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.
26 #include "md-bitmap.h"
79 #include "raid1-10.c"
84 write_sequnlock_irq(&(conf)->resync_lock); \
87 #define cmd_after(conf) write_seqlock_irq(&(conf)->resync_lock)
90 wait_event_cmd((conf)->wait_barrier, cond, cmd_before(conf, cmd), \
97 * for resync bio, r10bio pointer can be retrieved from the per-bio
102 return get_resync_pages(bio)->raid_bio; in get_resync_r10bio()
108 int size = offsetof(struct r10bio, devs[conf->geo.raid_disks]); in r10bio_pool_alloc()
143 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || in r10buf_pool_alloc()
144 test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) in r10buf_pool_alloc()
145 nalloc = conf->copies; /* resync */ in r10buf_pool_alloc()
150 if (!conf->have_replacement) in r10buf_pool_alloc()
161 for (j = nalloc ; j-- ; ) { in r10buf_pool_alloc()
165 bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); in r10buf_pool_alloc()
166 r10_bio->devs[j].bio = bio; in r10buf_pool_alloc()
167 if (!conf->have_replacement) in r10buf_pool_alloc()
172 bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); in r10buf_pool_alloc()
173 r10_bio->devs[j].repl_bio = bio; in r10buf_pool_alloc()
180 struct bio *rbio = r10_bio->devs[j].repl_bio; in r10buf_pool_alloc()
187 bio = r10_bio->devs[j].bio; in r10buf_pool_alloc()
190 &conf->mddev->recovery)) { in r10buf_pool_alloc()
198 rp->raid_bio = r10_bio; in r10buf_pool_alloc()
199 bio->bi_private = rp; in r10buf_pool_alloc()
202 rbio->bi_private = rp_repl; in r10buf_pool_alloc()
209 while (--j >= 0) in r10buf_pool_alloc()
215 if (r10_bio->devs[j].bio) in r10buf_pool_alloc()
216 bio_uninit(r10_bio->devs[j].bio); in r10buf_pool_alloc()
217 kfree(r10_bio->devs[j].bio); in r10buf_pool_alloc()
218 if (r10_bio->devs[j].repl_bio) in r10buf_pool_alloc()
219 bio_uninit(r10_bio->devs[j].repl_bio); in r10buf_pool_alloc()
220 kfree(r10_bio->devs[j].repl_bio); in r10buf_pool_alloc()
235 for (j = conf->copies; j--; ) { in r10buf_pool_free()
236 struct bio *bio = r10bio->devs[j].bio; in r10buf_pool_free()
245 bio = r10bio->devs[j].repl_bio; in r10buf_pool_free()
262 for (i = 0; i < conf->geo.raid_disks; i++) { in put_all_bios()
263 struct bio **bio = & r10_bio->devs[i].bio; in put_all_bios()
267 bio = &r10_bio->devs[i].repl_bio; in put_all_bios()
268 if (r10_bio->read_slot < 0 && !BIO_SPECIAL(*bio)) in put_all_bios()
276 struct r10conf *conf = r10_bio->mddev->private; in free_r10bio()
279 mempool_free(r10_bio, &conf->r10bio_pool); in free_r10bio()
284 struct r10conf *conf = r10_bio->mddev->private; in put_buf()
286 mempool_free(r10_bio, &conf->r10buf_pool); in put_buf()
293 if (wq_has_sleeper(&conf->wait_barrier)) in wake_up_barrier()
294 wake_up(&conf->wait_barrier); in wake_up_barrier()
300 struct mddev *mddev = r10_bio->mddev; in reschedule_retry()
301 struct r10conf *conf = mddev->private; in reschedule_retry()
303 spin_lock_irqsave(&conf->device_lock, flags); in reschedule_retry()
304 list_add(&r10_bio->retry_list, &conf->retry_list); in reschedule_retry()
305 conf->nr_queued ++; in reschedule_retry()
306 spin_unlock_irqrestore(&conf->device_lock, flags); in reschedule_retry()
309 wake_up(&conf->wait_barrier); in reschedule_retry()
311 md_wakeup_thread(mddev->thread); in reschedule_retry()
321 struct bio *bio = r10_bio->master_bio; in raid_end_bio_io()
322 struct r10conf *conf = r10_bio->mddev->private; in raid_end_bio_io()
324 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in raid_end_bio_io()
325 bio->bi_status = BLK_STS_IOERR; in raid_end_bio_io()
338 * Update disk head position estimator based on IRQ completion info.
342 struct r10conf *conf = r10_bio->mddev->private; in update_head_pos()
344 conf->mirrors[r10_bio->devs[slot].devnum].head_position = in update_head_pos()
345 r10_bio->devs[slot].addr + (r10_bio->sectors); in update_head_pos()
357 for (slot = 0; slot < conf->geo.raid_disks; slot++) { in find_bio_disk()
358 if (r10_bio->devs[slot].bio == bio) in find_bio_disk()
360 if (r10_bio->devs[slot].repl_bio == bio) { in find_bio_disk()
372 return r10_bio->devs[slot].devnum; in find_bio_disk()
377 int uptodate = !bio->bi_status; in raid10_end_read_request()
378 struct r10bio *r10_bio = bio->bi_private; in raid10_end_read_request()
381 struct r10conf *conf = r10_bio->mddev->private; in raid10_end_read_request()
383 slot = r10_bio->read_slot; in raid10_end_read_request()
384 rdev = r10_bio->devs[slot].rdev; in raid10_end_read_request()
397 * user-side. So if something waits for IO, then it will in raid10_end_read_request()
400 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_end_read_request()
407 if (!_enough(conf, test_bit(R10BIO_Previous, &r10_bio->state), in raid10_end_read_request()
408 rdev->raid_disk)) in raid10_end_read_request()
413 rdev_dec_pending(rdev, conf->mddev); in raid10_end_read_request()
416 * oops, read error - keep the refcount on the rdev in raid10_end_read_request()
419 mdname(conf->mddev), in raid10_end_read_request()
420 rdev->bdev, in raid10_end_read_request()
421 (unsigned long long)r10_bio->sector); in raid10_end_read_request()
422 set_bit(R10BIO_ReadError, &r10_bio->state); in raid10_end_read_request()
429 struct mddev *mddev = r10_bio->mddev; in close_write()
436 if (atomic_dec_and_test(&r10_bio->remaining)) { in one_write_done()
437 if (test_bit(R10BIO_WriteError, &r10_bio->state)) in one_write_done()
441 if (test_bit(R10BIO_MadeGood, &r10_bio->state)) in one_write_done()
451 struct r10bio *r10_bio = bio->bi_private; in raid10_end_write_request()
454 struct r10conf *conf = r10_bio->mddev->private; in raid10_end_write_request()
460 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; in raid10_end_write_request()
465 rdev = conf->mirrors[dev].replacement; in raid10_end_write_request()
469 rdev = conf->mirrors[dev].rdev; in raid10_end_write_request()
474 if (bio->bi_status && !discard_error) { in raid10_end_write_request()
479 md_error(rdev->mddev, rdev); in raid10_end_write_request()
481 set_bit(WriteErrorSeen, &rdev->flags); in raid10_end_write_request()
482 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in raid10_end_write_request()
484 &rdev->mddev->recovery); in raid10_end_write_request()
487 if (test_bit(FailFast, &rdev->flags) && in raid10_end_write_request()
488 (bio->bi_opf & MD_FAILFAST)) { in raid10_end_write_request()
489 md_error(rdev->mddev, rdev); in raid10_end_write_request()
496 if (!test_bit(Faulty, &rdev->flags)) in raid10_end_write_request()
497 set_bit(R10BIO_WriteError, &r10_bio->state); in raid10_end_write_request()
500 r10_bio->devs[slot].bio = NULL; in raid10_end_write_request()
512 * user-side. So if something waits for IO, then it will in raid10_end_write_request()
519 * before rdev->recovery_offset, but for simplicity we don't in raid10_end_write_request()
522 if (test_bit(In_sync, &rdev->flags) && in raid10_end_write_request()
523 !test_bit(Faulty, &rdev->flags)) in raid10_end_write_request()
524 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_end_write_request()
527 if (rdev_has_badblock(rdev, r10_bio->devs[slot].addr, in raid10_end_write_request()
528 r10_bio->sectors) && in raid10_end_write_request()
532 r10_bio->devs[slot].repl_bio = IO_MADE_GOOD; in raid10_end_write_request()
534 r10_bio->devs[slot].bio = IO_MADE_GOOD; in raid10_end_write_request()
536 set_bit(R10BIO_MadeGood, &r10_bio->state); in raid10_end_write_request()
547 rdev_dec_pending(rdev, conf->mddev); in raid10_end_write_request()
587 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; in __raid10_find_phys()
588 last_far_set_start *= geo->far_set_size; in __raid10_find_phys()
590 last_far_set_size = geo->far_set_size; in __raid10_find_phys()
591 last_far_set_size += (geo->raid_disks % geo->far_set_size); in __raid10_find_phys()
594 chunk = r10bio->sector >> geo->chunk_shift; in __raid10_find_phys()
595 sector = r10bio->sector & geo->chunk_mask; in __raid10_find_phys()
597 chunk *= geo->near_copies; in __raid10_find_phys()
599 dev = sector_div(stripe, geo->raid_disks); in __raid10_find_phys()
600 if (geo->far_offset) in __raid10_find_phys()
601 stripe *= geo->far_copies; in __raid10_find_phys()
603 sector += stripe << geo->chunk_shift; in __raid10_find_phys()
606 for (n = 0; n < geo->near_copies; n++) { in __raid10_find_phys()
610 r10bio->devs[slot].devnum = d; in __raid10_find_phys()
611 r10bio->devs[slot].addr = s; in __raid10_find_phys()
614 for (f = 1; f < geo->far_copies; f++) { in __raid10_find_phys()
615 set = d / geo->far_set_size; in __raid10_find_phys()
616 d += geo->near_copies; in __raid10_find_phys()
618 if ((geo->raid_disks % geo->far_set_size) && in __raid10_find_phys()
620 d -= last_far_set_start; in __raid10_find_phys()
624 d %= geo->far_set_size; in __raid10_find_phys()
625 d += geo->far_set_size * set; in __raid10_find_phys()
627 s += geo->stride; in __raid10_find_phys()
628 r10bio->devs[slot].devnum = d; in __raid10_find_phys()
629 r10bio->devs[slot].addr = s; in __raid10_find_phys()
633 if (dev >= geo->raid_disks) { in __raid10_find_phys()
635 sector += (geo->chunk_mask + 1); in __raid10_find_phys()
642 struct geom *geo = &conf->geo; in raid10_find_phys()
644 if (conf->reshape_progress != MaxSector && in raid10_find_phys()
645 ((r10bio->sector >= conf->reshape_progress) != in raid10_find_phys()
646 conf->mddev->reshape_backwards)) { in raid10_find_phys()
647 set_bit(R10BIO_Previous, &r10bio->state); in raid10_find_phys()
648 geo = &conf->prev; in raid10_find_phys()
650 clear_bit(R10BIO_Previous, &r10bio->state); in raid10_find_phys()
658 /* Never use conf->prev as this is only called during resync in raid10_find_virt()
661 struct geom *geo = &conf->geo; in raid10_find_virt()
662 int far_set_start = (dev / geo->far_set_size) * geo->far_set_size; in raid10_find_virt()
663 int far_set_size = geo->far_set_size; in raid10_find_virt()
666 if (geo->raid_disks % geo->far_set_size) { in raid10_find_virt()
667 last_far_set_start = (geo->raid_disks / geo->far_set_size) - 1; in raid10_find_virt()
668 last_far_set_start *= geo->far_set_size; in raid10_find_virt()
671 far_set_size = geo->far_set_size; in raid10_find_virt()
672 far_set_size += (geo->raid_disks % geo->far_set_size); in raid10_find_virt()
677 offset = sector & geo->chunk_mask; in raid10_find_virt()
678 if (geo->far_offset) { in raid10_find_virt()
679 int fc; in raid10_find_virt() local
680 chunk = sector >> geo->chunk_shift; in raid10_find_virt()
681 fc = sector_div(chunk, geo->far_copies); in raid10_find_virt()
682 dev -= fc * geo->near_copies; in raid10_find_virt()
686 while (sector >= geo->stride) { in raid10_find_virt()
687 sector -= geo->stride; in raid10_find_virt()
688 if (dev < (geo->near_copies + far_set_start)) in raid10_find_virt()
689 dev += far_set_size - geo->near_copies; in raid10_find_virt()
691 dev -= geo->near_copies; in raid10_find_virt()
693 chunk = sector >> geo->chunk_shift; in raid10_find_virt()
695 vchunk = chunk * geo->raid_disks + dev; in raid10_find_virt()
696 sector_div(vchunk, geo->near_copies); in raid10_find_virt()
697 return (vchunk << geo->chunk_shift) + offset; in raid10_find_virt()
702 * be done. There is a per-array 'next expected sequential IO' sector
703 * number - if this matches on the next IO then we use the last disk.
