Lines Matching +full:update +full:- +full:fc +full:- +full:fixup
1 // SPDX-License-Identifier: GPL-2.0-or-later
5 * Copyright (C) 2000-2004 Neil Brown
7 * RAID-10 support for md.
24 #include "md-bitmap.h"
78 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid10 " fmt, ##args); } while (0)
80 #include "raid1-10.c"
85 write_sequnlock_irq(&(conf)->resync_lock); \
88 #define cmd_after(conf) write_seqlock_irq(&(conf)->resync_lock)
91 wait_event_cmd((conf)->wait_barrier, cond, cmd_before(conf, cmd), \
98 * for resync bio, r10bio pointer can be retrieved from the per-bio
103 return get_resync_pages(bio)->raid_bio; in get_resync_r10bio()
109 int size = offsetof(struct r10bio, devs[conf->geo.raid_disks]); in r10bio_pool_alloc()
144 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || in r10buf_pool_alloc()
145 test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) in r10buf_pool_alloc()
146 nalloc = conf->copies; /* resync */ in r10buf_pool_alloc()
151 if (!conf->have_replacement) in r10buf_pool_alloc()
162 for (j = nalloc ; j-- ; ) { in r10buf_pool_alloc()
166 bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); in r10buf_pool_alloc()
167 r10_bio->devs[j].bio = bio; in r10buf_pool_alloc()
168 if (!conf->have_replacement) in r10buf_pool_alloc()
173 bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); in r10buf_pool_alloc()
174 r10_bio->devs[j].repl_bio = bio; in r10buf_pool_alloc()
181 struct bio *rbio = r10_bio->devs[j].repl_bio; in r10buf_pool_alloc()
188 bio = r10_bio->devs[j].bio; in r10buf_pool_alloc()
191 &conf->mddev->recovery)) { in r10buf_pool_alloc()
199 rp->raid_bio = r10_bio; in r10buf_pool_alloc()
200 bio->bi_private = rp; in r10buf_pool_alloc()
203 rbio->bi_private = rp_repl; in r10buf_pool_alloc()
210 while (--j >= 0) in r10buf_pool_alloc()
216 if (r10_bio->devs[j].bio) in r10buf_pool_alloc()
217 bio_uninit(r10_bio->devs[j].bio); in r10buf_pool_alloc()
218 kfree(r10_bio->devs[j].bio); in r10buf_pool_alloc()
219 if (r10_bio->devs[j].repl_bio) in r10buf_pool_alloc()
220 bio_uninit(r10_bio->devs[j].repl_bio); in r10buf_pool_alloc()
221 kfree(r10_bio->devs[j].repl_bio); in r10buf_pool_alloc()
236 for (j = conf->copies; j--; ) { in r10buf_pool_free()
237 struct bio *bio = r10bio->devs[j].bio; in r10buf_pool_free()
246 bio = r10bio->devs[j].repl_bio; in r10buf_pool_free()
263 for (i = 0; i < conf->geo.raid_disks; i++) { in put_all_bios()
264 struct bio **bio = & r10_bio->devs[i].bio; in put_all_bios()
268 bio = &r10_bio->devs[i].repl_bio; in put_all_bios()
269 if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) in put_all_bios()
277 struct r10conf *conf = r10_bio->mddev->private; in free_r10bio()
280 mempool_free(r10_bio, &conf->r10bio_pool); in free_r10bio()
285 struct r10conf *conf = r10_bio->mddev->private; in put_buf()
287 mempool_free(r10_bio, &conf->r10buf_pool); in put_buf()
294 if (wq_has_sleeper(&conf->wait_barrier)) in wake_up_barrier()
295 wake_up(&conf->wait_barrier); in wake_up_barrier()
301 struct mddev *mddev = r10_bio->mddev; in reschedule_retry()
302 struct r10conf *conf = mddev->private; in reschedule_retry()
304 spin_lock_irqsave(&conf->device_lock, flags); in reschedule_retry()
305 list_add(&r10_bio->retry_list, &conf->retry_list); in reschedule_retry()
306 conf->nr_queued ++; in reschedule_retry()
307 spin_unlock_irqrestore(&conf->device_lock, flags); in reschedule_retry()
310 wake_up(&conf->wait_barrier); in reschedule_retry()
312 md_wakeup_thread(mddev->thread); in reschedule_retry()
322 struct bio *bio = r10_bio->master_bio; in raid_end_bio_io()
323 struct r10conf *conf = r10_bio->mddev->private; in raid_end_bio_io()
325 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in raid_end_bio_io()
326 bio->bi_status = BLK_STS_IOERR; in raid_end_bio_io()
339 * Update disk head position estimator based on IRQ completion info.
343 struct r10conf *conf = r10_bio->mddev->private; in update_head_pos()
345 conf->mirrors[r10_bio->devs[slot].devnum].head_position = in update_head_pos()
346 r10_bio->devs[slot].addr + (r10_bio->sectors); in update_head_pos()
358 for (slot = 0; slot < conf->geo.raid_disks; slot++) { in find_bio_disk()
359 if (r10_bio->devs[slot].bio == bio) in find_bio_disk()
361 if (r10_bio->devs[slot].repl_bio == bio) { in find_bio_disk()
373 return r10_bio->devs[slot].devnum; in find_bio_disk()
378 int uptodate = !bio->bi_status; in raid10_end_read_request()
379 struct r10bio *r10_bio = bio->bi_private; in raid10_end_read_request()
382 struct r10conf *conf = r10_bio->mddev->private; in raid10_end_read_request()
384 slot = r10_bio->read_slot; in raid10_end_read_request()
385 rdev = r10_bio->devs[slot].rdev; in raid10_end_read_request()
398 * user-side. So if something waits for IO, then it will in raid10_end_read_request()
401 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_end_read_request()
408 if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), in raid10_end_read_request()
409 rdev->raid_disk)) in raid10_end_read_request()
414 rdev_dec_pending(rdev, conf->mddev); in raid10_end_read_request()
417 * oops, read error - keep the refcount on the rdev in raid10_end_read_request()
420 mdname(conf->mddev), in raid10_end_read_request()
421 rdev->bdev, in raid10_end_read_request()
422 (unsigned long long)r10_bio->sector); in raid10_end_read_request()
423 set_bit(R10BIO_ReadError, &r10_bio->state); in raid10_end_read_request()
430 md_write_end(r10_bio->mddev); in close_write()
435 if (atomic_dec_and_test(&r10_bio->remaining)) { in one_write_done()
436 if (test_bit(R10BIO_WriteError, &r10_bio->state)) in one_write_done()
440 if (test_bit(R10BIO_MadeGood, &r10_bio->state)) in one_write_done()
450 struct r10bio *r10_bio = bio->bi_private; in raid10_end_write_request()
453 struct r10conf *conf = r10_bio->mddev->private; in raid10_end_write_request()
459 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; in raid10_end_write_request()
464 rdev = conf->mirrors[dev].replacement; in raid10_end_write_request()
468 rdev = conf->mirrors[dev].rdev; in raid10_end_write_request()
473 if (bio->bi_status && !discard_error) { in raid10_end_write_request()
478 md_error(rdev->mddev, rdev); in raid10_end_write_request()
480 set_bit(WriteErrorSeen, &rdev->flags); in raid10_end_write_request()
481 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in raid10_end_write_request()
483 &rdev->mddev->recovery); in raid10_end_write_request()
486 if (test_bit(FailFast, &rdev->flags) && in raid10_end_write_request()
487 (bio->bi_opf & MD_FAILFAST)) { in raid10_end_write_request()
488 md_error(rdev->mddev, rdev); in raid10_end_write_request()
495 if (!test_bit(Faulty, &rdev->flags)) in raid10_end_write_request()
496 set_bit(R10BIO_WriteError, &r10_bio->state); in raid10_end_write_request()
499 r10_bio->devs[slot].bio = NULL; in raid10_end_write_request()
511 * user-side. So if something waits for IO, then it will in raid10_end_write_request()
522 * before rdev->recovery_offset, but for simplicity we don't in raid10_end_write_request()
525 if (test_bit(In_sync, &rdev->flags) && in raid10_end_write_request()
526 !test_bit(Faulty, &rdev->flags)) in raid10_end_write_request()
527 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_end_write_request()
531 r10_bio->devs[slot].addr, in raid10_end_write_request()
532 r10_bio->sectors, in raid10_end_write_request()
536 r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; in raid10_end_write_request()
538 r10_bio->devs[slot].bio = IO_MADE_GOOD; in raid10_end_write_request()
540 set_bit(R10BIO_MadeGood, &r10_bio->state); in raid10_end_write_request()
551 rdev_dec_pending(rdev, conf->mddev); in raid10_end_write_request()
591 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; in __raid10_find_phys()
592 last_far_set_start *= geo->far_set_size; in __raid10_find_phys()
594 last_far_set_size = geo->far_set_size; in __raid10_find_phys()
595 last_far_set_size += (geo->raid_disks % geo->far_set_size); in __raid10_find_phys()
598 chunk = r10bio->sector >> geo->chunk_shift; in __raid10_find_phys()
599 sector = r10bio->sector & geo->chunk_mask; in __raid10_find_phys()
601 chunk *= geo->near_copies; in __raid10_find_phys()
603 dev = sector_div(stripe, geo->raid_disks); in __raid10_find_phys()
604 if (geo->far_offset) in __raid10_find_phys()
605 stripe *= geo->far_copies; in __raid10_find_phys()
607 sector += stripe << geo->chunk_shift; in __raid10_find_phys()
610 for (n = 0; n < geo->near_copies; n++) { in __raid10_find_phys()
614 r10bio->devs[slot].devnum = d; in __raid10_find_phys()
615 r10bio->devs[slot].addr = s; in __raid10_find_phys()
618 for (f = 1; f < geo->far_copies; f++) { in __raid10_find_phys()
619 set = d / geo->far_set_size; in __raid10_find_phys()
620 d += geo->near_copies; in __raid10_find_phys()
622 if ((geo->raid_disks % geo->far_set_size) && in __raid10_find_phys()
624 d -= last_far_set_start; in __raid10_find_phys()
628 d %= geo->far_set_size; in __raid10_find_phys()
629 d += geo->far_set_size * set; in __raid10_find_phys()
631 s += geo->stride; in __raid10_find_phys()
632 r10bio->devs[slot].devnum = d; in __raid10_find_phys()
633 r10bio->devs[slot].addr = s; in __raid10_find_phys()
637 if (dev >= geo->raid_disks) { in __raid10_find_phys()
639 sector += (geo->chunk_mask + 1); in __raid10_find_phys()
646 struct geom *geo = &conf->geo; in raid10_find_phys()
648 if (conf->reshape_progress != MaxSector && in raid10_find_phys()
649 ((r10bio->sector >= conf->reshape_progress) != in raid10_find_phys()
650 conf->mddev->reshape_backwards)) { in raid10_find_phys()
651 set_bit(R10BIO_Previous, &r10bio->state); in raid10_find_phys()
652 geo = &conf->prev; in raid10_find_phys()
654 clear_bit(R10BIO_Previous, &r10bio->state); in raid10_find_phys()
662 /* Never use conf->prev as this is only called during resync in raid10_find_virt()
665 struct geom *geo = &conf->geo; in raid10_find_virt()
666 int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; in raid10_find_virt()
667 int far_set_size = geo->far_set_size; in raid10_find_virt()
670 if (geo->raid_disks % geo->far_set_size) { in raid10_find_virt()
671 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; in raid10_find_virt()
672 last_far_set_start *= geo->far_set_size; in raid10_find_virt()
675 far_set_size = geo->far_set_size; in raid10_find_virt()
676 far_set_size += (geo->raid_disks % geo->far_set_size); in raid10_find_virt()
681 offset = sector & geo->chunk_mask; in raid10_find_virt()
682 if (geo->far_offset) { in raid10_find_virt()
683 int fc; in raid10_find_virt() local
684 chunk = sector >> geo->chunk_shift; in raid10_find_virt()
685 fc = sector_div(chunk, geo->far_copies); in raid10_find_virt()
686 dev -= fc * geo->near_copies; in raid10_find_virt()
690 while (sector >= geo->stride) { in raid10_find_virt()
691 sector -= geo->stride; in raid10_find_virt()
692 if (dev < (geo->near_copies + far_set_start)) in raid10_find_virt()
693 dev += far_set_size - geo->near_copies; in raid10_find_virt()
695 dev -= geo->near_copies; in raid10_find_virt()
697 chunk = sector >> geo->chunk_shift; in raid10_find_virt()
699 vchunk = chunk * geo->raid_disks + dev; in raid10_find_virt()
700 sector_div(vchunk, geo->near_copies); in raid10_find_virt()
701 return (vchunk << geo->chunk_shift) + offset; in raid10_find_virt()
706 * be done. There is a per-array 'next expected sequential IO' sector
707 * number - if this matches on the next IO then we use the last disk.
