request()
2697 conf->fullsync = 0; in raid1_sync_request()
2699 md_bitmap_close_sync(mddev->bitmap); in raid1_sync_request()
2703 conf->cluster_sync_low = 0; in raid1_sync_request()
2704 conf->cluster_sync_high = 0; in raid1_sync_request()
2709 if (mddev->bitmap == NULL && in raid1_sync_request()
2710 mddev->recovery_cp == MaxSector && in raid1_sync_request()
2711 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid1_sync_request()
2712 conf->fullsync == 0) { in raid1_sync_request()
2714 return max_sector - sector_nr; in raid1_sync_request()
2719 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && in raid1_sync_request()
2720 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2727 * If there is non-resync activity waiting for a turn, then let it in raid1_sync_request()
2730 if (atomic_read(&conf->nr_waiting[idx])) in raid1_sync_request()
2737 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2738 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid1_sync_request()
2751 * and any others (which will be non-In_sync devices) for WRITE. in raid1_sync_request()
2756 r1_bio->mddev = mddev; in raid1_sync_request()
2757 r1_bio->sector = sector_nr; in raid1_sync_request()
2758 r1_bio->state = 0; in raid1_sync_request()
2759 set_bit(R1BIO_IsSync, &r1_bio->state); in raid1_sync_request()
2763 for (i = 0; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2765 bio = r1_bio->bios[i]; in raid1_sync_request()
2767 rdev = rcu_dereference(conf->mirrors[i].rdev); in raid1_sync_request()
2769 test_bit(Faulty, &rdev->flags)) { in raid1_sync_request()
2770 if (i < conf->raid_disks) in raid1_sync_request()
2772 } else if (!test_bit(In_sync, &rdev->flags)) { in raid1_sync_request()
2774 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2784 good_sectors = first_bad - sector_nr; in raid1_sync_request()
2786 bad_sectors -= (sector_nr - first_bad); in raid1_sync_request()
2793 if (test_bit(WriteMostly, &rdev->flags)) { in raid1_sync_request()
2801 bio->bi_end_io = end_sync_read; in raid1_sync_request()
2803 } else if (!test_bit(WriteErrorSeen, &rdev->flags) && in raid1_sync_request()
2804 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid1_sync_request()
2805 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { in raid1_sync_request()
2813 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2817 if (rdev && bio->bi_end_io) { in raid1_sync_request()
2818 atomic_inc(&rdev->nr_pending); in raid1_sync_request()
2819 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; in raid1_sync_request()
2820 bio_set_dev(bio, rdev->bdev); in raid1_sync_request()
2821 if (test_bit(FailFast, &rdev->flags)) in raid1_sync_request()
2822 bio->bi_opf |= MD_FAILFAST; in raid1_sync_request()
2828 r1_bio->read_disk = disk; in raid1_sync_request()
2835 for (i = 0 ; i < conf->raid_disks * 2 ; i++) in raid1_sync_request()
2836 if (r1_bio->bios[i]->bi_end_io == end_sync_write) { in raid1_sync_request()
2837 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_sync_request()
2842 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid1_sync_request()
2852 conf->recovery_disabled = mddev->recovery_disabled; in raid1_sync_request()
2853 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_sync_request()
2860 /* only resync enough to reach the next bad->good in raid1_sync_request()
2865 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) in raid1_sync_request()
2867 write_targets += read_targets-1; in raid1_sync_request()
2870 /* There is nowhere to write, so all non-sync in raid1_sync_request()
2871 * drives must be failed - so we are finished in raid1_sync_request()
2876 rv = max_sector - sector_nr; in raid1_sync_request()