704 * There is also a per-disk 'last know head position' sector that is
706 * completion handlers update this position correctly. If there is no
723 const sector_t this_sector = r10_bio->sector; in read_balance()
725 int sectors = r10_bio->sectors; in read_balance()
733 struct geom *geo = &conf->geo; in read_balance()
736 best_dist_slot = -1; in read_balance()
743 clear_bit(R10BIO_FailFast, &r10_bio->state); in read_balance()
745 if (raid1_should_read_first(conf->mddev, this_sector, sectors)) in read_balance()
748 for (slot = 0; slot < conf->copies ; slot++) { in read_balance()
755 if (r10_bio->devs[slot].bio == IO_BLOCKED) in read_balance()
757 disk = r10_bio->devs[slot].devnum; in read_balance()
758 rdev = conf->mirrors[disk].replacement; in read_balance()
759 if (rdev == NULL || test_bit(Faulty, &rdev->flags) || in read_balance()
760 r10_bio->devs[slot].addr + sectors > in read_balance()
761 rdev->recovery_offset) in read_balance()
762 rdev = conf->mirrors[disk].rdev; in read_balance()
764 test_bit(Faulty, &rdev->flags)) in read_balance()
766 if (!test_bit(In_sync, &rdev->flags) && in read_balance()
767 r10_bio->devs[slot].addr + sectors > rdev->recovery_offset) in read_balance()
770 dev_sector = r10_bio->devs[slot].addr; in read_balance()
781 bad_sectors -= (dev_sector - first_bad); in read_balance()
788 first_bad - dev_sector; in read_balance()
805 nonrot = bdev_nonrot(rdev->bdev); in read_balance()
807 pending = atomic_read(&rdev->nr_pending); in read_balance()
816 set_bit(R10BIO_FailFast, &r10_bio->state); in read_balance()
821 if (geo->near_copies > 1 && !pending) in read_balance()
825 else if (geo->far_copies > 1) in read_balance()
826 new_distance = r10_bio->devs[slot].addr; in read_balance()
828 new_distance = abs(r10_bio->devs[slot].addr - in read_balance()
829 conf->mirrors[disk].head_position); in read_balance()
837 if (slot >= conf->copies) { in read_balance()
848 atomic_inc(&rdev->nr_pending); in read_balance()
849 r10_bio->read_slot = slot; in read_balance()
862 spin_lock_irq(&conf->device_lock); in flush_pending_writes()
864 if (conf->pending_bio_list.head) { in flush_pending_writes()
868 bio = bio_list_get(&conf->pending_bio_list); in flush_pending_writes()
869 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
873 * current->state might be TASK_UNINTERRUPTIBLE which will in flush_pending_writes()
877 * is a false-positive. Silence the warning by resetting in flush_pending_writes()
883 raid1_prepare_flush_writes(conf->mddev); in flush_pending_writes()
884 wake_up(&conf->wait_barrier); in flush_pending_writes()
887 struct bio *next = bio->bi_next; in flush_pending_writes()
895 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
907 * We choose only to raise the barrier if no-one is waiting for the
922 write_seqlock_irq(&conf->resync_lock); in raise_barrier()
924 if (WARN_ON_ONCE(force && !conf->barrier)) in raise_barrier()
928 wait_event_barrier(conf, force || !conf->nr_waiting); in raise_barrier()
931 WRITE_ONCE(conf->barrier, conf->barrier + 1); in raise_barrier()
934 wait_event_barrier(conf, !atomic_read(&conf->nr_pending) && in raise_barrier()
935 conf->barrier < RESYNC_DEPTH); in raise_barrier()
937 write_sequnlock_irq(&conf->resync_lock); in raise_barrier()
944 write_seqlock_irqsave(&conf->resync_lock, flags); in lower_barrier()
945 WRITE_ONCE(conf->barrier, conf->barrier - 1); in lower_barrier()
946 write_sequnlock_irqrestore(&conf->resync_lock, flags); in lower_barrier()
947 wake_up(&conf->wait_barrier); in lower_barrier()
952 struct bio_list *bio_list = current->bio_list; in stop_waiting_barrier()
956 if (!conf->barrier) in stop_waiting_barrier()
961 * rising completely), and the pre-process bio queue isn't empty, then in stop_waiting_barrier()
965 if (atomic_read(&conf->nr_pending) && bio_list && in stop_waiting_barrier()
970 thread = rcu_dereference_protected(conf->mddev->thread, true); in stop_waiting_barrier()
976 if (thread->tsk == current) { in stop_waiting_barrier()
977 WARN_ON_ONCE(atomic_read(&conf->nr_pending) == 0); in stop_waiting_barrier()
986 unsigned int seq = read_seqbegin(&conf->resync_lock); in wait_barrier_nolock()
988 if (READ_ONCE(conf->barrier)) in wait_barrier_nolock()
991 atomic_inc(&conf->nr_pending); in wait_barrier_nolock()
992 if (!read_seqretry(&conf->resync_lock, seq)) in wait_barrier_nolock()
995 if (atomic_dec_and_test(&conf->nr_pending)) in wait_barrier_nolock()
1008 write_seqlock_irq(&conf->resync_lock); in wait_barrier()
1009 if (conf->barrier) { in wait_barrier()
1014 conf->nr_waiting++; in wait_barrier()
1015 mddev_add_trace_msg(conf->mddev, "raid10 wait barrier"); in wait_barrier()
1017 conf->nr_waiting--; in wait_barrier()
1019 if (!conf->nr_waiting) in wait_barrier()
1020 wake_up(&conf->wait_barrier); in wait_barrier()
1024 atomic_inc(&conf->nr_pending); in wait_barrier()
1025 write_sequnlock_irq(&conf->resync_lock); in wait_barrier()
1031 if ((atomic_dec_and_test(&conf->nr_pending)) || in allow_barrier()
1032 (conf->array_freeze_pending)) in allow_barrier()
1045 * pending IO requests to complete or be queued for re-try. in freeze_array()
1050 write_seqlock_irq(&conf->resync_lock); in freeze_array()
1051 conf->array_freeze_pending++; in freeze_array()
1052 WRITE_ONCE(conf->barrier, conf->barrier + 1); in freeze_array()
1053 conf->nr_waiting++; in freeze_array()
1054 wait_event_barrier_cmd(conf, atomic_read(&conf->nr_pending) == in freeze_array()
1055 conf->nr_queued + extra, flush_pending_writes(conf)); in freeze_array()
1056 conf->array_freeze_pending--; in freeze_array()
1057 write_sequnlock_irq(&conf->resync_lock); in freeze_array()
1063 write_seqlock_irq(&conf->resync_lock); in unfreeze_array()
1064 WRITE_ONCE(conf->barrier, conf->barrier - 1); in unfreeze_array()
1065 conf->nr_waiting--; in unfreeze_array()
1066 wake_up(&conf->wait_barrier); in unfreeze_array()
1067 write_sequnlock_irq(&conf->resync_lock); in unfreeze_array()
1073 if (!test_bit(MD_RECOVERY_RESHAPE, &rdev->mddev->recovery) || in choose_data_offset()
1074 test_bit(R10BIO_Previous, &r10_bio->state)) in choose_data_offset()
1075 return rdev->data_offset; in choose_data_offset()
1077 return rdev->new_data_offset; in choose_data_offset()
1083 struct mddev *mddev = plug->cb.data; in raid10_unplug()
1084 struct r10conf *conf = mddev->private; in raid10_unplug()
1088 spin_lock_irq(&conf->device_lock); in raid10_unplug()
1089 bio_list_merge(&conf->pending_bio_list, &plug->pending); in raid10_unplug()
1090 spin_unlock_irq(&conf->device_lock); in raid10_unplug()
1092 md_wakeup_thread(mddev->thread); in raid10_unplug()
1097 /* we aren't scheduling, so we can do the write-out directly. */ in raid10_unplug()
1098 bio = bio_list_get(&plug->pending); in raid10_unplug()
1103 struct bio *next = bio->bi_next; in raid10_unplug()
1122 if (!wait_barrier(conf, bio->bi_opf & REQ_NOWAIT)) { in regular_request_wait()
1126 while (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in regular_request_wait()
1127 bio->bi_iter.bi_sector < conf->reshape_progress && in regular_request_wait()
1128 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) { in regular_request_wait()
1130 if (bio->bi_opf & REQ_NOWAIT) { in regular_request_wait()
1134 mddev_add_trace_msg(conf->mddev, "raid10 wait reshape"); in regular_request_wait()
1135 wait_event(conf->wait_barrier, in regular_request_wait()
1136 conf->reshape_progress <= bio->bi_iter.bi_sector || in regular_request_wait()
1137 conf->reshape_progress >= bio->bi_iter.bi_sector + in regular_request_wait()
1147 struct r10conf *conf = mddev->private; in raid10_read_request()
1152 int slot = r10_bio->read_slot; in raid10_read_request()
1157 if (slot >= 0 && r10_bio->devs[slot].rdev) { in raid10_read_request()
1172 disk = r10_bio->devs[slot].devnum; in raid10_read_request()
1173 err_rdev = conf->mirrors[disk].rdev; in raid10_read_request()
1175 snprintf(b, sizeof(b), "%pg", err_rdev->bdev); in raid10_read_request()
1179 err_rdev = r10_bio->devs[slot].rdev; in raid10_read_request()
1183 if (!regular_request_wait(mddev, conf, bio, r10_bio->sectors)) in raid10_read_request()
1190 (unsigned long long)r10_bio->sector); in raid10_read_request()
1198 rdev->bdev, in raid10_read_request()
1199 (unsigned long long)r10_bio->sector); in raid10_read_request()
1202 gfp, &conf->bio_split); in raid10_read_request()
1212 r10_bio->master_bio = bio; in raid10_read_request()
1213 r10_bio->sectors = max_sectors; in raid10_read_request()
1215 slot = r10_bio->read_slot; in raid10_read_request()
1219 r10_bio->master_bio = bio; in raid10_read_request()
1221 read_bio = bio_alloc_clone(rdev->bdev, bio, gfp, &mddev->bio_set); in raid10_read_request()
1223 r10_bio->devs[slot].bio = read_bio; in raid10_read_request()
1224 r10_bio->devs[slot].rdev = rdev; in raid10_read_request()
1226 read_bio->bi_iter.bi_sector = r10_bio->devs[slot].addr + in raid10_read_request()
1228 read_bio->bi_end_io = raid10_end_read_request; in raid10_read_request()
1229 if (test_bit(FailFast, &rdev->flags) && in raid10_read_request()
1230 test_bit(R10BIO_FailFast, &r10_bio->state)) in raid10_read_request()
1231 read_bio->bi_opf |= MD_FAILFAST; in raid10_read_request()
1232 read_bio->bi_private = r10_bio; in raid10_read_request()
1233 mddev_trace_remap(mddev, read_bio, r10_bio->sector); in raid10_read_request()
1237 atomic_dec(&rdev->nr_pending); in raid10_read_request()
1238 bio->bi_status = errno_to_blk_status(error); in raid10_read_request()
1239 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_read_request()
1248 struct r10conf *conf = mddev->private; in raid10_write_one_disk()
1250 int devnum = r10_bio->devs[n_copy].devnum; in raid10_write_one_disk()
1253 rdev = replacement ? conf->mirrors[devnum].replacement : in raid10_write_one_disk()
1254 conf->mirrors[devnum].rdev; in raid10_write_one_disk()
1256 mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, &mddev->bio_set); in raid10_write_one_disk()
1258 r10_bio->devs[n_copy].repl_bio = mbio; in raid10_write_one_disk()
1260 r10_bio->devs[n_copy].bio = mbio; in raid10_write_one_disk()
1262 mbio->bi_iter.bi_sector = (r10_bio->devs[n_copy].addr + in raid10_write_one_disk()
1264 mbio->bi_end_io = raid10_end_write_request; in raid10_write_one_disk()
1266 &conf->mirrors[devnum].rdev->flags) in raid10_write_one_disk()
1268 mbio->bi_opf |= MD_FAILFAST; in raid10_write_one_disk()
1269 mbio->bi_private = r10_bio; in raid10_write_one_disk()
1270 mddev_trace_remap(mddev, mbio, r10_bio->sector); in raid10_write_one_disk()
1272 mbio->bi_bdev = (void *)rdev; in raid10_write_one_disk()
1274 atomic_inc(&r10_bio->remaining); in raid10_write_one_disk()
1276 if (!raid1_add_bio_to_plug(mddev, mbio, raid10_unplug, conf->copies)) { in raid10_write_one_disk()
1277 spin_lock_irqsave(&conf->device_lock, flags); in raid10_write_one_disk()
1278 bio_list_add(&conf->pending_bio_list, mbio); in raid10_write_one_disk()
1279 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_write_one_disk()
1280 md_wakeup_thread(mddev->thread); in raid10_write_one_disk()
1286 struct r10conf *conf = mddev->private; in wait_blocked_dev()
1292 for (i = 0; i < conf->copies; i++) { in wait_blocked_dev()
1295 rdev = conf->mirrors[i].rdev; in wait_blocked_dev()
1297 sector_t dev_sector = r10_bio->devs[i].addr; in wait_blocked_dev()
1303 if (test_bit(WriteErrorSeen, &rdev->flags) && in wait_blocked_dev()
1304 r10_bio->sectors && in wait_blocked_dev()
1306 r10_bio->sectors) < 0) in wait_blocked_dev()
1311 set_bit(BlockedBadBlocks, &rdev->flags); in wait_blocked_dev()
1315 atomic_inc(&rdev->nr_pending); in wait_blocked_dev()
1320 rrdev = conf->mirrors[i].replacement; in wait_blocked_dev()
1322 atomic_inc(&rrdev->nr_pending); in wait_blocked_dev()
1331 mddev_add_trace_msg(conf->mddev, in wait_blocked_dev()
1333 __func__, blocked_rdev->raid_disk); in wait_blocked_dev()
1343 struct r10conf *conf = mddev->private; in raid10_write_request()
1350 md_cluster_ops->area_resyncing(mddev, WRITE, in raid10_write_request()
1351 bio->bi_iter.bi_sector, in raid10_write_request()
1355 if (bio->bi_opf & REQ_NOWAIT) { in raid10_write_request()
1360 prepare_to_wait(&conf->wait_barrier, in raid10_write_request()
1362 if (!md_cluster_ops->area_resyncing(mddev, WRITE, in raid10_write_request()
1363 bio->bi_iter.bi_sector, bio_end_sector(bio))) in raid10_write_request()
1367 finish_wait(&conf->wait_barrier, &w); in raid10_write_request()
1370 sectors = r10_bio->sectors; in raid10_write_request()
1373 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in raid10_write_request()
1374 (mddev->reshape_backwards in raid10_write_request()
1375 ? (bio->bi_iter.bi_sector < conf->reshape_safe && in raid10_write_request()
1376 bio->bi_iter.bi_sector + sectors > conf->reshape_progress) in raid10_write_request()
1377 : (bio->bi_iter.bi_sector + sectors > conf->reshape_safe && in raid10_write_request()
1378 bio->bi_iter.bi_sector < conf->reshape_progress))) { in raid10_write_request()
1379 /* Need to update reshape_position in metadata */ in raid10_write_request()
1380 mddev->reshape_position = conf->reshape_progress; in raid10_write_request()
1381 set_mask_bits(&mddev->sb_flags, 0, in raid10_write_request()
1383 md_wakeup_thread(mddev->thread); in raid10_write_request()
1384 if (bio->bi_opf & REQ_NOWAIT) { in raid10_write_request()
1389 mddev_add_trace_msg(conf->mddev, in raid10_write_request()
1391 wait_event(mddev->sb_wait, in raid10_write_request()
1392 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)); in raid10_write_request()
