Lines Matching +full:data +full:- +full:mirror
1 // SPDX-License-Identifier: GPL-2.0-or-later
9 * RAID-1 management functions.
11 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
19 * - bitmap marked during normal i/o
20 * - bitmap used to skip nondirty blocks during sync
22 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
23 * - persistent bitmap code
38 #include "md-bitmap.h"
39 #include "md-cluster.h"
52 #include "raid1-10.c"
54 #define START(node) ((node)->start)
55 #define LAST(node) ((node)->last)
64 sector_t lo = r1_bio->sector; in check_and_add_serial()
65 sector_t hi = lo + r1_bio->sectors; in check_and_add_serial()
66 struct serial_in_rdev *serial = &rdev->serial[idx]; in check_and_add_serial()
68 spin_lock_irqsave(&serial->serial_lock, flags); in check_and_add_serial()
70 if (raid1_rb_iter_first(&serial->serial_rb, lo, hi)) in check_and_add_serial()
71 ret = -EBUSY; in check_and_add_serial()
73 si->start = lo; in check_and_add_serial()
74 si->last = hi; in check_and_add_serial()
75 raid1_rb_insert(si, &serial->serial_rb); in check_and_add_serial()
77 spin_unlock_irqrestore(&serial->serial_lock, flags); in check_and_add_serial()
84 struct mddev *mddev = rdev->mddev; in wait_for_serialization()
86 int idx = sector_to_idx(r1_bio->sector); in wait_for_serialization()
87 struct serial_in_rdev *serial = &rdev->serial[idx]; in wait_for_serialization()
89 if (WARN_ON(!mddev->serial_info_pool)) in wait_for_serialization()
91 si = mempool_alloc(mddev->serial_info_pool, GFP_NOIO); in wait_for_serialization()
92 wait_event(serial->serial_io_wait, in wait_for_serialization()
101 struct mddev *mddev = rdev->mddev; in remove_serial()
103 struct serial_in_rdev *serial = &rdev->serial[idx]; in remove_serial()
105 spin_lock_irqsave(&serial->serial_lock, flags); in remove_serial()
106 for (si = raid1_rb_iter_first(&serial->serial_rb, lo, hi); in remove_serial()
108 if (si->start == lo && si->last == hi) { in remove_serial()
109 raid1_rb_remove(si, &serial->serial_rb); in remove_serial()
110 mempool_free(si, mddev->serial_info_pool); in remove_serial()
117 spin_unlock_irqrestore(&serial->serial_lock, flags); in remove_serial()
118 wake_up(&serial->serial_io_wait); in remove_serial()
122 * for resync bio, r1bio pointer can be retrieved from the per-bio
127 return get_resync_pages(bio)->raid_bio; in get_resync_r1bio()
130 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) in r1bio_pool_alloc() argument
132 struct pool_info *pi = data; in r1bio_pool_alloc()
133 int size = offsetof(struct r1bio, bios[pi->raid_disks]); in r1bio_pool_alloc()
146 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) in r1buf_pool_alloc() argument
148 struct pool_info *pi = data; in r1buf_pool_alloc()
159 rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages), in r1buf_pool_alloc()
165 * Allocate bios : 1 for reading, n-1 for writing in r1buf_pool_alloc()
167 for (j = pi->raid_disks ; j-- ; ) { in r1buf_pool_alloc()
171 bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); in r1buf_pool_alloc()
172 r1_bio->bios[j] = bio; in r1buf_pool_alloc()
175 * Allocate RESYNC_PAGES data pages and attach them to in r1buf_pool_alloc()
177 * If this is a user-requested check/repair, allocate in r1buf_pool_alloc()
180 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) in r1buf_pool_alloc()
181 need_pages = pi->raid_disks; in r1buf_pool_alloc()
184 for (j = 0; j < pi->raid_disks; j++) { in r1buf_pool_alloc()
187 bio = r1_bio->bios[j]; in r1buf_pool_alloc()
197 rp->raid_bio = r1_bio; in r1buf_pool_alloc()