2882 if (max_sector > mddev->resync_max) in raid1_sync_request()
2883 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid1_sync_request()
2892 len = (max_sector - sector_nr) << 9; in raid1_sync_request()
2896 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2898 !conf->fullsync && in raid1_sync_request()
2899 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in raid1_sync_request()
2905 for (i = 0 ; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2908 bio = r1_bio->bios[i]; in raid1_sync_request()
2910 if (bio->bi_end_io) { in raid1_sync_request()
2922 sync_blocks -= (len>>9); in raid1_sync_request()
2925 r1_bio->sectors = nr_sectors; in raid1_sync_request()
2928 conf->cluster_sync_high < sector_nr + nr_sectors) { in raid1_sync_request()
2929 conf->cluster_sync_low = mddev->curr_resync_completed; in raid1_sync_request()
2930 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; in raid1_sync_request()
2932 md_cluster_ops->resync_info_update(mddev, in raid1_sync_request()
2933 conf->cluster_sync_low, in raid1_sync_request()
2934 conf->cluster_sync_high); in raid1_sync_request()
2937 /* For a user-requested sync, we read all readable devices and do a in raid1_sync_request()
2940 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2941 atomic_set(&r1_bio->remaining, read_targets); in raid1_sync_request()
2942 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { in raid1_sync_request()
2943 bio = r1_bio->bios[i]; in raid1_sync_request()
2944 if (bio->bi_end_io == end_sync_read) { in raid1_sync_request()
2945 read_targets--; in raid1_sync_request()
2948 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2953 atomic_set(&r1_bio->remaining, 1); in raid1_sync_request()
2954 bio = r1_bio->bios[r1_bio->read_disk]; in raid1_sync_request()
2957 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2968 return mddev->dev_sectors; in raid1_size()
2977 int err = -ENOMEM; in setup_conf()
2983 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2985 if (!conf->nr_pending) in setup_conf()
2988 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2990 if (!conf->nr_waiting) in setup_conf()
2993 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2995 if (!conf->nr_queued) in setup_conf()
2998 conf->barrier = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
3000 if (!conf->barrier) in setup_conf()
3003 conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), in setup_conf()
3004 mddev->raid_disks, 2), in setup_conf()
3006 if (!conf->mirrors) in setup_conf()
3009 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
3010 if (!conf->tmppage) in setup_conf()
3013 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); in setup_conf()
3014 if (!conf->poolinfo) in setup_conf()
3016 conf->poolinfo->raid_disks = mddev->raid_disks * 2; in setup_conf()
3017 err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, in setup_conf()
3018 rbio_pool_free, conf->poolinfo); in setup_conf()
3022 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
3026 conf->poolinfo->mddev = mddev; in setup_conf()
3028 err = -EINVAL; in setup_conf()
3029 spin_lock_init(&conf->device_lock); in setup_conf()
3031 int disk_idx = rdev->raid_disk; in setup_conf()
3032 if (disk_idx >= mddev->raid_disks in setup_conf()
3035 if (test_bit(Replacement, &rdev->flags)) in setup_conf()
3036 disk = conf->mirrors + mddev->raid_disks + disk_idx; in setup_conf()
3038 disk = conf->mirrors + disk_idx; in setup_conf()
3040 if (disk->rdev) in setup_conf()
3042 disk->rdev = rdev; in setup_conf()
3043 disk->head_position = 0; in setup_conf()
3044 disk->seq_start = MaxSector; in setup_conf()
3046 conf->raid_disks = mddev->raid_disks; in setup_conf()