1394 conf->reshape_safe = mddev->reshape_position; in raid10_write_request()
1407 r10_bio->read_slot = -1; /* make sure repl_bio gets freed */ in raid10_write_request()
1412 max_sectors = r10_bio->sectors; in raid10_write_request()
1414 for (i = 0; i < conf->copies; i++) { in raid10_write_request()
1415 int d = r10_bio->devs[i].devnum; in raid10_write_request()
1418 rdev = conf->mirrors[d].rdev; in raid10_write_request()
1419 rrdev = conf->mirrors[d].replacement; in raid10_write_request()
1420 if (rdev && (test_bit(Faulty, &rdev->flags))) in raid10_write_request()
1422 if (rrdev && (test_bit(Faulty, &rrdev->flags))) in raid10_write_request()
1425 r10_bio->devs[i].bio = NULL; in raid10_write_request()
1426 r10_bio->devs[i].repl_bio = NULL; in raid10_write_request()
1430 if (rdev && test_bit(WriteErrorSeen, &rdev->flags)) { in raid10_write_request()
1432 sector_t dev_sector = r10_bio->devs[i].addr; in raid10_write_request()
1440 bad_sectors -= (dev_sector - first_bad); in raid10_write_request()
1458 if (bio->bi_opf & REQ_ATOMIC) { in raid10_write_request()
1459 error = -EIO; in raid10_write_request()
1463 good_sectors = first_bad - dev_sector; in raid10_write_request()
1469 r10_bio->devs[i].bio = bio; in raid10_write_request()
1470 atomic_inc(&rdev->nr_pending); in raid10_write_request()
1473 r10_bio->devs[i].repl_bio = bio; in raid10_write_request()
1474 atomic_inc(&rrdev->nr_pending); in raid10_write_request()
1478 if (max_sectors < r10_bio->sectors) in raid10_write_request()
1479 r10_bio->sectors = max_sectors; in raid10_write_request()
1481 if (r10_bio->sectors < bio_sectors(bio)) { in raid10_write_request()
1482 struct bio *split = bio_split(bio, r10_bio->sectors, in raid10_write_request()
1483 GFP_NOIO, &conf->bio_split); in raid10_write_request()
1493 r10_bio->master_bio = bio; in raid10_write_request()
1497 r10_bio->master_bio = bio; in raid10_write_request()
1498 atomic_set(&r10_bio->remaining, 1); in raid10_write_request()
1500 for (i = 0; i < conf->copies; i++) { in raid10_write_request()
1501 if (r10_bio->devs[i].bio) in raid10_write_request()
1503 if (r10_bio->devs[i].repl_bio) in raid10_write_request()
1510 int d = r10_bio->devs[k].devnum; in raid10_write_request()
1511 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid10_write_request()
1512 struct md_rdev *rrdev = conf->mirrors[d].replacement; in raid10_write_request()
1514 if (r10_bio->devs[k].bio) { in raid10_write_request()
1516 r10_bio->devs[k].bio = NULL; in raid10_write_request()
1518 if (r10_bio->devs[k].repl_bio) { in raid10_write_request()
1520 r10_bio->devs[k].repl_bio = NULL; in raid10_write_request()
1524 bio->bi_status = errno_to_blk_status(error); in raid10_write_request()
1525 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_write_request()
1531 struct r10conf *conf = mddev->private; in __make_request()
1534 r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); in __make_request()
1536 r10_bio->master_bio = bio; in __make_request()
1537 r10_bio->sectors = sectors; in __make_request()
1539 r10_bio->mddev = mddev; in __make_request()
1540 r10_bio->sector = bio->bi_iter.bi_sector; in __make_request()
1541 r10_bio->state = 0; in __make_request()
1542 r10_bio->read_slot = -1; in __make_request()
1543 memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * in __make_request()
1544 conf->geo.raid_disks); in __make_request()
1554 struct r10conf *conf = r10bio->mddev->private; in raid_end_discard_bio()
1557 while (atomic_dec_and_test(&r10bio->remaining)) { in raid_end_discard_bio()
1561 if (!test_bit(R10BIO_Discard, &r10bio->state)) { in raid_end_discard_bio()
1562 first_r10bio = (struct r10bio *)r10bio->master_bio; in raid_end_discard_bio()
1566 md_write_end(r10bio->mddev); in raid_end_discard_bio()
1567 bio_endio(r10bio->master_bio); in raid_end_discard_bio()
1576 struct r10bio *r10_bio = bio->bi_private; in raid10_end_discard_request()
1577 struct r10conf *conf = r10_bio->mddev->private; in raid10_end_discard_request()
1585 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in raid10_end_discard_request()
1586 set_bit(R10BIO_Uptodate, &r10_bio->state); in raid10_end_discard_request()
1589 rdev = repl ? conf->mirrors[dev].replacement : in raid10_end_discard_request()
1590 conf->mirrors[dev].rdev; in raid10_end_discard_request()
1593 rdev_dec_pending(rdev, conf->mddev); in raid10_end_discard_request()
1604 struct r10conf *conf = mddev->private; in raid10_handle_discard()
1605 struct geom *geo = &conf->geo; in raid10_handle_discard()
1606 int far_copies = geo->far_copies; in raid10_handle_discard()
1623 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_handle_discard()
1624 return -EAGAIN; in raid10_handle_discard()
1626 if (!wait_barrier(conf, bio->bi_opf & REQ_NOWAIT)) { in raid10_handle_discard()
1635 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_handle_discard()
1638 if (geo->near_copies) in raid10_handle_discard()
1639 stripe_data_disks = geo->raid_disks / geo->near_copies + in raid10_handle_discard()
1640 geo->raid_disks % geo->near_copies; in raid10_handle_discard()
1642 stripe_data_disks = geo->raid_disks; in raid10_handle_discard()
1644 stripe_size = stripe_data_disks << geo->chunk_shift; in raid10_handle_discard()
1646 bio_start = bio->bi_iter.bi_sector; in raid10_handle_discard()
1665 split_size = stripe_size - remainder; in raid10_handle_discard()
1666 split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); in raid10_handle_discard()
1668 bio->bi_status = errno_to_blk_status(PTR_ERR(split)); in raid10_handle_discard()
1680 split_size = bio_sectors(bio) - remainder; in raid10_handle_discard()
1681 split = bio_split(bio, split_size, GFP_NOIO, &conf->bio_split); in raid10_handle_discard()
1683 bio->bi_status = errno_to_blk_status(PTR_ERR(split)); in raid10_handle_discard()
1695 bio_start = bio->bi_iter.bi_sector; in raid10_handle_discard()
1703 chunk = bio_start >> geo->chunk_shift; in raid10_handle_discard()
1704 chunk *= geo->near_copies; in raid10_handle_discard()
1706 start_disk_index = sector_div(first_stripe_index, geo->raid_disks); in raid10_handle_discard()
1707 if (geo->far_offset) in raid10_handle_discard()
1708 first_stripe_index *= geo->far_copies; in raid10_handle_discard()
1709 start_disk_offset = (bio_start & geo->chunk_mask) + in raid10_handle_discard()
1710 (first_stripe_index << geo->chunk_shift); in raid10_handle_discard()
1712 chunk = bio_end >> geo->chunk_shift; in raid10_handle_discard()
1713 chunk *= geo->near_copies; in raid10_handle_discard()
1715 end_disk_index = sector_div(last_stripe_index, geo->raid_disks); in raid10_handle_discard()
1716 if (geo->far_offset) in raid10_handle_discard()
1717 last_stripe_index *= geo->far_copies; in raid10_handle_discard()
1718 end_disk_offset = (bio_end & geo->chunk_mask) + in raid10_handle_discard()
1719 (last_stripe_index << geo->chunk_shift); in raid10_handle_discard()
1722 r10_bio = mempool_alloc(&conf->r10bio_pool, GFP_NOIO); in raid10_handle_discard()
1723 r10_bio->mddev = mddev; in raid10_handle_discard()
1724 r10_bio->state = 0; in raid10_handle_discard()
1725 r10_bio->sectors = 0; in raid10_handle_discard()
1726 memset(r10_bio->devs, 0, sizeof(r10_bio->devs[0]) * geo->raid_disks); in raid10_handle_discard()
1731 * Inspired by raid10_sync_request, we can use the first r10bio->master_bio in raid10_handle_discard()
1732 * to record the discard bio. Other r10bio->master_bio record the first in raid10_handle_discard()
1738 r10_bio->master_bio = bio; in raid10_handle_discard()
1739 set_bit(R10BIO_Discard, &r10_bio->state); in raid10_handle_discard()
1743 r10_bio->master_bio = (struct bio *)first_r10bio; in raid10_handle_discard()
1750 for (disk = 0; disk < geo->raid_disks; disk++) { in raid10_handle_discard()
1753 rdev = conf->mirrors[disk].rdev; in raid10_handle_discard()
1754 rrdev = conf->mirrors[disk].replacement; in raid10_handle_discard()
1755 r10_bio->devs[disk].bio = NULL; in raid10_handle_discard()
1756 r10_bio->devs[disk].repl_bio = NULL; in raid10_handle_discard()
1758 if (rdev && (test_bit(Faulty, &rdev->flags))) in raid10_handle_discard()
1760 if (rrdev && (test_bit(Faulty, &rrdev->flags))) in raid10_handle_discard()
1766 r10_bio->devs[disk].bio = bio; in raid10_handle_discard()
1767 atomic_inc(&rdev->nr_pending); in raid10_handle_discard()
1770 r10_bio->devs[disk].repl_bio = bio; in raid10_handle_discard()
1771 atomic_inc(&rrdev->nr_pending); in raid10_handle_discard()
1775 atomic_set(&r10_bio->remaining, 1); in raid10_handle_discard()
1776 for (disk = 0; disk < geo->raid_disks; disk++) { in raid10_handle_discard()
1793 dev_start = (first_stripe_index + 1) * mddev->chunk_sectors; in raid10_handle_discard()
1795 dev_start = first_stripe_index * mddev->chunk_sectors; in raid10_handle_discard()
1800 dev_end = (last_stripe_index + 1) * mddev->chunk_sectors; in raid10_handle_discard()
1802 dev_end = last_stripe_index * mddev->chunk_sectors; in raid10_handle_discard()
1810 * add rdev->nr_pending in the first loop. in raid10_handle_discard()
1812 if (r10_bio->devs[disk].bio) { in raid10_handle_discard()
1813 struct md_rdev *rdev = conf->mirrors[disk].rdev; in raid10_handle_discard()
1814 mbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, in raid10_handle_discard()
1815 &mddev->bio_set); in raid10_handle_discard()
1816 mbio->bi_end_io = raid10_end_discard_request; in raid10_handle_discard()
1817 mbio->bi_private = r10_bio; in raid10_handle_discard()
1818 r10_bio->devs[disk].bio = mbio; in raid10_handle_discard()
1819 r10_bio->devs[disk].devnum = disk; in raid10_handle_discard()
1820 atomic_inc(&r10_bio->remaining); in raid10_handle_discard()
1823 dev_end - dev_start); in raid10_handle_discard()
1826 if (r10_bio->devs[disk].repl_bio) { in raid10_handle_discard()
1827 struct md_rdev *rrdev = conf->mirrors[disk].replacement; in raid10_handle_discard()
1828 rbio = bio_alloc_clone(bio->bi_bdev, bio, GFP_NOIO, in raid10_handle_discard()
1829 &mddev->bio_set); in raid10_handle_discard()
1830 rbio->bi_end_io = raid10_end_discard_request; in raid10_handle_discard()
1831 rbio->bi_private = r10_bio; in raid10_handle_discard()
1832 r10_bio->devs[disk].repl_bio = rbio; in raid10_handle_discard()
1833 r10_bio->devs[disk].devnum = disk; in raid10_handle_discard()
1834 atomic_inc(&r10_bio->remaining); in raid10_handle_discard()
1837 dev_end - dev_start); in raid10_handle_discard()
1842 if (!geo->far_offset && --far_copies) { in raid10_handle_discard()
1843 first_stripe_index += geo->stride >> geo->chunk_shift; in raid10_handle_discard()
1844 start_disk_offset += geo->stride; in raid10_handle_discard()
1845 last_stripe_index += geo->stride >> geo->chunk_shift; in raid10_handle_discard()
1846 end_disk_offset += geo->stride; in raid10_handle_discard()
1847 atomic_inc(&first_r10bio->remaining); in raid10_handle_discard()
1858 return -EAGAIN; in raid10_handle_discard()
1863 struct r10conf *conf = mddev->private; in raid10_make_request()
1864 sector_t chunk_mask = (conf->geo.chunk_mask & conf->prev.chunk_mask); in raid10_make_request()
1868 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid10_make_request()
1882 if (unlikely((bio->bi_iter.bi_sector & chunk_mask) + in raid10_make_request()
1884 && (conf->geo.near_copies < conf->geo.raid_disks in raid10_make_request()
1885 || conf->prev.near_copies < in raid10_make_request()
1886 conf->prev.raid_disks))) in raid10_make_request()
1887 sectors = chunk_sects - in raid10_make_request()
1888 (bio->bi_iter.bi_sector & in raid10_make_request()
1889 (chunk_sects - 1)); in raid10_make_request()
1899 struct r10conf *conf = mddev->private; in raid10_status()
1902 lockdep_assert_held(&mddev->lock); in raid10_status()
1904 if (conf->geo.near_copies < conf->geo.raid_disks) in raid10_status()
1905 seq_printf(seq, " %dK chunks", mddev->chunk_sectors / 2); in raid10_status()
1906 if (conf->geo.near_copies > 1) in raid10_status()
1907 seq_printf(seq, " %d near-copies", conf->geo.near_copies); in raid10_status()
1908 if (conf->geo.far_copies > 1) { in raid10_status()
1909 if (conf->geo.far_offset) in raid10_status()
1910 seq_printf(seq, " %d offset-copies", conf->geo.far_copies); in raid10_status()
1912 seq_printf(seq, " %d far-copies", conf->geo.far_copies); in raid10_status()
1913 if (conf->geo.far_set_size != conf->geo.raid_disks) in raid10_status()
1914 seq_printf(seq, " %d devices per set", conf->geo.far_set_size); in raid10_status()
1916 seq_printf(seq, " [%d/%d] [", conf->geo.raid_disks, in raid10_status()
1917 conf->geo.raid_disks - mddev->degraded); in raid10_status()
1918 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_status()
1919 struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev); in raid10_status()
1921 seq_printf(seq, "%s", rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid10_status()
1937 disks = conf->prev.raid_disks; in _enough()
1938 ncopies = conf->prev.near_copies; in _enough()
1940 disks = conf->geo.raid_disks; in _enough()
1941 ncopies = conf->geo.near_copies; in _enough()
1945 int n = conf->copies; in _enough()
1948 while (n--) { in _enough()
1951 (rdev = conf->mirrors[this].rdev) && in _enough()
1952 test_bit(In_sync, &rdev->flags)) in _enough()
1969 * This is ensured if ->reconfig_mutex or ->device_lock in enough()
1977 * raid10_error() - RAID10 error handler.
1983 * - &MD_BROKEN flag is set in &mddev->flags.
1985 * - recovery is interrupted.
1986 * - &mddev->degraded is bumped.
1989 * &mddev->fail_last_dev is off.
