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()
1218 (unsigned long long)r10_bio->sector); in raid10_read_request()
1226 rdev->bdev, in raid10_read_request()
1227 (unsigned long long)r10_bio->sector); in raid10_read_request()
1230 gfp, &conf->bio_split); in raid10_read_request()
1236 r10_bio->master_bio = bio; in raid10_read_request()
1237 r10_bio->sectors = max_sectors; in raid10_read_request()
1239 slot = r10_bio->read_slot; in raid10_read_request()
1243 r10_bio->master_bio = bio; in raid10_read_request()
1245 read_bio = bio_alloc_clone(rdev->bdev, bio, gfp, &mddev->bio_set); in raid10_read_request()
1247 r10_bio->devs[slot].bio = read_bio; in raid10_read_request()
1248 r10_bio->devs[slot].rdev = rdev; in raid10_read_request()
1250 read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + in raid10_read_request()
1252 read_bio->bi_end_io = raid10_end_read_request; in raid10_read_request()
1253 read_bio->bi_opf = op | do_sync; in raid10_read_request()
1254 if (test_bit(FailFast, &rdev->flags) && in raid10_read_request()
1255 test_bit(R10BIO_FailFast, &r10_bio->state)) in raid10_read_request()
1256 read_bio->bi_opf |= MD_FAILFAST; in raid10_read_request()
1257 read_bio->bi_private = r10_bio; in raid10_read_request()
1259 if (mddev->gendisk) in raid10_read_request()
1260 trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), in raid10_read_request()
1261 r10_bio->sector); in raid10_read_request()
1271 const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; in raid10_write_one_disk()
1272 const blk_opf_t do_fua = bio->bi_opf & REQ_FUA; in raid10_write_one_disk()
1274 struct r10conf *conf = mddev->private; in raid10_write_one_disk()
1276 int devnum = r10_bio->devs[n_copy].devnum; in raid10_write_one_disk()
1280 rdev = conf->mirrors[devnum].replacement; in raid10_write_one_disk()
1284 rdev = conf->mirrors[devnum].rdev; in raid10_write_one_disk()
1287 rdev = conf->mirrors[devnum].rdev; in raid10_write_one_disk()
1289 mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, &mddev->bio_set); in raid10_write_one_disk()
1291 r10_bio->devs[n_copy].repl_bio = mbio; in raid10_write_one_disk()
1293 r10_bio->devs[n_copy].bio = mbio; in raid10_write_one_disk()
1295 mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + in raid10_write_one_disk()
1297 mbio->bi_end_io = raid10_end_write_request; in raid10_write_one_disk()
1298 mbio->bi_opf = op | do_sync | do_fua; in raid10_write_one_disk()
1300 &conf->mirrors[devnum].rdev->flags) in raid10_write_one_disk()
1302 mbio->bi_opf |= MD_FAILFAST; in raid10_write_one_disk()
1303 mbio->bi_private = r10_bio; in raid10_write_one_disk()
1305 if (conf->mddev->gendisk) in raid10_write_one_disk()
1306 trace_block_bio_remap(mbio, disk_devt(conf->mddev->gendisk), in raid10_write_one_disk()
1307 r10_bio->sector); in raid10_write_one_disk()
1309 mbio->bi_bdev = (void *)rdev; in raid10_write_one_disk()
1311 atomic_inc(&r10_bio->remaining); in raid10_write_one_disk()
1313 if (!raid1_add_bio_to_plug(mddev, mbio, raid10_unplug, conf->copies)) { in raid10_write_one_disk()
1314 spin_lock_irqsave(&conf->device_lock, flags); in raid10_write_one_disk()
1315 bio_list_add(&conf->pending_bio_list, mbio); in raid10_write_one_disk()
1316 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_write_one_disk()
1317 md_wakeup_thread(mddev->thread); in raid10_write_one_disk()
1326 rrdev = rcu_dereference(mirror->replacement); in dereference_rdev_and_rrdev()
1332 rdev = rcu_dereference(mirror->rdev); in dereference_rdev_and_rrdev()
1343 struct r10conf *conf = mddev->private; in wait_blocked_dev()
1349 for (i = 0; i < conf->copies; i++) { in wait_blocked_dev()
1352 rdev = dereference_rdev_and_rrdev(&conf->mirrors[i], &rrdev); in wait_blocked_dev()
1353 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { in wait_blocked_dev()
1354 atomic_inc(&rdev->nr_pending); in wait_blocked_dev()
1358 if (rrdev && unlikely(test_bit(Blocked, &rrdev->flags))) { in wait_blocked_dev()
1359 atomic_inc(&rrdev->nr_pending); in wait_blocked_dev()
1364 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { in wait_blocked_dev()
1366 sector_t dev_sector = r10_bio->devs[i].addr; in wait_blocked_dev()
1374 if (!r10_bio->sectors) in wait_blocked_dev()
1377 is_bad = is_badblock(rdev, dev_sector, r10_bio->sectors, in wait_blocked_dev()
1384 atomic_inc(&rdev->nr_pending); in wait_blocked_dev()
1385 set_bit(BlockedBadBlocks, &rdev->flags); in wait_blocked_dev()
1396 raid10_log(conf->mddev, "%s wait rdev %d blocked", in wait_blocked_dev()
1397 __func__, blocked_rdev->raid_disk); in wait_blocked_dev()
1407 struct r10conf *conf = mddev->private; in raid10_write_request()
1413 md_cluster_ops->area_resyncing(mddev, WRITE, in raid10_write_request()
1414 bio->bi_iter.bi_sector, in raid10_write_request()
1418 if (bio->bi_opf & REQ_NOWAIT) { in raid10_write_request()
1423 prepare_to_wait(&conf->wait_barrier, in raid10_write_request()
1425 if (!md_cluster_ops->area_resyncing(mddev, WRITE, in raid10_write_request()
1426 bio->bi_iter.bi_sector, bio_end_sector(bio))) in raid10_write_request()
1430 finish_wait(&conf->wait_barrier, &w); in raid10_write_request()
1433 sectors = r10_bio->sectors; in raid10_write_request()
1439 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in raid10_write_request()
1440 (mddev->reshape_backwards in raid10_write_request()
1441 ? (bio->bi_iter.bi_sector < conf->reshape_safe && in raid10_write_request()
1442 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) in raid10_write_request()
1443 : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && in raid10_write_request()
1444 bio->bi_iter.bi_sector < conf->reshape_progress))) { in raid10_write_request()
1445 /* Need to update reshape_position in metadata */ in raid10_write_request()
1446 mddev->reshape_position = conf->reshape_progress; in raid10_write_request()
1447 set_mask_bits(&mddev->sb_flags, 0, in raid10_write_request()
1449 md_wakeup_thread(mddev->thread); in raid10_write_request()
1450 if (bio->bi_opf & REQ_NOWAIT) { in raid10_write_request()
1455 raid10_log(conf->mddev, "wait reshape metadata"); in raid10_write_request()
1456 wait_event(mddev->sb_wait, in raid10_write_request()
1457 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); in raid10_write_request()
1459 conf->reshape_safe = mddev->reshape_position; in raid10_write_request()
1472 r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ in raid10_write_request()
1478 max_sectors = r10_bio->sectors; in raid10_write_request()
1480 for (i = 0; i < conf->copies; i++) { in raid10_write_request()
1481 int d = r10_bio->devs[i].devnum; in raid10_write_request()
1484 rdev = dereference_rdev_and_rrdev(&conf->mirrors[d], &rrdev); in raid10_write_request()
1485 if (rdev && (test_bit(Faulty, &rdev->flags))) in raid10_write_request()
1487 if (rrdev && (test_bit(Faulty, &rrdev->flags))) in raid10_write_request()
1490 r10_bio->devs[i].bio = NULL; in raid10_write_request()
1491 r10_bio->devs[i].repl_bio = NULL; in raid10_write_request()
1495 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { in raid10_write_request()
1497 sector_t dev_sector = r10_bio->devs[i].addr; in raid10_write_request()
1505 bad_sectors -= (dev_sector - first_bad); in raid10_write_request()
1514 int good_sectors = first_bad - dev_sector; in raid10_write_request()
1520 r10_bio->devs[i].bio = bio; in raid10_write_request()
1521 atomic_inc(&rdev->nr_pending); in raid10_write_request()
1524 r10_bio->devs[i].repl_bio = bio; in raid10_write_request()
1525 atomic_inc(&rrdev->nr_pending); in raid10_write_request()
1530 if (max_sectors < r10_bio->sectors) in raid10_write_request()
1531 r10_bio->sectors = max_sectors; in raid10_write_request()
1533 if (r10_bio->sectors < bio_sectors(bio)) { in raid10_write_request()
1534 struct bio *split = bio_split(bio, r10_bio->sectors, in raid10_write_request()
1535 GFP_NOIO, &conf->bio_split); in raid10_write_request()
1541 r10_bio->master_bio = bio; in raid10_write_request()
1545 r10_bio->master_bio = bio; in raid10_write_request()
1546 atomic_set(&r10_bio->remaining, 1); in raid10_write_request()
1548 for (i = 0; i < conf->copies; i++) { in raid10_write_request()
1549 if (r10_bio->devs[i].bio) in raid10_write_request()
1551 if (r10_bio->devs[i].repl_bio) in raid10_write_request()
1559 struct r10conf *conf = mddev->private; in __make_request()
1562 r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); in __make_request()
1564 r10_bio->master_bio = bio; in __make_request()
1565 r10_bio->sectors = sectors; in __make_request()
1567 r10_bio->mddev = mddev; in __make_request()
1568 r10_bio->sector = bio->bi_iter.bi_sector; in __make_request()
1569 r10_bio->state = 0; in __make_request()
1570 r10_bio->read_slot = -1; in __make_request()
1571 memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * in __make_request()
1572 conf->geo.raid_disks); in __make_request()
1582 struct r10conf *conf = r10bio->mddev->private; in raid_end_discard_bio()
1585 while (atomic_dec_and_test(&r10bio->remaining)) { in raid_end_discard_bio()
1589 if (!test_bit(R10BIO_Discard, &r10bio->state)) { in raid_end_discard_bio()
1590 first_r10bio = (struct r10bio *)r10bio->master_bio; in raid_end_discard_bio()
1594 md_write_end(r10bio->mddev); in raid_end_discard_bio()
1595 bio_endio(r10bio->master_bio); in raid_end_discard_bio()
1604 struct r10bio *r10_bio = bio->bi_private; in raid10_end_discard_request()
1605 struct r10conf *conf = r10_bio->mddev->private; in raid10_end_discard_request()
1613 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in raid10_end_discard_request()
1614 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_end_discard_request()
1618 rdev = conf->mirrors[dev].replacement; in raid10_end_discard_request()
1626 rdev = conf->mirrors[dev].rdev; in raid10_end_discard_request()
1630 rdev_dec_pending(rdev, conf->mddev); in raid10_end_discard_request()
1641 struct r10conf *conf = mddev->private; in raid10_handle_discard()
1642 struct geom *geo = &conf->geo; in raid10_handle_discard()
1643 int far_copies = geo->far_copies; in raid10_handle_discard()
1660 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_handle_discard()
1661 return -EAGAIN; in raid10_handle_discard()
1663 if (WARN_ON_ONCE(bio->bi_opf & REQ_NOWAIT)) { in raid10_handle_discard()
1673 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_handle_discard()
1676 if (geo->near_copies) in raid10_handle_discard()
1677 stripe_data_disks = geo->raid_disks / geo->near_copies + in raid10_handle_discard()
1678 geo->raid_disks % geo->near_copies; in raid10_handle_discard()
1680 stripe_data_disks = geo->raid_disks; in raid10_handle_discard()
1682 stripe_size = stripe_data_disks << geo->chunk_shift; in raid10_handle_discard()
1684 bio_start = bio->bi_iter.bi_sector; in raid10_handle_discard()
1703 split_size = stripe_size - remainder; in raid10_handle_discard()
1704 split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); in raid10_handle_discard()
1713 split_size = bio_sectors(bio) - remainder; in raid10_handle_discard()
1714 split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); in raid10_handle_discard()
1723 bio_start = bio->bi_iter.bi_sector; in raid10_handle_discard()
1731 chunk = bio_start >> geo->chunk_shift; in raid10_handle_discard()
1732 chunk *= geo->near_copies; in raid10_handle_discard()
1734 start_disk_index = sector_div(first_stripe_index, geo->raid_disks); in raid10_handle_discard()
1735 if (geo->far_offset) in raid10_handle_discard()
1736 first_stripe_index *= geo->far_copies; in raid10_handle_discard()