708 * There is also a per-disk 'last know head position' sector that is
710 * completion handlers update this position correctly. If there is no
727 const sector_t this_sector = r10_bio->sector; in read_balance()
729 int sectors = r10_bio->sectors; in read_balance()
737 struct geom *geo = &conf->geo; in read_balance()
741 best_dist_slot = -1; in read_balance()
748 clear_bit(R10BIO_FailFast, &r10_bio->state); in read_balance()
755 if ((conf->mddev->recovery_cp < MaxSector in read_balance()
756 && (this_sector + sectors >= conf->next_resync)) || in read_balance()
757 (mddev_is_clustered(conf->mddev) && in read_balance()
758 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, in read_balance()
762 for (slot = 0; slot < conf->copies ; slot++) { in read_balance()
769 if (r10_bio->devs[slot].bio == IO_BLOCKED) in read_balance()
771 disk = r10_bio->devs[slot].devnum; in read_balance()
772 rdev = rcu_dereference(conf->mirrors[disk].replacement); in read_balance()
773 if (rdev == NULL || test_bit(Faulty, &rdev->flags) || in read_balance()
774 r10_bio->devs[slot].addr + sectors > in read_balance()
775 rdev->recovery_offset) { in read_balance()
782 rdev = rcu_dereference(conf->mirrors[disk].rdev); in read_balance()
785 test_bit(Faulty, &rdev->flags)) in read_balance()
787 if (!test_bit(In_sync, &rdev->flags) && in read_balance()
788 r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) in read_balance()
791 dev_sector = r10_bio->devs[slot].addr; in read_balance()
802 bad_sectors -= (dev_sector - first_bad); in read_balance()
809 first_bad - dev_sector; in read_balance()
826 nonrot = bdev_nonrot(rdev->bdev); in read_balance()
828 pending = atomic_read(&rdev->nr_pending); in read_balance()
837 set_bit(R10BIO_FailFast, &r10_bio->state); in read_balance()
842 if (geo->near_copies > 1 && !pending) in read_balance()
846 else if (geo->far_copies > 1) in read_balance()
847 new_distance = r10_bio->devs[slot].addr; in read_balance()
849 new_distance = abs(r10_bio->devs[slot].addr - in read_balance()
850 conf->mirrors[disk].head_position); in read_balance()
858 if (slot >= conf->copies) { in read_balance()
869 atomic_inc(&rdev->nr_pending); in read_balance()
870 r10_bio->read_slot = slot; in read_balance()
884 spin_lock_irq(&conf->device_lock); in flush_pending_writes()
886 if (conf->pending_bio_list.head) { in flush_pending_writes()
890 bio = bio_list_get(&conf->pending_bio_list); in flush_pending_writes()
891 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
895 * current->state might be TASK_UNINTERRUPTIBLE which will in flush_pending_writes()
899 * is a false-positive. Silence the warning by resetting in flush_pending_writes()
905 raid1_prepare_flush_writes(conf->mddev->bitmap); in flush_pending_writes()
906 wake_up(&conf->wait_barrier); in flush_pending_writes()
909 struct bio *next = bio->bi_next; in flush_pending_writes()
917 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
929 * We choose only to raise the barrier if no-one is waiting for the
944 write_seqlock_irq(&conf->resync_lock); in raise_barrier()
946 if (WARN_ON_ONCE(force && !conf->barrier)) in raise_barrier()
950 wait_event_barrier(conf, force || !conf->nr_waiting); in raise_barrier()
953 WRITE_ONCE(conf->barrier, conf->barrier + 1); in raise_barrier()
956 wait_event_barrier(conf, !atomic_read(&conf->nr_pending) && in raise_barrier()
957 conf->barrier < RESYNC_DEPTH); in raise_barrier()
959 write_sequnlock_irq(&conf->resync_lock); in raise_barrier()
966 write_seqlock_irqsave(&conf->resync_lock, flags); in lower_barrier()
967 WRITE_ONCE(conf->barrier, conf->barrier - 1); in lower_barrier()
968 write_sequnlock_irqrestore(&conf->resync_lock, flags); in lower_barrier()
969 wake_up(&conf->wait_barrier); in lower_barrier()
974 struct bio_list *bio_list = current->bio_list; in stop_waiting_barrier()
978 if (!conf->barrier) in stop_waiting_barrier()
983 * rising completely), and the pre-process bio queue isn't empty, then in stop_waiting_barrier()
987 if (atomic_read(&conf->nr_pending) && bio_list && in stop_waiting_barrier()
992 thread = rcu_dereference_protected(conf->mddev->thread, true); in stop_waiting_barrier()
998 if (thread->tsk == current) { in stop_waiting_barrier()
999 WARN_ON_ONCE(atomic_read(&conf->nr_pending) == 0); in stop_waiting_barrier()
1008 unsigned int seq = read_seqbegin(&conf->resync_lock); in wait_barrier_nolock()
1010 if (READ_ONCE(conf->barrier)) in wait_barrier_nolock()
1013 atomic_inc(&conf->nr_pending); in wait_barrier_nolock()
1014 if (!read_seqretry(&conf->resync_lock, seq)) in wait_barrier_nolock()
1017 if (atomic_dec_and_test(&conf->nr_pending)) in wait_barrier_nolock()
1030 write_seqlock_irq(&conf->resync_lock); in wait_barrier()
1031 if (conf->barrier) { in wait_barrier()
1036 conf->nr_waiting++; in wait_barrier()
1037 raid10_log(conf->mddev, "wait barrier"); in wait_barrier()
1039 conf->nr_waiting--; in wait_barrier()
1041 if (!conf->nr_waiting) in wait_barrier()
1042 wake_up(&conf->wait_barrier); in wait_barrier()
1046 atomic_inc(&conf->nr_pending); in wait_barrier()
1047 write_sequnlock_irq(&conf->resync_lock); in wait_barrier()
1053 if ((atomic_dec_and_test(&conf->nr_pending)) || in allow_barrier()
1054 (conf->array_freeze_pending)) in allow_barrier()
1067 * pending IO requests to complete or be queued for re-try. in freeze_array()
1072 write_seqlock_irq(&conf->resync_lock); in freeze_array()
1073 conf->array_freeze_pending++; in freeze_array()
1074 WRITE_ONCE(conf->barrier, conf->barrier + 1); in freeze_array()
1075 conf->nr_waiting++; in freeze_array()
1076 wait_event_barrier_cmd(conf, atomic_read(&conf->nr_pending) == in freeze_array()
1077 conf->nr_queued + extra, flush_pending_writes(conf)); in freeze_array()
1078 conf->array_freeze_pending--; in freeze_array()
1079 write_sequnlock_irq(&conf->resync_lock); in freeze_array()
1085 write_seqlock_irq(&conf->resync_lock); in unfreeze_array()
1086 WRITE_ONCE(conf->barrier, conf->barrier - 1); in unfreeze_array()
1087 conf->nr_waiting--; in unfreeze_array()
1088 wake_up(&conf->wait_barrier); in unfreeze_array()
1089 write_sequnlock_irq(&conf->resync_lock); in unfreeze_array()
1095 if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || in choose_data_offset()
1096 test_bit(R10BIO_Previous, &r10_bio->state)) in choose_data_offset()
1097 return rdev->data_offset; in choose_data_offset()
1099 return rdev->new_data_offset; in choose_data_offset()
1105 struct mddev *mddev = plug->cb.data; in raid10_unplug()
1106 struct r10conf *conf = mddev->private; in raid10_unplug()
1110 spin_lock_irq(&conf->device_lock); in raid10_unplug()
1111 bio_list_merge(&conf->pending_bio_list, &plug->pending); in raid10_unplug()
1112 spin_unlock_irq(&conf->device_lock); in raid10_unplug()
1114 md_wakeup_thread(mddev->thread); in raid10_unplug()
1119 /* we aren't scheduling, so we can do the write-out directly. */ in raid10_unplug()
1120 bio = bio_list_get(&plug->pending); in raid10_unplug()
1121 raid1_prepare_flush_writes(mddev->bitmap); in raid10_unplug()
1125 struct bio *next = bio->bi_next; in raid10_unplug()
1144 if (!wait_barrier(conf, bio->bi_opf & REQ_NOWAIT)) { in regular_request_wait()
1148 while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in regular_request_wait()
1149 bio->bi_iter.bi_sector < conf->reshape_progress && in regular_request_wait()
1150 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { in regular_request_wait()
1152 if (bio->bi_opf & REQ_NOWAIT) { in regular_request_wait()
1156 raid10_log(conf->mddev, "wait reshape"); in regular_request_wait()
1157 wait_event(conf->wait_barrier, in regular_request_wait()
1158 conf->reshape_progress <= bio->bi_iter.bi_sector || in regular_request_wait()
1159 conf->reshape_progress >= bio->bi_iter.bi_sector + in regular_request_wait()
1169 struct r10conf *conf = mddev->private; in raid10_read_request()
1172 const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; in raid10_read_request()
1176 int slot = r10_bio->read_slot; in raid10_read_request()
1180 if (slot >= 0 && r10_bio->devs[slot].rdev) { in raid10_read_request()
1196 disk = r10_bio->devs[slot].devnum; in raid10_read_request()
1197 err_rdev = rcu_dereference(conf->mirrors[disk].rdev); in raid10_read_request()
1199 snprintf(b, sizeof(b), "%pg", err_rdev->bdev); in raid10_read_request()
1203 err_rdev = r10_bio->devs[slot].rdev; in raid10_read_request()
1208 if (!regular_request_wait(mddev, conf, bio, r10_bio->sectors)) in raid10_read_request()
1215 (unsigned long long)r10_bio->sector); in raid10_read_request()
1223 rdev->bdev, in raid10_read_request()
1224 (unsigned long long)r10_bio->sector); in raid10_read_request()
1227 gfp, &conf->bio_split); in raid10_read_request()
1233 r10_bio->master_bio = bio; in raid10_read_request()
1234 r10_bio->sectors = max_sectors; in raid10_read_request()
1236 slot = r10_bio->read_slot; in raid10_read_request()
1240 r10_bio->master_bio = bio; in raid10_read_request()
1242 read_bio = bio_alloc_clone(rdev->bdev, bio, gfp, &mddev->bio_set); in raid10_read_request()
1244 r10_bio->devs[slot].bio = read_bio; in raid10_read_request()
1245 r10_bio->devs[slot].rdev = rdev; in raid10_read_request()
1247 read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + in raid10_read_request()
1249 read_bio->bi_end_io = raid10_end_read_request; in raid10_read_request()
1250 read_bio->bi_opf = op | do_sync; in raid10_read_request()
1251 if (test_bit(FailFast, &rdev->flags) && in raid10_read_request()
1252 test_bit(R10BIO_FailFast, &r10_bio->state)) in raid10_read_request()
1253 read_bio->bi_opf |= MD_FAILFAST; in raid10_read_request()
1254 read_bio->bi_private = r10_bio; in raid10_read_request()
1256 if (mddev->gendisk) in raid10_read_request()
1257 trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), in raid10_read_request()
1258 r10_bio->sector); in raid10_read_request()
1268 const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; in raid10_write_one_disk()
1269 const blk_opf_t do_fua = bio->bi_opf & REQ_FUA; in raid10_write_one_disk()
1271 struct r10conf *conf = mddev->private; in raid10_write_one_disk()
1273 int devnum = r10_bio->devs[n_copy].devnum; in raid10_write_one_disk()
1277 rdev = conf->mirrors[devnum].replacement; in raid10_write_one_disk()
1281 rdev = conf->mirrors[devnum].rdev; in raid10_write_one_disk()
1284 rdev = conf->mirrors[devnum].rdev; in raid10_write_one_disk()
1286 mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, &mddev->bio_set); in raid10_write_one_disk()
1288 r10_bio->devs[n_copy].repl_bio = mbio; in raid10_write_one_disk()
1290 r10_bio->devs[n_copy].bio = mbio; in raid10_write_one_disk()
1292 mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + in raid10_write_one_disk()
1294 mbio->bi_end_io = raid10_end_write_request; in raid10_write_one_disk()
1295 mbio->bi_opf = op | do_sync | do_fua; in raid10_write_one_disk()
1297 &conf->mirrors[devnum].rdev->flags) in raid10_write_one_disk()
1299 mbio->bi_opf |= MD_FAILFAST; in raid10_write_one_disk()
1300 mbio->bi_private = r10_bio; in raid10_write_one_disk()
1302 if (conf->mddev->gendisk) in raid10_write_one_disk()
1303 trace_block_bio_remap(mbio, disk_devt(conf->mddev->gendisk), in raid10_write_one_disk()
1304 r10_bio->sector); in raid10_write_one_disk()
1306 mbio->bi_bdev = (void *)rdev; in raid10_write_one_disk()
1308 atomic_inc(&r10_bio->remaining); in raid10_write_one_disk()
1310 if (!raid1_add_bio_to_plug(mddev, mbio, raid10_unplug, conf->copies)) { in raid10_write_one_disk()
1311 spin_lock_irqsave(&conf->device_lock, flags); in raid10_write_one_disk()
1312 bio_list_add(&conf->pending_bio_list, mbio); in raid10_write_one_disk()
1313 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_write_one_disk()
1314 md_wakeup_thread(mddev->thread); in raid10_write_one_disk()
1323 rrdev = rcu_dereference(mirror->replacement); in dereference_rdev_and_rrdev()
1329 rdev = rcu_dereference(mirror->rdev); in dereference_rdev_and_rrdev()
1340 struct r10conf *conf = mddev->private; in wait_blocked_dev()
1346 for (i = 0; i < conf->copies; i++) { in wait_blocked_dev()
1349 rdev = dereference_rdev_and_rrdev(&conf->mirrors[i], &rrdev); in wait_blocked_dev()
1350 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { in wait_blocked_dev()
1351 atomic_inc(&rdev->nr_pending); in wait_blocked_dev()
1355 if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { in wait_blocked_dev()
1356 atomic_inc(&rrdev->nr_pending); in wait_blocked_dev()
1361 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { in wait_blocked_dev()
1363 sector_t dev_sector = r10_bio->devs[i].addr; in wait_blocked_dev()
1371 if (!r10_bio->sectors) in wait_blocked_dev()
1374 is_bad = is_badblock(rdev, dev_sector, r10_bio->sectors, in wait_blocked_dev()
1381 atomic_inc(&rdev->nr_pending); in wait_blocked_dev()
1382 set_bit(BlockedBadBlocks, &rdev->flags); in wait_blocked_dev()
1393 raid10_log(conf->mddev, "%s wait rdev %d blocked", in wait_blocked_dev()
1394 __func__, blocked_rdev->raid_disk); in wait_blocked_dev()
1404 struct r10conf *conf = mddev->private; in raid10_write_request()
1410 md_cluster_ops->area_resyncing(mddev, WRITE, in raid10_write_request()
1411 bio->bi_iter.bi_sector, in raid10_write_request()
1415 if (bio->bi_opf & REQ_NOWAIT) { in raid10_write_request()
1420 prepare_to_wait(&conf->wait_barrier, in raid10_write_request()
1422 if (!md_cluster_ops->area_resyncing(mddev, WRITE, in raid10_write_request()
1423 bio->bi_iter.bi_sector, bio_end_sector(bio))) in raid10_write_request()
1427 finish_wait(&conf->wait_barrier, &w); in raid10_write_request()