198 bio->bi_private = rp; in r1buf_pool_alloc()
201 r1_bio->master_bio = NULL; in r1buf_pool_alloc()
206 while (--j >= 0) in r1buf_pool_alloc()
210 while (++j < pi->raid_disks) { in r1buf_pool_alloc()
211 bio_uninit(r1_bio->bios[j]); in r1buf_pool_alloc()
212 kfree(r1_bio->bios[j]); in r1buf_pool_alloc()
217 rbio_pool_free(r1_bio, data); in r1buf_pool_alloc()
221 static void r1buf_pool_free(void *__r1_bio, void *data) in r1buf_pool_free() argument
223 struct pool_info *pi = data; in r1buf_pool_free()
228 for (i = pi->raid_disks; i--; ) { in r1buf_pool_free()
229 rp = get_resync_pages(r1bio->bios[i]); in r1buf_pool_free()
231 bio_uninit(r1bio->bios[i]); in r1buf_pool_free()
232 kfree(r1bio->bios[i]); in r1buf_pool_free()
238 rbio_pool_free(r1bio, data); in r1buf_pool_free()
245 for (i = 0; i < conf->raid_disks * 2; i++) { in put_all_bios()
246 struct bio **bio = r1_bio->bios + i; in put_all_bios()
255 struct r1conf *conf = r1_bio->mddev->private; in free_r1bio()
258 mempool_free(r1_bio, &conf->r1bio_pool); in free_r1bio()
263 struct r1conf *conf = r1_bio->mddev->private; in put_buf()
264 sector_t sect = r1_bio->sector; in put_buf()
267 for (i = 0; i < conf->raid_disks * 2; i++) { in put_buf()
268 struct bio *bio = r1_bio->bios[i]; in put_buf()
269 if (bio->bi_end_io) in put_buf()
270 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); in put_buf()
273 mempool_free(r1_bio, &conf->r1buf_pool); in put_buf()
281 struct mddev *mddev = r1_bio->mddev; in reschedule_retry()
282 struct r1conf *conf = mddev->private; in reschedule_retry()
285 idx = sector_to_idx(r1_bio->sector); in reschedule_retry()
286 spin_lock_irqsave(&conf->device_lock, flags); in reschedule_retry()
287 list_add(&r1_bio->retry_list, &conf->retry_list); in reschedule_retry()
288 atomic_inc(&conf->nr_queued[idx]); in reschedule_retry()
289 spin_unlock_irqrestore(&conf->device_lock, flags); in reschedule_retry()
291 wake_up(&conf->wait_barrier); in reschedule_retry()
292 md_wakeup_thread(mddev->thread); in reschedule_retry()
302 struct bio *bio = r1_bio->master_bio; in call_bio_endio()
304 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in call_bio_endio()
305 bio->bi_status = BLK_STS_IOERR; in call_bio_endio()
312 struct bio *bio = r1_bio->master_bio; in raid_end_bio_io()
313 struct r1conf *conf = r1_bio->mddev->private; in raid_end_bio_io()
314 sector_t sector = r1_bio->sector; in raid_end_bio_io()
317 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid_end_bio_io()
318 pr_debug("raid1: sync end %s on sectors %llu-%llu\n", in raid_end_bio_io()
320 (unsigned long long) bio->bi_iter.bi_sector, in raid_end_bio_io()
321 (unsigned long long) bio_end_sector(bio) - 1); in raid_end_bio_io()
329 * to go idle. All I/Os, even write-behind writes, are done. in raid_end_bio_io()
339 struct r1conf *conf = r1_bio->mddev->private; in update_head_pos()
341 conf->mirrors[disk].head_position = in update_head_pos()
342 r1_bio->sector + (r1_bio->sectors); in update_head_pos()
350 int mirror; in find_bio_disk() local
351 struct r1conf *conf = r1_bio->mddev->private; in find_bio_disk()
352 int raid_disks = conf->raid_disks; in find_bio_disk()
354 for (mirror = 0; mirror < raid_disks * 2; mirror++) in find_bio_disk()
355 if (r1_bio->bios[mirror] == bio) in find_bio_disk()
358 BUG_ON(mirror == raid_disks * 2); in find_bio_disk()
359 update_head_pos(mirror, r1_bio); in find_bio_disk()
361 return mirror; in find_bio_disk()
366 int uptodate = !bio->bi_status; in raid1_end_read_request()