1737 start_disk_offset = (bio_start & geo->chunk_mask) + in raid10_handle_discard()
1738 (first_stripe_index << geo->chunk_shift); in raid10_handle_discard()
1740 chunk = bio_end >> geo->chunk_shift; in raid10_handle_discard()
1741 chunk *= geo->near_copies; in raid10_handle_discard()
1743 end_disk_index = sector_div(last_stripe_index, geo->raid_disks); in raid10_handle_discard()
1744 if (geo->far_offset) in raid10_handle_discard()
1745 last_stripe_index *= geo->far_copies; in raid10_handle_discard()
1746 end_disk_offset = (bio_end & geo->chunk_mask) + in raid10_handle_discard()
1747 (last_stripe_index << geo->chunk_shift); in raid10_handle_discard()
1750 r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); in raid10_handle_discard()
1751 r10_bio->mddev = mddev; in raid10_handle_discard()
1752 r10_bio->state = 0; in raid10_handle_discard()
1753 r10_bio->sectors = 0; in raid10_handle_discard()
1754 memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * geo->raid_disks); in raid10_handle_discard()
1759 * Inspired by raid10_sync_request, we can use the first r10bio->master_bio in raid10_handle_discard()
1760 * to record the discard bio. Other r10bio->master_bio record the first in raid10_handle_discard()
1766 r10_bio->master_bio = bio; in raid10_handle_discard()
1767 set_bit(R10BIO_Discard, &r10_bio->state); in raid10_handle_discard()
1771 r10_bio->master_bio = (struct bio *)first_r10bio; in raid10_handle_discard()
1779 for (disk = 0; disk < geo->raid_disks; disk++) { in raid10_handle_discard()
1782 rdev = dereference_rdev_and_rrdev(&conf->mirrors[disk], &rrdev); in raid10_handle_discard()
1783 r10_bio->devs[disk].bio = NULL; in raid10_handle_discard()
1784 r10_bio->devs[disk].repl_bio = NULL; in raid10_handle_discard()
1786 if (rdev && (test_bit(Faulty, &rdev->flags))) in raid10_handle_discard()
1788 if (rrdev && (test_bit(Faulty, &rrdev->flags))) in raid10_handle_discard()
1794 r10_bio->devs[disk].bio = bio; in raid10_handle_discard()
1795 atomic_inc(&rdev->nr_pending); in raid10_handle_discard()
1798 r10_bio->devs[disk].repl_bio = bio; in raid10_handle_discard()
1799 atomic_inc(&rrdev->nr_pending); in raid10_handle_discard()
1804 atomic_set(&r10_bio->remaining, 1); in raid10_handle_discard()
1805 for (disk = 0; disk < geo->raid_disks; disk++) { in raid10_handle_discard()
1822 dev_start = (first_stripe_index + 1) * mddev->chunk_sectors; in raid10_handle_discard()
1824 dev_start = first_stripe_index * mddev->chunk_sectors; in raid10_handle_discard()
1829 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; in raid10_handle_discard()
1831 dev_end = last_stripe_index * mddev->chunk_sectors; in raid10_handle_discard()
1839 * add rdev->nr_pending in the first loop. in raid10_handle_discard()
1841 if (r10_bio->devs[disk].bio) { in raid10_handle_discard()
1842 struct md_rdev *rdev = conf->mirrors[disk].rdev; in raid10_handle_discard()
1843 mbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, in raid10_handle_discard()
1844 &mddev->bio_set); in raid10_handle_discard()
1845 mbio->bi_end_io = raid10_end_discard_request; in raid10_handle_discard()
1846 mbio->bi_private = r10_bio; in raid10_handle_discard()
1847 r10_bio->devs[disk].bio = mbio; in raid10_handle_discard()
1848 r10_bio->devs[disk].devnum = disk; in raid10_handle_discard()
1849 atomic_inc(&r10_bio->remaining); in raid10_handle_discard()
1852 dev_end - dev_start); in raid10_handle_discard()
1855 if (r10_bio->devs[disk].repl_bio) { in raid10_handle_discard()
1856 struct md_rdev *rrdev = conf->mirrors[disk].replacement; in raid10_handle_discard()
1857 rbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, in raid10_handle_discard()
1858 &mddev->bio_set); in raid10_handle_discard()
1859 rbio->bi_end_io = raid10_end_discard_request; in raid10_handle_discard()
1860 rbio->bi_private = r10_bio; in raid10_handle_discard()
1861 r10_bio->devs[disk].repl_bio = rbio; in raid10_handle_discard()
1862 r10_bio->devs[disk].devnum = disk; in raid10_handle_discard()
1863 atomic_inc(&r10_bio->remaining); in raid10_handle_discard()
1866 dev_end - dev_start); in raid10_handle_discard()
1871 if (!geo->far_offset && --far_copies) { in raid10_handle_discard()
1872 first_stripe_index += geo->stride >> geo->chunk_shift; in raid10_handle_discard()
1873 start_disk_offset += geo->stride; in raid10_handle_discard()
1874 last_stripe_index += geo->stride >> geo->chunk_shift; in raid10_handle_discard()
1875 end_disk_offset += geo->stride; in raid10_handle_discard()
1876 atomic_inc(&first_r10bio->remaining); in raid10_handle_discard()
1887 return -EAGAIN; in raid10_handle_discard()
1892 struct r10conf *conf = mddev->private; in raid10_make_request()
1893 sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); in raid10_make_request()
1897 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid10_make_request()
1912 if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + in raid10_make_request()
1914 && (conf->geo.near_copies < conf->geo.raid_disks in raid10_make_request()
1915 || conf->prev.near_copies < in raid10_make_request()
1916 conf->prev.raid_disks))) in raid10_make_request()
1917 sectors = chunk_sects - in raid10_make_request()
1918 (bio->bi_iter.bi_sector & in raid10_make_request()
1919 (chunk_sects - 1)); in raid10_make_request()
1929 struct r10conf *conf = mddev->private; in raid10_status()
1932 if (conf->geo.near_copies < conf->geo.raid_disks) in raid10_status()
1933 seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); in raid10_status()
1934 if (conf->geo.near_copies > 1) in raid10_status()
1935 seq_printf(seq, " %d near-copies", conf->geo.near_copies); in raid10_status()
1936 if (conf->geo.far_copies > 1) { in raid10_status()
1937 if (conf->geo.far_offset) in raid10_status()
1938 seq_printf(seq, " %d offset-copies", conf->geo.far_copies); in raid10_status()
1940 seq_printf(seq, " %d far-copies", conf->geo.far_copies); in raid10_status()
1941 if (conf->geo.far_set_size != conf->geo.raid_disks) in raid10_status()
1942 seq_printf(seq, " %d devices per set", conf->geo.far_set_size); in raid10_status()
1944 seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, in raid10_status()
1945 conf->geo.raid_disks - mddev->degraded); in raid10_status()
1947 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_status()
1948 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in raid10_status()
1949 seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid10_status()
1966 disks = conf->prev.raid_disks; in _enough()
1967 ncopies = conf->prev.near_copies; in _enough()
1969 disks = conf->geo.raid_disks; in _enough()
1970 ncopies = conf->geo.near_copies; in _enough()
1975 int n = conf->copies; in _enough()
1978 while (n--) { in _enough()
1981 (rdev = rcu_dereference(conf->mirrors[this].rdev)) && in _enough()
1982 test_bit(In_sync, &rdev->flags)) in _enough()
2000 * This is ensured if ->reconfig_mutex or ->device_lock in enough()
2008 * raid10_error() - RAID10 error handler.
2014 * - &MD_BROKEN flag is set in &mddev->flags.
2016 * - recovery is interrupted.
2017 * - &mddev->degraded is bumped.
2020 * &mddev->fail_last_dev is off.
2024 struct r10conf *conf = mddev->private; in raid10_error()
2027 spin_lock_irqsave(&conf->device_lock, flags); in raid10_error()
2029 if (test_bit(In_sync, &rdev->flags) && !enough(conf, rdev->raid_disk)) { in raid10_error()
2030 set_bit(MD_BROKEN, &mddev->flags); in raid10_error()
2032 if (!mddev->fail_last_dev) { in raid10_error()
2033 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_error()
2037 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid10_error()
2038 mddev->degraded++; in raid10_error()
2040 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid10_error()
2041 set_bit(Blocked, &rdev->flags); in raid10_error()
2042 set_bit(Faulty, &rdev->flags); in raid10_error()
2043 set_mask_bits(&mddev->sb_flags, 0, in raid10_error()
2045 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_error()
2048 mdname(mddev), rdev->bdev, in raid10_error()
2049 mdname(mddev), conf->geo.raid_disks - mddev->degraded); in raid10_error()
2062 pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, in print_conf()
2063 conf->geo.raid_disks); in print_conf()
2065 /* This is only called with ->reconfix_mutex held, so in print_conf()
2067 for (i = 0; i < conf->geo.raid_disks; i++) { in print_conf()
2068 rdev = conf->mirrors[i].rdev; in print_conf()
2071 i, !test_bit(In_sync, &rdev->flags), in print_conf()
2072 !test_bit(Faulty, &rdev->flags), in print_conf()
2073 rdev->bdev); in print_conf()
2082 mempool_exit(&conf->r10buf_pool); in close_sync()
2088 struct r10conf *conf = mddev->private; in raid10_spare_active()
2094 * Find all non-in_sync disks within the RAID10 configuration in raid10_spare_active()
2097 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_spare_active()
2098 tmp = conf->mirrors + i; in raid10_spare_active()
2099 if (tmp->replacement in raid10_spare_active()
2100 && tmp->replacement->recovery_offset == MaxSector in raid10_spare_active()
2101 && !test_bit(Faulty, &tmp->replacement->flags) in raid10_spare_active()
2102 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { in raid10_spare_active()
2104 if (!tmp->rdev in raid10_spare_active()
2105 || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) in raid10_spare_active()
2107 if (tmp->rdev) { in raid10_spare_active()
2110 * and never re-added. in raid10_spare_active()
2112 set_bit(Faulty, &tmp->rdev->flags); in raid10_spare_active()
2114 tmp->rdev->sysfs_state); in raid10_spare_active()
2116 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); in raid10_spare_active()
2117 } else if (tmp->rdev in raid10_spare_active()
2118 && tmp->rdev->recovery_offset == MaxSector in raid10_spare_active()
2119 && !test_bit(Faulty, &tmp->rdev->flags) in raid10_spare_active()
2120 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { in raid10_spare_active()
2122 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); in raid10_spare_active()
2125 spin_lock_irqsave(&conf->device_lock, flags); in raid10_spare_active()
2126 mddev->degraded -= count; in raid10_spare_active()
2127 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_spare_active()
2135 struct r10conf *conf = mddev->private; in raid10_add_disk()
2136 int err = -EEXIST; in raid10_add_disk()
2137 int mirror, repl_slot = -1; in raid10_add_disk()
2139 int last = conf->geo.raid_disks - 1; in raid10_add_disk()
2142 if (mddev->recovery_cp < MaxSector) in raid10_add_disk()
2143 /* only hot-add to in-sync arrays, as recovery is in raid10_add_disk()
2146 return -EBUSY; in raid10_add_disk()
2147 if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) in raid10_add_disk()
2148 return -EINVAL; in raid10_add_disk()
2151 return -ENXIO; in raid10_add_disk()
2153 if (rdev->raid_disk >= 0) in raid10_add_disk()
2154 first = last = rdev->raid_disk; in raid10_add_disk()
2156 if (rdev->saved_raid_disk >= first && in raid10_add_disk()
2157 rdev->saved_raid_disk < conf->geo.raid_disks && in raid10_add_disk()
2158 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid10_add_disk()
2159 mirror = rdev->saved_raid_disk; in raid10_add_disk()
2163 p = &conf->mirrors[mirror]; in raid10_add_disk()
2164 if (p->recovery_disabled == mddev->recovery_disabled) in raid10_add_disk()
2166 if (p->rdev) { in raid10_add_disk()
2167 if (test_bit(WantReplacement, &p->rdev->flags) && in raid10_add_disk()
2168 p->replacement == NULL && repl_slot < 0) in raid10_add_disk()
2173 if (mddev->gendisk) in raid10_add_disk()
2174 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid10_add_disk()
2175 rdev->data_offset << 9); in raid10_add_disk()