1430 sectors = r10_bio->sectors; in raid10_write_request()
1433 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in raid10_write_request()
1434 (mddev->reshape_backwards in raid10_write_request()
1435 ? (bio->bi_iter.bi_sector < conf->reshape_safe && in raid10_write_request()
1436 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) in raid10_write_request()
1437 : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && in raid10_write_request()
1438 bio->bi_iter.bi_sector < conf->reshape_progress))) { in raid10_write_request()
1439 /* Need to update reshape_position in metadata */ in raid10_write_request()
1440 mddev->reshape_position = conf->reshape_progress; in raid10_write_request()
1441 set_mask_bits(&mddev->sb_flags, 0, in raid10_write_request()
1443 md_wakeup_thread(mddev->thread); in raid10_write_request()
1444 if (bio->bi_opf & REQ_NOWAIT) { in raid10_write_request()
1449 raid10_log(conf->mddev, "wait reshape metadata"); in raid10_write_request()
1450 wait_event(mddev->sb_wait, in raid10_write_request()
1451 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); in raid10_write_request()
1453 conf->reshape_safe = mddev->reshape_position; in raid10_write_request()
1466 r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ in raid10_write_request()
1472 max_sectors = r10_bio->sectors; in raid10_write_request()
1474 for (i = 0; i < conf->copies; i++) { in raid10_write_request()
1475 int d = r10_bio->devs[i].devnum; in raid10_write_request()
1478 rdev = dereference_rdev_and_rrdev(&conf->mirrors[d], &rrdev); in raid10_write_request()
1479 if (rdev && (test_bit(Faulty, &rdev->flags))) in raid10_write_request()
1481 if (rrdev && (test_bit(Faulty, &rrdev->flags))) in raid10_write_request()
1484 r10_bio->devs[i].bio = NULL; in raid10_write_request()
1485 r10_bio->devs[i].repl_bio = NULL; in raid10_write_request()
1489 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { in raid10_write_request()
1491 sector_t dev_sector = r10_bio->devs[i].addr; in raid10_write_request()
1499 bad_sectors -= (dev_sector - first_bad); in raid10_write_request()
1508 int good_sectors = first_bad - dev_sector; in raid10_write_request()
1514 r10_bio->devs[i].bio = bio; in raid10_write_request()
1515 atomic_inc(&rdev->nr_pending); in raid10_write_request()
1518 r10_bio->devs[i].repl_bio = bio; in raid10_write_request()
1519 atomic_inc(&rrdev->nr_pending); in raid10_write_request()
1524 if (max_sectors < r10_bio->sectors) in raid10_write_request()
1525 r10_bio->sectors = max_sectors; in raid10_write_request()
1527 if (r10_bio->sectors < bio_sectors(bio)) { in raid10_write_request()
1528 struct bio *split = bio_split(bio, r10_bio->sectors, in raid10_write_request()
1529 GFP_NOIO, &conf->bio_split); in raid10_write_request()
1535 r10_bio->master_bio = bio; in raid10_write_request()
1539 r10_bio->master_bio = bio; in raid10_write_request()
1540 atomic_set(&r10_bio->remaining, 1); in raid10_write_request()
1542 for (i = 0; i < conf->copies; i++) { in raid10_write_request()
1543 if (r10_bio->devs[i].bio) in raid10_write_request()
1545 if (r10_bio->devs[i].repl_bio) in raid10_write_request()
1553 struct r10conf *conf = mddev->private; in __make_request()
1556 r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); in __make_request()
1558 r10_bio->master_bio = bio; in __make_request()
1559 r10_bio->sectors = sectors; in __make_request()
1561 r10_bio->mddev = mddev; in __make_request()
1562 r10_bio->sector = bio->bi_iter.bi_sector; in __make_request()
1563 r10_bio->state = 0; in __make_request()
1564 r10_bio->read_slot = -1; in __make_request()
1565 memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * in __make_request()
1566 conf->geo.raid_disks); in __make_request()
1576 struct r10conf *conf = r10bio->mddev->private; in raid_end_discard_bio()
1579 while (atomic_dec_and_test(&r10bio->remaining)) { in raid_end_discard_bio()
1583 if (!test_bit(R10BIO_Discard, &r10bio->state)) { in raid_end_discard_bio()
1584 first_r10bio = (struct r10bio *)r10bio->master_bio; in raid_end_discard_bio()
1588 md_write_end(r10bio->mddev); in raid_end_discard_bio()
1589 bio_endio(r10bio->master_bio); in raid_end_discard_bio()
1598 struct r10bio *r10_bio = bio->bi_private; in raid10_end_discard_request()
1599 struct r10conf *conf = r10_bio->mddev->private; in raid10_end_discard_request()
1607 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in raid10_end_discard_request()
1608 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_end_discard_request()
1612 rdev = conf->mirrors[dev].replacement; in raid10_end_discard_request()
1620 rdev = conf->mirrors[dev].rdev; in raid10_end_discard_request()
1624 rdev_dec_pending(rdev, conf->mddev); in raid10_end_discard_request()
1635 struct r10conf *conf = mddev->private; in raid10_handle_discard()
1636 struct geom *geo = &conf->geo; in raid10_handle_discard()
1637 int far_copies = geo->far_copies; in raid10_handle_discard()
1654 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_handle_discard()
1655 return -EAGAIN; in raid10_handle_discard()
1657 if (WARN_ON_ONCE(bio->bi_opf & REQ_NOWAIT)) { in raid10_handle_discard()
1667 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_handle_discard()
1670 if (geo->near_copies) in raid10_handle_discard()
1671 stripe_data_disks = geo->raid_disks / geo->near_copies + in raid10_handle_discard()
1672 geo->raid_disks % geo->near_copies; in raid10_handle_discard()
1674 stripe_data_disks = geo->raid_disks; in raid10_handle_discard()
1676 stripe_size = stripe_data_disks << geo->chunk_shift; in raid10_handle_discard()
1678 bio_start = bio->bi_iter.bi_sector; in raid10_handle_discard()
1697 split_size = stripe_size - remainder; in raid10_handle_discard()
1698 split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); in raid10_handle_discard()
1707 split_size = bio_sectors(bio) - remainder; in raid10_handle_discard()
1708 split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); in raid10_handle_discard()
1717 bio_start = bio->bi_iter.bi_sector; in raid10_handle_discard()
1725 chunk = bio_start >> geo->chunk_shift; in raid10_handle_discard()
1726 chunk *= geo->near_copies; in raid10_handle_discard()
1728 start_disk_index = sector_div(first_stripe_index, geo->raid_disks); in raid10_handle_discard()
1729 if (geo->far_offset) in raid10_handle_discard()
1730 first_stripe_index *= geo->far_copies; in raid10_handle_discard()
1731 start_disk_offset = (bio_start & geo->chunk_mask) + in raid10_handle_discard()
1732 (first_stripe_index << geo->chunk_shift); in raid10_handle_discard()
1734 chunk = bio_end >> geo->chunk_shift; in raid10_handle_discard()
1735 chunk *= geo->near_copies; in raid10_handle_discard()
1737 end_disk_index = sector_div(last_stripe_index, geo->raid_disks); in raid10_handle_discard()
1738 if (geo->far_offset) in raid10_handle_discard()
1739 last_stripe_index *= geo->far_copies; in raid10_handle_discard()
1740 end_disk_offset = (bio_end & geo->chunk_mask) + in raid10_handle_discard()
1741 (last_stripe_index << geo->chunk_shift); in raid10_handle_discard()
1744 r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); in raid10_handle_discard()
1745 r10_bio->mddev = mddev; in raid10_handle_discard()
1746 r10_bio->state = 0; in raid10_handle_discard()
1747 r10_bio->sectors = 0; in raid10_handle_discard()
1748 memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * geo->raid_disks); in raid10_handle_discard()
1753 * Inspired by raid10_sync_request, we can use the first r10bio->master_bio in raid10_handle_discard()
1754 * to record the discard bio. Other r10bio->master_bio record the first in raid10_handle_discard()
1759 r10_bio->master_bio = bio; in raid10_handle_discard()
1760 set_bit(R10BIO_Discard, &r10_bio->state); in raid10_handle_discard()
1764 r10_bio->master_bio = (struct bio *)first_r10bio; in raid10_handle_discard()
1772 for (disk = 0; disk < geo->raid_disks; disk++) { in raid10_handle_discard()
1775 rdev = dereference_rdev_and_rrdev(&conf->mirrors[disk], &rrdev); in raid10_handle_discard()
1776 r10_bio->devs[disk].bio = NULL; in raid10_handle_discard()
1777 r10_bio->devs[disk].repl_bio = NULL; in raid10_handle_discard()
1779 if (rdev && (test_bit(Faulty, &rdev->flags))) in raid10_handle_discard()
1781 if (rrdev && (test_bit(Faulty, &rrdev->flags))) in raid10_handle_discard()
1787 r10_bio->devs[disk].bio = bio; in raid10_handle_discard()
1788 atomic_inc(&rdev->nr_pending); in raid10_handle_discard()
1791 r10_bio->devs[disk].repl_bio = bio; in raid10_handle_discard()
1792 atomic_inc(&rrdev->nr_pending); in raid10_handle_discard()
1797 atomic_set(&r10_bio->remaining, 1); in raid10_handle_discard()
1798 for (disk = 0; disk < geo->raid_disks; disk++) { in raid10_handle_discard()
1815 dev_start = (first_stripe_index + 1) * mddev->chunk_sectors; in raid10_handle_discard()
1817 dev_start = first_stripe_index * mddev->chunk_sectors; in raid10_handle_discard()
1822 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; in raid10_handle_discard()
1824 dev_end = last_stripe_index * mddev->chunk_sectors; in raid10_handle_discard()
1832 * add rdev->nr_pending in the first loop. in raid10_handle_discard()
1834 if (r10_bio->devs[disk].bio) { in raid10_handle_discard()
1835 struct md_rdev *rdev = conf->mirrors[disk].rdev; in raid10_handle_discard()
1836 mbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, in raid10_handle_discard()
1837 &mddev->bio_set); in raid10_handle_discard()
1838 mbio->bi_end_io = raid10_end_discard_request; in raid10_handle_discard()
1839 mbio->bi_private = r10_bio; in raid10_handle_discard()
1840 r10_bio->devs[disk].bio = mbio; in raid10_handle_discard()
1841 r10_bio->devs[disk].devnum = disk; in raid10_handle_discard()
1842 atomic_inc(&r10_bio->remaining); in raid10_handle_discard()
1845 dev_end - dev_start); in raid10_handle_discard()
1848 if (r10_bio->devs[disk].repl_bio) { in raid10_handle_discard()
1849 struct md_rdev *rrdev = conf->mirrors[disk].replacement; in raid10_handle_discard()
1850 rbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, in raid10_handle_discard()
1851 &mddev->bio_set); in raid10_handle_discard()
1852 rbio->bi_end_io = raid10_end_discard_request; in raid10_handle_discard()
1853 rbio->bi_private = r10_bio; in raid10_handle_discard()
1854 r10_bio->devs[disk].repl_bio = rbio; in raid10_handle_discard()
1855 r10_bio->devs[disk].devnum = disk; in raid10_handle_discard()
1856 atomic_inc(&r10_bio->remaining); in raid10_handle_discard()
1859 dev_end - dev_start); in raid10_handle_discard()
1864 if (!geo->far_offset && --far_copies) { in raid10_handle_discard()
1865 first_stripe_index += geo->stride >> geo->chunk_shift; in raid10_handle_discard()
1866 start_disk_offset += geo->stride; in raid10_handle_discard()
1867 last_stripe_index += geo->stride >> geo->chunk_shift; in raid10_handle_discard()
1868 end_disk_offset += geo->stride; in raid10_handle_discard()
1869 atomic_inc(&first_r10bio->remaining); in raid10_handle_discard()
1880 return -EAGAIN; in raid10_handle_discard()
1885 struct r10conf *conf = mddev->private; in raid10_make_request()
1886 sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); in raid10_make_request()
1890 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid10_make_request()
1905 if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + in raid10_make_request()
1907 && (conf->geo.near_copies < conf->geo.raid_disks in raid10_make_request()
1908 || conf->prev.near_copies < in raid10_make_request()
1909 conf->prev.raid_disks))) in raid10_make_request()
1910 sectors = chunk_sects - in raid10_make_request()
1911 (bio->bi_iter.bi_sector & in raid10_make_request()
1912 (chunk_sects - 1)); in raid10_make_request()
1922 struct r10conf *conf = mddev->private; in raid10_status()
1925 if (conf->geo.near_copies < conf->geo.raid_disks) in raid10_status()
1926 seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); in raid10_status()
1927 if (conf->geo.near_copies > 1) in raid10_status()
1928 seq_printf(seq, " %d near-copies", conf->geo.near_copies); in raid10_status()
1929 if (conf->geo.far_copies > 1) { in raid10_status()
1930 if (conf->geo.far_offset) in raid10_status()
1931 seq_printf(seq, " %d offset-copies", conf->geo.far_copies); in raid10_status()
1933 seq_printf(seq, " %d far-copies", conf->geo.far_copies); in raid10_status()
1934 if (conf->geo.far_set_size != conf->geo.raid_disks) in raid10_status()
1935 seq_printf(seq, " %d devices per set", conf->geo.far_set_size); in raid10_status()
1937 seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, in raid10_status()
1938 conf->geo.raid_disks - mddev->degraded); in raid10_status()
1940 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_status()
1941 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid10_status()
1942 seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid10_status()
1959 disks = conf->prev.raid_disks; in _enough()
1960 ncopies = conf->prev.near_copies; in _enough()
1962 disks = conf->geo.raid_disks; in _enough()
1963 ncopies = conf->geo.near_copies; in _enough()
1968 int n = conf->copies; in _enough()
1971 while (n--) { in _enough()
1974 (rdev = rcu_dereference(conf->mirrors[this].rdev)) && in _enough()
1975 test_bit(In_sync, &rdev->flags)) in _enough()
1993 * This is ensured if ->reconfig_mutex or ->device_lock in enough()
2001 * raid10_error() - RAID10 error handler.
2007 * - &MD_BROKEN flag is set in &mddev->flags.
2009 * - recovery is interrupted.
2010 * - &mddev->degraded is bumped.
2013 * &mddev->fail_last_dev is off.