367 struct r1bio *r1_bio = bio->bi_private; in raid1_end_read_request()
368 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_read_request()
369 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; in raid1_end_read_request()
372 * this branch is our 'one mirror IO has finished' event handler: in raid1_end_read_request()
374 update_head_pos(r1_bio->read_disk, r1_bio); in raid1_end_read_request()
377 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_read_request()
378 else if (test_bit(FailFast, &rdev->flags) && in raid1_end_read_request()
379 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_end_read_request()
380 /* This was a fail-fast read so we definitely in raid1_end_read_request()
389 spin_lock_irqsave(&conf->device_lock, flags); in raid1_end_read_request()
390 if (r1_bio->mddev->degraded == conf->raid_disks || in raid1_end_read_request()
391 (r1_bio->mddev->degraded == conf->raid_disks-1 && in raid1_end_read_request()
392 test_bit(In_sync, &rdev->flags))) in raid1_end_read_request()
394 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_end_read_request()
399 rdev_dec_pending(rdev, conf->mddev); in raid1_end_read_request()
405 mdname(conf->mddev), in raid1_end_read_request()
406 rdev->bdev, in raid1_end_read_request()
407 (unsigned long long)r1_bio->sector); in raid1_end_read_request()
408 set_bit(R1BIO_ReadError, &r1_bio->state); in raid1_end_read_request()
416 struct mddev *mddev = r1_bio->mddev; in close_write()
419 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in close_write()
420 bio_free_pages(r1_bio->behind_master_bio); in close_write()
421 bio_put(r1_bio->behind_master_bio); in close_write()
422 r1_bio->behind_master_bio = NULL; in close_write()
425 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) in close_write()
426 mddev->bitmap_ops->end_behind_write(mddev); in close_write()
432 if (!atomic_dec_and_test(&r1_bio->remaining)) in r1_bio_write_done()
435 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in r1_bio_write_done()
439 if (test_bit(R1BIO_MadeGood, &r1_bio->state)) in r1_bio_write_done()
448 struct r1bio *r1_bio = bio->bi_private; in raid1_end_write_request()
449 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); in raid1_end_write_request()
450 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_write_request()
452 int mirror = find_bio_disk(r1_bio, bio); in raid1_end_write_request() local
453 struct md_rdev *rdev = conf->mirrors[mirror].rdev; in raid1_end_write_request()
455 sector_t lo = r1_bio->sector; in raid1_end_write_request()
456 sector_t hi = r1_bio->sector + r1_bio->sectors; in raid1_end_write_request()
458 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; in raid1_end_write_request()
461 * 'one mirror IO has finished' event handler: in raid1_end_write_request()
463 if (bio->bi_status && !discard_error) { in raid1_end_write_request()
464 set_bit(WriteErrorSeen, &rdev->flags); in raid1_end_write_request()
465 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in raid1_end_write_request()
467 conf->mddev->recovery); in raid1_end_write_request()
469 if (test_bit(FailFast, &rdev->flags) && in raid1_end_write_request()
470 (bio->bi_opf & MD_FAILFAST) && in raid1_end_write_request()
472 !test_bit(WriteMostly, &rdev->flags)) { in raid1_end_write_request()
473 md_error(r1_bio->mddev, rdev); in raid1_end_write_request()
480 if (!test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
481 set_bit(R1BIO_WriteError, &r1_bio->state); in raid1_end_write_request()
484 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