1993 struct r10conf *conf = mddev->private; in raid10_error()
1996 spin_lock_irqsave(&conf->device_lock, flags); in raid10_error()
1998 if (test_bit(In_sync, &rdev->flags) && !enough(conf, rdev->raid_disk)) { in raid10_error()
1999 set_bit(MD_BROKEN, &mddev->flags); in raid10_error()
2001 if (!mddev->fail_last_dev) { in raid10_error()
2002 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_error()
2006 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid10_error()
2007 mddev->degraded++; in raid10_error()
2009 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid10_error()
2010 set_bit(Blocked, &rdev->flags); in raid10_error()
2011 set_bit(Faulty, &rdev->flags); in raid10_error()
2012 set_mask_bits(&mddev->sb_flags, 0, in raid10_error()
2014 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_error()
2017 mdname(mddev), rdev->bdev, in raid10_error()
2018 mdname(mddev), conf->geo.raid_disks - mddev->degraded); in raid10_error()
2031 pr_debug(" --- wd:%d rd:%d\n", conf->geo.raid_disks - conf->mddev->degraded, in print_conf()
2032 conf->geo.raid_disks); in print_conf()
2034 lockdep_assert_held(&conf->mddev->reconfig_mutex); in print_conf()
2035 for (i = 0; i < conf->geo.raid_disks; i++) { in print_conf()
2036 rdev = conf->mirrors[i].rdev; in print_conf()
2039 i, !test_bit(In_sync, &rdev->flags), in print_conf()
2040 !test_bit(Faulty, &rdev->flags), in print_conf()
2041 rdev->bdev); in print_conf()
2050 mempool_exit(&conf->r10buf_pool); in close_sync()
2056 struct r10conf *conf = mddev->private; in raid10_spare_active()
2062 * Find all non-in_sync disks within the RAID10 configuration in raid10_spare_active()
2065 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_spare_active()
2066 tmp = conf->mirrors + i; in raid10_spare_active()
2067 if (tmp->replacement in raid10_spare_active()
2068 && tmp->replacement->recovery_offset == MaxSector in raid10_spare_active()
2069 && !test_bit(Faulty, &tmp->replacement->flags) in raid10_spare_active()
2070 && !test_and_set_bit(In_sync, &tmp->replacement->flags)) { in raid10_spare_active()
2072 if (!tmp->rdev in raid10_spare_active()
2073 || !test_and_clear_bit(In_sync, &tmp->rdev->flags)) in raid10_spare_active()
2075 if (tmp->rdev) { in raid10_spare_active()
2078 * and never re-added. in raid10_spare_active()
2080 set_bit(Faulty, &tmp->rdev->flags); in raid10_spare_active()
2082 tmp->rdev->sysfs_state); in raid10_spare_active()
2084 sysfs_notify_dirent_safe(tmp->replacement->sysfs_state); in raid10_spare_active()
2085 } else if (tmp->rdev in raid10_spare_active()
2086 && tmp->rdev->recovery_offset == MaxSector in raid10_spare_active()
2087 && !test_bit(Faulty, &tmp->rdev->flags) in raid10_spare_active()
2088 && !test_and_set_bit(In_sync, &tmp->rdev->flags)) { in raid10_spare_active()
2090 sysfs_notify_dirent_safe(tmp->rdev->sysfs_state); in raid10_spare_active()
2093 spin_lock_irqsave(&conf->device_lock, flags); in raid10_spare_active()
2094 mddev->degraded -= count; in raid10_spare_active()
2095 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10_spare_active()
2103 struct r10conf *conf = mddev->private; in raid10_add_disk()
2104 int err = -EEXIST; in raid10_add_disk()
2105 int mirror, repl_slot = -1; in raid10_add_disk()
2107 int last = conf->geo.raid_disks - 1; in raid10_add_disk()
2110 if (mddev->recovery_cp < MaxSector) in raid10_add_disk()
2111 /* only hot-add to in-sync arrays, as recovery is in raid10_add_disk()
2114 return -EBUSY; in raid10_add_disk()
2115 if (rdev->saved_raid_disk < 0 && !_enough(conf, 1, -1)) in raid10_add_disk()
2116 return -EINVAL; in raid10_add_disk()
2118 if (rdev->raid_disk >= 0) in raid10_add_disk()
2119 first = last = rdev->raid_disk; in raid10_add_disk()
2121 if (rdev->saved_raid_disk >= first && in raid10_add_disk()
2122 rdev->saved_raid_disk < conf->geo.raid_disks && in raid10_add_disk()
2123 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid10_add_disk()
2124 mirror = rdev->saved_raid_disk; in raid10_add_disk()
2128 p = &conf->mirrors[mirror]; in raid10_add_disk()
2129 if (p->recovery_disabled == mddev->recovery_disabled) in raid10_add_disk()
2131 if (p->rdev) { in raid10_add_disk()
2132 if (test_bit(WantReplacement, &p->rdev->flags) && in raid10_add_disk()
2133 p->replacement == NULL && repl_slot < 0) in raid10_add_disk()
2141 p->head_position = 0; in raid10_add_disk()
2142 p->recovery_disabled = mddev->recovery_disabled - 1; in raid10_add_disk()
2143 rdev->raid_disk = mirror; in raid10_add_disk()
2145 if (rdev->saved_raid_disk != mirror) in raid10_add_disk()
2146 conf->fullsync = 1; in raid10_add_disk()
2147 WRITE_ONCE(p->rdev, rdev); in raid10_add_disk()
2152 p = &conf->mirrors[repl_slot]; in raid10_add_disk()
2153 clear_bit(In_sync, &rdev->flags); in raid10_add_disk()
2154 set_bit(Replacement, &rdev->flags); in raid10_add_disk()
2155 rdev->raid_disk = repl_slot; in raid10_add_disk()
2159 conf->fullsync = 1; in raid10_add_disk()
2160 WRITE_ONCE(p->replacement, rdev); in raid10_add_disk()
2169 struct r10conf *conf = mddev->private; in raid10_remove_disk()
2171 int number = rdev->raid_disk; in raid10_remove_disk()
2176 if (unlikely(number >= mddev->raid_disks)) in raid10_remove_disk()
2178 p = conf->mirrors + number; in raid10_remove_disk()
2179 if (rdev == p->rdev) in raid10_remove_disk()
2180 rdevp = &p->rdev; in raid10_remove_disk()
2181 else if (rdev == p->replacement) in raid10_remove_disk()
2182 rdevp = &p->replacement; in raid10_remove_disk()
2186 if (test_bit(In_sync, &rdev->flags) || in raid10_remove_disk()
2187 atomic_read(&rdev->nr_pending)) { in raid10_remove_disk()
2188 err = -EBUSY; in raid10_remove_disk()
2191 /* Only remove non-faulty devices if recovery in raid10_remove_disk()
2194 if (!test_bit(Faulty, &rdev->flags) && in raid10_remove_disk()
2195 mddev->recovery_disabled != p->recovery_disabled && in raid10_remove_disk()
2196 (!p->replacement || p->replacement == rdev) && in raid10_remove_disk()
2197 number < conf->geo.raid_disks && in raid10_remove_disk()
2198 enough(conf, -1)) { in raid10_remove_disk()
2199 err = -EBUSY; in raid10_remove_disk()
2203 if (p->replacement) { in raid10_remove_disk()
2205 WRITE_ONCE(p->rdev, p->replacement); in raid10_remove_disk()
2206 clear_bit(Replacement, &p->replacement->flags); in raid10_remove_disk()
2207 WRITE_ONCE(p->replacement, NULL); in raid10_remove_disk()
2210 clear_bit(WantReplacement, &rdev->flags); in raid10_remove_disk()
2221 struct r10conf *conf = r10_bio->mddev->private; in __end_sync_read()
2223 if (!bio->bi_status) in __end_sync_read()
2224 set_bit(R10BIO_Uptodate, &r10_bio->state); in __end_sync_read()
2229 atomic_add(r10_bio->sectors, in __end_sync_read()
2230 &conf->mirrors[d].rdev->corrected_errors); in __end_sync_read()
2235 rdev_dec_pending(conf->mirrors[d].rdev, conf->mddev); in __end_sync_read()
2236 if (test_bit(R10BIO_IsRecover, &r10_bio->state) || in __end_sync_read()
2237 atomic_dec_and_test(&r10_bio->remaining)) { in __end_sync_read()
2248 struct r10conf *conf = r10_bio->mddev->private; in end_sync_read()
2257 struct r10bio *r10_bio = bio->bi_private; in end_reshape_read()
2259 __end_sync_read(r10_bio, bio, r10_bio->read_slot); in end_reshape_read()
2264 struct mddev *mddev = r10_bio->mddev; in end_sync_request()
2266 while (atomic_dec_and_test(&r10_bio->remaining)) { in end_sync_request()
2267 if (r10_bio->master_bio == NULL) { in end_sync_request()
2269 sector_t s = r10_bio->sectors; in end_sync_request()
2270 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in end_sync_request()
2271 test_bit(R10BIO_WriteError, &r10_bio->state)) in end_sync_request()
2278 struct r10bio *r10_bio2 = (struct r10bio *)r10_bio->master_bio; in end_sync_request()
2279 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in end_sync_request()
2280 test_bit(R10BIO_WriteError, &r10_bio->state)) in end_sync_request()
2292 struct mddev *mddev = r10_bio->mddev; in end_sync_write()
2293 struct r10conf *conf = mddev->private; in end_sync_write()
2301 rdev = conf->mirrors[d].replacement; in end_sync_write()
2303 rdev = conf->mirrors[d].rdev; in end_sync_write()
2305 if (bio->bi_status) { in end_sync_write()
2309 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
2310 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
2312 &rdev->mddev->recovery); in end_sync_write()
2313 set_bit(R10BIO_WriteError, &r10_bio->state); in end_sync_write()
2315 } else if (rdev_has_badblock(rdev, r10_bio->devs[slot].addr, in end_sync_write()
2316 r10_bio->sectors)) { in end_sync_write()
2317 set_bit(R10BIO_MadeGood, &r10_bio->state); in end_sync_write()
2338 * We check if all blocks are in-sync and only write to blocks that
2343 struct r10conf *conf = mddev->private; in sync_request_write()
2349 atomic_set(&r10_bio->remaining, 1); in sync_request_write()
2352 for (i=0; i<conf->copies; i++) in sync_request_write()
2353 if (!r10_bio->devs[i].bio->bi_status) in sync_request_write()
2356 if (i == conf->copies) in sync_request_write()
2360 fbio = r10_bio->devs[i].bio; in sync_request_write()
2361 fbio->bi_iter.bi_size = r10_bio->sectors << 9; in sync_request_write()
2362 fbio->bi_iter.bi_idx = 0; in sync_request_write()
2363 fpages = get_resync_pages(fbio)->pages; in sync_request_write()
2365 vcnt = (r10_bio->sectors + (PAGE_SIZE >> 9) - 1) >> (PAGE_SHIFT - 9); in sync_request_write()
2367 for (i=0 ; i < conf->copies ; i++) { in sync_request_write()
2372 tbio = r10_bio->devs[i].bio; in sync_request_write()
2374 if (tbio->bi_end_io != end_sync_read) in sync_request_write()
2379 tpages = get_resync_pages(tbio)->pages; in sync_request_write()
2380 d = r10_bio->devs[i].devnum; in sync_request_write()
2381 rdev = conf->mirrors[d].rdev; in sync_request_write()
2382 if (!r10_bio->devs[i].bio->bi_status) { in sync_request_write()
2387 int sectors = r10_bio->sectors; in sync_request_write()
2396 sectors -= len/512; in sync_request_write()
2400 atomic64_add(r10_bio->sectors, &mddev->resync_mismatches); in sync_request_write()
2401 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) in sync_request_write()
2404 } else if (test_bit(FailFast, &rdev->flags)) { in sync_request_write()
2406 md_error(rdev->mddev, rdev); in sync_request_write()
2411 * First we need to fixup bv_offset, bv_len and in sync_request_write()
2415 bio_reset(tbio, conf->mirrors[d].rdev->bdev, REQ_OP_WRITE); in sync_request_write()
2417 md_bio_reset_resync_pages(tbio, rp, fbio->bi_iter.bi_size); in sync_request_write()
2419 rp->raid_bio = r10_bio; in sync_request_write()
2420 tbio->bi_private = rp; in sync_request_write()
2421 tbio->bi_iter.bi_sector = r10_bio->devs[i].addr; in sync_request_write()
2422 tbio->bi_end_io = end_sync_write; in sync_request_write()
2426 atomic_inc(&conf->mirrors[d].rdev->nr_pending); in sync_request_write()
2427 atomic_inc(&r10_bio->remaining); in sync_request_write()
2428 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(tbio)); in sync_request_write()
2430 if (test_bit(FailFast, &conf->mirrors[d].rdev->flags)) in sync_request_write()
2431 tbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2432 tbio->bi_iter.bi_sector += conf->mirrors[d].rdev->data_offset; in sync_request_write()
2439 for (i = 0; i < conf->copies; i++) { in sync_request_write()
2442 tbio = r10_bio->devs[i].repl_bio; in sync_request_write()
2443 if (!tbio || !tbio->bi_end_io) in sync_request_write()
2445 if (r10_bio->devs[i].bio->bi_end_io != end_sync_write in sync_request_write()
2446 && r10_bio->devs[i].bio != fbio) in sync_request_write()
2448 d = r10_bio->devs[i].devnum; in sync_request_write()
2449 atomic_inc(&r10_bio->remaining); in sync_request_write()
2450 md_sync_acct(conf->mirrors[d].replacement->bdev, in sync_request_write()
2456 if (atomic_dec_and_test(&r10_bio->remaining)) { in sync_request_write()
2457 md_done_sync(mddev, r10_bio->sectors, 1); in sync_request_write()
2465 * We recover all non-is_sync drives by finding the virtual address of
2467 * There is a separate r10_bio for each non-in_sync drive.
2475 * We repeat the read in smaller page-sized sections. in fix_recovery_read_error()
2481 struct mddev *mddev = r10_bio->mddev; in fix_recovery_read_error()
2482 struct r10conf *conf = mddev->private; in fix_recovery_read_error()
2483 struct bio *bio = r10_bio->devs[0].bio; in fix_recovery_read_error()
2485 int sectors = r10_bio->sectors; in fix_recovery_read_error()
2487 int dr = r10_bio->devs[0].devnum; in fix_recovery_read_error()
2488 int dw = r10_bio->devs[1].devnum; in fix_recovery_read_error()
2489 struct page **pages = get_resync_pages(bio)->pages; in fix_recovery_read_error()
2500 rdev = conf->mirrors[dr].rdev; in fix_recovery_read_error()
2501 addr = r10_bio->devs[0].addr + sect; in fix_recovery_read_error()
2508 rdev = conf->mirrors[dw].rdev; in fix_recovery_read_error()
2509 addr = r10_bio->devs[1].addr + sect; in fix_recovery_read_error()
2516 set_bit(WriteErrorSeen, &rdev->flags); in fix_recovery_read_error()
2518 &rdev->flags)) in fix_recovery_read_error()
2520 &rdev->mddev->recovery); in fix_recovery_read_error()
2524 /* We don't worry if we cannot set a bad block - in fix_recovery_read_error()
2530 if (rdev != conf->mirrors[dw].rdev) { in fix_recovery_read_error()
2532 struct md_rdev *rdev2 = conf->mirrors[dw].rdev; in fix_recovery_read_error()
2533 addr = r10_bio->devs[1].addr + sect; in fix_recovery_read_error()
2540 conf->mirrors[dw].recovery_disabled in fix_recovery_read_error()
2541 = mddev->recovery_disabled; in fix_recovery_read_error()
2543 &mddev->recovery); in fix_recovery_read_error()
2549 sectors -= s; in fix_recovery_read_error()
2557 struct r10conf *conf = mddev->private; in recovery_request_write()
2559 struct bio *wbio = r10_bio->devs[1].bio; in recovery_request_write()
2560 struct bio *wbio2 = r10_bio->devs[1].repl_bio; in recovery_request_write()
2562 /* Need to test wbio2->bi_end_io before we call in recovery_request_write()
2566 if (wbio2 && !wbio2->bi_end_io) in recovery_request_write()
2569 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) { in recovery_request_write()
2571 if (wbio->bi_end_io) in recovery_request_write()
2582 d = r10_bio->devs[1].devnum; in recovery_request_write()
2583 if (wbio->bi_end_io) { in recovery_request_write()
2584 atomic_inc(&conf->mirrors[d].rdev->nr_pending); in recovery_request_write()
2585 md_sync_acct(conf->mirrors[d].rdev->bdev, bio_sectors(wbio)); in recovery_request_write()
2589 atomic_inc(&conf->mirrors[d].replacement->nr_pending); in recovery_request_write()
2590 md_sync_acct(conf->mirrors[d].replacement->bdev, in recovery_request_write()
2600 (op == REQ_OP_READ || test_bit(WriteErrorSeen, &rdev->flags))) in r10_sync_page_io()
2601 return -1; in r10_sync_page_io()
2606 set_bit(WriteErrorSeen, &rdev->flags); in r10_sync_page_io()
2607 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in r10_sync_page_io()
2609 &rdev->mddev->recovery); in r10_sync_page_io()
2611 /* need to record an error - either for the block or the device */ in r10_sync_page_io()
2613 md_error(rdev->mddev, rdev); in r10_sync_page_io()
2627 int sect = 0; /* Offset from r10_bio->sector */ in fix_read_error()