2177 p->head_position = 0; in raid10_add_disk()
2178 p->recovery_disabled = mddev->recovery_disabled - 1; in raid10_add_disk()
2179 rdev->raid_disk = mirror; in raid10_add_disk()
2181 if (rdev->saved_raid_disk != mirror) in raid10_add_disk()
2182 conf->fullsync = 1; in raid10_add_disk()
2183 rcu_assign_pointer(p->rdev, rdev); in raid10_add_disk()
2188 p = &conf->mirrors[repl_slot]; in raid10_add_disk()
2189 clear_bit(In_sync, &rdev->flags); in raid10_add_disk()
2190 set_bit(Replacement, &rdev->flags); in raid10_add_disk()
2191 rdev->raid_disk = repl_slot; in raid10_add_disk()
2193 if (mddev->gendisk) in raid10_add_disk()
2194 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid10_add_disk()
2195 rdev->data_offset << 9); in raid10_add_disk()
2196 conf->fullsync = 1; in raid10_add_disk()
2197 rcu_assign_pointer(p->replacement, rdev); in raid10_add_disk()
2206 struct r10conf *conf = mddev->private; in raid10_remove_disk()
2208 int number = rdev->raid_disk; in raid10_remove_disk()
2213 if (unlikely(number >= mddev->raid_disks)) in raid10_remove_disk()
2215 p = conf->mirrors + number; in raid10_remove_disk()
2216 if (rdev == p->rdev) in raid10_remove_disk()
2217 rdevp = &p->rdev; in raid10_remove_disk()
2218 else if (rdev == p->replacement) in raid10_remove_disk()
2219 rdevp = &p->replacement; in raid10_remove_disk()
2223 if (test_bit(In_sync, &rdev->flags) || in raid10_remove_disk()
2224 atomic_read(&rdev->nr_pending)) { in raid10_remove_disk()
2225 err = -EBUSY; in raid10_remove_disk()
2228 /* Only remove non-faulty devices if recovery in raid10_remove_disk()
2231 if (!test_bit(Faulty, &rdev->flags) && in raid10_remove_disk()
2232 mddev->recovery_disabled != p->recovery_disabled && in raid10_remove_disk()
2233 (!p->replacement || p->replacement == rdev) && in raid10_remove_disk()
2234 number < conf->geo.raid_disks && in raid10_remove_disk()
2235 enough(conf, -1)) { in raid10_remove_disk()
2236 err = -EBUSY; in raid10_remove_disk()
2240 if (!test_bit(RemoveSynchronized, &rdev->flags)) { in raid10_remove_disk()
2242 if (atomic_read(&rdev->nr_pending)) { in raid10_remove_disk()
2244 err = -EBUSY; in raid10_remove_disk()
2249 if (p->replacement) { in raid10_remove_disk()
2251 p->rdev = p->replacement; in raid10_remove_disk()
2252 clear_bit(Replacement, &p->replacement->flags); in raid10_remove_disk()
2254 * but will never see neither -- if they are careful. in raid10_remove_disk()
2256 p->replacement = NULL; in raid10_remove_disk()
2259 clear_bit(WantReplacement, &rdev->flags); in raid10_remove_disk()
2270 struct r10conf *conf = r10_bio->mddev->private; in __end_sync_read()
2272 if (!bio->bi_status) in __end_sync_read()
2273 set_bit(R10BIO_Uptodate, &r10_bio->state); in __end_sync_read()
2278 atomic_add(r10_bio->sectors, in __end_sync_read()
2279 &conf->mirrors[d].rdev->corrected_errors); in __end_sync_read()
2284 rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); in __end_sync_read()
2285 if (test_bit(R10BIO_IsRecover, &r10_bio->state) || in __end_sync_read()
2286 atomic_dec_and_test(&r10_bio->remaining)) { in __end_sync_read()
2297 struct r10conf *conf = r10_bio->mddev->private; in end_sync_read()
2306 struct r10bio *r10_bio = bio->bi_private; in end_reshape_read()
2308 __end_sync_read(r10_bio, bio, r10_bio->read_slot); in end_reshape_read()
2313 struct mddev *mddev = r10_bio->mddev; in end_sync_request()
2315 while (atomic_dec_and_test(&r10_bio->remaining)) { in end_sync_request()
2316 if (r10_bio->master_bio == NULL) { in end_sync_request()
2318 sector_t s = r10_bio->sectors; in end_sync_request()
2319 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in end_sync_request()
2320 test_bit(R10BIO_WriteError, &r10_bio->state)) in end_sync_request()
2327 struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; in end_sync_request()
2328 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in end_sync_request()
2329 test_bit(R10BIO_WriteError, &r10_bio->state)) in end_sync_request()
2341 struct mddev *mddev = r10_bio->mddev; in end_sync_write()
2342 struct r10conf *conf = mddev->private; in end_sync_write()
2352 rdev = conf->mirrors[d].replacement; in end_sync_write()
2354 rdev = conf->mirrors[d].rdev; in end_sync_write()
2356 if (bio->bi_status) { in end_sync_write()
2360 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
2361 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
2363 &rdev->mddev->recovery); in end_sync_write()
2364 set_bit(R10BIO_WriteError, &r10_bio->state); in end_sync_write()
2367 r10_bio->devs[slot].addr, in end_sync_write()
2368 r10_bio->sectors, in end_sync_write()
2370 set_bit(R10BIO_MadeGood, &r10_bio->state); in end_sync_write()
2390 * We check if all blocks are in-sync and only write to blocks that
2395 struct r10conf *conf = mddev->private; in sync_request_write()
2401 atomic_set(&r10_bio->remaining, 1); in sync_request_write()
2404 for (i=0; i<conf->copies; i++) in sync_request_write()
2405 if (!r10_bio->devs[i].bio->bi_status) in sync_request_write()
2408 if (i == conf->copies) in sync_request_write()
2412 fbio = r10_bio->devs[i].bio; in sync_request_write()
2413 fbio->bi_iter.bi_size = r10_bio->sectors << 9; in sync_request_write()
2414 fbio->bi_iter.bi_idx = 0; in sync_request_write()
2415 fpages = get_resync_pages(fbio)->pages; in sync_request_write()
2417 vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); in sync_request_write()
2419 for (i=0 ; i < conf->copies ; i++) { in sync_request_write()
2424 tbio = r10_bio->devs[i].bio; in sync_request_write()
2426 if (tbio->bi_end_io != end_sync_read) in sync_request_write()
2431 tpages = get_resync_pages(tbio)->pages; in sync_request_write()
2432 d = r10_bio->devs[i].devnum; in sync_request_write()
2433 rdev = conf->mirrors[d].rdev; in sync_request_write()
2434 if (!r10_bio->devs[i].bio->bi_status) { in sync_request_write()
2439 int sectors = r10_bio->sectors; in sync_request_write()
2448 sectors -= len/512; in sync_request_write()
2452 atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); in sync_request_write()
2453 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) in sync_request_write()
2456 } else if (test_bit(FailFast, &rdev->flags)) { in sync_request_write()
2458 md_error(rdev->mddev, rdev); in sync_request_write()
2463 * First we need to fixup bv_offset, bv_len and in sync_request_write()
2467 bio_reset(tbio, conf->mirrors[d].rdev->bdev, REQ_OP_WRITE); in sync_request_write()
2469 md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size); in sync_request_write()
2471 rp->raid_bio = r10_bio; in sync_request_write()
2472 tbio->bi_private = rp; in sync_request_write()
2473 tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; in sync_request_write()
2474 tbio->bi_end_io = end_sync_write; in sync_request_write()
2478 atomic_inc(&conf->mirrors[d].rdev->nr_pending); in sync_request_write()
2479 atomic_inc(&r10_bio->remaining); in sync_request_write()
2480 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); in sync_request_write()
2482 if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) in sync_request_write()
2483 tbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2484 tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; in sync_request_write()
2491 for (i = 0; i < conf->copies; i++) { in sync_request_write()
2494 tbio = r10_bio->devs[i].repl_bio; in sync_request_write()
2495 if (!tbio || !tbio->bi_end_io) in sync_request_write()
2497 if (r10_bio->devs[i].bio->bi_end_io != end_sync_write in sync_request_write()
2498 && r10_bio->devs[i].bio != fbio) in sync_request_write()
2500 d = r10_bio->devs[i].devnum; in sync_request_write()
2501 atomic_inc(&r10_bio->remaining); in sync_request_write()
2502 md_sync_acct(conf->mirrors[d].replacement->bdev, in sync_request_write()
2508 if (atomic_dec_and_test(&r10_bio->remaining)) { in sync_request_write()
2509 md_done_sync(mddev, r10_bio->sectors, 1); in sync_request_write()
2517 * We recover all non-is_sync drives by finding the virtual address of
2519 * There is a separate r10_bio for each non-in_sync drive.
2527 * We repeat the read in smaller page-sized sections. in fix_recovery_read_error()
2533 struct mddev *mddev = r10_bio->mddev; in fix_recovery_read_error()
2534 struct r10conf *conf = mddev->private; in fix_recovery_read_error()
2535 struct bio *bio = r10_bio->devs[0].bio; in fix_recovery_read_error()
2537 int sectors = r10_bio->sectors; in fix_recovery_read_error()
2539 int dr = r10_bio->devs[0].devnum; in fix_recovery_read_error()
2540 int dw = r10_bio->devs[1].devnum; in fix_recovery_read_error()
2541 struct page **pages = get_resync_pages(bio)->pages; in fix_recovery_read_error()
2552 rdev = conf->mirrors[dr].rdev; in fix_recovery_read_error()
2553 addr = r10_bio->devs[0].addr + sect, in fix_recovery_read_error()
2560 rdev = conf->mirrors[dw].rdev; in fix_recovery_read_error()
2561 addr = r10_bio->devs[1].addr + sect; in fix_recovery_read_error()
2568 set_bit(WriteErrorSeen, &rdev->flags); in fix_recovery_read_error()
2570 &rdev->flags)) in fix_recovery_read_error()
2572 &rdev->mddev->recovery); in fix_recovery_read_error()
2576 /* We don't worry if we cannot set a bad block - in fix_recovery_read_error()
2582 if (rdev != conf->mirrors[dw].rdev) { in fix_recovery_read_error()
2584 struct md_rdev *rdev2 = conf->mirrors[dw].rdev; in fix_recovery_read_error()
2585 addr = r10_bio->devs[1].addr + sect; in fix_recovery_read_error()
2592 conf->mirrors[dw].recovery_disabled in fix_recovery_read_error()
2593 = mddev->recovery_disabled; in fix_recovery_read_error()
2595 &mddev->recovery); in fix_recovery_read_error()
2601 sectors -= s; in fix_recovery_read_error()
2609 struct r10conf *conf = mddev->private; in recovery_request_write()
2611 struct bio *wbio = r10_bio->devs[1].bio; in recovery_request_write()
2612 struct bio *wbio2 = r10_bio->devs[1].repl_bio; in recovery_request_write()
2614 /* Need to test wbio2->bi_end_io before we call in recovery_request_write()
2618 if (wbio2 && !wbio2->bi_end_io) in recovery_request_write()
2621 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { in recovery_request_write()
2623 if (wbio->bi_end_io) in recovery_request_write()
2634 d = r10_bio->devs[1].devnum; in recovery_request_write()
2635 if (wbio->bi_end_io) { in recovery_request_write()
2636 atomic_inc(&conf->mirrors[d].rdev->nr_pending); in recovery_request_write()
2637 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); in recovery_request_write()
2641 atomic_inc(&conf->mirrors[d].replacement->nr_pending); in recovery_request_write()
2642 md_sync_acct(conf->mirrors[d].replacement->bdev, in recovery_request_write()
2658 unsigned int read_errors = atomic_read(&rdev->read_errors); in check_decay_read_errors()
2662 if (rdev->last_read_error == 0) { in check_decay_read_errors()
2664 rdev->last_read_error = cur_time_mon; in check_decay_read_errors()
2668 hours_since_last = (long)(cur_time_mon - in check_decay_read_errors()
2669 rdev->last_read_error) / 3600; in check_decay_read_errors()
2671 rdev->last_read_error = cur_time_mon; in check_decay_read_errors()
2679 atomic_set(&rdev->read_errors, 0); in check_decay_read_errors()
2681 atomic_set(&rdev->read_errors, read_errors >> hours_since_last); in check_decay_read_errors()
2691 && (op == REQ_OP_READ || test_bit(WriteErrorSeen, &rdev->flags))) in r10_sync_page_io()
2692 return -1; in r10_sync_page_io()
2697 set_bit(WriteErrorSeen, &rdev->flags); in r10_sync_page_io()
2698 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in r10_sync_page_io()
2700 &rdev->mddev->recovery); in r10_sync_page_io()