2017 struct r10conf *conf = mddev->private; in raid10_error()
2020 spin_lock_irqsave(&conf->device_lock, flags); in raid10_error()
2022 if (test_bit(In_sync, &rdev->flags) && !enough(conf, rdev->raid_disk)) { in raid10_error()
2023 set_bit(MD_BROKEN, &mddev->flags); in raid10_error()
2025 if (!mddev->fail_last_dev) { in raid10_error()
2026 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_error()
2030 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid10_error()
2031 mddev->degraded++; in raid10_error()
2033 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid10_error()
2034 set_bit(Blocked, &rdev->flags); in raid10_error()
2035 set_bit(Faulty, &rdev->flags); in raid10_error()
2036 set_mask_bits(&mddev->sb_flags, 0, in raid10_error()
2038 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_error()
2041 mdname(mddev), rdev->bdev, in raid10_error()
2042 mdname(mddev), conf->geo.raid_disks - mddev->degraded); in raid10_error()
2055 pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, in print_conf()
2056 conf->geo.raid_disks); in print_conf()
2058 /* This is only called with ->reconfix_mutex held, so in print_conf()
2060 for (i = 0; i < conf->geo.raid_disks; i++) { in print_conf()
2061 rdev = conf->mirrors[i].rdev; in print_conf()
2064 i, !test_bit(In_sync, &rdev->flags), in print_conf()
2065 !test_bit(Faulty, &rdev->flags), in print_conf()
2066 rdev->bdev); in print_conf()
2075 mempool_exit(&conf->r10buf_pool); in close_sync()
2081 struct r10conf *conf = mddev->private; in raid10_spare_active()
2087 * Find all non-in_sync disks within the RAID10 configuration in raid10_spare_active()
2090 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_spare_active()
2091 tmp = conf->mirrors + i; in raid10_spare_active()
2092 if (tmp->replacement in raid10_spare_active()
2093 && tmp->replacement->recovery_offset == MaxSector in raid10_spare_active()
2094 && !test_bit(Faulty, &tmp->replacement->flags) in raid10_spare_active()
2095 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { in raid10_spare_active()
2097 if (!tmp->rdev in raid10_spare_active()
2098 || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) in raid10_spare_active()
2100 if (tmp->rdev) { in raid10_spare_active()
2103 * and never re-added. in raid10_spare_active()
2105 set_bit(Faulty, &tmp->rdev->flags); in raid10_spare_active()
2107 tmp->rdev->sysfs_state); in raid10_spare_active()
2109 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); in raid10_spare_active()
2110 } else if (tmp->rdev in raid10_spare_active()
2111 && tmp->rdev->recovery_offset == MaxSector in raid10_spare_active()
2112 && !test_bit(Faulty, &tmp->rdev->flags) in raid10_spare_active()
2113 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { in raid10_spare_active()
2115 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); in raid10_spare_active()
2118 spin_lock_irqsave(&conf->device_lock, flags); in raid10_spare_active()
2119 mddev->degraded -= count; in raid10_spare_active()
2120 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_spare_active()
2128 struct r10conf *conf = mddev->private; in raid10_add_disk()
2129 int err = -EEXIST; in raid10_add_disk()
2130 int mirror, repl_slot = -1; in raid10_add_disk()
2132 int last = conf->geo.raid_disks - 1; in raid10_add_disk()
2135 if (mddev->recovery_cp < MaxSector) in raid10_add_disk()
2136 /* only hot-add to in-sync arrays, as recovery is in raid10_add_disk()
2139 return -EBUSY; in raid10_add_disk()
2140 if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) in raid10_add_disk()
2141 return -EINVAL; in raid10_add_disk()
2144 return -ENXIO; in raid10_add_disk()
2146 if (rdev->raid_disk >= 0) in raid10_add_disk()
2147 first = last = rdev->raid_disk; in raid10_add_disk()
2149 if (rdev->saved_raid_disk >= first && in raid10_add_disk()
2150 rdev->saved_raid_disk < conf->geo.raid_disks && in raid10_add_disk()
2151 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid10_add_disk()
2152 mirror = rdev->saved_raid_disk; in raid10_add_disk()
2156 p = &conf->mirrors[mirror]; in raid10_add_disk()
2157 if (p->recovery_disabled == mddev->recovery_disabled) in raid10_add_disk()
2159 if (p->rdev) { in raid10_add_disk()
2160 if (test_bit(WantReplacement, &p->rdev->flags) && in raid10_add_disk()
2161 p->replacement == NULL && repl_slot < 0) in raid10_add_disk()
2166 if (mddev->gendisk) in raid10_add_disk()
2167 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid10_add_disk()
2168 rdev->data_offset << 9); in raid10_add_disk()
2170 p->head_position = 0; in raid10_add_disk()
2171 p->recovery_disabled = mddev->recovery_disabled - 1; in raid10_add_disk()
2172 rdev->raid_disk = mirror; in raid10_add_disk()
2174 if (rdev->saved_raid_disk != mirror) in raid10_add_disk()
2175 conf->fullsync = 1; in raid10_add_disk()
2176 rcu_assign_pointer(p->rdev, rdev); in raid10_add_disk()
2181 p = &conf->mirrors[repl_slot]; in raid10_add_disk()
2182 clear_bit(In_sync, &rdev->flags); in raid10_add_disk()
2183 set_bit(Replacement, &rdev->flags); in raid10_add_disk()
2184 rdev->raid_disk = repl_slot; in raid10_add_disk()
2186 if (mddev->gendisk) in raid10_add_disk()
2187 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid10_add_disk()
2188 rdev->data_offset << 9); in raid10_add_disk()
2189 conf->fullsync = 1; in raid10_add_disk()
2190 rcu_assign_pointer(p->replacement, rdev); in raid10_add_disk()
2199 struct r10conf *conf = mddev->private; in raid10_remove_disk()
2201 int number = rdev->raid_disk; in raid10_remove_disk()
2206 if (unlikely(number >= mddev->raid_disks)) in raid10_remove_disk()
2208 p = conf->mirrors + number; in raid10_remove_disk()
2209 if (rdev == p->rdev) in raid10_remove_disk()
2210 rdevp = &p->rdev; in raid10_remove_disk()
2211 else if (rdev == p->replacement) in raid10_remove_disk()
2212 rdevp = &p->replacement; in raid10_remove_disk()
2216 if (test_bit(In_sync, &rdev->flags) || in raid10_remove_disk()
2217 atomic_read(&rdev->nr_pending)) { in raid10_remove_disk()
2218 err = -EBUSY; in raid10_remove_disk()
2221 /* Only remove non-faulty devices if recovery in raid10_remove_disk()
2224 if (!test_bit(Faulty, &rdev->flags) && in raid10_remove_disk()
2225 mddev->recovery_disabled != p->recovery_disabled && in raid10_remove_disk()
2226 (!p->replacement || p->replacement == rdev) && in raid10_remove_disk()
2227 number < conf->geo.raid_disks && in raid10_remove_disk()
2228 enough(conf, -1)) { in raid10_remove_disk()
2229 err = -EBUSY; in raid10_remove_disk()
2233 if (!test_bit(RemoveSynchronized, &rdev->flags)) { in raid10_remove_disk()
2235 if (atomic_read(&rdev->nr_pending)) { in raid10_remove_disk()
2237 err = -EBUSY; in raid10_remove_disk()
2242 if (p->replacement) { in raid10_remove_disk()
2244 p->rdev = p->replacement; in raid10_remove_disk()
2245 clear_bit(Replacement, &p->replacement->flags); in raid10_remove_disk()
2247 * but will never see neither -- if they are careful. in raid10_remove_disk()
2249 p->replacement = NULL; in raid10_remove_disk()
2252 clear_bit(WantReplacement, &rdev->flags); in raid10_remove_disk()
2263 struct r10conf *conf = r10_bio->mddev->private; in __end_sync_read()
2265 if (!bio->bi_status) in __end_sync_read()
2266 set_bit(R10BIO_Uptodate, &r10_bio->state); in __end_sync_read()
2271 atomic_add(r10_bio->sectors, in __end_sync_read()
2272 &conf->mirrors[d].rdev->corrected_errors); in __end_sync_read()
2277 rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); in __end_sync_read()
2278 if (test_bit(R10BIO_IsRecover, &r10_bio->state) || in __end_sync_read()
2279 atomic_dec_and_test(&r10_bio->remaining)) { in __end_sync_read()
2290 struct r10conf *conf = r10_bio->mddev->private; in end_sync_read()
2299 struct r10bio *r10_bio = bio->bi_private; in end_reshape_read()
2301 __end_sync_read(r10_bio, bio, r10_bio->read_slot); in end_reshape_read()
2306 struct mddev *mddev = r10_bio->mddev; in end_sync_request()
2308 while (atomic_dec_and_test(&r10_bio->remaining)) { in end_sync_request()
2309 if (r10_bio->master_bio == NULL) { in end_sync_request()
2311 sector_t s = r10_bio->sectors; in end_sync_request()
2312 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in end_sync_request()
2313 test_bit(R10BIO_WriteError, &r10_bio->state)) in end_sync_request()
2320 struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; in end_sync_request()
2321 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in end_sync_request()
2322 test_bit(R10BIO_WriteError, &r10_bio->state)) in end_sync_request()
2334 struct mddev *mddev = r10_bio->mddev; in end_sync_write()
2335 struct r10conf *conf = mddev->private; in end_sync_write()
2345 rdev = conf->mirrors[d].replacement; in end_sync_write()
2347 rdev = conf->mirrors[d].rdev; in end_sync_write()
2349 if (bio->bi_status) { in end_sync_write()
2353 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
2354 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
2356 &rdev->mddev->recovery); in end_sync_write()
2357 set_bit(R10BIO_WriteError, &r10_bio->state); in end_sync_write()
2360 r10_bio->devs[slot].addr, in end_sync_write()
2361 r10_bio->sectors, in end_sync_write()
2363 set_bit(R10BIO_MadeGood, &r10_bio->state); in end_sync_write()
2383 * We check if all blocks are in-sync and only write to blocks that
2388 struct r10conf *conf = mddev->private; in sync_request_write()
2394 atomic_set(&r10_bio->remaining, 1); in sync_request_write()
2397 for (i=0; i<conf->copies; i++) in sync_request_write()
2398 if (!r10_bio->devs[i].bio->bi_status) in sync_request_write()
2401 if (i == conf->copies) in sync_request_write()
2405 fbio = r10_bio->devs[i].bio; in sync_request_write()
2406 fbio->bi_iter.bi_size = r10_bio->sectors << 9; in sync_request_write()
2407 fbio->bi_iter.bi_idx = 0; in sync_request_write()
2408 fpages = get_resync_pages(fbio)->pages; in sync_request_write()
2410 vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); in sync_request_write()
2412 for (i=0 ; i < conf->copies ; i++) { in sync_request_write()
2417 tbio = r10_bio->devs[i].bio; in sync_request_write()
2419 if (tbio->bi_end_io != end_sync_read) in sync_request_write()
2424 tpages = get_resync_pages(tbio)->pages; in sync_request_write()
2425 d = r10_bio->devs[i].devnum; in sync_request_write()
2426 rdev = conf->mirrors[d].rdev; in sync_request_write()
2427 if (!r10_bio->devs[i].bio->bi_status) { in sync_request_write()
2432 int sectors = r10_bio->sectors; in sync_request_write()
2441 sectors -= len/512; in sync_request_write()
2445 atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); in sync_request_write()
2446 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) in sync_request_write()
2449 } else if (test_bit(FailFast, &rdev->flags)) { in sync_request_write()
2451 md_error(rdev->mddev, rdev); in sync_request_write()
2456 * First we need to fixup bv_offset, bv_len and in sync_request_write()
2460 bio_reset(tbio, conf->mirrors[d].rdev->bdev, REQ_OP_WRITE); in sync_request_write()
2462 md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size); in sync_request_write()
2464 rp->raid_bio = r10_bio; in sync_request_write()
2465 tbio->bi_private = rp; in sync_request_write()
2466 tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; in sync_request_write()
2467 tbio->bi_end_io = end_sync_write; in sync_request_write()
2471 atomic_inc(&conf->mirrors[d].rdev->nr_pending); in sync_request_write()
2472 atomic_inc(&r10_bio->remaining); in sync_request_write()
2473 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); in sync_request_write()
2475 if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) in sync_request_write()
2476 tbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2477 tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; in sync_request_write()
2484 for (i = 0; i < conf->copies; i++) { in sync_request_write()
2487 tbio = r10_bio->devs[i].repl_bio; in sync_request_write()
2488 if (!tbio || !tbio->bi_end_io) in sync_request_write()
2490 if (r10_bio->devs[i].bio->bi_end_io != end_sync_write in sync_request_write()
2491 && r10_bio->devs[i].bio != fbio) in sync_request_write()
2493 d = r10_bio->devs[i].devnum; in sync_request_write()
2494 atomic_inc(&r10_bio->remaining); in sync_request_write()
2495 md_sync_acct(conf->mirrors[d].replacement->bdev, in sync_request_write()
2501 if (atomic_dec_and_test(&r10_bio->remaining)) { in sync_request_write()
2502 md_done_sync(mddev, r10_bio->sectors, 1); in sync_request_write()
2510 * We recover all non-is_sync drives by finding the virtual address of
2512 * There is a separate r10_bio for each non-in_sync drive.
2520 * We repeat the read in smaller page-sized sections. in fix_recovery_read_error()
2526 struct mddev *mddev = r10_bio->mddev; in fix_recovery_read_error()
2527 struct r10conf *conf = mddev->private; in fix_recovery_read_error()
2528 struct bio *bio = r10_bio->devs[0].bio; in fix_recovery_read_error()
2530 int sectors = r10_bio->sectors; in fix_recovery_read_error()
2532 int dr = r10_bio->devs[0].devnum; in fix_recovery_read_error()
2533 int dw = r10_bio->devs[1].devnum; in fix_recovery_read_error()
2534 struct page **pages = get_resync_pages(bio)->pages; in fix_recovery_read_error()
2545 rdev = conf->mirrors[dr].rdev; in fix_recovery_read_error()
2546 addr = r10_bio->devs[0].addr + sect, in fix_recovery_read_error()
2553 rdev = conf->mirrors[dw].rdev; in fix_recovery_read_error()
2554 addr = r10_bio->devs[1].addr + sect; in fix_recovery_read_error()
2561 set_bit(WriteErrorSeen, &rdev->flags); in fix_recovery_read_error()
2563 &rdev->flags)) in fix_recovery_read_error()
2565 &rdev->mddev->recovery); in fix_recovery_read_error()
2569 /* We don't worry if we cannot set a bad block - in fix_recovery_read_error()
2575 if (rdev != conf->mirrors[dw].rdev) { in fix_recovery_read_error()
2577 struct md_rdev *rdev2 = conf->mirrors[dw].rdev; in fix_recovery_read_error()
2578 addr = r10_bio->devs[1].addr + sect; in fix_recovery_read_error()
2585 conf->mirrors[dw].recovery_disabled in fix_recovery_read_error()
2586 = mddev->recovery_disabled; in fix_recovery_read_error()
2588 &mddev->recovery); in fix_recovery_read_error()
2594 sectors -= s; in fix_recovery_read_error()
2602 struct r10conf *conf = mddev->private; in recovery_request_write()
2604 struct bio *wbio = r10_bio->devs[1].bio; in recovery_request_write()
2605 struct bio *wbio2 = r10_bio->devs[1].repl_bio; in recovery_request_write()
2607 /* Need to test wbio2->bi_end_io before we call in recovery_request_write()
2611 if (wbio2 && !wbio2->bi_end_io) in recovery_request_write()
2614 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { in recovery_request_write()
2616 if (wbio->bi_end_io) in recovery_request_write()
2627 d = r10_bio->devs[1].devnum; in recovery_request_write()
2628 if (wbio->bi_end_io) { in recovery_request_write()
2629 atomic_inc(&conf->mirrors[d].rdev->nr_pending); in recovery_request_write()
2630 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); in recovery_request_write()
2634 atomic_inc(&conf->mirrors[d].replacement->nr_pending); in recovery_request_write()
2635 md_sync_acct(conf->mirrors[d].replacement->bdev, in recovery_request_write()
2651 unsigned int read_errors = atomic_read(&rdev->read_errors); in check_decay_read_errors()
2655 if (rdev->last_read_error == 0) { in check_decay_read_errors()
2657 rdev->last_read_error = cur_time_mon; in check_decay_read_errors()
2661 hours_since_last = (long)(cur_time_mon - in check_decay_read_errors()
2662 rdev->last_read_error) / 3600; in check_decay_read_errors()
2664 rdev->last_read_error = cur_time_mon; in check_decay_read_errors()
2672 atomic_set(&rdev->read_errors, 0); in check_decay_read_errors()