2628 int sectors = r10_bio->sectors, slot = r10_bio->read_slot; in fix_read_error()
2630 int d = r10_bio->devs[slot].devnum; in fix_read_error()
2635 rdev = conf->mirrors[d].rdev; in fix_read_error()
2637 if (test_bit(Faulty, &rdev->flags)) in fix_read_error()
2643 r10_bio->devs[slot].bio = IO_BLOCKED; in fix_read_error()
2657 d = r10_bio->devs[sl].devnum; in fix_read_error()
2658 rdev = conf->mirrors[d].rdev; in fix_read_error()
2660 test_bit(In_sync, &rdev->flags) && in fix_read_error()
2661 !test_bit(Faulty, &rdev->flags) && in fix_read_error()
2663 r10_bio->devs[sl].addr + sect, in fix_read_error()
2665 atomic_inc(&rdev->nr_pending); in fix_read_error()
2667 r10_bio->devs[sl].addr + in fix_read_error()
2670 conf->tmppage, in fix_read_error()
2677 if (sl == conf->copies) in fix_read_error()
2686 int dn = r10_bio->devs[slot].devnum; in fix_read_error()
2687 rdev = conf->mirrors[dn].rdev; in fix_read_error()
2691 r10_bio->devs[slot].addr in fix_read_error()
2695 r10_bio->devs[slot].bio in fix_read_error()
2702 /* write it back and re-read */ in fix_read_error()
2705 sl = conf->copies; in fix_read_error()
2706 sl--; in fix_read_error()
2707 d = r10_bio->devs[sl].devnum; in fix_read_error()
2708 rdev = conf->mirrors[d].rdev; in fix_read_error()
2710 test_bit(Faulty, &rdev->flags) || in fix_read_error()
2711 !test_bit(In_sync, &rdev->flags)) in fix_read_error()
2714 atomic_inc(&rdev->nr_pending); in fix_read_error()
2716 r10_bio->devs[sl].addr + in fix_read_error()
2718 s, conf->tmppage, REQ_OP_WRITE) in fix_read_error()
2727 rdev->bdev); in fix_read_error()
2730 rdev->bdev); in fix_read_error()
2737 sl = conf->copies; in fix_read_error()
2738 sl--; in fix_read_error()
2739 d = r10_bio->devs[sl].devnum; in fix_read_error()
2740 rdev = conf->mirrors[d].rdev; in fix_read_error()
2742 test_bit(Faulty, &rdev->flags) || in fix_read_error()
2743 !test_bit(In_sync, &rdev->flags)) in fix_read_error()
2746 atomic_inc(&rdev->nr_pending); in fix_read_error()
2748 r10_bio->devs[sl].addr + in fix_read_error()
2750 s, conf->tmppage, REQ_OP_READ)) { in fix_read_error()
2758 rdev->bdev); in fix_read_error()
2761 rdev->bdev); in fix_read_error()
2769 rdev->bdev); in fix_read_error()
2770 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2776 sectors -= s; in fix_read_error()
2783 struct bio *bio = r10_bio->master_bio; in narrow_write_error()
2784 struct mddev *mddev = r10_bio->mddev; in narrow_write_error()
2785 struct r10conf *conf = mddev->private; in narrow_write_error()
2786 struct md_rdev *rdev = conf->mirrors[r10_bio->devs[i].devnum].rdev; in narrow_write_error()
2801 int sect_to_write = r10_bio->sectors; in narrow_write_error()
2804 if (rdev->badblocks.shift < 0) in narrow_write_error()
2807 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2808 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2809 sector = r10_bio->sector; in narrow_write_error()
2810 sectors = ((r10_bio->sector + block_sectors) in narrow_write_error()
2811 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2812 - sector; in narrow_write_error()
2820 wbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, in narrow_write_error()
2821 &mddev->bio_set); in narrow_write_error()
2822 bio_trim(wbio, sector - bio->bi_iter.bi_sector, sectors); in narrow_write_error()
2823 wsector = r10_bio->devs[i].addr + (sector - r10_bio->sector); in narrow_write_error()
2824 wbio->bi_iter.bi_sector = wsector + in narrow_write_error()
2826 wbio->bi_opf = REQ_OP_WRITE; in narrow_write_error()
2835 sect_to_write -= sectors; in narrow_write_error()
2844 int slot = r10_bio->read_slot; in handle_read_error()
2846 struct r10conf *conf = mddev->private; in handle_read_error()
2847 struct md_rdev *rdev = r10_bio->devs[slot].rdev; in handle_read_error()
2852 * other devices. When we find one, we re-write in handle_read_error()
2857 bio = r10_bio->devs[slot].bio; in handle_read_error()
2859 r10_bio->devs[slot].bio = NULL; in handle_read_error()
2861 if (mddev->ro) in handle_read_error()
2862 r10_bio->devs[slot].bio = IO_BLOCKED; in handle_read_error()
2863 else if (!test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2871 r10_bio->state = 0; in handle_read_error()
2872 raid10_read_request(mddev, r10_bio->master_bio, r10_bio, false); in handle_read_error()
2874 * allow_barrier after re-submit to ensure no sync io in handle_read_error()
2891 if (test_bit(R10BIO_IsSync, &r10_bio->state) || in handle_write_completed()
2892 test_bit(R10BIO_IsRecover, &r10_bio->state)) { in handle_write_completed()
2893 for (m = 0; m < conf->copies; m++) { in handle_write_completed()
2894 int dev = r10_bio->devs[m].devnum; in handle_write_completed()
2895 rdev = conf->mirrors[dev].rdev; in handle_write_completed()
2896 if (r10_bio->devs[m].bio == NULL || in handle_write_completed()
2897 r10_bio->devs[m].bio->bi_end_io == NULL) in handle_write_completed()
2899 if (!r10_bio->devs[m].bio->bi_status) { in handle_write_completed()
2902 r10_bio->devs[m].addr, in handle_write_completed()
2903 r10_bio->sectors, 0); in handle_write_completed()
2907 r10_bio->devs[m].addr, in handle_write_completed()
2908 r10_bio->sectors, 0)) in handle_write_completed()
2909 md_error(conf->mddev, rdev); in handle_write_completed()
2911 rdev = conf->mirrors[dev].replacement; in handle_write_completed()
2912 if (r10_bio->devs[m].repl_bio == NULL || in handle_write_completed()
2913 r10_bio->devs[m].repl_bio->bi_end_io == NULL) in handle_write_completed()
2916 if (!r10_bio->devs[m].repl_bio->bi_status) { in handle_write_completed()
2919 r10_bio->devs[m].addr, in handle_write_completed()
2920 r10_bio->sectors, 0); in handle_write_completed()
2924 r10_bio->devs[m].addr, in handle_write_completed()
2925 r10_bio->sectors, 0)) in handle_write_completed()
2926 md_error(conf->mddev, rdev); in handle_write_completed()
2932 for (m = 0; m < conf->copies; m++) { in handle_write_completed()
2933 int dev = r10_bio->devs[m].devnum; in handle_write_completed()
2934 struct bio *bio = r10_bio->devs[m].bio; in handle_write_completed()
2935 rdev = conf->mirrors[dev].rdev; in handle_write_completed()
2939 r10_bio->devs[m].addr, in handle_write_completed()
2940 r10_bio->sectors, 0); in handle_write_completed()
2941 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
2942 } else if (bio != NULL && bio->bi_status) { in handle_write_completed()
2945 md_error(conf->mddev, rdev); in handle_write_completed()
2946 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
2948 bio = r10_bio->devs[m].repl_bio; in handle_write_completed()
2949 rdev = conf->mirrors[dev].replacement; in handle_write_completed()
2953 r10_bio->devs[m].addr, in handle_write_completed()
2954 r10_bio->sectors, 0); in handle_write_completed()
2955 rdev_dec_pending(rdev, conf->mddev); in handle_write_completed()
2959 spin_lock_irq(&conf->device_lock); in handle_write_completed()
2960 list_add(&r10_bio->retry_list, &conf->bio_end_io_list); in handle_write_completed()
2961 conf->nr_queued++; in handle_write_completed()
2962 spin_unlock_irq(&conf->device_lock); in handle_write_completed()
2967 wake_up(&conf->wait_barrier); in handle_write_completed()
2968 md_wakeup_thread(conf->mddev->thread); in handle_write_completed()
2971 &r10_bio->state)) in handle_write_completed()
2980 struct mddev *mddev = thread->mddev; in raid10d()
2983 struct r10conf *conf = mddev->private; in raid10d()
2984 struct list_head *head = &conf->retry_list; in raid10d()
2989 if (!list_empty_careful(&conf->bio_end_io_list) && in raid10d()
2990 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid10d()
2992 spin_lock_irqsave(&conf->device_lock, flags); in raid10d()
2993 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid10d()
2994 while (!list_empty(&conf->bio_end_io_list)) { in raid10d()
2995 list_move(conf->bio_end_io_list.prev, &tmp); in raid10d()
2996 conf->nr_queued--; in raid10d()
2999 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3003 list_del(&r10_bio->retry_list); in raid10d()
3006 &r10_bio->state)) in raid10d()
3017 spin_lock_irqsave(&conf->device_lock, flags); in raid10d()
3019 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3022 r10_bio = list_entry(head->prev, struct r10bio, retry_list); in raid10d()
3023 list_del(head->prev); in raid10d()
3024 conf->nr_queued--; in raid10d()
3025 spin_unlock_irqrestore(&conf->device_lock, flags); in raid10d()
3027 mddev = r10_bio->mddev; in raid10d()
3028 conf = mddev->private; in raid10d()
3029 if (test_bit(R10BIO_MadeGood, &r10_bio->state) || in raid10d()
3030 test_bit(R10BIO_WriteError, &r10_bio->state)) in raid10d()
3032 else if (test_bit(R10BIO_IsReshape, &r10_bio->state)) in raid10d()
3034 else if (test_bit(R10BIO_IsSync, &r10_bio->state)) in raid10d()
3036 else if (test_bit(R10BIO_IsRecover, &r10_bio->state)) in raid10d()
3038 else if (test_bit(R10BIO_ReadError, &r10_bio->state)) in raid10d()
3044 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid10d()
3055 BUG_ON(mempool_initialized(&conf->r10buf_pool)); in init_resync()
3056 conf->have_replacement = 0; in init_resync()
3057 for (i = 0; i < conf->geo.raid_disks; i++) in init_resync()
3058 if (conf->mirrors[i].replacement) in init_resync()
3059 conf->have_replacement = 1; in init_resync()
3060 ret = mempool_init(&conf->r10buf_pool, buffs, in init_resync()
3064 conf->next_resync = 0; in init_resync()
3070 struct r10bio *r10bio = mempool_alloc(&conf->r10buf_pool, GFP_NOIO); in raid10_alloc_init_r10buf()
3076 if (test_bit(MD_RECOVERY_SYNC, &conf->mddev->recovery) || in raid10_alloc_init_r10buf()
3077 test_bit(MD_RECOVERY_RESHAPE, &conf->mddev->recovery)) in raid10_alloc_init_r10buf()
3078 nalloc = conf->copies; /* resync */ in raid10_alloc_init_r10buf()
3083 bio = r10bio->devs[i].bio; in raid10_alloc_init_r10buf()
3084 rp = bio->bi_private; in raid10_alloc_init_r10buf()
3086 bio->bi_private = rp; in raid10_alloc_init_r10buf()
3087 bio = r10bio->devs[i].repl_bio; in raid10_alloc_init_r10buf()
3089 rp = bio->bi_private; in raid10_alloc_init_r10buf()
3091 bio->bi_private = rp; in raid10_alloc_init_r10buf()
3118 chunks = conf->geo.raid_disks / conf->geo.near_copies; in raid10_set_cluster_sync_high()
3119 if (conf->geo.raid_disks % conf->geo.near_copies == 0) in raid10_set_cluster_sync_high()
3123 window_size = (chunks + extra_chunk) * conf->mddev->chunk_sectors; in raid10_set_cluster_sync_high()
3131 conf->cluster_sync_high = conf->cluster_sync_low + window_size; in raid10_set_cluster_sync_high()
3137 * We need to make sure that no normal I/O request - particularly write
3138 * requests - conflict with active sync requests.
3144 * We differentiate by looking at MD_RECOVERY_SYNC in mddev->recovery.
3147 * and update if there are differences. If only one copy is live,
3150 * value for each non-in_sync drive, and over-write.
3153 * given address, one for each out-of-sync device. We model this by allocating
3154 * a number of r10_bio structures, one for each out-of-sync device.
3160 * This link is counted in ->remaining. When the r10_bio that points to NULL
3169 struct r10conf *conf = mddev->private; in raid10_sync_request()
3178 sector_t chunk_mask = conf->geo.chunk_mask; in raid10_sync_request()
3180 int error_disk = -1; in raid10_sync_request()
3186 if (mddev->bitmap == NULL && in raid10_sync_request()
3187 mddev->recovery_cp == MaxSector && in raid10_sync_request()
3188 mddev->reshape_position == MaxSector && in raid10_sync_request()
3189 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid10_sync_request()
3190 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid10_sync_request()
3191 !test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) && in raid10_sync_request()
3192 conf->fullsync == 0) { in raid10_sync_request()
3194 return mddev->dev_sectors - sector_nr; in raid10_sync_request()
3197 if (!mempool_initialized(&conf->r10buf_pool)) in raid10_sync_request()
3203 conf->cluster_sync_low = 0; in raid10_sync_request()
3204 conf->cluster_sync_high = 0; in raid10_sync_request()
3211 * mddev->curr_resync, but for recovery, in raid10_sync_request()
3215 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) { in raid10_sync_request()
3221 if (mddev->curr_resync < max_sector) { /* aborted */ in raid10_sync_request()
3222 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) in raid10_sync_request()
3223 mddev->bitmap_ops->end_sync(mddev, in raid10_sync_request()
3224 mddev->curr_resync, in raid10_sync_request()
3226 else for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3228 raid10_find_virt(conf, mddev->curr_resync, i); in raid10_sync_request()
3230 mddev->bitmap_ops->end_sync(mddev, sect, in raid10_sync_request()
3235 if ((!mddev->bitmap || conf->fullsync) in raid10_sync_request()
3236 && conf->have_replacement in raid10_sync_request()
3237 && test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3241 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3243 conf->mirrors[i].replacement; in raid10_sync_request()
3246 rdev->recovery_offset = MaxSector; in raid10_sync_request()
3249 conf->fullsync = 0; in raid10_sync_request()
3251 mddev->bitmap_ops->close_sync(mddev); in raid10_sync_request()
3257 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)) in raid10_sync_request()
3260 if (chunks_skipped >= conf->geo.raid_disks) { in raid10_sync_request()
3262 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ? "resync" : "recovery"); in raid10_sync_request()
3264 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3269 conf->mirrors[error_disk].recovery_disabled = in raid10_sync_request()
3270 mddev->recovery_disabled; in raid10_sync_request()
3278 return (max_sector - sector_nr) + sectors_skipped; in raid10_sync_request()
3281 if (max_sector > mddev->resync_max) in raid10_sync_request()
3282 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid10_sync_request()
3284 /* make sure whole request will fit in a chunk - if chunks in raid10_sync_request()
3287 if (conf->geo.near_copies < conf->geo.raid_disks && in raid10_sync_request()
3292 * If there is non-resync activity waiting for a turn, then let it in raid10_sync_request()
3295 if (conf->nr_waiting) in raid10_sync_request()
3308 /* First, we decide what to do and set ->bi_end_io in raid10_sync_request()
3313 max_sync = RESYNC_PAGES << (PAGE_SHIFT-9); in raid10_sync_request()
3314 if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3319 for (i = 0 ; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3325 struct raid10_info *mirror = &conf->mirrors[i]; in raid10_sync_request()
3328 mrdev = mirror->rdev; in raid10_sync_request()
3329 mreplace = mirror->replacement; in raid10_sync_request()
3331 if (mrdev && (test_bit(Faulty, &mrdev->flags) || in raid10_sync_request()
3332 test_bit(In_sync, &mrdev->flags))) in raid10_sync_request()
3334 if (mreplace && test_bit(Faulty, &mreplace->flags)) in raid10_sync_request()
3344 if (sect >= mddev->resync_max_sectors) in raid10_sync_request()
3345 /* last stripe is not complete - don't in raid10_sync_request()
3353 must_sync = mddev->bitmap_ops->start_sync(mddev, sect, in raid10_sync_request()