2702 /* need to record an error - either for the block or the device */ in r10_sync_page_io()
2704 md_error(rdev->mddev, rdev); in r10_sync_page_io()
2718 int sect = 0; /* Offset from r10_bio->sector */ in fix_read_error()
2719 int sectors = r10_bio->sectors, slot = r10_bio->read_slot; in fix_read_error()
2721 int max_read_errors = atomic_read(&mddev->max_corr_read_errors); in fix_read_error()
2722 int d = r10_bio->devs[slot].devnum; in fix_read_error()
2727 rdev = conf->mirrors[d].rdev; in fix_read_error()
2729 if (test_bit(Faulty, &rdev->flags)) in fix_read_error()
2735 atomic_inc(&rdev->read_errors); in fix_read_error()
2736 if (atomic_read(&rdev->read_errors) > max_read_errors) { in fix_read_error()
2738 mdname(mddev), rdev->bdev, in fix_read_error()
2739 atomic_read(&rdev->read_errors), max_read_errors); in fix_read_error()
2741 mdname(mddev), rdev->bdev); in fix_read_error()
2743 r10_bio->devs[slot].bio = IO_BLOCKED; in fix_read_error()
2761 d = r10_bio->devs[sl].devnum; in fix_read_error()
2762 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2764 test_bit(In_sync, &rdev->flags) && in fix_read_error()
2765 !test_bit(Faulty, &rdev->flags) && in fix_read_error()
2766 is_badblock(rdev, r10_bio->devs[sl].addr + sect, s, in fix_read_error()
2768 atomic_inc(&rdev->nr_pending); in fix_read_error()
2771 r10_bio->devs[sl].addr + in fix_read_error()
2774 conf->tmppage, in fix_read_error()
2782 if (sl == conf->copies) in fix_read_error()
2792 int dn = r10_bio->devs[slot].devnum; in fix_read_error()
2793 rdev = conf->mirrors[dn].rdev; in fix_read_error()
2797 r10_bio->devs[slot].addr in fix_read_error()
2801 r10_bio->devs[slot].bio in fix_read_error()
2808 /* write it back and re-read */ in fix_read_error()
2812 sl = conf->copies; in fix_read_error()
2813 sl--; in fix_read_error()
2814 d = r10_bio->devs[sl].devnum; in fix_read_error()
2815 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2817 test_bit(Faulty, &rdev->flags) || in fix_read_error()
2818 !test_bit(In_sync, &rdev->flags)) in fix_read_error()
2821 atomic_inc(&rdev->nr_pending); in fix_read_error()
2824 r10_bio->devs[sl].addr + in fix_read_error()
2826 s, conf->tmppage, REQ_OP_WRITE) in fix_read_error()
2835 rdev->bdev); in fix_read_error()
2838 rdev->bdev); in fix_read_error()
2846 sl = conf->copies; in fix_read_error()
2847 sl--; in fix_read_error()
2848 d = r10_bio->devs[sl].devnum; in fix_read_error()
2849 rdev = rcu_dereference(conf->mirrors[d].rdev); in fix_read_error()
2851 test_bit(Faulty, &rdev->flags) || in fix_read_error()
2852 !test_bit(In_sync, &rdev->flags)) in fix_read_error()
2855 atomic_inc(&rdev->nr_pending); in fix_read_error()
2858 r10_bio->devs[sl].addr + in fix_read_error()
2860 s, conf->tmppage, REQ_OP_READ)) { in fix_read_error()
2868 rdev->bdev); in fix_read_error()
2871 rdev->bdev); in fix_read_error()
2879 rdev->bdev); in fix_read_error()
2880 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2888 sectors -= s; in fix_read_error()
2895 struct bio *bio = r10_bio->master_bio; in narrow_write_error()
2896 struct mddev *mddev = r10_bio->mddev; in narrow_write_error()
2897 struct r10conf *conf = mddev->private; in narrow_write_error()
2898 struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; in narrow_write_error()
2913 int sect_to_write = r10_bio->sectors; in narrow_write_error()
2916 if (rdev->badblocks.shift < 0) in narrow_write_error()
2919 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2920 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2921 sector = r10_bio->sector; in narrow_write_error()
2922 sectors = ((r10_bio->sector + block_sectors) in narrow_write_error()
2923 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2924 - sector; in narrow_write_error()
2932 wbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, in narrow_write_error()
2933 &mddev->bio_set); in narrow_write_error()
2934 bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); in narrow_write_error()
2935 wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); in narrow_write_error()
2936 wbio->bi_iter.bi_sector = wsector + in narrow_write_error()
2938 wbio->bi_opf = REQ_OP_WRITE; in narrow_write_error()
2947 sect_to_write -= sectors; in narrow_write_error()
2956 int slot = r10_bio->read_slot; in handle_read_error()
2958 struct r10conf *conf = mddev->private; in handle_read_error()
2959 struct md_rdev *rdev = r10_bio->devs[slot].rdev; in handle_read_error()
2964 * other devices. When we find one, we re-write in handle_read_error()
2969 bio = r10_bio->devs[slot].bio; in handle_read_error()
2971 r10_bio->devs[slot].bio = NULL; in handle_read_error()
2973 if (mddev->ro) in handle_read_error()
2974 r10_bio->devs[slot].bio = IO_BLOCKED; in handle_read_error()
2975 else if (!test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2983 r10_bio->state = 0; in handle_read_error()
2984 raid10_read_request(mddev, r10_bio->master_bio, r10_bio, false); in handle_read_error()
2986 * allow_barrier after re-submit to ensure no sync io in handle_read_error()
3003 if (test_bit(R10BIO_IsSync, &r10_bio->state) || in handle_write_completed()
3004 test_bit(R10BIO_IsRecover, &r10_bio->state)) { in handle_write_completed()
3005 for (m = 0; m < conf->copies; m++) { in handle_write_completed()
3006 int dev = r10_bio->devs[m].devnum; in handle_write_completed()
3007 rdev = conf->mirrors[dev].rdev; in handle_write_completed()
3008 if (r10_bio->devs[m].bio == NULL || in handle_write_completed()
3009 r10_bio->devs[m].bio->bi_end_io == NULL) in handle_write_completed()
3011 if (!r10_bio->devs[m].bio->bi_status) { in handle_write_completed()
3014 r10_bio->devs[m].addr, in handle_write_completed()
3015 r10_bio->sectors, 0); in handle_write_completed()
3019 r10_bio->devs[m].addr, in handle_write_completed()
3020 r10_bio->sectors, 0)) in handle_write_completed()
3021 md_error(conf->mddev, rdev); in handle_write_completed()
3023 rdev = conf->mirrors[dev].replacement; in handle_write_completed()
3024 if (r10_bio->devs[m].repl_bio == NULL || in handle_write_completed()
3025 r10_bio->devs[m].repl_bio->bi_end_io == NULL) in handle_write_completed()
3028 if (!r10_bio->devs[m].repl_bio->bi_status) { in handle_write_completed()
3031 r10_bio->devs[m].addr, in handle_write_completed()
3032 r10_bio->sectors, 0); in handle_write_completed()
3036 r10_bio->devs[m].addr, in handle_write_completed()
3037 r10_bio->sectors, 0)) in handle_write_completed()
3038 md_error(conf->mddev, rdev); in handle_write_completed()
3044 for (m = 0; m < conf->copies; m++) { in handle_write_completed()
3045 int dev = r10_bio->devs[m].devnum; in handle_write_completed()
3046 struct bio *bio = r10_bio->devs[m].bio; in handle_write_completed()
3047 rdev = conf->mirrors[dev].rdev; in handle_write_completed()
3051 r10_bio->devs[m].addr, in handle_write_completed()
3052 r10_bio->sectors, 0); in handle_write_completed()
3053 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
3054 } else if (bio != NULL && bio->bi_status) { in handle_write_completed()
3057 md_error(conf->mddev, rdev); in handle_write_completed()
3058 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
3060 bio = r10_bio->devs[m].repl_bio; in handle_write_completed()
3061 rdev = conf->mirrors[dev].replacement; in handle_write_completed()
3065 r10_bio->devs[m].addr, in handle_write_completed()
3066 r10_bio->sectors, 0); in handle_write_completed()
3067 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
3071 spin_lock_irq(&conf->device_lock); in handle_write_completed()
3072 list_add(&r10_bio->retry_list, &conf->bio_end_io_list); in handle_write_completed()
3073 conf->nr_queued++; in handle_write_completed()
3074 spin_unlock_irq(&conf->device_lock); in handle_write_completed()
3079 wake_up(&conf->wait_barrier); in handle_write_completed()
3080 md_wakeup_thread(conf->mddev->thread); in handle_write_completed()
3083 &r10_bio->state)) in handle_write_completed()
3092 struct mddev *mddev = thread->mddev; in raid10d()
3095 struct r10conf *conf = mddev->private; in raid10d()
3096 struct list_head *head = &conf->retry_list; in raid10d()
3101 if (!list_empty_careful(&conf->bio_end_io_list) && in raid10d()
3102 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid10d()
3104 spin_lock_irqsave(&conf->device_lock, flags); in raid10d()
3105 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid10d()
3106 while (!list_empty(&conf->bio_end_io_list)) { in raid10d()
3107 list_move(conf->bio_end_io_list.prev, &tmp); in raid10d()
3108 conf->nr_queued--; in raid10d()
3111 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3115 list_del(&r10_bio->retry_list); in raid10d()
3118 &r10_bio->state)) in raid10d()
3129 spin_lock_irqsave(&conf->device_lock, flags); in raid10d()
3131 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3134 r10_bio = list_entry(head->prev, struct r10bio, retry_list); in raid10d()
3135 list_del(head->prev); in raid10d()
3136 conf->nr_queued--; in raid10d()
3137 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3139 mddev = r10_bio->mddev; in raid10d()
3140 conf = mddev->private; in raid10d()
3141 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in raid10d()
3142 test_bit(R10BIO_WriteError, &r10_bio->state)) in raid10d()
3144 else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) in raid10d()
3146 else if (test_bit(R10BIO_IsSync, &r10_bio->state)) in raid10d()
3148 else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) in raid10d()
3150 else if (test_bit(R10BIO_ReadError, &r10_bio->state)) in raid10d()
3156 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid10d()
3167 BUG_ON(mempool_initialized(&conf->r10buf_pool)); in init_resync()
3168 conf->have_replacement = 0; in init_resync()
3169 for (i = 0; i < conf->geo.raid_disks; i++) in init_resync()
3170 if (conf->mirrors[i].replacement) in init_resync()
3171 conf->have_replacement = 1; in init_resync()
3172 ret = mempool_init(&conf->r10buf_pool, buffs, in init_resync()
3176 conf->next_resync = 0; in init_resync()
3182 struct r10bio *r10bio = mempool_alloc(&conf->r10buf_pool, GFP_NOIO); in raid10_alloc_init_r10buf()
3188 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || in raid10_alloc_init_r10buf()
3189 test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) in raid10_alloc_init_r10buf()
3190 nalloc = conf->copies; /* resync */ in raid10_alloc_init_r10buf()
3195 bio = r10bio->devs[i].bio; in raid10_alloc_init_r10buf()
3196 rp = bio->bi_private; in raid10_alloc_init_r10buf()
3198 bio->bi_private = rp; in raid10_alloc_init_r10buf()
3199 bio = r10bio->devs[i].repl_bio; in raid10_alloc_init_r10buf()
3201 rp = bio->bi_private; in raid10_alloc_init_r10buf()
3203 bio->bi_private = rp; in raid10_alloc_init_r10buf()
3230 chunks = conf->geo.raid_disks / conf->geo.near_copies; in raid10_set_cluster_sync_high()
3231 if (conf->geo.raid_disks % conf->geo.near_copies == 0) in raid10_set_cluster_sync_high()
3235 window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors; in raid10_set_cluster_sync_high()
3243 conf->cluster_sync_high = conf->cluster_sync_low + window_size; in raid10_set_cluster_sync_high()
3249 * We need to make sure that no normal I/O request - particularly write
3250 * requests - conflict with active sync requests.
3256 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
3259 * and update if there are differences. If only one copy is live,
3262 * value for each non-in_sync drive, and over-write.
3265 * given address, one for each out-of-sync device. We model this by allocating
3266 * a number of r10_bio structures, one for each out-of-sync device.