3047 conf->mddev = mddev; in setup_conf()
3048 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
3049 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
3051 spin_lock_init(&conf->resync_lock); in setup_conf()
3052 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
3054 bio_list_init(&conf->pending_bio_list); in setup_conf()
3055 conf->recovery_disabled = mddev->recovery_disabled - 1; in setup_conf()
3057 err = -EIO; in setup_conf()
3058 for (i = 0; i < conf->raid_disks * 2; i++) { in setup_conf()
3060 disk = conf->mirrors + i; in setup_conf()
3062 if (i < conf->raid_disks && in setup_conf()
3063 disk[conf->raid_disks].rdev) { in setup_conf()
3065 if (!disk->rdev) { in setup_conf()
3069 disk->rdev = in setup_conf()
3070 disk[conf->raid_disks].rdev; in setup_conf()
3071 disk[conf->raid_disks].rdev = NULL; in setup_conf()
3072 } else if (!test_bit(In_sync, &disk->rdev->flags)) in setup_conf()
3073 /* Original is not in_sync - bad */ in setup_conf()
3077 if (!disk->rdev || in setup_conf()
3078 !test_bit(In_sync, &disk->rdev->flags)) { in setup_conf()
3079 disk->head_position = 0; in setup_conf()
3080 if (disk->rdev && in setup_conf()
3081 (disk->rdev->saved_raid_disk < 0)) in setup_conf()
3082 conf->fullsync = 1; in setup_conf()
3086 err = -ENOMEM; in setup_conf()
3087 conf->thread = md_register_thread(raid1d, mddev, "raid1"); in setup_conf()
3088 if (!conf->thread) in setup_conf()
3095 mempool_exit(&conf->r1bio_pool); in setup_conf()
3096 kfree(conf->mirrors); in setup_conf()
3097 safe_put_page(conf->tmppage); in setup_conf()
3098 kfree(conf->poolinfo); in setup_conf()
3099 kfree(conf->nr_pending); in setup_conf()
3100 kfree(conf->nr_waiting); in setup_conf()
3101 kfree(conf->nr_queued); in setup_conf()
3102 kfree(conf->barrier); in setup_conf()
3103 bioset_exit(&conf->bio_split); in setup_conf()
3117 if (mddev->level != 1) { in raid1_run()
3119 mdname(mddev), mddev->level); in raid1_run()
3120 return -EIO; in raid1_run()
3122 if (mddev->reshape_position != MaxSector) { in raid1_run()
3125 return -EIO; in raid1_run()
3128 return -ENOMEM; in raid1_run()
3134 if (mddev->private == NULL) in raid1_run()
3137 conf = mddev->private; in raid1_run()
3142 if (mddev->queue) in raid1_run()
3143 blk_queue_max_write_zeroes_sectors(mddev->queue, 0); in raid1_run()
3146 if (!mddev->gendisk) in raid1_run()
3148 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_run()
3149 rdev->data_offset << 9); in raid1_run()
3152 mddev->degraded = 0; in raid1_run()
3153 for (i = 0; i < conf->raid_disks; i++) in raid1_run()
3154 if (conf->mirrors[i].rdev == NULL || in raid1_run()
3155 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || in raid1_run()
3156 test_bit(Faulty, &conf->mirrors[i].rdev->flags)) in raid1_run()
3157 mddev->degraded++; in raid1_run()
3161 if (conf->raid_disks - mddev->degraded < 1) { in raid1_run()
3162 ret = -EINVAL; in raid1_run()
3166 if (conf->raid_disks - mddev->degraded == 1) in raid1_run()
3167 mddev->recovery_cp = MaxSector; in raid1_run()
3169 if (mddev->recovery_cp != MaxSector) in raid1_run()
3170 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", in raid1_run()
3173 mdname(mddev), mddev->raid_disks - mddev->degraded, in raid1_run()
3174 mddev->raid_disks); in raid1_run()
3179 mddev->thread = conf->thread; in raid1_run()
3180 conf->thread = NULL; in raid1_run()
3181 mddev->private = conf; in raid1_run()
3182 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid1_run()
3188 md_unregister_thread(&mddev->thread); in raid1_run()
3202 mempool_exit(&conf->r1bio_pool); in raid1_free()
3203 kfree(conf->mirrors); in raid1_free()
3204 safe_put_page(conf->tmppage); in raid1_free()