2674 atomic_set(&rdev->read_errors, read_errors >> hours_since_last); in check_decay_read_errors()
2684 && (op == REQ_OP_READ || test_bit(WriteErrorSeen, &rdev->flags))) in r10_sync_page_io()
2685 return -1; in r10_sync_page_io()
2690 set_bit(WriteErrorSeen, &rdev->flags); in r10_sync_page_io()
2691 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in r10_sync_page_io()
2693 &rdev->mddev->recovery); in r10_sync_page_io()
2695 /* need to record an error - either for the block or the device */ in r10_sync_page_io()
2697 md_error(rdev->mddev, rdev); in r10_sync_page_io()
2711 int sect = 0; /* Offset from r10_bio->sector */ in fix_read_error()
2712 int sectors = r10_bio->sectors, slot = r10_bio->read_slot; in fix_read_error()
2714 int max_read_errors = atomic_read(&mddev->max_corr_read_errors); in fix_read_error()
2715 int d = r10_bio->devs[slot].devnum; in fix_read_error()
2720 rdev = conf->mirrors[d].rdev; in fix_read_error()
2722 if (test_bit(Faulty, &rdev->flags)) in fix_read_error()
2728 atomic_inc(&rdev->read_errors); in fix_read_error()
2729 if (atomic_read(&rdev->read_errors) > max_read_errors) { in fix_read_error()
2731 mdname(mddev), rdev->bdev, in fix_read_error()
2732 atomic_read(&rdev->read_errors), max_read_errors); in fix_read_error()
2734 mdname(mddev), rdev->bdev); in fix_read_error()
2736 r10_bio->devs[slot].bio = IO_BLOCKED; in fix_read_error()
2754 d = r10_bio->devs[sl].devnum; in fix_read_error()
2755 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2757 test_bit(In_sync, &rdev->flags) && in fix_read_error()
2758 !test_bit(Faulty, &rdev->flags) && in fix_read_error()
2759 is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, in fix_read_error()
2761 atomic_inc(&rdev->nr_pending); in fix_read_error()
2764 r10_bio->devs[sl].addr + in fix_read_error()
2767 conf->tmppage, in fix_read_error()
2775 if (sl == conf->copies) in fix_read_error()
2785 int dn = r10_bio->devs[slot].devnum; in fix_read_error()
2786 rdev = conf->mirrors[dn].rdev; in fix_read_error()
2790 r10_bio->devs[slot].addr in fix_read_error()
2794 r10_bio->devs[slot].bio in fix_read_error()
2801 /* write it back and re-read */ in fix_read_error()
2805 sl = conf->copies; in fix_read_error()
2806 sl--; in fix_read_error()
2807 d = r10_bio->devs[sl].devnum; in fix_read_error()
2808 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2810 test_bit(Faulty, &rdev->flags) || in fix_read_error()
2811 !test_bit(In_sync, &rdev->flags)) in fix_read_error()
2814 atomic_inc(&rdev->nr_pending); in fix_read_error()
2817 r10_bio->devs[sl].addr + in fix_read_error()
2819 s, conf->tmppage, REQ_OP_WRITE) in fix_read_error()
2828 rdev->bdev); in fix_read_error()
2831 rdev->bdev); in fix_read_error()
2839 sl = conf->copies; in fix_read_error()
2840 sl--; in fix_read_error()
2841 d = r10_bio->devs[sl].devnum; in fix_read_error()
2842 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2844 test_bit(Faulty, &rdev->flags) || in fix_read_error()
2845 !test_bit(In_sync, &rdev->flags)) in fix_read_error()
2848 atomic_inc(&rdev->nr_pending); in fix_read_error()
2851 r10_bio->devs[sl].addr + in fix_read_error()
2853 s, conf->tmppage, REQ_OP_READ)) { in fix_read_error()
2861 rdev->bdev); in fix_read_error()
2864 rdev->bdev); in fix_read_error()
2872 rdev->bdev); in fix_read_error()
2873 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2881 sectors -= s; in fix_read_error()
2888 struct bio *bio = r10_bio->master_bio; in narrow_write_error()
2889 struct mddev *mddev = r10_bio->mddev; in narrow_write_error()
2890 struct r10conf *conf = mddev->private; in narrow_write_error()
2891 struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; in narrow_write_error()
2906 int sect_to_write = r10_bio->sectors; in narrow_write_error()
2909 if (rdev->badblocks.shift < 0) in narrow_write_error()
2912 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2913 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2914 sector = r10_bio->sector; in narrow_write_error()
2915 sectors = ((r10_bio->sector + block_sectors) in narrow_write_error()
2916 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2917 - sector; in narrow_write_error()
2925 wbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, in narrow_write_error()
2926 &mddev->bio_set); in narrow_write_error()
2927 bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); in narrow_write_error()
2928 wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); in narrow_write_error()
2929 wbio->bi_iter.bi_sector = wsector + in narrow_write_error()
2931 wbio->bi_opf = REQ_OP_WRITE; in narrow_write_error()
2940 sect_to_write -= sectors; in narrow_write_error()
2949 int slot = r10_bio->read_slot; in handle_read_error()
2951 struct r10conf *conf = mddev->private; in handle_read_error()
2952 struct md_rdev *rdev = r10_bio->devs[slot].rdev; in handle_read_error()
2957 * other devices. When we find one, we re-write in handle_read_error()
2962 bio = r10_bio->devs[slot].bio; in handle_read_error()
2964 r10_bio->devs[slot].bio = NULL; in handle_read_error()
2966 if (mddev->ro) in handle_read_error()
2967 r10_bio->devs[slot].bio = IO_BLOCKED; in handle_read_error()
2968 else if (!test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2976 r10_bio->state = 0; in handle_read_error()
2977 raid10_read_request(mddev, r10_bio->master_bio, r10_bio, false); in handle_read_error()
2979 * allow_barrier after re-submit to ensure no sync io in handle_read_error()
2996 if (test_bit(R10BIO_IsSync, &r10_bio->state) || in handle_write_completed()
2997 test_bit(R10BIO_IsRecover, &r10_bio->state)) { in handle_write_completed()
2998 for (m = 0; m < conf->copies; m++) { in handle_write_completed()
2999 int dev = r10_bio->devs[m].devnum; in handle_write_completed()
3000 rdev = conf->mirrors[dev].rdev; in handle_write_completed()
3001 if (r10_bio->devs[m].bio == NULL || in handle_write_completed()
3002 r10_bio->devs[m].bio->bi_end_io == NULL) in handle_write_completed()
3004 if (!r10_bio->devs[m].bio->bi_status) { in handle_write_completed()
3007 r10_bio->devs[m].addr, in handle_write_completed()
3008 r10_bio->sectors, 0); in handle_write_completed()
3012 r10_bio->devs[m].addr, in handle_write_completed()
3013 r10_bio->sectors, 0)) in handle_write_completed()
3014 md_error(conf->mddev, rdev); in handle_write_completed()
3016 rdev = conf->mirrors[dev].replacement; in handle_write_completed()
3017 if (r10_bio->devs[m].repl_bio == NULL || in handle_write_completed()
3018 r10_bio->devs[m].repl_bio->bi_end_io == NULL) in handle_write_completed()
3021 if (!r10_bio->devs[m].repl_bio->bi_status) { in handle_write_completed()
3024 r10_bio->devs[m].addr, in handle_write_completed()
3025 r10_bio->sectors, 0); in handle_write_completed()
3029 r10_bio->devs[m].addr, in handle_write_completed()
3030 r10_bio->sectors, 0)) in handle_write_completed()
3031 md_error(conf->mddev, rdev); in handle_write_completed()
3037 for (m = 0; m < conf->copies; m++) { in handle_write_completed()
3038 int dev = r10_bio->devs[m].devnum; in handle_write_completed()
3039 struct bio *bio = r10_bio->devs[m].bio; in handle_write_completed()
3040 rdev = conf->mirrors[dev].rdev; in handle_write_completed()
3044 r10_bio->devs[m].addr, in handle_write_completed()
3045 r10_bio->sectors, 0); in handle_write_completed()
3046 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
3047 } else if (bio != NULL && bio->bi_status) { in handle_write_completed()
3050 md_error(conf->mddev, rdev); in handle_write_completed()
3051 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
3053 bio = r10_bio->devs[m].repl_bio; in handle_write_completed()
3054 rdev = conf->mirrors[dev].replacement; in handle_write_completed()
3058 r10_bio->devs[m].addr, in handle_write_completed()
3059 r10_bio->sectors, 0); in handle_write_completed()
3060 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
3064 spin_lock_irq(&conf->device_lock); in handle_write_completed()
3065 list_add(&r10_bio->retry_list, &conf->bio_end_io_list); in handle_write_completed()
3066 conf->nr_queued++; in handle_write_completed()
3067 spin_unlock_irq(&conf->device_lock); in handle_write_completed()
3072 wake_up(&conf->wait_barrier); in handle_write_completed()
3073 md_wakeup_thread(conf->mddev->thread); in handle_write_completed()
3076 &r10_bio->state)) in handle_write_completed()
3085 struct mddev *mddev = thread->mddev; in raid10d()
3088 struct r10conf *conf = mddev->private; in raid10d()
3089 struct list_head *head = &conf->retry_list; in raid10d()
3094 if (!list_empty_careful(&conf->bio_end_io_list) && in raid10d()
3095 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid10d()
3097 spin_lock_irqsave(&conf->device_lock, flags); in raid10d()
3098 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid10d()
3099 while (!list_empty(&conf->bio_end_io_list)) { in raid10d()
3100 list_move(conf->bio_end_io_list.prev, &tmp); in raid10d()
3101 conf->nr_queued--; in raid10d()
3104 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3108 list_del(&r10_bio->retry_list); in raid10d()
3111 &r10_bio->state)) in raid10d()
3122 spin_lock_irqsave(&conf->device_lock, flags); in raid10d()
3124 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3127 r10_bio = list_entry(head->prev, struct r10bio, retry_list); in raid10d()
3128 list_del(head->prev); in raid10d()
3129 conf->nr_queued--; in raid10d()
3130 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3132 mddev = r10_bio->mddev; in raid10d()
3133 conf = mddev->private; in raid10d()
3134 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in raid10d()
3135 test_bit(R10BIO_WriteError, &r10_bio->state)) in raid10d()
3137 else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) in raid10d()
3139 else if (test_bit(R10BIO_IsSync, &r10_bio->state)) in raid10d()
3141 else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) in raid10d()
3143 else if (test_bit(R10BIO_ReadError, &r10_bio->state)) in raid10d()
3149 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid10d()
3160 BUG_ON(mempool_initialized(&conf->r10buf_pool)); in init_resync()
3161 conf->have_replacement = 0; in init_resync()
3162 for (i = 0; i < conf->geo.raid_disks; i++) in init_resync()
3163 if (conf->mirrors[i].replacement) in init_resync()
3164 conf->have_replacement = 1; in init_resync()
3165 ret = mempool_init(&conf->r10buf_pool, buffs, in init_resync()
3169 conf->next_resync = 0; in init_resync()
3175 struct r10bio *r10bio = mempool_alloc(&conf->r10buf_pool, GFP_NOIO); in raid10_alloc_init_r10buf()
3181 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || in raid10_alloc_init_r10buf()
3182 test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) in raid10_alloc_init_r10buf()
3183 nalloc = conf->copies; /* resync */ in raid10_alloc_init_r10buf()
3188 bio = r10bio->devs[i].bio; in raid10_alloc_init_r10buf()
3189 rp = bio->bi_private; in raid10_alloc_init_r10buf()
3191 bio->bi_private = rp; in raid10_alloc_init_r10buf()
3192 bio = r10bio->devs[i].repl_bio; in raid10_alloc_init_r10buf()
3194 rp = bio->bi_private; in raid10_alloc_init_r10buf()
3196 bio->bi_private = rp; in raid10_alloc_init_r10buf()
3223 chunks = conf->geo.raid_disks / conf->geo.near_copies; in raid10_set_cluster_sync_high()
3224 if (conf->geo.raid_disks % conf->geo.near_copies == 0) in raid10_set_cluster_sync_high()
3228 window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors; in raid10_set_cluster_sync_high()
3236 conf->cluster_sync_high = conf->cluster_sync_low + window_size; in raid10_set_cluster_sync_high()
3242 * We need to make sure that no normal I/O request - particularly write
3243 * requests - conflict with active sync requests.
3249 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
3252 * and update if there are differences. If only one copy is live,
3255 * value for each non-in_sync drive, and over-write.
3258 * given address, one for each out-of-sync device. We model this by allocating
3259 * a number of r10_bio structures, one for each out-of-sync device.
3265 * This link is counted in ->remaining. When the r10_bio that points to NULL
3274 struct r10conf *conf = mddev->private; in raid10_sync_request()
3283 sector_t chunk_mask = conf->geo.chunk_mask; in raid10_sync_request()
3285 int error_disk = -1; in raid10_sync_request()
3291 if (mddev->bitmap == NULL && in raid10_sync_request()
3292 mddev->recovery_cp == MaxSector && in raid10_sync_request()
3293 mddev->reshape_position == MaxSector && in raid10_sync_request()
3294 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid10_sync_request()
3295 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid10_sync_request()
3296 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in raid10_sync_request()
3297 conf->fullsync == 0) { in raid10_sync_request()
3299 return mddev->dev_sectors - sector_nr; in raid10_sync_request()
3302 if (!mempool_initialized(&conf->r10buf_pool)) in raid10_sync_request()
3307 max_sector = mddev->dev_sectors; in raid10_sync_request()
3308 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || in raid10_sync_request()
3309 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_sync_request()
3310 max_sector = mddev->resync_max_sectors; in raid10_sync_request()
3312 conf->cluster_sync_low = 0; in raid10_sync_request()
3313 conf->cluster_sync_high = 0; in raid10_sync_request()
3320 * mddev->curr_resync, but for recovery, in raid10_sync_request()
3324 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { in raid10_sync_request()
3330 if (mddev->curr_resync < max_sector) { /* aborted */ in raid10_sync_request()
3331 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) in raid10_sync_request()
3332 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, in raid10_sync_request()
3334 else for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3336 raid10_find_virt(conf, mddev->curr_resync, i); in raid10_sync_request()
3337 md_bitmap_end_sync(mddev->bitmap, sect, in raid10_sync_request()
3342 if ((!mddev->bitmap || conf->fullsync) in raid10_sync_request()
3343 && conf->have_replacement in raid10_sync_request()
3344 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3349 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3351 rcu_dereference(conf->mirrors[i].replacement); in raid10_sync_request()
3353 rdev->recovery_offset = MaxSector; in raid10_sync_request()
3357 conf->fullsync = 0; in raid10_sync_request()
3359 md_bitmap_close_sync(mddev->bitmap); in raid10_sync_request()
3365 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_sync_request()
3368 if (chunks_skipped >= conf->geo.raid_disks) { in raid10_sync_request()
3370 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? "resync" : "recovery"); in raid10_sync_request()
3372 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3377 conf->mirrors[error_disk].recovery_disabled = in raid10_sync_request()
3378 mddev->recovery_disabled; in raid10_sync_request()
3386 return (max_sector - sector_nr) + sectors_skipped; in raid10_sync_request()
3389 if (max_sector > mddev->resync_max) in raid10_sync_request()
3390 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid10_sync_request()
3392 /* make sure whole request will fit in a chunk - if chunks in raid10_sync_request()
3395 if (conf->geo.near_copies < conf->geo.raid_disks && in raid10_sync_request()
3400 * If there is non-resync activity waiting for a turn, then let it in raid10_sync_request()
3403 if (conf->nr_waiting) in raid10_sync_request()
3416 /* First, we decide what to do and set ->bi_end_io in raid10_sync_request()
3421 max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); in raid10_sync_request()