3272 * This link is counted in ->remaining. When the r10_bio that points to NULL
3281 struct r10conf *conf = mddev->private; in raid10_sync_request()
3290 sector_t chunk_mask = conf->geo.chunk_mask; in raid10_sync_request()
3292 int error_disk = -1; in raid10_sync_request()
3298 if (mddev->bitmap == NULL && in raid10_sync_request()
3299 mddev->recovery_cp == MaxSector && in raid10_sync_request()
3300 mddev->reshape_position == MaxSector && in raid10_sync_request()
3301 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid10_sync_request()
3302 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid10_sync_request()
3303 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in raid10_sync_request()
3304 conf->fullsync == 0) { in raid10_sync_request()
3306 return mddev->dev_sectors - sector_nr; in raid10_sync_request()
3309 if (!mempool_initialized(&conf->r10buf_pool)) in raid10_sync_request()
3314 max_sector = mddev->dev_sectors; in raid10_sync_request()
3315 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) || in raid10_sync_request()
3316 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_sync_request()
3317 max_sector = mddev->resync_max_sectors; in raid10_sync_request()
3319 conf->cluster_sync_low = 0; in raid10_sync_request()
3320 conf->cluster_sync_high = 0; in raid10_sync_request()
3327 * mddev->curr_resync, but for recovery, in raid10_sync_request()
3331 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { in raid10_sync_request()
3337 if (mddev->curr_resync < max_sector) { /* aborted */ in raid10_sync_request()
3338 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) in raid10_sync_request()
3339 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, in raid10_sync_request()
3341 else for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3343 raid10_find_virt(conf, mddev->curr_resync, i); in raid10_sync_request()
3344 md_bitmap_end_sync(mddev->bitmap, sect, in raid10_sync_request()
3349 if ((!mddev->bitmap || conf->fullsync) in raid10_sync_request()
3350 && conf->have_replacement in raid10_sync_request()
3351 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3356 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3358 rcu_dereference(conf->mirrors[i].replacement); in raid10_sync_request()
3360 rdev->recovery_offset = MaxSector; in raid10_sync_request()
3364 conf->fullsync = 0; in raid10_sync_request()
3366 md_bitmap_close_sync(mddev->bitmap); in raid10_sync_request()
3372 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_sync_request()
3375 if (chunks_skipped >= conf->geo.raid_disks) { in raid10_sync_request()
3377 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? "resync" : "recovery"); in raid10_sync_request()
3379 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3384 conf->mirrors[error_disk].recovery_disabled = in raid10_sync_request()
3385 mddev->recovery_disabled; in raid10_sync_request()
3393 return (max_sector - sector_nr) + sectors_skipped; in raid10_sync_request()
3396 if (max_sector > mddev->resync_max) in raid10_sync_request()
3397 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid10_sync_request()
3399 /* make sure whole request will fit in a chunk - if chunks in raid10_sync_request()
3402 if (conf->geo.near_copies < conf->geo.raid_disks && in raid10_sync_request()
3407 * If there is non-resync activity waiting for a turn, then let it in raid10_sync_request()
3410 if (conf->nr_waiting) in raid10_sync_request()
3423 /* First, we decide what to do and set ->bi_end_io in raid10_sync_request()
3428 max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); in raid10_sync_request()
3429 if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3434 for (i = 0 ; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3440 struct raid10_info *mirror = &conf->mirrors[i]; in raid10_sync_request()
3444 mrdev = rcu_dereference(mirror->rdev); in raid10_sync_request()
3445 mreplace = rcu_dereference(mirror->replacement); in raid10_sync_request()
3447 if (mrdev && (test_bit(Faulty, &mrdev->flags) || in raid10_sync_request()
3448 test_bit(In_sync, &mrdev->flags))) in raid10_sync_request()
3450 if (mreplace && test_bit(Faulty, &mreplace->flags)) in raid10_sync_request()
3462 if (sect >= mddev->resync_max_sectors) { in raid10_sync_request()
3463 /* last stripe is not complete - don't in raid10_sync_request()
3473 must_sync = md_bitmap_start_sync(mddev->bitmap, sect, in raid10_sync_request()
3479 !conf->fullsync) { in raid10_sync_request()
3483 chunks_skipped = -1; in raid10_sync_request()
3488 atomic_inc(&mrdev->nr_pending); in raid10_sync_request()
3490 atomic_inc(&mreplace->nr_pending); in raid10_sync_request()
3494 r10_bio->state = 0; in raid10_sync_request()
3496 atomic_set(&r10_bio->remaining, 0); in raid10_sync_request()
3498 r10_bio->master_bio = (struct bio*)rb2; in raid10_sync_request()
3500 atomic_inc(&rb2->remaining); in raid10_sync_request()
3501 r10_bio->mddev = mddev; in raid10_sync_request()
3502 set_bit(R10BIO_IsRecover, &r10_bio->state); in raid10_sync_request()
3503 r10_bio->sector = sect; in raid10_sync_request()
3511 for (j = 0; j < conf->geo.raid_disks; j++) { in raid10_sync_request()
3513 conf->mirrors[j].rdev); in raid10_sync_request()
3514 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { in raid10_sync_request()
3520 must_sync = md_bitmap_start_sync(mddev->bitmap, sect, in raid10_sync_request()
3524 for (j=0; j<conf->copies;j++) { in raid10_sync_request()
3526 int d = r10_bio->devs[j].devnum; in raid10_sync_request()
3529 rcu_dereference(conf->mirrors[d].rdev); in raid10_sync_request()
3533 !test_bit(In_sync, &rdev->flags)) in raid10_sync_request()
3537 sector = r10_bio->devs[j].addr; in raid10_sync_request()
3542 max_sync = first_bad - sector; in raid10_sync_request()
3544 bad_sectors -= (sector in raid10_sync_request()
3545 - first_bad); in raid10_sync_request()
3551 bio = r10_bio->devs[0].bio; in raid10_sync_request()
3552 bio->bi_next = biolist; in raid10_sync_request()
3554 bio->bi_end_io = end_sync_read; in raid10_sync_request()
3555 bio->bi_opf = REQ_OP_READ; in raid10_sync_request()
3556 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3557 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3558 from_addr = r10_bio->devs[j].addr; in raid10_sync_request()
3559 bio->bi_iter.bi_sector = from_addr + in raid10_sync_request()
3560 rdev->data_offset; in raid10_sync_request()
3561 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3562 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3565 for (k=0; k<conf->copies; k++) in raid10_sync_request()
3566 if (r10_bio->devs[k].devnum == i) in raid10_sync_request()
3568 BUG_ON(k == conf->copies); in raid10_sync_request()
3569 to_addr = r10_bio->devs[k].addr; in raid10_sync_request()
3570 r10_bio->devs[0].devnum = d; in raid10_sync_request()
3571 r10_bio->devs[0].addr = from_addr; in raid10_sync_request()
3572 r10_bio->devs[1].devnum = i; in raid10_sync_request()
3573 r10_bio->devs[1].addr = to_addr; in raid10_sync_request()
3576 bio = r10_bio->devs[1].bio; in raid10_sync_request()
3577 bio->bi_next = biolist; in raid10_sync_request()
3579 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3580 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3581 bio->bi_iter.bi_sector = to_addr in raid10_sync_request()
3582 + mrdev->data_offset; in raid10_sync_request()
3583 bio_set_dev(bio, mrdev->bdev); in raid10_sync_request()
3584 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3586 r10_bio->devs[1].bio->bi_end_io = NULL; in raid10_sync_request()
3589 bio = r10_bio->devs[1].repl_bio; in raid10_sync_request()
3591 bio->bi_end_io = NULL; in raid10_sync_request()
3598 bio->bi_next = biolist; in raid10_sync_request()
3600 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3601 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3602 bio->bi_iter.bi_sector = to_addr + in raid10_sync_request()
3603 mreplace->data_offset; in raid10_sync_request()
3604 bio_set_dev(bio, mreplace->bdev); in raid10_sync_request()
3605 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3609 if (j == conf->copies) { in raid10_sync_request()
3617 for (k = 0; k < conf->copies; k++) in raid10_sync_request()
3618 if (r10_bio->devs[k].devnum == i) in raid10_sync_request()
3621 &mrdev->flags) in raid10_sync_request()
3624 r10_bio->devs[k].addr, in raid10_sync_request()
3630 r10_bio->devs[k].addr, in raid10_sync_request()
3636 &mddev->recovery)) in raid10_sync_request()
3639 mirror->recovery_disabled in raid10_sync_request()
3640 = mddev->recovery_disabled; in raid10_sync_request()
3646 atomic_dec(&rb2->remaining); in raid10_sync_request()
3658 if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { in raid10_sync_request()
3664 for (; j < conf->copies; j++) { in raid10_sync_request()
3665 int d = r10_bio->devs[j].devnum; in raid10_sync_request()
3666 if (conf->mirrors[d].rdev && in raid10_sync_request()
3668 &conf->mirrors[d].rdev->flags)) in raid10_sync_request()
3672 r10_bio->devs[0].bio->bi_opf in raid10_sync_request()
3679 r10_bio = (struct r10bio*) rb2->master_bio; in raid10_sync_request()
3680 rb2->master_bio = NULL; in raid10_sync_request()
3690 * Since curr_resync_completed could probably not update in in raid10_sync_request()
3696 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, in raid10_sync_request()
3698 (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid10_sync_request()
3700 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, in raid10_sync_request()
3701 &sync_blocks, mddev->degraded) && in raid10_sync_request()
3702 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, in raid10_sync_request()
3703 &mddev->recovery)) { in raid10_sync_request()
3711 r10_bio->state = 0; in raid10_sync_request()
3713 r10_bio->mddev = mddev; in raid10_sync_request()
3714 atomic_set(&r10_bio->remaining, 0); in raid10_sync_request()
3716 conf->next_resync = sector_nr; in raid10_sync_request()
3718 r10_bio->master_bio = NULL; in raid10_sync_request()
3719 r10_bio->sector = sector_nr; in raid10_sync_request()
3720 set_bit(R10BIO_IsSync, &r10_bio->state); in raid10_sync_request()
3722 r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; in raid10_sync_request()
3724 for (i = 0; i < conf->copies; i++) { in raid10_sync_request()
3725 int d = r10_bio->devs[i].devnum; in raid10_sync_request()
3730 if (r10_bio->devs[i].repl_bio) in raid10_sync_request()
3731 r10_bio->devs[i].repl_bio->bi_end_io = NULL; in raid10_sync_request()
3733 bio = r10_bio->devs[i].bio; in raid10_sync_request()
3734 bio->bi_status = BLK_STS_IOERR; in raid10_sync_request()
3736 rdev = rcu_dereference(conf->mirrors[d].rdev); in raid10_sync_request()
3737 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { in raid10_sync_request()
3741 sector = r10_bio->devs[i].addr; in raid10_sync_request()
3745 max_sync = first_bad - sector; in raid10_sync_request()
3747 bad_sectors -= (sector - first_bad); in raid10_sync_request()
3754 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3755 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3756 bio->bi_next = biolist; in raid10_sync_request()
3758 bio->bi_end_io = end_sync_read; in raid10_sync_request()
3759 bio->bi_opf = REQ_OP_READ; in raid10_sync_request()
3760 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3761 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3762 bio->bi_iter.bi_sector = sector + rdev->data_offset; in raid10_sync_request()
3763 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3766 rdev = rcu_dereference(conf->mirrors[d].replacement); in raid10_sync_request()
3767 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { in raid10_sync_request()