495 * to user-side. So if something waits for IO, then it in raid1_end_write_request()
498 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
503 * such device for properly reading the data back (we could in raid1_end_write_request()
505 * before rdev->recovery_offset, but for simplicity we don't in raid1_end_write_request()
508 if (test_bit(In_sync, &rdev->flags) && in raid1_end_write_request()
509 !test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
510 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_write_request()
513 if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) && in raid1_end_write_request()
515 r1_bio->bios[mirror] = IO_MADE_GOOD; in raid1_end_write_request()
516 set_bit(R1BIO_MadeGood, &r1_bio->state); in raid1_end_write_request()
521 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_end_write_request()
523 if (test_bit(WriteMostly, &rdev->flags)) in raid1_end_write_request()
524 atomic_dec(&r1_bio->behind_remaining); in raid1_end_write_request()
528 * has safely reached all non-writemostly in raid1_end_write_request()
530 * gets done only once -- we don't ever want to return in raid1_end_write_request()
531 * -EIO here, instead we'll wait in raid1_end_write_request()
533 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && in raid1_end_write_request()
534 test_bit(R1BIO_Uptodate, &r1_bio->state)) { in raid1_end_write_request()
536 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid1_end_write_request()
537 struct bio *mbio = r1_bio->master_bio; in raid1_end_write_request()
539 " %llu-%llu\n", in raid1_end_write_request()
540 (unsigned long long) mbio->bi_iter.bi_sector, in raid1_end_write_request()
541 (unsigned long long) bio_end_sector(mbio) - 1); in raid1_end_write_request()
545 } else if (rdev->mddev->serialize_policy) in raid1_end_write_request()
547 if (r1_bio->bios[mirror] == NULL) in raid1_end_write_request()
548 rdev_dec_pending(rdev, conf->mddev); in raid1_end_write_request()
570 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - in align_to_barrier_unit_end()
582 struct raid1_info *info = &conf->mirrors[disk]; in update_read_sectors()
584 atomic_inc(&info->rdev->nr_pending); in update_read_sectors()
585 if (info->next_seq_sect != this_sector) in update_read_sectors()
586 info->seq_start = this_sector; in update_read_sectors()
587 info->next_seq_sect = this_sector + len; in update_read_sectors()
593 sector_t this_sector = r1_bio->sector; in choose_first_rdev()
594 int len = r1_bio->sectors; in choose_first_rdev()
597 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in choose_first_rdev()
601 if (r1_bio->bios[disk] == IO_BLOCKED) in choose_first_rdev()
604 rdev = conf->mirrors[disk].rdev; in choose_first_rdev()
605 if (!rdev || test_bit(Faulty, &rdev->flags)) in choose_first_rdev()
617 return -1; in choose_first_rdev()
622 return !test_bit(In_sync, &rdev->flags) && in rdev_in_recovery()
623 rdev->recovery_offset < r1_bio->sector + r1_bio->sectors; in rdev_in_recovery()
629 sector_t this_sector = r1_bio->sector; in choose_bb_rdev()
630 int best_disk = -1; in choose_bb_rdev()
634 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in choose_bb_rdev()
639 if (r1_bio->bios[disk] == IO_BLOCKED) in choose_bb_rdev()
642 rdev = conf->mirrors[disk].rdev; in choose_bb_rdev()
643 if (!rdev || test_bit(Faulty, &rdev->flags) || in choose_bb_rdev()
645 test_bit(WriteMostly, &rdev->flags)) in choose_bb_rdev()
649 len = r1_bio->sectors; in choose_bb_rdev()
657 if (best_disk != -1) { in choose_bb_rdev()
668 sector_t this_sector = r1_bio->sector; in choose_slow_rdev()