3422 if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3427 for (i = 0 ; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3433 struct raid10_info *mirror = &conf->mirrors[i]; in raid10_sync_request()
3437 mrdev = rcu_dereference(mirror->rdev); in raid10_sync_request()
3438 mreplace = rcu_dereference(mirror->replacement); in raid10_sync_request()
3440 if (mrdev && (test_bit(Faulty, &mrdev->flags) || in raid10_sync_request()
3441 test_bit(In_sync, &mrdev->flags))) in raid10_sync_request()
3443 if (mreplace && test_bit(Faulty, &mreplace->flags)) in raid10_sync_request()
3455 if (sect >= mddev->resync_max_sectors) { in raid10_sync_request()
3456 /* last stripe is not complete - don't in raid10_sync_request()
3466 must_sync = md_bitmap_start_sync(mddev->bitmap, sect, in raid10_sync_request()
3472 !conf->fullsync) { in raid10_sync_request()
3476 chunks_skipped = -1; in raid10_sync_request()
3481 atomic_inc(&mrdev->nr_pending); in raid10_sync_request()
3483 atomic_inc(&mreplace->nr_pending); in raid10_sync_request()
3487 r10_bio->state = 0; in raid10_sync_request()
3489 atomic_set(&r10_bio->remaining, 0); in raid10_sync_request()
3491 r10_bio->master_bio = (struct bio*)rb2; in raid10_sync_request()
3493 atomic_inc(&rb2->remaining); in raid10_sync_request()
3494 r10_bio->mddev = mddev; in raid10_sync_request()
3495 set_bit(R10BIO_IsRecover, &r10_bio->state); in raid10_sync_request()
3496 r10_bio->sector = sect; in raid10_sync_request()
3504 for (j = 0; j < conf->geo.raid_disks; j++) { in raid10_sync_request()
3506 conf->mirrors[j].rdev); in raid10_sync_request()
3507 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { in raid10_sync_request()
3513 must_sync = md_bitmap_start_sync(mddev->bitmap, sect, in raid10_sync_request()
3517 for (j=0; j<conf->copies;j++) { in raid10_sync_request()
3519 int d = r10_bio->devs[j].devnum; in raid10_sync_request()
3522 rcu_dereference(conf->mirrors[d].rdev); in raid10_sync_request()
3526 !test_bit(In_sync, &rdev->flags)) in raid10_sync_request()
3530 sector = r10_bio->devs[j].addr; in raid10_sync_request()
3535 max_sync = first_bad - sector; in raid10_sync_request()
3537 bad_sectors -= (sector in raid10_sync_request()
3538 - first_bad); in raid10_sync_request()
3544 bio = r10_bio->devs[0].bio; in raid10_sync_request()
3545 bio->bi_next = biolist; in raid10_sync_request()
3547 bio->bi_end_io = end_sync_read; in raid10_sync_request()
3548 bio->bi_opf = REQ_OP_READ; in raid10_sync_request()
3549 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3550 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3551 from_addr = r10_bio->devs[j].addr; in raid10_sync_request()
3552 bio->bi_iter.bi_sector = from_addr + in raid10_sync_request()
3553 rdev->data_offset; in raid10_sync_request()
3554 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3555 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3558 for (k=0; k<conf->copies; k++) in raid10_sync_request()
3559 if (r10_bio->devs[k].devnum == i) in raid10_sync_request()
3561 BUG_ON(k == conf->copies); in raid10_sync_request()
3562 to_addr = r10_bio->devs[k].addr; in raid10_sync_request()
3563 r10_bio->devs[0].devnum = d; in raid10_sync_request()
3564 r10_bio->devs[0].addr = from_addr; in raid10_sync_request()
3565 r10_bio->devs[1].devnum = i; in raid10_sync_request()
3566 r10_bio->devs[1].addr = to_addr; in raid10_sync_request()
3569 bio = r10_bio->devs[1].bio; in raid10_sync_request()
3570 bio->bi_next = biolist; in raid10_sync_request()
3572 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3573 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3574 bio->bi_iter.bi_sector = to_addr in raid10_sync_request()
3575 + mrdev->data_offset; in raid10_sync_request()
3576 bio_set_dev(bio, mrdev->bdev); in raid10_sync_request()
3577 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3579 r10_bio->devs[1].bio->bi_end_io = NULL; in raid10_sync_request()
3582 bio = r10_bio->devs[1].repl_bio; in raid10_sync_request()
3584 bio->bi_end_io = NULL; in raid10_sync_request()
3591 bio->bi_next = biolist; in raid10_sync_request()
3593 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3594 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3595 bio->bi_iter.bi_sector = to_addr + in raid10_sync_request()
3596 mreplace->data_offset; in raid10_sync_request()
3597 bio_set_dev(bio, mreplace->bdev); in raid10_sync_request()
3598 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3602 if (j == conf->copies) { in raid10_sync_request()
3610 for (k = 0; k < conf->copies; k++) in raid10_sync_request()
3611 if (r10_bio->devs[k].devnum == i) in raid10_sync_request()
3614 &mrdev->flags) in raid10_sync_request()
3617 r10_bio->devs[k].addr, in raid10_sync_request()
3623 r10_bio->devs[k].addr, in raid10_sync_request()
3629 &mddev->recovery)) in raid10_sync_request()
3632 mirror->recovery_disabled in raid10_sync_request()
3633 = mddev->recovery_disabled; in raid10_sync_request()
3639 atomic_dec(&rb2->remaining); in raid10_sync_request()
3651 if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { in raid10_sync_request()
3657 for (; j < conf->copies; j++) { in raid10_sync_request()
3658 int d = r10_bio->devs[j].devnum; in raid10_sync_request()
3659 if (conf->mirrors[d].rdev && in raid10_sync_request()
3661 &conf->mirrors[d].rdev->flags)) in raid10_sync_request()
3665 r10_bio->devs[0].bio->bi_opf in raid10_sync_request()
3672 r10_bio = (struct r10bio*) rb2->master_bio; in raid10_sync_request()
3673 rb2->master_bio = NULL; in raid10_sync_request()
3683 * Since curr_resync_completed could probably not update in in raid10_sync_request()
3689 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, in raid10_sync_request()
3691 (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid10_sync_request()
3693 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, in raid10_sync_request()
3694 &sync_blocks, mddev->degraded) && in raid10_sync_request()
3695 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, in raid10_sync_request()
3696 &mddev->recovery)) { in raid10_sync_request()
3704 r10_bio->state = 0; in raid10_sync_request()
3706 r10_bio->mddev = mddev; in raid10_sync_request()
3707 atomic_set(&r10_bio->remaining, 0); in raid10_sync_request()
3709 conf->next_resync = sector_nr; in raid10_sync_request()
3711 r10_bio->master_bio = NULL; in raid10_sync_request()
3712 r10_bio->sector = sector_nr; in raid10_sync_request()
3713 set_bit(R10BIO_IsSync, &r10_bio->state); in raid10_sync_request()
3715 r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; in raid10_sync_request()
3717 for (i = 0; i < conf->copies; i++) { in raid10_sync_request()
3718 int d = r10_bio->devs[i].devnum; in raid10_sync_request()
3723 if (r10_bio->devs[i].repl_bio) in raid10_sync_request()
3724 r10_bio->devs[i].repl_bio->bi_end_io = NULL; in raid10_sync_request()
3726 bio = r10_bio->devs[i].bio; in raid10_sync_request()
3727 bio->bi_status = BLK_STS_IOERR; in raid10_sync_request()
3729 rdev = rcu_dereference(conf->mirrors[d].rdev); in raid10_sync_request()
3730 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { in raid10_sync_request()
3734 sector = r10_bio->devs[i].addr; in raid10_sync_request()
3738 max_sync = first_bad - sector; in raid10_sync_request()
3740 bad_sectors -= (sector - first_bad); in raid10_sync_request()
3747 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3748 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3749 bio->bi_next = biolist; in raid10_sync_request()
3751 bio->bi_end_io = end_sync_read; in raid10_sync_request()
3752 bio->bi_opf = REQ_OP_READ; in raid10_sync_request()
3753 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3754 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3755 bio->bi_iter.bi_sector = sector + rdev->data_offset; in raid10_sync_request()
3756 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3759 rdev = rcu_dereference(conf->mirrors[d].replacement); in raid10_sync_request()
3760 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { in raid10_sync_request()
3764 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3767 bio = r10_bio->devs[i].repl_bio; in raid10_sync_request()
3768 bio->bi_status = BLK_STS_IOERR; in raid10_sync_request()
3770 sector = r10_bio->devs[i].addr; in raid10_sync_request()
3771 bio->bi_next = biolist; in raid10_sync_request()
3773 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3774 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3775 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3776 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3777 bio->bi_iter.bi_sector = sector + rdev->data_offset; in raid10_sync_request()
3778 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3784 for (i=0; i<conf->copies; i++) { in raid10_sync_request()
3785 int d = r10_bio->devs[i].devnum; in raid10_sync_request()
3786 if (r10_bio->devs[i].bio->bi_end_io) in raid10_sync_request()
3787 rdev_dec_pending(conf->mirrors[d].rdev, in raid10_sync_request()
3789 if (r10_bio->devs[i].repl_bio && in raid10_sync_request()
3790 r10_bio->devs[i].repl_bio->bi_end_io) in raid10_sync_request()
3792 conf->mirrors[d].replacement, in raid10_sync_request()
3808 len = (max_sector - sector_nr) << 9; in raid10_sync_request()
3811 for (bio= biolist ; bio ; bio=bio->bi_next) { in raid10_sync_request()
3815 bio->bi_status = BLK_STS_RESOURCE; in raid10_sync_request()
3823 r10_bio->sectors = nr_sectors; in raid10_sync_request()
3826 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3828 if (conf->cluster_sync_high < sector_nr + nr_sectors) { in raid10_sync_request()
3829 conf->cluster_sync_low = mddev->curr_resync_completed; in raid10_sync_request()
3832 md_cluster_ops->resync_info_update(mddev, in raid10_sync_request()
3833 conf->cluster_sync_low, in raid10_sync_request()
3834 conf->cluster_sync_high); in raid10_sync_request()
3841 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3849 if (conf->cluster_sync_high < sect_va1 + nr_sectors) { in raid10_sync_request()
3856 mddev->curr_resync_completed, i); in raid10_sync_request()
3858 if (conf->cluster_sync_low == 0 || in raid10_sync_request()
3859 conf->cluster_sync_low > sect_va2) in raid10_sync_request()
3860 conf->cluster_sync_low = sect_va2; in raid10_sync_request()
3865 md_cluster_ops->resync_info_update(mddev, in raid10_sync_request()
3866 conf->cluster_sync_low, in raid10_sync_request()
3867 conf->cluster_sync_high); in raid10_sync_request()
3873 biolist = biolist->bi_next; in raid10_sync_request()
3875 bio->bi_next = NULL; in raid10_sync_request()
3877 r10_bio->sectors = nr_sectors; in raid10_sync_request()
3879 if (bio->bi_end_io == end_sync_read) { in raid10_sync_request()
3881 bio->bi_status = 0; in raid10_sync_request()
3894 /* There is nowhere to write, so all non-sync in raid10_sync_request()
3901 sectors_skipped += (max_sector - sector_nr); in raid10_sync_request()
3911 struct r10conf *conf = mddev->private; in raid10_size()
3914 raid_disks = min(conf->geo.raid_disks, in raid10_size()
3915 conf->prev.raid_disks); in raid10_size()
3917 sectors = conf->dev_sectors; in raid10_size()
3919 size = sectors >> conf->geo.chunk_shift; in raid10_size()
3920 sector_div(size, conf->geo.far_copies); in raid10_size()
3922 sector_div(size, conf->geo.near_copies); in raid10_size()
3924 return size << conf->geo.chunk_shift; in raid10_size()
3929 /* Calculate the number of sectors-per-device that will in calc_sectors()
3930 * actually be used, and set conf->dev_sectors and in calc_sectors()
3931 * conf->stride in calc_sectors()
3934 size = size >> conf->geo.chunk_shift; in calc_sectors()
3935 sector_div(size, conf->geo.far_copies); in calc_sectors()
3936 size = size * conf->geo.raid_disks; in calc_sectors()
3937 sector_div(size, conf->geo.near_copies); in calc_sectors()
3940 size = size * conf->copies; in calc_sectors()
3945 size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); in calc_sectors()
3947 conf->dev_sectors = size << conf->geo.chunk_shift; in calc_sectors()
3949 if (conf->geo.far_offset) in calc_sectors()
3950 conf->geo.stride = 1 << conf->geo.chunk_shift; in calc_sectors()
3952 sector_div(size, conf->geo.far_copies); in calc_sectors()
3953 conf->geo.stride = size << conf->geo.chunk_shift; in calc_sectors()
3960 int nc, fc, fo; in setup_geo() local
3964 layout = mddev->layout; in setup_geo()
3965 chunk = mddev->chunk_sectors; in setup_geo()
3966 disks = mddev->raid_disks - mddev->delta_disks; in setup_geo()
3969 layout = mddev->new_layout; in setup_geo()
3970 chunk = mddev->new_chunk_sectors; in setup_geo()
3971 disks = mddev->raid_disks; in setup_geo()
3974 case geo_start: /* new when starting reshape - raid_disks not in setup_geo()
3976 layout = mddev->new_layout; in setup_geo()
3977 chunk = mddev->new_chunk_sectors; in setup_geo()
3978 disks = mddev->raid_disks + mddev->delta_disks; in setup_geo()
3982 return -1; in setup_geo()
3985 return -2; in setup_geo()
3987 fc = (layout >> 8) & 255; in setup_geo()
3989 geo->raid_disks = disks; in setup_geo()
3990 geo->near_copies = nc; in setup_geo()
3991 geo->far_copies = fc; in setup_geo()
3992 geo->far_offset = fo; in setup_geo()
3995 geo->far_set_size = disks; in setup_geo()
3997 case 1: /* "improved" layout which was buggy. Hopefully no-one is in setup_geo()
3999 geo->far_set_size = disks/fc; in setup_geo()
4000 WARN(geo->far_set_size < fc, in setup_geo()
4001 "This RAID10 layout does not provide data safety - please backup and create new array\n"); in setup_geo()
4004 geo->far_set_size = fc * nc; in setup_geo()
4007 return -1; in setup_geo()
4009 geo->chunk_mask = chunk - 1; in setup_geo()
4010 geo->chunk_shift = ffz(~chunk); in setup_geo()
4011 return nc*fc; in setup_geo()
4019 mempool_exit(&conf->r10bio_pool); in raid10_free_conf()
4020 kfree(conf->mirrors); in raid10_free_conf()
4021 kfree(conf->mirrors_old); in raid10_free_conf()
4022 kfree(conf->mirrors_new); in raid10_free_conf()
4023 safe_put_page(conf->tmppage); in raid10_free_conf()
4024 bioset_exit(&conf->bio_split); in raid10_free_conf()
4031 int err = -EINVAL; in setup_conf()
4037 if (copies == -2) { in setup_conf()
4043 if (copies < 2 || copies > mddev->raid_disks) { in setup_conf()
4045 mdname(mddev), mddev->new_layout); in setup_conf()
4049 err = -ENOMEM; in setup_conf()
4055 conf->mirrors = kcalloc(mddev->raid_disks + max(0, -mddev->delta_disks), in setup_conf()
4058 if (!conf->mirrors) in setup_conf()
4061 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
4062 if (!conf->tmppage) in setup_conf()
4065 conf->geo = geo; in setup_conf()
4066 conf->copies = copies; in setup_conf()
4067 err = mempool_init(&conf->r10bio_pool, NR_RAID_BIOS, r10bio_pool_alloc, in setup_conf()
4072 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
4076 calc_sectors(conf, mddev->dev_sectors); in setup_conf()
4077 if (mddev->reshape_position == MaxSector) { in setup_conf()
4078 conf->prev = conf->geo; in setup_conf()
4079 conf->reshape_progress = MaxSector; in setup_conf()
4081 if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { in setup_conf()
4082 err = -EINVAL; in setup_conf()
4085 conf->reshape_progress = mddev->reshape_position; in setup_conf()
4086 if (conf->prev.far_offset) in setup_conf()
4087 conf->prev.stride = 1 << conf->prev.chunk_shift; in setup_conf()
4090 conf->prev.stride = conf->dev_sectors; in setup_conf()
4092 conf->reshape_safe = conf->reshape_progress; in setup_conf()
4093 spin_lock_init(&conf->device_lock); in setup_conf()
4094 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
4095 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
4097 seqlock_init(&conf->resync_lock); in setup_conf()
4098 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
4099 atomic_set(&conf->nr_pending, 0); in setup_conf()