3360 !conf->fullsync) { in raid10_sync_request()
3364 chunks_skipped = -1; in raid10_sync_request()
3368 atomic_inc(&mrdev->nr_pending); in raid10_sync_request()
3370 atomic_inc(&mreplace->nr_pending); in raid10_sync_request()
3373 r10_bio->state = 0; in raid10_sync_request()
3375 atomic_set(&r10_bio->remaining, 0); in raid10_sync_request()
3377 r10_bio->master_bio = (struct bio*)rb2; in raid10_sync_request()
3379 atomic_inc(&rb2->remaining); in raid10_sync_request()
3380 r10_bio->mddev = mddev; in raid10_sync_request()
3381 set_bit(R10BIO_IsRecover, &r10_bio->state); in raid10_sync_request()
3382 r10_bio->sector = sect; in raid10_sync_request()
3389 for (j = 0; j < conf->geo.raid_disks; j++) { in raid10_sync_request()
3390 struct md_rdev *rdev = conf->mirrors[j].rdev; in raid10_sync_request()
3392 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) { in raid10_sync_request()
3398 must_sync = mddev->bitmap_ops->start_sync(mddev, sect, in raid10_sync_request()
3402 for (j=0; j<conf->copies;j++) { in raid10_sync_request()
3404 int d = r10_bio->devs[j].devnum; in raid10_sync_request()
3406 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid10_sync_request()
3410 !test_bit(In_sync, &rdev->flags)) in raid10_sync_request()
3414 sector = r10_bio->devs[j].addr; in raid10_sync_request()
3419 max_sync = first_bad - sector; in raid10_sync_request()
3421 bad_sectors -= (sector in raid10_sync_request()
3422 - first_bad); in raid10_sync_request()
3428 bio = r10_bio->devs[0].bio; in raid10_sync_request()
3429 bio->bi_next = biolist; in raid10_sync_request()
3431 bio->bi_end_io = end_sync_read; in raid10_sync_request()
3432 bio->bi_opf = REQ_OP_READ; in raid10_sync_request()
3433 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3434 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3435 from_addr = r10_bio->devs[j].addr; in raid10_sync_request()
3436 bio->bi_iter.bi_sector = from_addr + in raid10_sync_request()
3437 rdev->data_offset; in raid10_sync_request()
3438 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3439 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3442 for (k=0; k<conf->copies; k++) in raid10_sync_request()
3443 if (r10_bio->devs[k].devnum == i) in raid10_sync_request()
3445 BUG_ON(k == conf->copies); in raid10_sync_request()
3446 to_addr = r10_bio->devs[k].addr; in raid10_sync_request()
3447 r10_bio->devs[0].devnum = d; in raid10_sync_request()
3448 r10_bio->devs[0].addr = from_addr; in raid10_sync_request()
3449 r10_bio->devs[1].devnum = i; in raid10_sync_request()
3450 r10_bio->devs[1].addr = to_addr; in raid10_sync_request()
3453 bio = r10_bio->devs[1].bio; in raid10_sync_request()
3454 bio->bi_next = biolist; in raid10_sync_request()
3456 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3457 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3458 bio->bi_iter.bi_sector = to_addr in raid10_sync_request()
3459 + mrdev->data_offset; in raid10_sync_request()
3460 bio_set_dev(bio, mrdev->bdev); in raid10_sync_request()
3461 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3463 r10_bio->devs[1].bio->bi_end_io = NULL; in raid10_sync_request()
3466 bio = r10_bio->devs[1].repl_bio; in raid10_sync_request()
3468 bio->bi_end_io = NULL; in raid10_sync_request()
3475 bio->bi_next = biolist; in raid10_sync_request()
3477 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3478 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3479 bio->bi_iter.bi_sector = to_addr + in raid10_sync_request()
3480 mreplace->data_offset; in raid10_sync_request()
3481 bio_set_dev(bio, mreplace->bdev); in raid10_sync_request()
3482 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3485 if (j == conf->copies) { in raid10_sync_request()
3493 for (k = 0; k < conf->copies; k++) in raid10_sync_request()
3494 if (r10_bio->devs[k].devnum == i) in raid10_sync_request()
3497 &mrdev->flags) in raid10_sync_request()
3500 r10_bio->devs[k].addr, in raid10_sync_request()
3506 r10_bio->devs[k].addr, in raid10_sync_request()
3512 &mddev->recovery)) in raid10_sync_request()
3515 mirror->recovery_disabled in raid10_sync_request()
3516 = mddev->recovery_disabled; in raid10_sync_request()
3522 atomic_dec(&rb2->remaining); in raid10_sync_request()
3534 if (r10_bio->devs[0].bio->bi_opf & MD_FAILFAST) { in raid10_sync_request()
3540 for (; j < conf->copies; j++) { in raid10_sync_request()
3541 int d = r10_bio->devs[j].devnum; in raid10_sync_request()
3542 if (conf->mirrors[d].rdev && in raid10_sync_request()
3544 &conf->mirrors[d].rdev->flags)) in raid10_sync_request()
3548 r10_bio->devs[0].bio->bi_opf in raid10_sync_request()
3555 r10_bio = (struct r10bio*) rb2->master_bio; in raid10_sync_request()
3556 rb2->master_bio = NULL; in raid10_sync_request()
3566 * Since curr_resync_completed could probably not update in in raid10_sync_request()
3572 mddev->bitmap_ops->cond_end_sync(mddev, sector_nr, in raid10_sync_request()
3574 (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid10_sync_request()
3576 if (!mddev->bitmap_ops->start_sync(mddev, sector_nr, in raid10_sync_request()
3578 mddev->degraded) && in raid10_sync_request()
3579 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, in raid10_sync_request()
3580 &mddev->recovery)) { in raid10_sync_request()
3588 r10_bio->state = 0; in raid10_sync_request()
3590 r10_bio->mddev = mddev; in raid10_sync_request()
3591 atomic_set(&r10_bio->remaining, 0); in raid10_sync_request()
3593 conf->next_resync = sector_nr; in raid10_sync_request()
3595 r10_bio->master_bio = NULL; in raid10_sync_request()
3596 r10_bio->sector = sector_nr; in raid10_sync_request()
3597 set_bit(R10BIO_IsSync, &r10_bio->state); in raid10_sync_request()
3599 r10_bio->sectors = (sector_nr | chunk_mask) - sector_nr + 1; in raid10_sync_request()
3601 for (i = 0; i < conf->copies; i++) { in raid10_sync_request()
3602 int d = r10_bio->devs[i].devnum; in raid10_sync_request()
3607 if (r10_bio->devs[i].repl_bio) in raid10_sync_request()
3608 r10_bio->devs[i].repl_bio->bi_end_io = NULL; in raid10_sync_request()
3610 bio = r10_bio->devs[i].bio; in raid10_sync_request()
3611 bio->bi_status = BLK_STS_IOERR; in raid10_sync_request()
3612 rdev = conf->mirrors[d].rdev; in raid10_sync_request()
3613 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) in raid10_sync_request()
3616 sector = r10_bio->devs[i].addr; in raid10_sync_request()
3620 max_sync = first_bad - sector; in raid10_sync_request()
3622 bad_sectors -= (sector - first_bad); in raid10_sync_request()
3628 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3629 atomic_inc(&r10_bio->remaining); in raid10_sync_request()
3630 bio->bi_next = biolist; in raid10_sync_request()
3632 bio->bi_end_io = end_sync_read; in raid10_sync_request()
3633 bio->bi_opf = REQ_OP_READ; in raid10_sync_request()
3634 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3635 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3636 bio->bi_iter.bi_sector = sector + rdev->data_offset; in raid10_sync_request()
3637 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3640 rdev = conf->mirrors[d].replacement; in raid10_sync_request()
3641 if (rdev == NULL || test_bit(Faulty, &rdev->flags)) in raid10_sync_request()
3644 atomic_inc(&rdev->nr_pending); in raid10_sync_request()
3647 bio = r10_bio->devs[i].repl_bio; in raid10_sync_request()
3648 bio->bi_status = BLK_STS_IOERR; in raid10_sync_request()
3650 sector = r10_bio->devs[i].addr; in raid10_sync_request()
3651 bio->bi_next = biolist; in raid10_sync_request()
3653 bio->bi_end_io = end_sync_write; in raid10_sync_request()
3654 bio->bi_opf = REQ_OP_WRITE; in raid10_sync_request()
3655 if (test_bit(FailFast, &rdev->flags)) in raid10_sync_request()
3656 bio->bi_opf |= MD_FAILFAST; in raid10_sync_request()
3657 bio->bi_iter.bi_sector = sector + rdev->data_offset; in raid10_sync_request()
3658 bio_set_dev(bio, rdev->bdev); in raid10_sync_request()
3663 for (i=0; i<conf->copies; i++) { in raid10_sync_request()
3664 int d = r10_bio->devs[i].devnum; in raid10_sync_request()
3665 if (r10_bio->devs[i].bio->bi_end_io) in raid10_sync_request()
3666 rdev_dec_pending(conf->mirrors[d].rdev, in raid10_sync_request()
3668 if (r10_bio->devs[i].repl_bio && in raid10_sync_request()
3669 r10_bio->devs[i].repl_bio->bi_end_io) in raid10_sync_request()
3671 conf->mirrors[d].replacement, in raid10_sync_request()
3687 len = (max_sector - sector_nr) << 9; in raid10_sync_request()
3690 for (bio= biolist ; bio ; bio=bio->bi_next) { in raid10_sync_request()
3694 bio->bi_status = BLK_STS_RESOURCE; in raid10_sync_request()
3702 r10_bio->sectors = nr_sectors; in raid10_sync_request()
3705 test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) { in raid10_sync_request()
3707 if (conf->cluster_sync_high < sector_nr + nr_sectors) { in raid10_sync_request()
3708 conf->cluster_sync_low = mddev->curr_resync_completed; in raid10_sync_request()
3711 md_cluster_ops->resync_info_update(mddev, in raid10_sync_request()
3712 conf->cluster_sync_low, in raid10_sync_request()
3713 conf->cluster_sync_high); in raid10_sync_request()
3720 for (i = 0; i < conf->geo.raid_disks; i++) { in raid10_sync_request()
3728 if (conf->cluster_sync_high < sect_va1 + nr_sectors) { in raid10_sync_request()
3735 mddev->curr_resync_completed, i); in raid10_sync_request()
3737 if (conf->cluster_sync_low == 0 || in raid10_sync_request()
3738 conf->cluster_sync_low > sect_va2) in raid10_sync_request()
3739 conf->cluster_sync_low = sect_va2; in raid10_sync_request()
3744 md_cluster_ops->resync_info_update(mddev, in raid10_sync_request()
3745 conf->cluster_sync_low, in raid10_sync_request()
3746 conf->cluster_sync_high); in raid10_sync_request()
3752 biolist = biolist->bi_next; in raid10_sync_request()
3754 bio->bi_next = NULL; in raid10_sync_request()
3756 r10_bio->sectors = nr_sectors; in raid10_sync_request()
3758 if (bio->bi_end_io == end_sync_read) { in raid10_sync_request()
3760 bio->bi_status = 0; in raid10_sync_request()
3773 /* There is nowhere to write, so all non-sync in raid10_sync_request()
3780 sectors_skipped += (max_sector - sector_nr); in raid10_sync_request()
3790 struct r10conf *conf = mddev->private; in raid10_size()
3793 raid_disks = min(conf->geo.raid_disks, in raid10_size()
3794 conf->prev.raid_disks); in raid10_size()
3796 sectors = conf->dev_sectors; in raid10_size()
3798 size = sectors >> conf->geo.chunk_shift; in raid10_size()
3799 sector_div(size, conf->geo.far_copies); in raid10_size()
3801 sector_div(size, conf->geo.near_copies); in raid10_size()
3803 return size << conf->geo.chunk_shift; in raid10_size()
3808 /* Calculate the number of sectors-per-device that will in calc_sectors()
3809 * actually be used, and set conf->dev_sectors and in calc_sectors()
3810 * conf->stride in calc_sectors()
3813 size = size >> conf->geo.chunk_shift; in calc_sectors()
3814 sector_div(size, conf->geo.far_copies); in calc_sectors()
3815 size = size * conf->geo.raid_disks; in calc_sectors()
3816 sector_div(size, conf->geo.near_copies); in calc_sectors()
3819 size = size * conf->copies; in calc_sectors()
3824 size = DIV_ROUND_UP_SECTOR_T(size, conf->geo.raid_disks); in calc_sectors()
3826 conf->dev_sectors = size << conf->geo.chunk_shift; in calc_sectors()
3828 if (conf->geo.far_offset) in calc_sectors()
3829 conf->geo.stride = 1 << conf->geo.chunk_shift; in calc_sectors()
3831 sector_div(size, conf->geo.far_copies); in calc_sectors()
3832 conf->geo.stride = size << conf->geo.chunk_shift; in calc_sectors()
3839 int nc, fc, fo; in setup_geo() local
3843 layout = mddev->layout; in setup_geo()
3844 chunk = mddev->chunk_sectors; in setup_geo()
3845 disks = mddev->raid_disks - mddev->delta_disks; in setup_geo()
3848 layout = mddev->new_layout; in setup_geo()
3849 chunk = mddev->new_chunk_sectors; in setup_geo()
3850 disks = mddev->raid_disks; in setup_geo()
3853 case geo_start: /* new when starting reshape - raid_disks not in setup_geo()
3855 layout = mddev->new_layout; in setup_geo()
3856 chunk = mddev->new_chunk_sectors; in setup_geo()
3857 disks = mddev->raid_disks + mddev->delta_disks; in setup_geo()
3861 return -1; in setup_geo()
3864 return -2; in setup_geo()
3866 fc = (layout >> 8) & 255; in setup_geo()
3868 geo->raid_disks = disks; in setup_geo()
3869 geo->near_copies = nc; in setup_geo()
3870 geo->far_copies = fc; in setup_geo()
3871 geo->far_offset = fo; in setup_geo()
3874 geo->far_set_size = disks; in setup_geo()
3876 case 1: /* "improved" layout which was buggy. Hopefully no-one is in setup_geo()
3878 geo->far_set_size = disks/fc; in setup_geo()
3879 WARN(geo->far_set_size < fc, in setup_geo()
3880 "This RAID10 layout does not provide data safety - please backup and create new array\n"); in setup_geo()
3883 geo->far_set_size = fc * nc; in setup_geo()
3886 return -1; in setup_geo()
3888 geo->chunk_mask = chunk - 1; in setup_geo()
3889 geo->chunk_shift = ffz(~chunk); in setup_geo()
3890 return nc*fc; in setup_geo()
3898 mempool_exit(&conf->r10bio_pool); in raid10_free_conf()
3899 kfree(conf->mirrors); in raid10_free_conf()
3900 kfree(conf->mirrors_old); in raid10_free_conf()
3901 kfree(conf->mirrors_new); in raid10_free_conf()
3902 safe_put_page(conf->tmppage); in raid10_free_conf()
3903 bioset_exit(&conf->bio_split); in raid10_free_conf()
3910 int err = -EINVAL; in setup_conf()
3916 if (copies == -2) { in setup_conf()
3922 if (copies < 2 || copies > mddev->raid_disks) { in setup_conf()
3924 mdname(mddev), mddev->new_layout); in setup_conf()
3928 err = -ENOMEM; in setup_conf()
3934 conf->mirrors = kcalloc(mddev->raid_disks + max(0, -mddev->delta_disks), in setup_conf()
3937 if (!conf->mirrors) in setup_conf()
3940 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
3941 if (!conf->tmppage) in setup_conf()
3944 conf->geo = geo; in setup_conf()
3945 conf->copies = copies; in setup_conf()
3946 err = mempool_init(&conf->r10bio_pool, NR_RAID_BIOS, r10bio_pool_alloc, in setup_conf()
3951 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
3955 calc_sectors(conf, mddev->dev_sectors); in setup_conf()
3956 if (mddev->reshape_position == MaxSector) { in setup_conf()
3957 conf->prev = conf->geo; in setup_conf()
3958 conf->reshape_progress = MaxSector; in setup_conf()
3960 if (setup_geo(&conf->prev, mddev, geo_old) != conf->copies) { in setup_conf()
3961 err = -EINVAL; in setup_conf()
3964 conf->reshape_progress = mddev->reshape_position; in setup_conf()
3965 if (conf->prev.far_offset) in setup_conf()
3966 conf->prev.stride = 1 << conf->prev.chunk_shift; in setup_conf()
3969 conf->prev.stride = conf->dev_sectors; in setup_conf()
3971 conf->reshape_safe = conf->reshape_progress; in setup_conf()
3972 spin_lock_init(&conf->device_lock); in setup_conf()
3973 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
3974 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
3976 seqlock_init(&conf->resync_lock); in setup_conf()
3977 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
3978 atomic_set(&conf->nr_pending, 0); in setup_conf()
3980 err = -ENOMEM; in setup_conf()
3981 rcu_assign_pointer(conf->thread, in setup_conf()