3771 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3774 bio = r10_bio->devs[i].repl_bio; in raid10_sync_request()
3775 bio->bi_status = BLK_STS_IOERR; in raid10_sync_request()
3777 sector = r10_bio->devs[i].addr; in raid10_sync_request()
3778 bio->bi_next = biolist; in raid10_sync_request()
3780 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3781 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3782 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3783 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3784 bio->bi_iter.bi_sector = sector + rdev->data_offset; in raid10_sync_request()
3785 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3791 for (i=0; i<conf->copies; i++) { in raid10_sync_request()
3792 int d = r10_bio->devs[i].devnum; in raid10_sync_request()
3793 if (r10_bio->devs[i].bio->bi_end_io) in raid10_sync_request()
3794 rdev_dec_pending(conf->mirrors[d].rdev, in raid10_sync_request()
3796 if (r10_bio->devs[i].repl_bio && in raid10_sync_request()
3797 r10_bio->devs[i].repl_bio->bi_end_io) in raid10_sync_request()
3799 conf->mirrors[d].replacement, in raid10_sync_request()
3815 len = (max_sector - sector_nr) << 9; in raid10_sync_request()
3818 for (bio= biolist ; bio ; bio=bio->bi_next) { in raid10_sync_request()
3822 bio->bi_status = BLK_STS_RESOURCE; in raid10_sync_request()
3830 r10_bio->sectors = nr_sectors; in raid10_sync_request()
3833 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3835 if (conf->cluster_sync_high < sector_nr + nr_sectors) { in raid10_sync_request()
3836 conf->cluster_sync_low = mddev->curr_resync_completed; in raid10_sync_request()
3839 md_cluster_ops->resync_info_update(mddev, in raid10_sync_request()
3840 conf->cluster_sync_low, in raid10_sync_request()
3841 conf->cluster_sync_high); in raid10_sync_request()
3848 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3856 if (conf->cluster_sync_high < sect_va1 + nr_sectors) { in raid10_sync_request()
3863 mddev->curr_resync_completed, i); in raid10_sync_request()
3865 if (conf->cluster_sync_low == 0 || in raid10_sync_request()
3866 conf->cluster_sync_low > sect_va2) in raid10_sync_request()
3867 conf->cluster_sync_low = sect_va2; in raid10_sync_request()
3872 md_cluster_ops->resync_info_update(mddev, in raid10_sync_request()
3873 conf->cluster_sync_low, in raid10_sync_request()
3874 conf->cluster_sync_high); in raid10_sync_request()
3880 biolist = biolist->bi_next; in raid10_sync_request()
3882 bio->bi_next = NULL; in raid10_sync_request()
3884 r10_bio->sectors = nr_sectors; in raid10_sync_request()
3886 if (bio->bi_end_io == end_sync_read) { in raid10_sync_request()
3888 bio->bi_status = 0; in raid10_sync_request()
3901 /* There is nowhere to write, so all non-sync in raid10_sync_request()
3908 sectors_skipped += (max_sector - sector_nr); in raid10_sync_request()
3918 struct r10conf *conf = mddev->private; in raid10_size()
3921 raid_disks = min(conf->geo.raid_disks, in raid10_size()
3922 conf->prev.raid_disks); in raid10_size()
3924 sectors = conf->dev_sectors; in raid10_size()
3926 size = sectors >> conf->geo.chunk_shift; in raid10_size()
3927 sector_div(size, conf->geo.far_copies); in raid10_size()
3929 sector_div(size, conf->geo.near_copies); in raid10_size()
3931 return size << conf->geo.chunk_shift; in raid10_size()
3936 /* Calculate the number of sectors-per-device that will in calc_sectors()
3937 * actually be used, and set conf->dev_sectors and in calc_sectors()
3938 * conf->stride in calc_sectors()
3941 size = size >> conf->geo.chunk_shift; in calc_sectors()
3942 sector_div(size, conf->geo.far_copies); in calc_sectors()
3943 size = size * conf->geo.raid_disks; in calc_sectors()
3944 sector_div(size, conf->geo.near_copies); in calc_sectors()
3947 size = size * conf->copies; in calc_sectors()
3952 size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); in calc_sectors()
3954 conf->dev_sectors = size << conf->geo.chunk_shift; in calc_sectors()
3956 if (conf->geo.far_offset) in calc_sectors()
3957 conf->geo.stride = 1 << conf->geo.chunk_shift; in calc_sectors()
3959 sector_div(size, conf->geo.far_copies); in calc_sectors()
3960 conf->geo.stride = size << conf->geo.chunk_shift; in calc_sectors()
3967 int nc, fc, fo; in setup_geo() local
3971 layout = mddev->layout; in setup_geo()
3972 chunk = mddev->chunk_sectors; in setup_geo()
3973 disks = mddev->raid_disks - mddev->delta_disks; 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; in setup_geo()
3981 case geo_start: /* new when starting reshape - raid_disks not in setup_geo()
3983 layout = mddev->new_layout; in setup_geo()
3984 chunk = mddev->new_chunk_sectors; in setup_geo()
3985 disks = mddev->raid_disks + mddev->delta_disks; in setup_geo()
3989 return -1; in setup_geo()
3992 return -2; in setup_geo()
3994 fc = (layout >> 8) & 255; in setup_geo()
3996 geo->raid_disks = disks; in setup_geo()
3997 geo->near_copies = nc; in setup_geo()
3998 geo->far_copies = fc; in setup_geo()
3999 geo->far_offset = fo; in setup_geo()
4002 geo->far_set_size = disks; in setup_geo()
4004 case 1: /* "improved" layout which was buggy. Hopefully no-one is in setup_geo()
4006 geo->far_set_size = disks/fc; in setup_geo()
4007 WARN(geo->far_set_size < fc, in setup_geo()
4008 "This RAID10 layout does not provide data safety - please backup and create new array\n"); in setup_geo()
4011 geo->far_set_size = fc * nc; in setup_geo()
4014 return -1; in setup_geo()
4016 geo->chunk_mask = chunk - 1; in setup_geo()
4017 geo->chunk_shift = ffz(~chunk); in setup_geo()
4018 return nc*fc; in setup_geo()
4026 mempool_exit(&conf->r10bio_pool); in raid10_free_conf()
4027 kfree(conf->mirrors); in raid10_free_conf()
4028 kfree(conf->mirrors_old); in raid10_free_conf()
4029 kfree(conf->mirrors_new); in raid10_free_conf()
4030 safe_put_page(conf->tmppage); in raid10_free_conf()
4031 bioset_exit(&conf->bio_split); in raid10_free_conf()
4038 int err = -EINVAL; in setup_conf()
4044 if (copies == -2) { in setup_conf()
4050 if (copies < 2 || copies > mddev->raid_disks) { in setup_conf()
4052 mdname(mddev), mddev->new_layout); in setup_conf()
4056 err = -ENOMEM; in setup_conf()
4062 conf->mirrors = kcalloc(mddev->raid_disks + max(0, -mddev->delta_disks), in setup_conf()
4065 if (!conf->mirrors) in setup_conf()
4068 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
4069 if (!conf->tmppage) in setup_conf()
4072 conf->geo = geo; in setup_conf()
4073 conf->copies = copies; in setup_conf()
4074 err = mempool_init(&conf->r10bio_pool, NR_RAID_BIOS, r10bio_pool_alloc, in setup_conf()
4079 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
4083 calc_sectors(conf, mddev->dev_sectors); in setup_conf()
4084 if (mddev->reshape_position == MaxSector) { in setup_conf()
4085 conf->prev = conf->geo; in setup_conf()
4086 conf->reshape_progress = MaxSector; in setup_conf()
4088 if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { in setup_conf()
4089 err = -EINVAL; in setup_conf()
4092 conf->reshape_progress = mddev->reshape_position; in setup_conf()
4093 if (conf->prev.far_offset) in setup_conf()
4094 conf->prev.stride = 1 << conf->prev.chunk_shift; in setup_conf()
4097 conf->prev.stride = conf->dev_sectors; in setup_conf()
4099 conf->reshape_safe = conf->reshape_progress; in setup_conf()
4100 spin_lock_init(&conf->device_lock); in setup_conf()
4101 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
4102 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
4104 seqlock_init(&conf->resync_lock); in setup_conf()
4105 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
4106 atomic_set(&conf->nr_pending, 0); in setup_conf()
4108 err = -ENOMEM; in setup_conf()
4109 rcu_assign_pointer(conf->thread, in setup_conf()
4111 if (!conf->thread) in setup_conf()
4114 conf->mddev = mddev; in setup_conf()
4124 int raid_disks = conf->geo.raid_disks; in raid10_set_io_opt()
4126 if (!(conf->geo.raid_disks % conf->geo.near_copies)) in raid10_set_io_opt()
4127 raid_disks /= conf->geo.near_copies; in raid10_set_io_opt()
4128 blk_queue_io_opt(conf->mddev->queue, (conf->mddev->chunk_sectors << 9) * in raid10_set_io_opt()
4143 return -ENOMEM; in raid10_run()
4145 if (mddev->private == NULL) { in raid10_run()
4149 mddev->private = conf; in raid10_run()
4151 conf = mddev->private; in raid10_run()
4155 rcu_assign_pointer(mddev->thread, conf->thread); in raid10_run()
4156 rcu_assign_pointer(conf->thread, NULL); in raid10_run()
4158 if (mddev_is_clustered(conf->mddev)) { in raid10_run()
4159 int fc, fo; in raid10_run() local
4161 fc = (mddev->layout >> 8) & 255; in raid10_run()
4162 fo = mddev->layout & (1<<16); in raid10_run()
4163 if (fc > 1 || fo > 0) { in raid10_run()
4170 if (mddev->queue) { in raid10_run()
4171 blk_queue_max_write_zeroes_sectors(mddev->queue, 0); in raid10_run()
4172 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); in raid10_run()
4179 disk_idx = rdev->raid_disk; in raid10_run()
4182 if (disk_idx >= conf->geo.raid_disks && in raid10_run()
4183 disk_idx >= conf->prev.raid_disks) in raid10_run()
4185 disk = conf->mirrors + disk_idx; in raid10_run()
4187 if (test_bit(Replacement, &rdev->flags)) { in raid10_run()
4188 if (disk->replacement) in raid10_run()
4190 disk->replacement = rdev; in raid10_run()
4192 if (disk->rdev) in raid10_run()
4194 disk->rdev = rdev; in raid10_run()
4196 diff = (rdev->new_data_offset - rdev->data_offset); in raid10_run()
4197 if (!mddev->reshape_backwards) in raid10_run()
4198 diff = -diff; in raid10_run()
4204 if (mddev->gendisk) in raid10_run()
4205 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid10_run()
4206 rdev->data_offset << 9); in raid10_run()
4208 disk->head_position = 0; in raid10_run()
4213 if (!enough(conf, -1)) { in raid10_run()
4219 if (conf->reshape_progress != MaxSector) { in raid10_run()
4221 if (conf->geo.far_copies != 1 && in raid10_run()
4222 conf->geo.far_offset == 0) in raid10_run()
4224 if (conf->prev.far_copies != 1 && in raid10_run()
4225 conf->prev.far_offset == 0) in raid10_run()
4229 mddev->degraded = 0; in raid10_run()
4231 i < conf->geo.raid_disks in raid10_run()
4232 || i < conf->prev.raid_disks; in raid10_run()
4235 disk = conf->mirrors + i; in raid10_run()
4237 if (!disk->rdev && disk->replacement) { in raid10_run()
4238 /* The replacement is all we have - use it */ in raid10_run()
4239 disk->rdev = disk->replacement; in raid10_run()
4240 disk->replacement = NULL; in raid10_run()
4241 clear_bit(Replacement, &disk->rdev->flags); in raid10_run()
4244 if (!disk->rdev || in raid10_run()
4245 !test_bit(In_sync, &disk->rdev->flags)) { in raid10_run()
4246 disk->head_position = 0; in raid10_run()
4247 mddev->degraded++; in raid10_run()
4248 if (disk->rdev && in raid10_run()
4249 disk->rdev->saved_raid_disk < 0) in raid10_run()
4250 conf->fullsync = 1; in raid10_run()
4253 if (disk->replacement && in raid10_run()
4254 !test_bit(In_sync, &disk->replacement->flags) && in raid10_run()
4255 disk->replacement->saved_raid_disk < 0) { in raid10_run()
4256 conf->fullsync = 1; in raid10_run()
4259 disk->recovery_disabled = mddev->recovery_disabled - 1; in raid10_run()
4262 if (mddev->recovery_cp != MaxSector) in raid10_run()
4263 pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", in raid10_run()
4266 mdname(mddev), conf->geo.raid_disks - mddev->degraded, in raid10_run()
4267 conf->geo.raid_disks); in raid10_run()
4271 mddev->dev_sectors = conf->dev_sectors; in raid10_run()
4274 mddev->resync_max_sectors = size; in raid10_run()
4275 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid10_run()
4280 if (conf->reshape_progress != MaxSector) { in raid10_run()
4283 before_length = ((1 << conf->prev.chunk_shift) * in raid10_run()