4101 err = -ENOMEM; in setup_conf()
4102 rcu_assign_pointer(conf->thread, in setup_conf()
4104 if (!conf->thread) in setup_conf()
4107 conf->mddev = mddev; in setup_conf()
4117 int raid_disks = conf->geo.raid_disks; in raid10_set_io_opt()
4119 if (!(conf->geo.raid_disks % conf->geo.near_copies)) in raid10_set_io_opt()
4120 raid_disks /= conf->geo.near_copies; in raid10_set_io_opt()
4121 blk_queue_io_opt(conf->mddev->queue, (conf->mddev->chunk_sectors << 9) * in raid10_set_io_opt()
4136 return -ENOMEM; in raid10_run()
4138 if (mddev->private == NULL) { in raid10_run()
4142 mddev->private = conf; in raid10_run()
4144 conf = mddev->private; in raid10_run()
4148 rcu_assign_pointer(mddev->thread, conf->thread); in raid10_run()
4149 rcu_assign_pointer(conf->thread, NULL); in raid10_run()
4151 if (mddev_is_clustered(conf->mddev)) { in raid10_run()
4152 int fc, fo; in raid10_run() local
4154 fc = (mddev->layout >> 8) & 255; in raid10_run()
4155 fo = mddev->layout & (1<<16); in raid10_run()
4156 if (fc > 1 || fo > 0) { in raid10_run()
4163 if (mddev->queue) { in raid10_run()
4164 blk_queue_max_write_zeroes_sectors(mddev->queue, 0); in raid10_run()
4165 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); in raid10_run()
4172 disk_idx = rdev->raid_disk; in raid10_run()
4175 if (disk_idx >= conf->geo.raid_disks && in raid10_run()
4176 disk_idx >= conf->prev.raid_disks) in raid10_run()
4178 disk = conf->mirrors + disk_idx; in raid10_run()
4180 if (test_bit(Replacement, &rdev->flags)) { in raid10_run()
4181 if (disk->replacement) in raid10_run()
4183 disk->replacement = rdev; in raid10_run()
4185 if (disk->rdev) in raid10_run()
4187 disk->rdev = rdev; in raid10_run()
4189 diff = (rdev->new_data_offset - rdev->data_offset); in raid10_run()
4190 if (!mddev->reshape_backwards) in raid10_run()
4191 diff = -diff; in raid10_run()
4197 if (mddev->gendisk) in raid10_run()
4198 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid10_run()
4199 rdev->data_offset << 9); in raid10_run()
4201 disk->head_position = 0; in raid10_run()
4206 if (!enough(conf, -1)) { in raid10_run()
4212 if (conf->reshape_progress != MaxSector) { in raid10_run()
4214 if (conf->geo.far_copies != 1 && in raid10_run()
4215 conf->geo.far_offset == 0) in raid10_run()
4217 if (conf->prev.far_copies != 1 && in raid10_run()
4218 conf->prev.far_offset == 0) in raid10_run()
4222 mddev->degraded = 0; in raid10_run()
4224 i < conf->geo.raid_disks in raid10_run()
4225 || i < conf->prev.raid_disks; in raid10_run()
4228 disk = conf->mirrors + i; in raid10_run()
4230 if (!disk->rdev && disk->replacement) { in raid10_run()
4231 /* The replacement is all we have - use it */ in raid10_run()
4232 disk->rdev = disk->replacement; in raid10_run()
4233 disk->replacement = NULL; in raid10_run()
4234 clear_bit(Replacement, &disk->rdev->flags); in raid10_run()
4237 if (!disk->rdev || in raid10_run()
4238 !test_bit(In_sync, &disk->rdev->flags)) { in raid10_run()
4239 disk->head_position = 0; in raid10_run()
4240 mddev->degraded++; in raid10_run()
4241 if (disk->rdev && in raid10_run()
4242 disk->rdev->saved_raid_disk < 0) in raid10_run()
4243 conf->fullsync = 1; in raid10_run()
4246 if (disk->replacement && in raid10_run()
4247 !test_bit(In_sync, &disk->replacement->flags) && in raid10_run()
4248 disk->replacement->saved_raid_disk < 0) { in raid10_run()
4249 conf->fullsync = 1; in raid10_run()
4252 disk->recovery_disabled = mddev->recovery_disabled - 1; in raid10_run()
4255 if (mddev->recovery_cp != MaxSector) in raid10_run()
4256 pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", in raid10_run()
4259 mdname(mddev), conf->geo.raid_disks - mddev->degraded, in raid10_run()
4260 conf->geo.raid_disks); in raid10_run()
4264 mddev->dev_sectors = conf->dev_sectors; in raid10_run()
4267 mddev->resync_max_sectors = size; in raid10_run()
4268 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid10_run()
4273 if (conf->reshape_progress != MaxSector) { in raid10_run()
4276 before_length = ((1 << conf->prev.chunk_shift) * in raid10_run()
4277 conf->prev.far_copies); in raid10_run()
4278 after_length = ((1 << conf->geo.chunk_shift) * in raid10_run()
4279 conf->geo.far_copies); in raid10_run()
4286 conf->offset_diff = min_offset_diff; in raid10_run()
4288 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); in raid10_run()
4289 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); in raid10_run()
4290 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); in raid10_run()
4291 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); in raid10_run()
4292 rcu_assign_pointer(mddev->sync_thread, in raid10_run()
4294 if (!mddev->sync_thread) in raid10_run()
4301 md_unregister_thread(mddev, &mddev->thread); in raid10_run()
4303 mddev->private = NULL; in raid10_run()
4305 return -EIO; in raid10_run()
4315 struct r10conf *conf = mddev->private; in raid10_quiesce()
4337 struct r10conf *conf = mddev->private; in raid10_resize()
4340 if (mddev->reshape_position != MaxSector) in raid10_resize()
4341 return -EBUSY; in raid10_resize()
4343 if (conf->geo.far_copies > 1 && !conf->geo.far_offset) in raid10_resize()
4344 return -EINVAL; in raid10_resize()
4348 if (mddev->external_size && in raid10_resize()
4349 mddev->array_sectors > size) in raid10_resize()
4350 return -EINVAL; in raid10_resize()
4351 if (mddev->bitmap) { in raid10_resize()
4352 int ret = md_bitmap_resize(mddev->bitmap, size, 0, 0); in raid10_resize()
4357 if (sectors > mddev->dev_sectors && in raid10_resize()
4358 mddev->recovery_cp > oldsize) { in raid10_resize()
4359 mddev->recovery_cp = oldsize; in raid10_resize()
4360 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_resize()
4363 mddev->dev_sectors = conf->dev_sectors; in raid10_resize()
4364 mddev->resync_max_sectors = size; in raid10_resize()
4373 if (mddev->degraded > 0) { in raid10_takeover_raid0()
4376 return ERR_PTR(-EINVAL); in raid10_takeover_raid0()
4381 mddev->new_level = 10; in raid10_takeover_raid0()
4383 mddev->new_layout = (1<<8) + 2; in raid10_takeover_raid0()
4384 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid10_takeover_raid0()
4385 mddev->delta_disks = mddev->raid_disks; in raid10_takeover_raid0()
4386 mddev->raid_disks *= 2; in raid10_takeover_raid0()
4388 mddev->recovery_cp = MaxSector; in raid10_takeover_raid0()
4389 mddev->dev_sectors = size; in raid10_takeover_raid0()
4394 if (rdev->raid_disk >= 0) { in raid10_takeover_raid0()
4395 rdev->new_raid_disk = rdev->raid_disk * 2; in raid10_takeover_raid0()
4396 rdev->sectors = size; in raid10_takeover_raid0()
4408 * raid0 - providing it has only two drives in raid10_takeover()
4410 if (mddev->level == 0) { in raid10_takeover()
4412 raid0_conf = mddev->private; in raid10_takeover()
4413 if (raid0_conf->nr_strip_zones > 1) { in raid10_takeover()
4416 return ERR_PTR(-EINVAL); in raid10_takeover()
4419 raid0_conf->strip_zone->zone_end, in raid10_takeover()
4420 raid0_conf->strip_zone->nb_dev); in raid10_takeover()
4422 return ERR_PTR(-EINVAL); in raid10_takeover()
4428 * - layout (to ->new_layout) in raid10_check_reshape()
4429 * - chunk size (to ->new_chunk_sectors) in raid10_check_reshape()
4430 * - raid_disks (by delta_disks) in raid10_check_reshape()
4441 struct r10conf *conf = mddev->private; in raid10_check_reshape()
4444 if (conf->geo.far_copies != 1 && !conf->geo.far_offset) in raid10_check_reshape()
4445 return -EINVAL; in raid10_check_reshape()
4447 if (setup_geo(&geo, mddev, geo_start) != conf->copies) in raid10_check_reshape()
4449 return -EINVAL; in raid10_check_reshape()
4452 return -EINVAL; in raid10_check_reshape()
4454 if (mddev->array_sectors & geo.chunk_mask) in raid10_check_reshape()
4456 return -EINVAL; in raid10_check_reshape()
4458 if (!enough(conf, -1)) in raid10_check_reshape()
4459 return -EINVAL; in raid10_check_reshape()
4461 kfree(conf->mirrors_new); in raid10_check_reshape()
4462 conf->mirrors_new = NULL; in raid10_check_reshape()
4463 if (mddev->delta_disks > 0) { in raid10_check_reshape()
4465 conf->mirrors_new = in raid10_check_reshape()
4466 kcalloc(mddev->raid_disks + mddev->delta_disks, in raid10_check_reshape()
4469 if (!conf->mirrors_new) in raid10_check_reshape()
4470 return -ENOMEM; in raid10_check_reshape()
4477 * - start an array
4478 * - remove non-faulty devices
4479 * - add a spare
4480 * - allow a reshape
4485 * of the two sections, and some non-in_sync devices may
4496 for (i = 0; i < conf->prev.raid_disks; i++) { in calc_degraded()
4497 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in calc_degraded()
4498 if (!rdev || test_bit(Faulty, &rdev->flags)) in calc_degraded()
4500 else if (!test_bit(In_sync, &rdev->flags)) in calc_degraded()
4508 if (conf->geo.raid_disks == conf->prev.raid_disks) in calc_degraded()
4512 for (i = 0; i < conf->geo.raid_disks; i++) { in calc_degraded()
4513 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in calc_degraded()
4514 if (!rdev || test_bit(Faulty, &rdev->flags)) in calc_degraded()
4516 else if (!test_bit(In_sync, &rdev->flags)) { in calc_degraded()
4522 if (conf->geo.raid_disks <= conf->prev.raid_disks) in calc_degraded()
4539 * array non-degraded. We also require that the difference in raid10_start_reshape()
4540 * between old and new data_offset - on each device - is in raid10_start_reshape()
4541 * enough that we never risk over-writing. in raid10_start_reshape()
4548 struct r10conf *conf = mddev->private; in raid10_start_reshape()
4553 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) in raid10_start_reshape()
4554 return -EBUSY; in raid10_start_reshape()
4556 if (setup_geo(&new, mddev, geo_start) != conf->copies) in raid10_start_reshape()
4557 return -EINVAL; in raid10_start_reshape()
4559 before_length = ((1 << conf->prev.chunk_shift) * in raid10_start_reshape()
4560 conf->prev.far_copies); in raid10_start_reshape()
4561 after_length = ((1 << conf->geo.chunk_shift) * in raid10_start_reshape()
4562 conf->geo.far_copies); in raid10_start_reshape()
4565 if (!test_bit(In_sync, &rdev->flags) in raid10_start_reshape()
4566 && !test_bit(Faulty, &rdev->flags)) in raid10_start_reshape()
4568 if (rdev->raid_disk >= 0) { in raid10_start_reshape()
4569 long long diff = (rdev->new_data_offset in raid10_start_reshape()
4570 - rdev->data_offset); in raid10_start_reshape()
4571 if (!mddev->reshape_backwards) in raid10_start_reshape()
4572 diff = -diff; in raid10_start_reshape()
4582 return -EINVAL; in raid10_start_reshape()
4584 if (spares < mddev->delta_disks) in raid10_start_reshape()
4585 return -EINVAL; in raid10_start_reshape()
4587 conf->offset_diff = min_offset_diff; in raid10_start_reshape()
4588 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4589 if (conf->mirrors_new) { in raid10_start_reshape()
4590 memcpy(conf->mirrors_new, conf->mirrors, in raid10_start_reshape()
4591 sizeof(struct raid10_info)*conf->prev.raid_disks); in raid10_start_reshape()
4593 kfree(conf->mirrors_old); in raid10_start_reshape()
4594 conf->mirrors_old = conf->mirrors; in raid10_start_reshape()
4595 conf->mirrors = conf->mirrors_new; in raid10_start_reshape()
4596 conf->mirrors_new = NULL; in raid10_start_reshape()
4598 setup_geo(&conf->geo, mddev, geo_start); in raid10_start_reshape()
4600 if (mddev->reshape_backwards) { in raid10_start_reshape()
4602 if (size < mddev->array_sectors) { in raid10_start_reshape()
4603 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4606 return -EINVAL; in raid10_start_reshape()
4608 mddev->resync_max_sectors = size; in raid10_start_reshape()
4609 conf->reshape_progress = size; in raid10_start_reshape()
4611 conf->reshape_progress = 0; in raid10_start_reshape()
4612 conf->reshape_safe = conf->reshape_progress; in raid10_start_reshape()
4613 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4615 if (mddev->delta_disks && mddev->bitmap) { in raid10_start_reshape()
4620 newsize = raid10_size(mddev, 0, conf->geo.raid_disks); in raid10_start_reshape()
4623 ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid10_start_reshape()
4631 if (rdev->raid_disk > -1 && in raid10_start_reshape()
4632 !test_bit(Faulty, &rdev->flags)) in raid10_start_reshape()
4633 sb = page_address(rdev->sb_page); in raid10_start_reshape()
4641 if ((sb && (le32_to_cpu(sb->feature_map) & in raid10_start_reshape()
4645 ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid10_start_reshape()
4649 ret = md_cluster_ops->resize_bitmaps(mddev, newsize, oldsize); in raid10_start_reshape()
4651 md_bitmap_resize(mddev->bitmap, oldsize, 0, 0); in raid10_start_reshape()
4656 if (mddev->delta_disks > 0) { in raid10_start_reshape()
4658 if (rdev->raid_disk < 0 && in raid10_start_reshape()
4659 !test_bit(Faulty, &rdev->flags)) { in raid10_start_reshape()
4661 if (rdev->raid_disk >= in raid10_start_reshape()
4662 conf->prev.raid_disks) in raid10_start_reshape()
4663 set_bit(In_sync, &rdev->flags); in raid10_start_reshape()
4665 rdev->recovery_offset = 0; in raid10_start_reshape()
4670 } else if (rdev->raid_disk >= conf->prev.raid_disks in raid10_start_reshape()
4671 && !test_bit(Faulty, &rdev->flags)) { in raid10_start_reshape()
4673 set_bit(In_sync, &rdev->flags); in raid10_start_reshape()
4677 * ->degraded is measured against the larger of the in raid10_start_reshape()
4680 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4681 mddev->degraded = calc_degraded(conf); in raid10_start_reshape()
4682 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4683 mddev->raid_disks = conf->geo.raid_disks; in raid10_start_reshape()
4684 mddev->reshape_position = conf->reshape_progress; in raid10_start_reshape()
4685 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid10_start_reshape()
4687 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); in raid10_start_reshape()
4688 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); in raid10_start_reshape()
4689 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); in raid10_start_reshape()
4690 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); in raid10_start_reshape()
4691 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); in raid10_start_reshape()
4693 rcu_assign_pointer(mddev->sync_thread, in raid10_start_reshape()
4695 if (!mddev->sync_thread) { in raid10_start_reshape()
4696 ret = -EAGAIN; in raid10_start_reshape()
4699 conf->reshape_checkpoint = jiffies; in raid10_start_reshape()
4700 md_wakeup_thread(mddev->sync_thread); in raid10_start_reshape()
4705 mddev->recovery = 0; in raid10_start_reshape()
4706 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4707 conf->geo = conf->prev; in raid10_start_reshape()
4708 mddev->raid_disks = conf->geo.raid_disks; in raid10_start_reshape()
4710 rdev->new_data_offset = rdev->data_offset; in raid10_start_reshape()
4712 conf->reshape_progress = MaxSector; in raid10_start_reshape()
4713 conf->reshape_safe = MaxSector; in raid10_start_reshape()
4714 mddev->reshape_position = MaxSector; in raid10_start_reshape()
4715 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4719 /* Calculate the last device-address that could contain
4720 * any block from the chunk that includes the array-address 's'
4722 * i.e. the address returned will be chunk-aligned and after
4727 s = (s | geo->chunk_mask) + 1; in last_dev_address()
4728 s >>= geo->chunk_shift; in last_dev_address()
4729 s *= geo->near_copies; in last_dev_address()
4730 s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); in last_dev_address()
4731 s *= geo->far_copies; in last_dev_address()
4732 s <<= geo->chunk_shift; in last_dev_address()
4736 /* Calculate the first device-address that could contain
4737 * any block from the chunk that includes the array-address 's'.