3983 if (!conf->thread) in setup_conf()
3986 conf->mddev = mddev; in setup_conf()
3996 unsigned int raid_disks = conf->geo.raid_disks; in raid10_nr_stripes()
3998 if (conf->geo.raid_disks % conf->geo.near_copies) in raid10_nr_stripes()
4000 return raid_disks / conf->geo.near_copies; in raid10_nr_stripes()
4005 struct r10conf *conf = mddev->private; in raid10_set_queue_limits()
4011 lim.io_min = mddev->chunk_sectors << 9; in raid10_set_queue_limits()
4017 return queue_limits_set(mddev->gendisk->queue, &lim); in raid10_set_queue_limits()
4029 int ret = -EIO; in raid10_run()
4031 if (mddev->private == NULL) { in raid10_run()
4035 mddev->private = conf; in raid10_run()
4037 conf = mddev->private; in raid10_run()
4041 rcu_assign_pointer(mddev->thread, conf->thread); in raid10_run()
4042 rcu_assign_pointer(conf->thread, NULL); in raid10_run()
4044 if (mddev_is_clustered(conf->mddev)) { in raid10_run()
4045 int fc, fo; in raid10_run() local
4047 fc = (mddev->layout >> 8) & 255; in raid10_run()
4048 fo = mddev->layout & (1<<16); in raid10_run()
4049 if (fc > 1 || fo > 0) { in raid10_run()
4059 disk_idx = rdev->raid_disk; in raid10_run()
4062 if (disk_idx >= conf->geo.raid_disks && in raid10_run()
4063 disk_idx >= conf->prev.raid_disks) in raid10_run()
4065 disk = conf->mirrors + disk_idx; in raid10_run()
4067 if (test_bit(Replacement, &rdev->flags)) { in raid10_run()
4068 if (disk->replacement) in raid10_run()
4070 disk->replacement = rdev; in raid10_run()
4072 if (disk->rdev) in raid10_run()
4074 disk->rdev = rdev; in raid10_run()
4076 diff = (rdev->new_data_offset - rdev->data_offset); in raid10_run()
4077 if (!mddev->reshape_backwards) in raid10_run()
4078 diff = -diff; in raid10_run()
4084 disk->head_position = 0; in raid10_run()
4088 if (!mddev_is_dm(conf->mddev)) { in raid10_run()
4098 if (!enough(conf, -1)) { in raid10_run()
4104 if (conf->reshape_progress != MaxSector) { in raid10_run()
4106 if (conf->geo.far_copies != 1 && in raid10_run()
4107 conf->geo.far_offset == 0) in raid10_run()
4109 if (conf->prev.far_copies != 1 && in raid10_run()
4110 conf->prev.far_offset == 0) in raid10_run()
4114 mddev->degraded = 0; in raid10_run()
4116 i < conf->geo.raid_disks in raid10_run()
4117 || i < conf->prev.raid_disks; in raid10_run()
4120 disk = conf->mirrors + i; in raid10_run()
4122 if (!disk->rdev && disk->replacement) { in raid10_run()
4123 /* The replacement is all we have - use it */ in raid10_run()
4124 disk->rdev = disk->replacement; in raid10_run()
4125 disk->replacement = NULL; in raid10_run()
4126 clear_bit(Replacement, &disk->rdev->flags); in raid10_run()
4129 if (!disk->rdev || in raid10_run()
4130 !test_bit(In_sync, &disk->rdev->flags)) { in raid10_run()
4131 disk->head_position = 0; in raid10_run()
4132 mddev->degraded++; in raid10_run()
4133 if (disk->rdev && in raid10_run()
4134 disk->rdev->saved_raid_disk < 0) in raid10_run()
4135 conf->fullsync = 1; in raid10_run()
4138 if (disk->replacement && in raid10_run()
4139 !test_bit(In_sync, &disk->replacement->flags) && in raid10_run()
4140 disk->replacement->saved_raid_disk < 0) { in raid10_run()
4141 conf->fullsync = 1; in raid10_run()
4144 disk->recovery_disabled = mddev->recovery_disabled - 1; in raid10_run()
4147 if (mddev->recovery_cp != MaxSector) in raid10_run()
4148 pr_notice("md/raid10:%s: not clean -- starting background reconstruction\n", in raid10_run()
4151 mdname(mddev), conf->geo.raid_disks - mddev->degraded, in raid10_run()
4152 conf->geo.raid_disks); in raid10_run()
4156 mddev->dev_sectors = conf->dev_sectors; in raid10_run()
4159 mddev->resync_max_sectors = size; in raid10_run()
4160 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid10_run()
4165 if (conf->reshape_progress != MaxSector) { in raid10_run()
4168 before_length = ((1 << conf->prev.chunk_shift) * in raid10_run()
4169 conf->prev.far_copies); in raid10_run()
4170 after_length = ((1 << conf->geo.chunk_shift) * in raid10_run()
4171 conf->geo.far_copies); in raid10_run()
4178 conf->offset_diff = min_offset_diff; in raid10_run()
4180 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); in raid10_run()
4181 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); in raid10_run()
4182 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); in raid10_run()
4183 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_run()
4189 md_unregister_thread(mddev, &mddev->thread); in raid10_run()
4191 mddev->private = NULL; in raid10_run()
4203 struct r10conf *conf = mddev->private; in raid10_quiesce()
4225 struct r10conf *conf = mddev->private; in raid10_resize()
4229 if (mddev->reshape_position != MaxSector) in raid10_resize()
4230 return -EBUSY; in raid10_resize()
4232 if (conf->geo.far_copies > 1 && !conf->geo.far_offset) in raid10_resize()
4233 return -EINVAL; in raid10_resize()
4237 if (mddev->external_size && in raid10_resize()
4238 mddev->array_sectors > size) in raid10_resize()
4239 return -EINVAL; in raid10_resize()
4241 ret = mddev->bitmap_ops->resize(mddev, size, 0, false); in raid10_resize()
4246 if (sectors > mddev->dev_sectors && in raid10_resize()
4247 mddev->recovery_cp > oldsize) { in raid10_resize()
4248 mddev->recovery_cp = oldsize; in raid10_resize()
4249 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_resize()
4252 mddev->dev_sectors = conf->dev_sectors; in raid10_resize()
4253 mddev->resync_max_sectors = size; in raid10_resize()
4262 if (mddev->degraded > 0) { in raid10_takeover_raid0()
4265 return ERR_PTR(-EINVAL); in raid10_takeover_raid0()
4270 mddev->new_level = 10; in raid10_takeover_raid0()
4272 mddev->new_layout = (1<<8) + 2; in raid10_takeover_raid0()
4273 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid10_takeover_raid0()
4274 mddev->delta_disks = mddev->raid_disks; in raid10_takeover_raid0()
4275 mddev->raid_disks *= 2; in raid10_takeover_raid0()
4277 mddev->recovery_cp = MaxSector; in raid10_takeover_raid0()
4278 mddev->dev_sectors = size; in raid10_takeover_raid0()
4283 if (rdev->raid_disk >= 0) { in raid10_takeover_raid0()
4284 rdev->new_raid_disk = rdev->raid_disk * 2; in raid10_takeover_raid0()
4285 rdev->sectors = size; in raid10_takeover_raid0()
4297 * raid0 - providing it has only two drives in raid10_takeover()
4299 if (mddev->level == 0) { in raid10_takeover()
4301 raid0_conf = mddev->private; in raid10_takeover()
4302 if (raid0_conf->nr_strip_zones > 1) { in raid10_takeover()
4305 return ERR_PTR(-EINVAL); in raid10_takeover()
4308 raid0_conf->strip_zone->zone_end, in raid10_takeover()
4309 raid0_conf->strip_zone->nb_dev); in raid10_takeover()
4311 return ERR_PTR(-EINVAL); in raid10_takeover()
4317 * - layout (to ->new_layout) in raid10_check_reshape()
4318 * - chunk size (to ->new_chunk_sectors) in raid10_check_reshape()
4319 * - raid_disks (by delta_disks) in raid10_check_reshape()
4330 struct r10conf *conf = mddev->private; in raid10_check_reshape()
4333 if (conf->geo.far_copies != 1 && !conf->geo.far_offset) in raid10_check_reshape()
4334 return -EINVAL; in raid10_check_reshape()
4336 if (setup_geo(&geo, mddev, geo_start) != conf->copies) in raid10_check_reshape()
4338 return -EINVAL; in raid10_check_reshape()
4341 return -EINVAL; in raid10_check_reshape()
4343 if (mddev->array_sectors & geo.chunk_mask) in raid10_check_reshape()
4345 return -EINVAL; in raid10_check_reshape()
4347 if (!enough(conf, -1)) in raid10_check_reshape()
4348 return -EINVAL; in raid10_check_reshape()
4350 kfree(conf->mirrors_new); in raid10_check_reshape()
4351 conf->mirrors_new = NULL; in raid10_check_reshape()
4352 if (mddev->delta_disks > 0) { in raid10_check_reshape()
4354 conf->mirrors_new = in raid10_check_reshape()
4355 kcalloc(mddev->raid_disks + mddev->delta_disks, in raid10_check_reshape()
4358 if (!conf->mirrors_new) in raid10_check_reshape()
4359 return -ENOMEM; in raid10_check_reshape()
4366 * - start an array
4367 * - remove non-faulty devices
4368 * - add a spare
4369 * - allow a reshape
4374 * of the two sections, and some non-in_sync devices may
4384 for (i = 0; i < conf->prev.raid_disks; i++) { in calc_degraded()
4385 struct md_rdev *rdev = conf->mirrors[i].rdev; in calc_degraded()
4387 if (!rdev || test_bit(Faulty, &rdev->flags)) in calc_degraded()
4389 else if (!test_bit(In_sync, &rdev->flags)) in calc_degraded()
4396 if (conf->geo.raid_disks == conf->prev.raid_disks) in calc_degraded()
4399 for (i = 0; i < conf->geo.raid_disks; i++) { in calc_degraded()
4400 struct md_rdev *rdev = conf->mirrors[i].rdev; in calc_degraded()
4402 if (!rdev || test_bit(Faulty, &rdev->flags)) in calc_degraded()
4404 else if (!test_bit(In_sync, &rdev->flags)) { in calc_degraded()
4410 if (conf->geo.raid_disks <= conf->prev.raid_disks) in calc_degraded()
4426 * array non-degraded. We also require that the difference in raid10_start_reshape()
4427 * between old and new data_offset - on each device - is in raid10_start_reshape()
4428 * enough that we never risk over-writing. in raid10_start_reshape()
4435 struct r10conf *conf = mddev->private; in raid10_start_reshape()
4440 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery)) in raid10_start_reshape()
4441 return -EBUSY; in raid10_start_reshape()
4443 if (setup_geo(&new, mddev, geo_start) != conf->copies) in raid10_start_reshape()
4444 return -EINVAL; in raid10_start_reshape()
4446 before_length = ((1 << conf->prev.chunk_shift) * in raid10_start_reshape()
4447 conf->prev.far_copies); in raid10_start_reshape()
4448 after_length = ((1 << conf->geo.chunk_shift) * in raid10_start_reshape()
4449 conf->geo.far_copies); in raid10_start_reshape()
4452 if (!test_bit(In_sync, &rdev->flags) in raid10_start_reshape()
4453 && !test_bit(Faulty, &rdev->flags)) in raid10_start_reshape()
4455 if (rdev->raid_disk >= 0) { in raid10_start_reshape()
4456 long long diff = (rdev->new_data_offset in raid10_start_reshape()
4457 - rdev->data_offset); in raid10_start_reshape()
4458 if (!mddev->reshape_backwards) in raid10_start_reshape()
4459 diff = -diff; in raid10_start_reshape()
4469 return -EINVAL; in raid10_start_reshape()
4471 if (spares < mddev->delta_disks) in raid10_start_reshape()
4472 return -EINVAL; in raid10_start_reshape()
4474 conf->offset_diff = min_offset_diff; in raid10_start_reshape()
4475 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4476 if (conf->mirrors_new) { in raid10_start_reshape()
4477 memcpy(conf->mirrors_new, conf->mirrors, in raid10_start_reshape()
4478 sizeof(struct raid10_info)*conf->prev.raid_disks); in raid10_start_reshape()
4480 kfree(conf->mirrors_old); in raid10_start_reshape()
4481 conf->mirrors_old = conf->mirrors; in raid10_start_reshape()
4482 conf->mirrors = conf->mirrors_new; in raid10_start_reshape()
4483 conf->mirrors_new = NULL; in raid10_start_reshape()
4485 setup_geo(&conf->geo, mddev, geo_start); in raid10_start_reshape()
4487 if (mddev->reshape_backwards) { in raid10_start_reshape()
4489 if (size < mddev->array_sectors) { in raid10_start_reshape()
4490 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4493 return -EINVAL; in raid10_start_reshape()
4495 mddev->resync_max_sectors = size; in raid10_start_reshape()
4496 conf->reshape_progress = size; in raid10_start_reshape()
4498 conf->reshape_progress = 0; in raid10_start_reshape()
4499 conf->reshape_safe = conf->reshape_progress; in raid10_start_reshape()
4500 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4502 if (mddev->delta_disks && mddev->bitmap) { in raid10_start_reshape()
4507 newsize = raid10_size(mddev, 0, conf->geo.raid_disks); in raid10_start_reshape()
4510 ret = mddev->bitmap_ops->resize(mddev, newsize, 0, false); in raid10_start_reshape()
4518 if (rdev->raid_disk > -1 && in raid10_start_reshape()
4519 !test_bit(Faulty, &rdev->flags)) in raid10_start_reshape()
4520 sb = page_address(rdev->sb_page); in raid10_start_reshape()
4525 * call bitmap_ops->resize again since it should be called when in raid10_start_reshape()
4528 if ((sb && (le32_to_cpu(sb->feature_map) & in raid10_start_reshape()
4532 ret = mddev->bitmap_ops->resize(mddev, newsize, 0, false); in raid10_start_reshape()
4536 ret = md_cluster_ops->resize_bitmaps(mddev, newsize, oldsize); in raid10_start_reshape()
4538 mddev->bitmap_ops->resize(mddev, oldsize, 0, false); in raid10_start_reshape()
4543 if (mddev->delta_disks > 0) { in raid10_start_reshape()
4545 if (rdev->raid_disk < 0 && in raid10_start_reshape()
4546 !test_bit(Faulty, &rdev->flags)) { in raid10_start_reshape()
4548 if (rdev->raid_disk >= in raid10_start_reshape()
4549 conf->prev.raid_disks) in raid10_start_reshape()
4550 set_bit(In_sync, &rdev->flags); in raid10_start_reshape()
4552 rdev->recovery_offset = 0; in raid10_start_reshape()
4557 } else if (rdev->raid_disk >= conf->prev.raid_disks in raid10_start_reshape()
4558 && !test_bit(Faulty, &rdev->flags)) { in raid10_start_reshape()
4560 set_bit(In_sync, &rdev->flags); in raid10_start_reshape()
4564 * ->degraded is measured against the larger of the in raid10_start_reshape()
4567 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4568 mddev->degraded = calc_degraded(conf); in raid10_start_reshape()
4569 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4570 mddev->raid_disks = conf->geo.raid_disks; in raid10_start_reshape()
4571 mddev->reshape_position = conf->reshape_progress; in raid10_start_reshape()
4572 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid10_start_reshape()
4574 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery); in raid10_start_reshape()
4575 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery); in raid10_start_reshape()
4576 clear_bit(MD_RECOVERY_DONE, &mddev->recovery); in raid10_start_reshape()
4577 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery); in raid10_start_reshape()
4578 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_start_reshape()
4579 conf->reshape_checkpoint = jiffies; in raid10_start_reshape()
4584 mddev->recovery = 0; in raid10_start_reshape()
4585 spin_lock_irq(&conf->device_lock); in raid10_start_reshape()
4586 conf->geo = conf->prev; in raid10_start_reshape()
4587 mddev->raid_disks = conf->geo.raid_disks; in raid10_start_reshape()
4589 rdev->new_data_offset = rdev->data_offset; in raid10_start_reshape()
4591 conf->reshape_progress = MaxSector; in raid10_start_reshape()
4592 conf->reshape_safe = MaxSector; in raid10_start_reshape()
4593 mddev->reshape_position = MaxSector; in raid10_start_reshape()
4594 spin_unlock_irq(&conf->device_lock); in raid10_start_reshape()
4598 /* Calculate the last device-address that could contain
4599 * any block from the chunk that includes the array-address 's'
4601 * i.e. the address returned will be chunk-aligned and after
4606 s = (s | geo->chunk_mask) + 1; in last_dev_address()
4607 s >>= geo->chunk_shift; in last_dev_address()
4608 s *= geo->near_copies; in last_dev_address()
4609 s = DIV_ROUND_UP_SECTOR_T(s, geo->raid_disks); in last_dev_address()
4610 s *= geo->far_copies; in last_dev_address()
4611 s <<= geo->chunk_shift; in last_dev_address()
4615 /* Calculate the first device-address that could contain
4616 * any block from the chunk that includes the array-address 's'.