3205 kfree(conf->poolinfo); in raid1_free()
3206 kfree(conf->nr_pending); in raid1_free()
3207 kfree(conf->nr_waiting); in raid1_free()
3208 kfree(conf->nr_queued); in raid1_free()
3209 kfree(conf->barrier); in raid1_free()
3210 bioset_exit(&conf->bio_split); in raid1_free()
3224 if (mddev->external_size && in raid1_resize()
3225 mddev->array_sectors > newsize) in raid1_resize()
3226 return -EINVAL; in raid1_resize()
3227 if (mddev->bitmap) { in raid1_resize()
3228 int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid1_resize()
3233 if (sectors > mddev->dev_sectors && in raid1_resize()
3234 mddev->recovery_cp > mddev->dev_sectors) { in raid1_resize()
3235 mddev->recovery_cp = mddev->dev_sectors; in raid1_resize()
3236 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_resize()
3238 mddev->dev_sectors = sectors; in raid1_resize()
3239 mddev->resync_max_sectors = sectors; in raid1_resize()
3247 * 2/ resize conf->mirrors in raid1_reshape()
3251 * Then resize conf->mirrors and swap in the new r1bio pool. in raid1_reshape()
3259 struct r1conf *conf = mddev->private; in raid1_reshape()
3269 if (mddev->chunk_sectors != mddev->new_chunk_sectors || in raid1_reshape()
3270 mddev->layout != mddev->new_layout || in raid1_reshape()
3271 mddev->level != mddev->new_level) { in raid1_reshape()
3272 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid1_reshape()
3273 mddev->new_layout = mddev->layout; in raid1_reshape()
3274 mddev->new_level = mddev->level; in raid1_reshape()
3275 return -EINVAL; in raid1_reshape()
3281 raid_disks = mddev->raid_disks + mddev->delta_disks; in raid1_reshape()
3283 if (raid_disks < conf->raid_disks) { in raid1_reshape()
3285 for (d= 0; d < conf->raid_disks; d++) in raid1_reshape()
3286 if (conf->mirrors[d].rdev) in raid1_reshape()
3289 return -EBUSY; in raid1_reshape()
3294 return -ENOMEM; in raid1_reshape()
3295 newpoolinfo->mddev = mddev; in raid1_reshape()
3296 newpoolinfo->raid_disks = raid_disks * 2; in raid1_reshape()
3310 return -ENOMEM; in raid1_reshape()
3316 oldpool = conf->r1bio_pool; in raid1_reshape()
3317 conf->r1bio_pool = newpool; in raid1_reshape()
3319 for (d = d2 = 0; d < conf->raid_disks; d++) { in raid1_reshape()
3320 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid1_reshape()
3321 if (rdev && rdev->raid_disk != d2) { in raid1_reshape()
3323 rdev->raid_disk = d2; in raid1_reshape()
3327 mdname(mddev), rdev->raid_disk); in raid1_reshape()
3332 kfree(conf->mirrors); in raid1_reshape()
3333 conf->mirrors = newmirrors; in raid1_reshape()
3334 kfree(conf->poolinfo); in raid1_reshape()
3335 conf->poolinfo = newpoolinfo; in raid1_reshape()
3337 spin_lock_irqsave(&conf->device_lock, flags); in raid1_reshape()
3338 mddev->degraded += (raid_disks - conf->raid_disks); in raid1_reshape()
3339 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_reshape()
3340 conf->raid_disks = mddev->raid_disks = raid_disks; in raid1_reshape()
3341 mddev->delta_disks = 0; in raid1_reshape()
3345 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); in raid1_reshape()
3346 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_reshape()
3347 md_wakeup_thread(mddev->thread); in raid1_reshape()
3355 struct r1conf *conf = mddev->private; in raid1_quiesce()
3368 if (mddev->level == 5 && mddev->raid_disks == 2) { in raid1_takeover()
3370 mddev->new_level = 1; in raid1_takeover()
3371 mddev->new_layout = 0; in raid1_takeover()
3372 mddev->new_chunk_sectors = 0; in raid1_takeover()
3376 conf->array_frozen = 1; in raid1_takeover()
3382 return ERR_PTR(-EINVAL); in raid1_takeover()
3420 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3421 MODULE_ALIAS("md-raid1");
3422 MODULE_ALIAS("md-level-1");