4742 s >>= geo->chunk_shift; in first_dev_address()
4743 s *= geo->near_copies; in first_dev_address()
4744 sector_div(s, geo->raid_disks); in first_dev_address()
4745 s *= geo->far_copies; in first_dev_address()
4746 s <<= geo->chunk_shift; in first_dev_address()
4757 * a chunk, but that is not a problem as mddev->reshape_position in reshape_request()
4762 * we need to flush all reshape requests and update the metadata. in reshape_request()
4773 * we must update the metadata to increase the 'safe' position. in reshape_request()
4780 * (conf->offset_diff - always positive) allows a bit of slack, in reshape_request()
4785 * The means one for each copy for write-out and an extra one for in reshape_request()
4786 * read-in. in reshape_request()
4787 * We store the read-in bio in ->master_bio and the others in in reshape_request()
4788 * ->devs[x].bio and ->devs[x].repl_bio. in reshape_request()
4790 struct r10conf *conf = mddev->private; in reshape_request()
4805 if (mddev->reshape_backwards && in reshape_request()
4806 conf->reshape_progress < raid10_size(mddev, 0, 0)) { in reshape_request()
4808 - conf->reshape_progress); in reshape_request()
4809 } else if (!mddev->reshape_backwards && in reshape_request()
4810 conf->reshape_progress > 0) in reshape_request()
4811 sector_nr = conf->reshape_progress; in reshape_request()
4813 mddev->curr_resync_completed = sector_nr; in reshape_request()
4814 sysfs_notify_dirent_safe(mddev->sysfs_completed); in reshape_request()
4821 * as that doesn't work well for ->reshape_backwards. in reshape_request()
4822 * So just use ->reshape_progress. in reshape_request()
4824 if (mddev->reshape_backwards) { in reshape_request()
4828 next = first_dev_address(conf->reshape_progress - 1, in reshape_request()
4829 &conf->geo); in reshape_request()
4834 safe = last_dev_address(conf->reshape_safe - 1, in reshape_request()
4835 &conf->prev); in reshape_request()
4837 if (next + conf->offset_diff < safe) in reshape_request()
4840 last = conf->reshape_progress - 1; in reshape_request()
4841 sector_nr = last & ~(sector_t)(conf->geo.chunk_mask in reshape_request()
4842 & conf->prev.chunk_mask); in reshape_request()
4844 sector_nr = last + 1 - RESYNC_SECTORS; in reshape_request()
4849 next = last_dev_address(conf->reshape_progress, &conf->geo); in reshape_request()
4854 safe = first_dev_address(conf->reshape_safe, &conf->prev); in reshape_request()
4856 /* Need to update metadata if 'next' might be beyond 'safe' in reshape_request()
4859 if (next > safe + conf->offset_diff) in reshape_request()
4862 sector_nr = conf->reshape_progress; in reshape_request()
4863 last = sector_nr | (conf->geo.chunk_mask in reshape_request()
4864 & conf->prev.chunk_mask); in reshape_request()
4867 last = sector_nr + RESYNC_SECTORS - 1; in reshape_request()
4871 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { in reshape_request()
4872 /* Need to update reshape_position in metadata */ in reshape_request()
4874 mddev->reshape_position = conf->reshape_progress; in reshape_request()
4875 if (mddev->reshape_backwards) in reshape_request()
4876 mddev->curr_resync_completed = raid10_size(mddev, 0, 0) in reshape_request()
4877 - conf->reshape_progress; in reshape_request()
4879 mddev->curr_resync_completed = conf->reshape_progress; in reshape_request()
4880 conf->reshape_checkpoint = jiffies; in reshape_request()
4881 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in reshape_request()
4882 md_wakeup_thread(mddev->thread); in reshape_request()
4883 wait_event(mddev->sb_wait, mddev->sb_flags == 0 || in reshape_request()
4884 test_bit(MD_RECOVERY_INTR, &mddev->recovery)); in reshape_request()
4885 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { in reshape_request()
4889 conf->reshape_safe = mddev->reshape_position; in reshape_request()
4897 r10_bio->state = 0; in reshape_request()
4899 atomic_set(&r10_bio->remaining, 0); in reshape_request()
4900 r10_bio->mddev = mddev; in reshape_request()
4901 r10_bio->sector = sector_nr; in reshape_request()
4902 set_bit(R10BIO_IsReshape, &r10_bio->state); in reshape_request()
4903 r10_bio->sectors = last - sector_nr + 1; in reshape_request()
4905 BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); in reshape_request()
4912 mempool_free(r10_bio, &conf->r10buf_pool); in reshape_request()
4913 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in reshape_request()
4917 read_bio = bio_alloc_bioset(rdev->bdev, RESYNC_PAGES, REQ_OP_READ, in reshape_request()
4918 GFP_KERNEL, &mddev->bio_set); in reshape_request()
4919 read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr in reshape_request()
4920 + rdev->data_offset); in reshape_request()
4921 read_bio->bi_private = r10_bio; in reshape_request()
4922 read_bio->bi_end_io = end_reshape_read; in reshape_request()
4923 r10_bio->master_bio = read_bio; in reshape_request()
4924 r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; in reshape_request()
4930 if (mddev_is_clustered(mddev) && conf->cluster_sync_high <= sector_nr) { in reshape_request()
4934 conf->cluster_sync_low = sector_nr; in reshape_request()
4935 conf->cluster_sync_high = sector_nr + CLUSTER_RESYNC_WINDOW_SECTORS; in reshape_request()
4936 sb = page_address(rdev->sb_page); in reshape_request()
4938 sb_reshape_pos = le64_to_cpu(sb->reshape_position); in reshape_request()
4942 * update cluster_sync_low until it has finished reshape. in reshape_request()
4944 if (sb_reshape_pos < conf->cluster_sync_low) in reshape_request()
4945 conf->cluster_sync_low = sb_reshape_pos; in reshape_request()
4948 md_cluster_ops->resync_info_update(mddev, conf->cluster_sync_low, in reshape_request()
4949 conf->cluster_sync_high); in reshape_request()
4953 __raid10_find_phys(&conf->geo, r10_bio); in reshape_request()
4956 read_bio->bi_next = NULL; in reshape_request()
4959 for (s = 0; s < conf->copies*2; s++) { in reshape_request()
4961 int d = r10_bio->devs[s/2].devnum; in reshape_request()
4964 rdev2 = rcu_dereference(conf->mirrors[d].replacement); in reshape_request()
4965 b = r10_bio->devs[s/2].repl_bio; in reshape_request()
4967 rdev2 = rcu_dereference(conf->mirrors[d].rdev); in reshape_request()
4968 b = r10_bio->devs[s/2].bio; in reshape_request()
4970 if (!rdev2 || test_bit(Faulty, &rdev2->flags)) in reshape_request()
4973 bio_set_dev(b, rdev2->bdev); in reshape_request()
4974 b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + in reshape_request()
4975 rdev2->new_data_offset; in reshape_request()
4976 b->bi_end_io = end_reshape_write; in reshape_request()
4977 b->bi_opf = REQ_OP_WRITE; in reshape_request()
4978 b->bi_next = blist; in reshape_request()
4985 pages = get_resync_pages(r10_bio->devs[0].bio)->pages; in reshape_request()
4988 int len = (max_sectors - s) << 9; in reshape_request()
4991 for (bio = blist; bio ; bio = bio->bi_next) { in reshape_request()
4993 bio->bi_status = BLK_STS_RESOURCE; in reshape_request()
5002 r10_bio->sectors = nr_sectors; in reshape_request()
5005 md_sync_acct_bio(read_bio, r10_bio->sectors); in reshape_request()
5006 atomic_inc(&r10_bio->remaining); in reshape_request()
5007 read_bio->bi_next = NULL; in reshape_request()
5016 * update reshape_progress in reshape_request()
5018 if (mddev->reshape_backwards) in reshape_request()
5019 conf->reshape_progress -= sectors_done; in reshape_request()
5021 conf->reshape_progress += sectors_done; in reshape_request()
5033 * If we got a read error then we do sync 1-page reads from in reshape_request_write()
5034 * elsewhere until we find the data - or give up. in reshape_request_write()
5036 struct r10conf *conf = mddev->private; in reshape_request_write()
5039 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in reshape_request_write()
5042 md_done_sync(mddev, r10_bio->sectors, 0); in reshape_request_write()
5049 atomic_set(&r10_bio->remaining, 1); in reshape_request_write()
5050 for (s = 0; s < conf->copies*2; s++) { in reshape_request_write()
5052 int d = r10_bio->devs[s/2].devnum; in reshape_request_write()
5056 rdev = rcu_dereference(conf->mirrors[d].replacement); in reshape_request_write()
5057 b = r10_bio->devs[s/2].repl_bio; in reshape_request_write()
5059 rdev = rcu_dereference(conf->mirrors[d].rdev); in reshape_request_write()
5060 b = r10_bio->devs[s/2].bio; in reshape_request_write()
5062 if (!rdev || test_bit(Faulty, &rdev->flags)) { in reshape_request_write()
5066 atomic_inc(&rdev->nr_pending); in reshape_request_write()
5068 md_sync_acct_bio(b, r10_bio->sectors); in reshape_request_write()
5069 atomic_inc(&r10_bio->remaining); in reshape_request_write()
5070 b->bi_next = NULL; in reshape_request_write()
5078 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) in end_reshape()
5081 spin_lock_irq(&conf->device_lock); in end_reshape()
5082 conf->prev = conf->geo; in end_reshape()
5083 md_finish_reshape(conf->mddev); in end_reshape()
5085 conf->reshape_progress = MaxSector; in end_reshape()
5086 conf->reshape_safe = MaxSector; in end_reshape()
5087 spin_unlock_irq(&conf->device_lock); in end_reshape()
5089 if (conf->mddev->queue) in end_reshape()
5091 conf->fullsync = 0; in end_reshape()
5096 struct r10conf *conf = mddev->private; in raid10_update_reshape_pos()
5099 md_cluster_ops->resync_info_get(mddev, &lo, &hi); in raid10_update_reshape_pos()
5100 if (((mddev->reshape_position <= hi) && (mddev->reshape_position >= lo)) in raid10_update_reshape_pos()
5101 || mddev->reshape_position == MaxSector) in raid10_update_reshape_pos()
5102 conf->reshape_progress = mddev->reshape_position; in raid10_update_reshape_pos()
5111 int sectors = r10_bio->sectors; in handle_reshape_read_error()
5112 struct r10conf *conf = mddev->private; in handle_reshape_read_error()
5118 r10b = kmalloc(struct_size(r10b, devs, conf->copies), GFP_NOIO); in handle_reshape_read_error()
5120 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in handle_reshape_read_error()
5121 return -ENOMEM; in handle_reshape_read_error()
5125 pages = get_resync_pages(r10_bio->devs[0].bio)->pages; in handle_reshape_read_error()
5127 r10b->sector = r10_bio->sector; in handle_reshape_read_error()
5128 __raid10_find_phys(&conf->prev, r10b); in handle_reshape_read_error()
5140 int d = r10b->devs[slot].devnum; in handle_reshape_read_error()
5141 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); in handle_reshape_read_error()
5144 test_bit(Faulty, &rdev->flags) || in handle_reshape_read_error()
5145 !test_bit(In_sync, &rdev->flags)) in handle_reshape_read_error()
5148 addr = r10b->devs[slot].addr + idx * PAGE_SIZE; in handle_reshape_read_error()
5149 atomic_inc(&rdev->nr_pending); in handle_reshape_read_error()
5162 if (slot >= conf->copies) in handle_reshape_read_error()
5171 &mddev->recovery); in handle_reshape_read_error()
5173 return -EIO; in handle_reshape_read_error()
5175 sectors -= s; in handle_reshape_read_error()
5185 struct mddev *mddev = r10_bio->mddev; in end_reshape_write()
5186 struct r10conf *conf = mddev->private; in end_reshape_write()
5194 rdev = conf->mirrors[d].replacement; in end_reshape_write()
5197 rdev = conf->mirrors[d].rdev; in end_reshape_write()
5200 if (bio->bi_status) { in end_reshape_write()
5211 if (!atomic_dec_and_test(&r10_bio->remaining)) in end_reshape_request()
5213 md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); in end_reshape_request()
5214 bio_put(r10_bio->master_bio); in end_reshape_request()
5220 struct r10conf *conf = mddev->private; in raid10_finish_reshape()
5222 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) in raid10_finish_reshape()
5225 if (mddev->delta_disks > 0) { in raid10_finish_reshape()
5226 if (mddev->recovery_cp > mddev->resync_max_sectors) { in raid10_finish_reshape()
5227 mddev->recovery_cp = mddev->resync_max_sectors; in raid10_finish_reshape()
5228 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_finish_reshape()
5230 mddev->resync_max_sectors = mddev->array_sectors; in raid10_finish_reshape()
5234 for (d = conf->geo.raid_disks ; in raid10_finish_reshape()
5235 d < conf->geo.raid_disks - mddev->delta_disks; in raid10_finish_reshape()
5237 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); in raid10_finish_reshape()
5239 clear_bit(In_sync, &rdev->flags); in raid10_finish_reshape()
5240 rdev = rcu_dereference(conf->mirrors[d].replacement); in raid10_finish_reshape()
5242 clear_bit(In_sync, &rdev->flags); in raid10_finish_reshape()
5246 mddev->layout = mddev->new_layout; in raid10_finish_reshape()
5247 mddev->chunk_sectors = 1 << conf->geo.chunk_shift; in raid10_finish_reshape()
5248 mddev->reshape_position = MaxSector; in raid10_finish_reshape()
5249 mddev->delta_disks = 0; in raid10_finish_reshape()
5250 mddev->reshape_backwards = 0; in raid10_finish_reshape()
5291 MODULE_ALIAS("md-personality-9"); /* RAID10 */
5292 MODULE_ALIAS("md-raid10");
5293 MODULE_ALIAS("md-level-10");