4621 s >>= geo->chunk_shift; in first_dev_address()
4622 s *= geo->near_copies; in first_dev_address()
4623 sector_div(s, geo->raid_disks); in first_dev_address()
4624 s *= geo->far_copies; in first_dev_address()
4625 s <<= geo->chunk_shift; in first_dev_address()
4636 * a chunk, but that is not a problem as mddev->reshape_position in reshape_request()
4641 * we need to flush all reshape requests and update the metadata. in reshape_request()
4652 * we must update the metadata to increase the 'safe' position. in reshape_request()
4659 * (conf->offset_diff - always positive) allows a bit of slack, in reshape_request()
4664 * The means one for each copy for write-out and an extra one for in reshape_request()
4665 * read-in. in reshape_request()
4666 * We store the read-in bio in ->master_bio and the others in in reshape_request()
4667 * ->devs[x].bio and ->devs[x].repl_bio. in reshape_request()
4669 struct r10conf *conf = mddev->private; in reshape_request()
4684 if (mddev->reshape_backwards && in reshape_request()
4685 conf->reshape_progress < raid10_size(mddev, 0, 0)) { in reshape_request()
4687 - conf->reshape_progress); in reshape_request()
4688 } else if (!mddev->reshape_backwards && in reshape_request()
4689 conf->reshape_progress > 0) in reshape_request()
4690 sector_nr = conf->reshape_progress; in reshape_request()
4692 mddev->curr_resync_completed = sector_nr; in reshape_request()
4693 sysfs_notify_dirent_safe(mddev->sysfs_completed); in reshape_request()
4700 * as that doesn't work well for ->reshape_backwards. in reshape_request()
4701 * So just use ->reshape_progress. in reshape_request()
4703 if (mddev->reshape_backwards) { in reshape_request()
4707 next = first_dev_address(conf->reshape_progress - 1, in reshape_request()
4708 &conf->geo); in reshape_request()
4713 safe = last_dev_address(conf->reshape_safe - 1, in reshape_request()
4714 &conf->prev); in reshape_request()
4716 if (next + conf->offset_diff < safe) in reshape_request()
4719 last = conf->reshape_progress - 1; in reshape_request()
4720 sector_nr = last & ~(sector_t)(conf->geo.chunk_mask in reshape_request()
4721 & conf->prev.chunk_mask); in reshape_request()
4723 sector_nr = last + 1 - RESYNC_SECTORS; in reshape_request()
4728 next = last_dev_address(conf->reshape_progress, &conf->geo); in reshape_request()
4733 safe = first_dev_address(conf->reshape_safe, &conf->prev); in reshape_request()
4735 /* Need to update metadata if 'next' might be beyond 'safe' in reshape_request()
4738 if (next > safe + conf->offset_diff) in reshape_request()
4741 sector_nr = conf->reshape_progress; in reshape_request()
4742 last = sector_nr | (conf->geo.chunk_mask in reshape_request()
4743 & conf->prev.chunk_mask); in reshape_request()
4746 last = sector_nr + RESYNC_SECTORS - 1; in reshape_request()
4750 time_after(jiffies, conf->reshape_checkpoint + 10*HZ)) { in reshape_request()
4751 /* Need to update reshape_position in metadata */ in reshape_request()
4753 mddev->reshape_position = conf->reshape_progress; in reshape_request()
4754 if (mddev->reshape_backwards) in reshape_request()
4755 mddev->curr_resync_completed = raid10_size(mddev, 0, 0) in reshape_request()
4756 - conf->reshape_progress; in reshape_request()
4758 mddev->curr_resync_completed = conf->reshape_progress; in reshape_request()
4759 conf->reshape_checkpoint = jiffies; in reshape_request()
4760 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in reshape_request()
4761 md_wakeup_thread(mddev->thread); in reshape_request()
4762 wait_event(mddev->sb_wait, mddev->sb_flags == 0 || in reshape_request()
4763 test_bit(MD_RECOVERY_INTR, &mddev->recovery)); in reshape_request()
4764 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) { in reshape_request()
4768 conf->reshape_safe = mddev->reshape_position; in reshape_request()
4776 r10_bio->state = 0; in reshape_request()
4778 atomic_set(&r10_bio->remaining, 0); in reshape_request()
4779 r10_bio->mddev = mddev; in reshape_request()
4780 r10_bio->sector = sector_nr; in reshape_request()
4781 set_bit(R10BIO_IsReshape, &r10_bio->state); in reshape_request()
4782 r10_bio->sectors = last - sector_nr + 1; in reshape_request()
4784 BUG_ON(!test_bit(R10BIO_Previous, &r10_bio->state)); in reshape_request()
4791 mempool_free(r10_bio, &conf->r10buf_pool); in reshape_request()
4792 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in reshape_request()
4796 read_bio = bio_alloc_bioset(rdev->bdev, RESYNC_PAGES, REQ_OP_READ, in reshape_request()
4797 GFP_KERNEL, &mddev->bio_set); in reshape_request()
4798 read_bio->bi_iter.bi_sector = (r10_bio->devs[r10_bio->read_slot].addr in reshape_request()
4799 + rdev->data_offset); in reshape_request()
4800 read_bio->bi_private = r10_bio; in reshape_request()
4801 read_bio->bi_end_io = end_reshape_read; in reshape_request()
4802 r10_bio->master_bio = read_bio; in reshape_request()
4803 r10_bio->read_slot = r10_bio->devs[r10_bio->read_slot].devnum; in reshape_request()
4809 if (mddev_is_clustered(mddev) && conf->cluster_sync_high <= sector_nr) { in reshape_request()
4813 conf->cluster_sync_low = sector_nr; in reshape_request()
4814 conf->cluster_sync_high = sector_nr + CLUSTER_RESYNC_WINDOW_SECTORS; in reshape_request()
4815 sb = page_address(rdev->sb_page); in reshape_request()
4817 sb_reshape_pos = le64_to_cpu(sb->reshape_position); in reshape_request()
4821 * update cluster_sync_low until it has finished reshape. in reshape_request()
4823 if (sb_reshape_pos < conf->cluster_sync_low) in reshape_request()
4824 conf->cluster_sync_low = sb_reshape_pos; in reshape_request()
4827 md_cluster_ops->resync_info_update(mddev, conf->cluster_sync_low, in reshape_request()
4828 conf->cluster_sync_high); in reshape_request()
4832 __raid10_find_phys(&conf->geo, r10_bio); in reshape_request()
4835 read_bio->bi_next = NULL; in reshape_request()
4837 for (s = 0; s < conf->copies*2; s++) { in reshape_request()
4839 int d = r10_bio->devs[s/2].devnum; in reshape_request()
4842 rdev2 = conf->mirrors[d].replacement; in reshape_request()
4843 b = r10_bio->devs[s/2].repl_bio; in reshape_request()
4845 rdev2 = conf->mirrors[d].rdev; in reshape_request()
4846 b = r10_bio->devs[s/2].bio; in reshape_request()
4848 if (!rdev2 || test_bit(Faulty, &rdev2->flags)) in reshape_request()
4851 bio_set_dev(b, rdev2->bdev); in reshape_request()
4852 b->bi_iter.bi_sector = r10_bio->devs[s/2].addr + in reshape_request()
4853 rdev2->new_data_offset; in reshape_request()
4854 b->bi_end_io = end_reshape_write; in reshape_request()
4855 b->bi_opf = REQ_OP_WRITE; in reshape_request()
4856 b->bi_next = blist; in reshape_request()
4863 pages = get_resync_pages(r10_bio->devs[0].bio)->pages; in reshape_request()
4866 int len = (max_sectors - s) << 9; in reshape_request()
4869 for (bio = blist; bio ; bio = bio->bi_next) { in reshape_request()
4871 bio->bi_status = BLK_STS_RESOURCE; in reshape_request()
4879 r10_bio->sectors = nr_sectors; in reshape_request()
4882 md_sync_acct_bio(read_bio, r10_bio->sectors); in reshape_request()
4883 atomic_inc(&r10_bio->remaining); in reshape_request()
4884 read_bio->bi_next = NULL; in reshape_request()
4893 * update reshape_progress in reshape_request()
4895 if (mddev->reshape_backwards) in reshape_request()
4896 conf->reshape_progress -= sectors_done; in reshape_request()
4898 conf->reshape_progress += sectors_done; in reshape_request()
4910 * If we got a read error then we do sync 1-page reads from in reshape_request_write()
4911 * elsewhere until we find the data - or give up. in reshape_request_write()
4913 struct r10conf *conf = mddev->private; in reshape_request_write()
4916 if (!test_bit(R10BIO_Uptodate, &r10_bio->state)) in reshape_request_write()
4919 md_done_sync(mddev, r10_bio->sectors, 0); in reshape_request_write()
4926 atomic_set(&r10_bio->remaining, 1); in reshape_request_write()
4927 for (s = 0; s < conf->copies*2; s++) { in reshape_request_write()
4929 int d = r10_bio->devs[s/2].devnum; in reshape_request_write()
4932 rdev = conf->mirrors[d].replacement; in reshape_request_write()
4933 b = r10_bio->devs[s/2].repl_bio; in reshape_request_write()
4935 rdev = conf->mirrors[d].rdev; in reshape_request_write()
4936 b = r10_bio->devs[s/2].bio; in reshape_request_write()
4938 if (!rdev || test_bit(Faulty, &rdev->flags)) in reshape_request_write()
4941 atomic_inc(&rdev->nr_pending); in reshape_request_write()
4942 md_sync_acct_bio(b, r10_bio->sectors); in reshape_request_write()
4943 atomic_inc(&r10_bio->remaining); in reshape_request_write()
4944 b->bi_next = NULL; in reshape_request_write()
4952 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) in end_reshape()
4955 spin_lock_irq(&conf->device_lock); in end_reshape()
4956 conf->prev = conf->geo; in end_reshape()
4957 md_finish_reshape(conf->mddev); in end_reshape()
4959 conf->reshape_progress = MaxSector; in end_reshape()
4960 conf->reshape_safe = MaxSector; in end_reshape()
4961 spin_unlock_irq(&conf->device_lock); in end_reshape()
4963 mddev_update_io_opt(conf->mddev, raid10_nr_stripes(conf)); in end_reshape()
4964 conf->fullsync = 0; in end_reshape()
4969 struct r10conf *conf = mddev->private; in raid10_update_reshape_pos()
4972 md_cluster_ops->resync_info_get(mddev, &lo, &hi); in raid10_update_reshape_pos()
4973 if (((mddev->reshape_position <= hi) && (mddev->reshape_position >= lo)) in raid10_update_reshape_pos()
4974 || mddev->reshape_position == MaxSector) in raid10_update_reshape_pos()
4975 conf->reshape_progress = mddev->reshape_position; in raid10_update_reshape_pos()
4984 int sectors = r10_bio->sectors; in handle_reshape_read_error()
4985 struct r10conf *conf = mddev->private; in handle_reshape_read_error()
4991 r10b = kmalloc(struct_size(r10b, devs, conf->copies), GFP_NOIO); in handle_reshape_read_error()
4993 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in handle_reshape_read_error()
4994 return -ENOMEM; in handle_reshape_read_error()
4998 pages = get_resync_pages(r10_bio->devs[0].bio)->pages; in handle_reshape_read_error()
5000 r10b->sector = r10_bio->sector; in handle_reshape_read_error()
5001 __raid10_find_phys(&conf->prev, r10b); in handle_reshape_read_error()
5012 int d = r10b->devs[slot].devnum; in handle_reshape_read_error()
5013 struct md_rdev *rdev = conf->mirrors[d].rdev; in handle_reshape_read_error()
5016 test_bit(Faulty, &rdev->flags) || in handle_reshape_read_error()
5017 !test_bit(In_sync, &rdev->flags)) in handle_reshape_read_error()
5020 addr = r10b->devs[slot].addr + idx * PAGE_SIZE; in handle_reshape_read_error()
5021 atomic_inc(&rdev->nr_pending); in handle_reshape_read_error()
5032 if (slot >= conf->copies) in handle_reshape_read_error()
5040 &mddev->recovery); in handle_reshape_read_error()
5042 return -EIO; in handle_reshape_read_error()
5044 sectors -= s; in handle_reshape_read_error()
5054 struct mddev *mddev = r10_bio->mddev; in end_reshape_write()
5055 struct r10conf *conf = mddev->private; in end_reshape_write()
5062 rdev = repl ? conf->mirrors[d].replacement : in end_reshape_write()
5063 conf->mirrors[d].rdev; in end_reshape_write()
5065 if (bio->bi_status) { in end_reshape_write()
5076 if (!atomic_dec_and_test(&r10_bio->remaining)) in end_reshape_request()
5078 md_done_sync(r10_bio->mddev, r10_bio->sectors, 1); in end_reshape_request()
5079 bio_put(r10_bio->master_bio); in end_reshape_request()
5085 struct r10conf *conf = mddev->private; in raid10_finish_reshape()
5087 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) in raid10_finish_reshape()
5090 if (mddev->delta_disks > 0) { in raid10_finish_reshape()
5091 if (mddev->recovery_cp > mddev->resync_max_sectors) { in raid10_finish_reshape()
5092 mddev->recovery_cp = mddev->resync_max_sectors; in raid10_finish_reshape()
5093 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid10_finish_reshape()
5095 mddev->resync_max_sectors = mddev->array_sectors; in raid10_finish_reshape()
5098 for (d = conf->geo.raid_disks ; in raid10_finish_reshape()
5099 d < conf->geo.raid_disks - mddev->delta_disks; in raid10_finish_reshape()
5101 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid10_finish_reshape()
5103 clear_bit(In_sync, &rdev->flags); in raid10_finish_reshape()
5104 rdev = conf->mirrors[d].replacement; in raid10_finish_reshape()
5106 clear_bit(In_sync, &rdev->flags); in raid10_finish_reshape()
5109 mddev->layout = mddev->new_layout; in raid10_finish_reshape()
5110 mddev->chunk_sectors = 1 << conf->geo.chunk_shift; in raid10_finish_reshape()
5111 mddev->reshape_position = MaxSector; in raid10_finish_reshape()
5112 mddev->delta_disks = 0; in raid10_finish_reshape()
5113 mddev->reshape_backwards = 0; in raid10_finish_reshape()
5154 MODULE_ALIAS("md-personality-9"); /* RAID10 */
5155 MODULE_ALIAS("md-raid10");
5156 MODULE_ALIAS("md-level-10");