4284 conf->prev.far_copies); in raid10_run()
4285 after_length = ((1 << conf->geo.chunk_shift) * in raid10_run()
4286 conf->geo.far_copies); in raid10_run()
4293 conf->offset_diff = min_offset_diff; in raid10_run()
4295 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); in raid10_run()
4296 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); in raid10_run()
4297 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); in raid10_run()
4298 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); in raid10_run()
4299 rcu_assign_pointer(mddev->sync_thread, in raid10_run()
4301 if (!mddev->sync_thread) in raid10_run()
4308 md_unregister_thread(mddev, &mddev->thread); in raid10_run()
4310 mddev->private = NULL; in raid10_run()
4312 return -EIO; in raid10_run()
4322 struct r10conf *conf = mddev->private; in raid10_quiesce()
4344 struct r10conf *conf = mddev->private; in raid10_resize()
4347 if (mddev->reshape_position != MaxSector) in raid10_resize()
4348 return -EBUSY; in raid10_resize()
4350 if (conf->geo.far_copies > 1 && !conf->geo.far_offset) in raid10_resize()
4351 return -EINVAL; in raid10_resize()
4355 if (mddev->external_size && in raid10_resize()
4356 mddev->array_sectors > size) in raid10_resize()
4357 return -EINVAL; in raid10_resize()
4358 if (mddev->bitmap) { in raid10_resize()
4359 int ret = md_bitmap_resize(mddev->bitmap, size, 0, 0); in raid10_resize()
4364 if (sectors > mddev->dev_sectors && in raid10_resize()
4365 mddev->recovery_cp > oldsize) { in raid10_resize()
4366 mddev->recovery_cp = oldsize; in raid10_resize()
4367 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_resize()
4370 mddev->dev_sectors = conf->dev_sectors; in raid10_resize()
4371 mddev->resync_max_sectors = size; in raid10_resize()
4380 if (mddev->degraded > 0) { in raid10_takeover_raid0()
4383 return ERR_PTR(-EINVAL); in raid10_takeover_raid0()
4388 mddev->new_level = 10; in raid10_takeover_raid0()
4390 mddev->new_layout = (1<<8) + 2; in raid10_takeover_raid0()
4391 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid10_takeover_raid0()
4392 mddev->delta_disks = mddev->raid_disks; in raid10_takeover_raid0()
4393 mddev->raid_disks *= 2; in raid10_takeover_raid0()
4395 mddev->recovery_cp = MaxSector; in raid10_takeover_raid0()
4396 mddev->dev_sectors = size; in raid10_takeover_raid0()
4401 if (rdev->raid_disk >= 0) { in raid10_takeover_raid0()
4402 rdev->new_raid_disk = rdev->raid_disk * 2; in raid10_takeover_raid0()
4403 rdev->sectors = size; in raid10_takeover_raid0()
4415 * raid0 - providing it has only two drives in raid10_takeover()
4417 if (mddev->level == 0) { in raid10_takeover()
4419 raid0_conf = mddev->private; in raid10_takeover()
4420 if (raid0_conf->nr_strip_zones > 1) { in raid10_takeover()
4423 return ERR_PTR(-EINVAL); in raid10_takeover()
4426 raid0_conf->strip_zone->zone_end, in raid10_takeover()
4427 raid0_conf->strip_zone->nb_dev); in raid10_takeover()
4429 return ERR_PTR(-EINVAL); in raid10_takeover()
4435 * - layout (to ->new_layout) in raid10_check_reshape()
4436 * - chunk size (to ->new_chunk_sectors) in raid10_check_reshape()
4437 * - raid_disks (by delta_disks) in raid10_check_reshape()
4448 struct r10conf *conf = mddev->private; in raid10_check_reshape()
4451 if (conf->geo.far_copies != 1 && !conf->geo.far_offset) in raid10_check_reshape()
4452 return -EINVAL; in raid10_check_reshape()
4454 if (setup_geo(&geo, mddev, geo_start) != conf->copies) in raid10_check_reshape()
4456 return -EINVAL; in raid10_check_reshape()
4459 return -EINVAL; in raid10_check_reshape()
4461 if (mddev->array_sectors & geo.chunk_mask) in raid10_check_reshape()
4463 return -EINVAL; in raid10_check_reshape()
4465 if (!enough(conf, -1)) in raid10_check_reshape()
4466 return -EINVAL; in raid10_check_reshape()
4468 kfree(conf->mirrors_new); in raid10_check_reshape()
4469 conf->mirrors_new = NULL; in raid10_check_reshape()
4470 if (mddev->delta_disks > 0) { in raid10_check_reshape()
4472 conf->mirrors_new = in raid10_check_reshape()
4473 kcalloc(mddev->raid_disks + mddev->delta_disks, in raid10_check_reshape()
4476 if (!conf->mirrors_new) in raid10_check_reshape()
4477 return -ENOMEM; in raid10_check_reshape()
4484 * - start an array
4485 * - remove non-faulty devices
4486 * - add a spare
4487 * - allow a reshape
4492 * of the two sections, and some non-in_sync devices may
4503 for (i = 0; i < conf->prev.raid_disks; i++) { in calc_degraded()
4504 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in calc_degraded()
4505 if (!rdev || test_bit(Faulty, &rdev->flags)) in calc_degraded()
4507 else if (!test_bit(In_sync, &rdev->flags)) in calc_degraded()
4515 if (conf->geo.raid_disks == conf->prev.raid_disks) in calc_degraded()
4519 for (i = 0; i < conf->geo.raid_disks; i++) { in calc_degraded()
4520 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev); in calc_degraded()
4521 if (!rdev || test_bit(Faulty, &rdev->flags)) in calc_degraded()
4523 else if (!test_bit(In_sync, &rdev->flags)) { in calc_degraded()
4529 if (conf->geo.raid_disks <= conf->prev.raid_disks) in calc_degraded()
4546 * array non-degraded. We also require that the difference in raid10_start_reshape()
4547 * between old and new data_offset - on each device - is in raid10_start_reshape()
4548 * enough that we never risk over-writing. in raid10_start_reshape()
4555 struct r10conf *conf = mddev->private; in raid10_start_reshape()
4560 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) in raid10_start_reshape()
4561 return -EBUSY; in raid10_start_reshape()
4563 if (setup_geo(&new, mddev, geo_start) != conf->copies) in raid10_start_reshape()
4564 return -EINVAL; in raid10_start_reshape()
4566 before_length = ((1 << conf->prev.chunk_shift) * in raid10_start_reshape()
4567 conf->prev.far_copies); in raid10_start_reshape()
4568 after_length = ((1 << conf->geo.chunk_shift) * in raid10_start_reshape()
4569 conf->geo.far_copies); in raid10_start_reshape()
4572 if (!test_bit(In_sync, &rdev->flags) in raid10_start_reshape()
4573 && !test_bit(Faulty, &rdev->flags)) in raid10_start_reshape()
4575 if (rdev->raid_disk >= 0) { in raid10_start_reshape()
4576 long long diff = (rdev->new_data_offset in raid10_start_reshape()
4577 - rdev->data_offset); in raid10_start_reshape()
4578 if (!mddev->reshape_backwards) in raid10_start_reshape()
4579 diff = -diff; in raid10_start_reshape()
4589 return -EINVAL; in raid10_start_reshape()
4591 if (spares < mddev->delta_disks) in raid10_start_reshape()
4592 return -EINVAL; in raid10_start_reshape()
4594 conf->offset_diff = min_offset_diff; in raid10_start_reshape()
4595 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4596 if (conf->mirrors_new) { in raid10_start_reshape()
4597 memcpy(conf->mirrors_new, conf->mirrors, in raid10_start_reshape()
4598 sizeof(struct raid10_info)*conf->prev.raid_disks); in raid10_start_reshape()
4600 kfree(conf->mirrors_old); in raid10_start_reshape()
4601 conf->mirrors_old = conf->mirrors; in raid10_start_reshape()
4602 conf->mirrors = conf->mirrors_new; in raid10_start_reshape()
4603 conf->mirrors_new = NULL; in raid10_start_reshape()
4605 setup_geo(&conf->geo, mddev, geo_start); in raid10_start_reshape()
4607 if (mddev->reshape_backwards) { in raid10_start_reshape()
4609 if (size < mddev->array_sectors) { in raid10_start_reshape()
4610 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4613 return -EINVAL; in raid10_start_reshape()
4615 mddev->resync_max_sectors = size; in raid10_start_reshape()
4616 conf->reshape_progress = size; in raid10_start_reshape()
4618 conf->reshape_progress = 0; in raid10_start_reshape()
4619 conf->reshape_safe = conf->reshape_progress; in raid10_start_reshape()
4620 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4622 if (mddev->delta_disks && mddev->bitmap) { in raid10_start_reshape()
4627 newsize = raid10_size(mddev, 0, conf->geo.raid_disks); in raid10_start_reshape()
4630 ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid10_start_reshape()
4638 if (rdev->raid_disk > -1 && in raid10_start_reshape()
4639 !test_bit(Faulty, &rdev->flags)) in raid10_start_reshape()
4640 sb = page_address(rdev->sb_page); in raid10_start_reshape()
4648 if ((sb && (le32_to_cpu(sb->feature_map) & in raid10_start_reshape()
4652 ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid10_start_reshape()
4656 ret = md_cluster_ops->resize_bitmaps(mddev, newsize, oldsize); in raid10_start_reshape()
4658 md_bitmap_resize(mddev->bitmap, oldsize, 0, 0); in raid10_start_reshape()
4663 if (mddev->delta_disks > 0) { in raid10_start_reshape()
4665 if (rdev->raid_disk < 0 && in raid10_start_reshape()
4666 !test_bit(Faulty, &rdev->flags)) { in raid10_start_reshape()
4668 if (rdev->raid_disk >= in raid10_start_reshape()
4669 conf->prev.raid_disks) in raid10_start_reshape()
4670 set_bit(In_sync, &rdev->flags); in raid10_start_reshape()
4672 rdev->recovery_offset = 0; in raid10_start_reshape()
4677 } else if (rdev->raid_disk >= conf->prev.raid_disks in raid10_start_reshape()
4678 && !test_bit(Faulty, &rdev->flags)) { in raid10_start_reshape()
4680 set_bit(In_sync, &rdev->flags); in raid10_start_reshape()
4684 * ->degraded is measured against the larger of the in raid10_start_reshape()
4687 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4688 mddev->degraded = calc_degraded(conf); in raid10_start_reshape()
4689 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4690 mddev->raid_disks = conf->geo.raid_disks; in raid10_start_reshape()
4691 mddev->reshape_position = conf->reshape_progress; in raid10_start_reshape()
4692 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid10_start_reshape()
4694 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); in raid10_start_reshape()
4695 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); in raid10_start_reshape()
4696 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); in raid10_start_reshape()
4697 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); in raid10_start_reshape()
4698 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery); in raid10_start_reshape()
4700 rcu_assign_pointer(mddev->sync_thread, in raid10_start_reshape()
4702 if (!mddev->sync_thread) { in raid10_start_reshape()
4703 ret = -EAGAIN; in raid10_start_reshape()
4706 conf->reshape_checkpoint = jiffies; in raid10_start_reshape()
4707 md_wakeup_thread(mddev->sync_thread); in raid10_start_reshape()
4712 mddev->recovery = 0; in raid10_start_reshape()
4713 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4714 conf->geo = conf->prev; in raid10_start_reshape()
4715 mddev->raid_disks = conf->geo.raid_disks; in raid10_start_reshape()
4717 rdev->new_data_offset = rdev->data_offset; in raid10_start_reshape()
4719 conf->reshape_progress = MaxSector; in raid10_start_reshape()
4720 conf->reshape_safe = MaxSector; in raid10_start_reshape()
4721 mddev->reshape_position = MaxSector; in raid10_start_reshape()
4722 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4726 /* Calculate the last device-address that could contain
4727 * any block from the chunk that includes the array-address 's'
4729 * i.e. the address returned will be chunk-aligned and after
4734 s = (s | geo->chunk_mask) + 1; in last_dev_address()
4735 s >>= geo->chunk_shift; in last_dev_address()
4736 s *= geo->near_copies; in last_dev_address()
4737 s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); in last_dev_address()
4738 s *= geo->far_copies; in last_dev_address()
4739 s <<= geo->chunk_shift; in last_dev_address()
4743 /* Calculate the first device-address that could contain
4744 * any block from the chunk that includes the array-address 's'.
4749 s >>= geo->chunk_shift; in first_dev_address()
4750 s *= geo->near_copies; in first_dev_address()
4751 sector_div(s, geo->raid_disks); in first_dev_address()
4752 s *= geo->far_copies; in first_dev_address()
4753 s <<= geo->chunk_shift; in first_dev_address()
4764 * a chunk, but that is not a problem as mddev->reshape_position in reshape_request()
4769 * we need to flush all reshape requests and update the metadata. in reshape_request()
4780 * we must update the metadata to increase the 'safe' position. in reshape_request()
4787 * (conf->offset_diff - always positive) allows a bit of slack, in reshape_request()
4792 * The means one for each copy for write-out and an extra one for in reshape_request()
4793 * read-in. in reshape_request()
4794 * We store the read-in bio in ->master_bio and the others in in reshape_request()
4795 * ->devs[x].bio and ->devs[x].repl_bio. in reshape_request()
4797 struct r10conf *conf = mddev->private; in reshape_request()
4812 if (mddev->reshape_backwards && in reshape_request()
4813 conf->reshape_progress < raid10_size(mddev, 0, 0)) { in reshape_request()
4815 - conf->reshape_progress); in reshape_request()
4816 } else if (!mddev->reshape_backwards && in reshape_request()
4817 conf->reshape_progress > 0) in reshape_request()
4818 sector_nr = conf->reshape_progress; in reshape_request()
4820 mddev->curr_resync_completed = sector_nr; in reshape_request()
4821 sysfs_notify_dirent_safe(mddev->sysfs_completed); in reshape_request()
4828 * as that doesn't work well for ->reshape_backwards. in reshape_request()
4829 * So just use ->reshape_progress. in reshape_request()
4831 if (mddev->reshape_backwards) { in reshape_request()
4835 next = first_dev_address(conf->reshape_progress - 1, in reshape_request()
4836 &conf->geo); in reshape_request()
4841 safe = last_dev_address(conf->reshape_safe - 1, in reshape_request()
4842 &conf->prev); in reshape_request()
4844 if (next + conf->offset_diff < safe) in reshape_request()
4847 last = conf->reshape_progress - 1; in reshape_request()
4848 sector_nr = last & ~(sector_t)(conf->geo.chunk_mask in reshape_request()
4849 & conf->prev.chunk_mask); in reshape_request()
4851 sector_nr = last + 1 - RESYNC_SECTORS; in reshape_request()
4856 next = last_dev_address(conf->reshape_progress, &conf->geo); in reshape_request()
4861 safe = first_dev_address(conf->reshape_safe, &conf->prev); in reshape_request()
4863 /* Need to update metadata if 'next' might be beyond 'safe' in reshape_request()
4866 if (next > safe + conf->offset_diff) in reshape_request()
4869 sector_nr = conf->reshape_progress; in reshape_request()
4870 last = sector_nr | (conf->geo.chunk_mask in reshape_request()
4871 & conf->prev.chunk_mask); in reshape_request()
4874 last = sector_nr + RESYNC_SECTORS - 1; in reshape_request()
4878 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { in reshape_request()
4879 /* Need to update reshape_position in metadata */ in reshape_request()
4881 mddev->reshape_position = conf->reshape_progress; in reshape_request()
4882 if (mddev->reshape_backwards) in reshape_request()
4883 mddev->curr_resync_completed = raid10_size(mddev, 0, 0) in reshape_request()
4884 - conf->reshape_progress; in reshape_request()
4886 mddev->curr_resync_completed = conf->reshape_progress; in reshape_request()
4887 conf->reshape_checkpoint = jiffies; in reshape_request()
4888 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in reshape_request()
4889 md_wakeup_thread(mddev->thread); in reshape_request()
4890 wait_event(mddev->sb_wait, mddev->sb_flags == 0 || in reshape_request()
4891 test_bit(MD_RECOVERY_INTR, &mddev->recovery)); in reshape_request()
4892 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { in reshape_request()
4896 conf->reshape_safe = mddev->reshape_position; in reshape_request()
4904 r10_bio->state = 0; in reshape_request()
4906 atomic_set(&r10_bio->remaining, 0); in reshape_request()
4907 r10_bio->mddev = mddev; in reshape_request()
4908 r10_bio->sector = sector_nr; in reshape_request()
4909 set_bit(R10BIO_IsReshape, &r10_bio->state); in reshape_request()
4910 r10_bio->sectors = last - sector_nr + 1; in reshape_request()
4912 BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); in reshape_request()
4919 mempool_free(r10_bio, &conf->r10buf_pool); in reshape_request()
4920 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in reshape_request()
4924 read_bio = bio_alloc_bioset(rdev->bdev, RESYNC_PAGES, REQ_OP_READ, in reshape_request()
4925 GFP_KERNEL, &mddev->bio_set); in reshape_request()
4926 read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr in reshape_request()
4927 + rdev->data_offset); in reshape_request()
4928 read_bio->bi_private = r10_bio; in reshape_request()
4929 read_bio->bi_end_io = end_reshape_read; in reshape_request()
4930 r10_bio->master_bio = read_bio; in reshape_request()
4931 r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; in reshape_request()
4937 if (mddev_is_clustered(mddev) && conf->cluster_sync_high <= sector_nr) { in reshape_request()
4941 conf->cluster_sync_low = sector_nr; in reshape_request()
4942 conf->cluster_sync_high = sector_nr + CLUSTER_RESYNC_WINDOW_SECTORS; in reshape_request()
4943 sb = page_address(rdev->sb_page); in reshape_request()
4945 sb_reshape_pos = le64_to_cpu(sb->reshape_position); in reshape_request()
4949 * update cluster_sync_low until it has finished reshape. in reshape_request()
4951 if (sb_reshape_pos < conf->cluster_sync_low) in reshape_request()
4952 conf->cluster_sync_low = sb_reshape_pos; in reshape_request()
4955 md_cluster_ops->resync_info_update(mddev, conf->cluster_sync_low, in reshape_request()
4956 conf->cluster_sync_high); in reshape_request()
4960 __raid10_find_phys(&conf->geo, r10_bio); in reshape_request()
4963 read_bio->bi_next = NULL; in reshape_request()
4966 for (s = 0; s < conf->copies*2; s++) { in reshape_request()
4968 int d = r10_bio->devs[s/2].devnum; in reshape_request()
4971 rdev2 = rcu_dereference(conf->mirrors[d].replacement); in reshape_request()
4972 b = r10_bio->devs[s/2].repl_bio; in reshape_request()
4974 rdev2 = rcu_dereference(conf->mirrors[d].rdev); in reshape_request()
4975 b = r10_bio->devs[s/2].bio; in reshape_request()
4977 if (!rdev2 || test_bit(Faulty, &rdev2->flags)) in reshape_request()
4980 bio_set_dev(b, rdev2->bdev); in reshape_request()
4981 b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + in reshape_request()
4982 rdev2->new_data_offset; in reshape_request()
4983 b->bi_end_io = end_reshape_write; in reshape_request()
4984 b->bi_opf = REQ_OP_WRITE; in reshape_request()
4985 b->bi_next = blist; in reshape_request()
4992 pages = get_resync_pages(r10_bio->devs[0].bio)->pages; in reshape_request()
4995 int len = (max_sectors - s) << 9; in reshape_request()
4998 for (bio = blist; bio ; bio = bio->bi_next) { in reshape_request()
5000 bio->bi_status = BLK_STS_RESOURCE; in reshape_request()
5009 r10_bio->sectors = nr_sectors; in reshape_request()
5012 md_sync_acct_bio(read_bio, r10_bio->sectors); in reshape_request()
5013 atomic_inc(&r10_bio->remaining); in reshape_request()
5014 read_bio->bi_next = NULL; in reshape_request()
5023 * update reshape_progress in reshape_request()
5025 if (mddev->reshape_backwards) in reshape_request()
5026 conf->reshape_progress -= sectors_done; in reshape_request()
5028 conf->reshape_progress += sectors_done; in reshape_request()
5040 * If we got a read error then we do sync 1-page reads from in reshape_request_write()
5041 * elsewhere until we find the data - or give up. in reshape_request_write()
5043 struct r10conf *conf = mddev->private; in reshape_request_write()
5046 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in reshape_request_write()
5049 md_done_sync(mddev, r10_bio->sectors, 0); in reshape_request_write()
5056 atomic_set(&r10_bio->remaining, 1); in reshape_request_write()
5057 for (s = 0; s < conf->copies*2; s++) { in reshape_request_write()
5059 int d = r10_bio->devs[s/2].devnum; in reshape_request_write()
5063 rdev = rcu_dereference(conf->mirrors[d].replacement); in reshape_request_write()
5064 b = r10_bio->devs[s/2].repl_bio; in reshape_request_write()
5066 rdev = rcu_dereference(conf->mirrors[d].rdev); in reshape_request_write()
5067 b = r10_bio->devs[s/2].bio; in reshape_request_write()
5069 if (!rdev || test_bit(Faulty, &rdev->flags)) { in reshape_request_write()
5073 atomic_inc(&rdev->nr_pending); in reshape_request_write()
5075 md_sync_acct_bio(b, r10_bio->sectors); in reshape_request_write()
5076 atomic_inc(&r10_bio->remaining); in reshape_request_write()
5077 b->bi_next = NULL; in reshape_request_write()
5085 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) in end_reshape()
5088 spin_lock_irq(&conf->device_lock); in end_reshape()
5089 conf->prev = conf->geo; in end_reshape()
5090 md_finish_reshape(conf->mddev); in end_reshape()
5092 conf->reshape_progress = MaxSector; in end_reshape()
5093 conf->reshape_safe = MaxSector; in end_reshape()
5094 spin_unlock_irq(&conf->device_lock); in end_reshape()
5096 if (conf->mddev->queue) in end_reshape()
5098 conf->fullsync = 0; in end_reshape()
5103 struct r10conf *conf = mddev->private; in raid10_update_reshape_pos()
5106 md_cluster_ops->resync_info_get(mddev, &lo, &hi); in raid10_update_reshape_pos()
5107 if (((mddev->reshape_position <= hi) && (mddev->reshape_position >= lo)) in raid10_update_reshape_pos()
5108 || mddev->reshape_position == MaxSector) in raid10_update_reshape_pos()
5109 conf->reshape_progress = mddev->reshape_position; in raid10_update_reshape_pos()
5118 int sectors = r10_bio->sectors; in handle_reshape_read_error()
5119 struct r10conf *conf = mddev->private; in handle_reshape_read_error()
5125 r10b = kmalloc(struct_size(r10b, devs, conf->copies), GFP_NOIO); in handle_reshape_read_error()
5127 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in handle_reshape_read_error()
5128 return -ENOMEM; in handle_reshape_read_error()
5132 pages = get_resync_pages(r10_bio->devs[0].bio)->pages; in handle_reshape_read_error()
5134 r10b->sector = r10_bio->sector; in handle_reshape_read_error()
5135 __raid10_find_phys(&conf->prev, r10b); in handle_reshape_read_error()
5147 int d = r10b->devs[slot].devnum; in handle_reshape_read_error()
5148 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); in handle_reshape_read_error()
5151 test_bit(Faulty, &rdev->flags) || in handle_reshape_read_error()
5152 !test_bit(In_sync, &rdev->flags)) in handle_reshape_read_error()
5155 addr = r10b->devs[slot].addr + idx * PAGE_SIZE; in handle_reshape_read_error()
5156 atomic_inc(&rdev->nr_pending); in handle_reshape_read_error()
5169 if (slot >= conf->copies) in handle_reshape_read_error()
5178 &mddev->recovery); in handle_reshape_read_error()
5180 return -EIO; in handle_reshape_read_error()
5182 sectors -= s; in handle_reshape_read_error()
5192 struct mddev *mddev = r10_bio->mddev; in end_reshape_write()
5193 struct r10conf *conf = mddev->private; in end_reshape_write()
5201 rdev = conf->mirrors[d].replacement; in end_reshape_write()
5204 rdev = conf->mirrors[d].rdev; in end_reshape_write()
5207 if (bio->bi_status) { in end_reshape_write()
5218 if (!atomic_dec_and_test(&r10_bio->remaining)) in end_reshape_request()
5220 md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); in end_reshape_request()
5221 bio_put(r10_bio->master_bio); in end_reshape_request()
5227 struct r10conf *conf = mddev->private; in raid10_finish_reshape()
5229 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) in raid10_finish_reshape()
5232 if (mddev->delta_disks > 0) { in raid10_finish_reshape()
5233 if (mddev->recovery_cp > mddev->resync_max_sectors) { in raid10_finish_reshape()
5234 mddev->recovery_cp = mddev->resync_max_sectors; in raid10_finish_reshape()
5235 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_finish_reshape()
5237 mddev->resync_max_sectors = mddev->array_sectors; in raid10_finish_reshape()
5241 for (d = conf->geo.raid_disks ; in raid10_finish_reshape()
5242 d < conf->geo.raid_disks - mddev->delta_disks; in raid10_finish_reshape()
5244 struct md_rdev *rdev = rcu_dereference(conf->mirrors[d].rdev); in raid10_finish_reshape()
5246 clear_bit(In_sync, &rdev->flags); in raid10_finish_reshape()
5247 rdev = rcu_dereference(conf->mirrors[d].replacement); in raid10_finish_reshape()
5249 clear_bit(In_sync, &rdev->flags); in raid10_finish_reshape()
5253 mddev->layout = mddev->new_layout; in raid10_finish_reshape()
5254 mddev->chunk_sectors = 1 << conf->geo.chunk_shift; in raid10_finish_reshape()
5255 mddev->reshape_position = MaxSector; in raid10_finish_reshape()
5256 mddev->delta_disks = 0; in raid10_finish_reshape()
5257 mddev->reshape_backwards = 0; in raid10_finish_reshape()
5298 MODULE_ALIAS("md-personality-9"); /* RAID10 */
5299 MODULE_ALIAS("md-raid10");
5300 MODULE_ALIAS("md-level-10");