669 int bb_disk = -1; in choose_slow_rdev()
673 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in choose_slow_rdev()
678 if (r1_bio->bios[disk] == IO_BLOCKED) in choose_slow_rdev()
681 rdev = conf->mirrors[disk].rdev; in choose_slow_rdev()
682 if (!rdev || test_bit(Faulty, &rdev->flags) || in choose_slow_rdev()
683 !test_bit(WriteMostly, &rdev->flags) || in choose_slow_rdev()
688 len = r1_bio->sectors; in choose_slow_rdev()
690 if (read_len == r1_bio->sectors) { in choose_slow_rdev()
706 if (bb_disk != -1) { in choose_slow_rdev()
717 return conf->mirrors[disk].next_seq_sect == r1_bio->sector || in is_sequential()
718 conf->mirrors[disk].head_position == r1_bio->sector; in is_sequential()
727 struct raid1_info *mirror = &conf->mirrors[disk]; in should_choose_next() local
730 if (!test_bit(Nonrot, &mirror->rdev->flags)) in should_choose_next()
733 opt_iosize = bdev_io_opt(mirror->rdev->bdev) >> 9; in should_choose_next()
734 return opt_iosize > 0 && mirror->seq_start != MaxSector && in should_choose_next()
735 mirror->next_seq_sect > opt_iosize && in should_choose_next()
736 mirror->next_seq_sect - opt_iosize >= mirror->seq_start; in should_choose_next()
741 if (!rdev || test_bit(Faulty, &rdev->flags)) in rdev_readable()
748 if (test_bit(WriteMostly, &rdev->flags)) in rdev_readable()
752 if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors)) in rdev_readable()
771 .closest_dist_disk = -1, in choose_best_rdev()
773 .min_pending_disk = -1, in choose_best_rdev()
775 .sequential_disk = -1, in choose_best_rdev()
778 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in choose_best_rdev()
783 if (r1_bio->bios[disk] == IO_BLOCKED) in choose_best_rdev()
786 rdev = conf->mirrors[disk].rdev; in choose_best_rdev()
792 set_bit(R1BIO_FailFast, &r1_bio->state); in choose_best_rdev()
794 pending = atomic_read(&rdev->nr_pending); in choose_best_rdev()
795 dist = abs(r1_bio->sector - conf->mirrors[disk].head_position); in choose_best_rdev()
829 if (ctl.sequential_disk != -1 && ctl.min_pending != 0) in choose_best_rdev()
834 * non-rotational, choose the disk with less pending request even the in choose_best_rdev()
836 * mixed ratation/non-rotational disks depending on workload. in choose_best_rdev()
838 if (ctl.min_pending_disk != -1 && in choose_best_rdev()
839 (READ_ONCE(conf->nonrot_disks) || ctl.min_pending == 0)) in choose_best_rdev()
868 clear_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
870 if (raid1_should_read_first(conf->mddev, r1_bio->sector, in read_balance()
871 r1_bio->sectors)) in read_balance()
876 *max_sectors = r1_bio->sectors; in read_balance()
877 update_read_sectors(conf, disk, r1_bio->sector, in read_balance()
878 r1_bio->sectors); in read_balance()
896 if (wq_has_sleeper(&conf->wait_barrier)) in wake_up_barrier()
897 wake_up(&conf->wait_barrier); in wake_up_barrier()
903 raid1_prepare_flush_writes(conf->mddev); in flush_bio_list()
907 struct bio *next = bio->bi_next; in flush_bio_list()
920 spin_lock_irq(&conf->device_lock); in flush_pending_writes()
922 if (conf->pending_bio_list.head) { in flush_pending_writes()
926 bio = bio_list_get(&conf->pending_bio_list); in flush_pending_writes()
927 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
931 * current->state might be TASK_UNINTERRUPTIBLE which will in flush_pending_writes()
935 * is a false-positive. Silence the warning by resetting in flush_pending_writes()
943 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
955 * We choose only to raise the barrier if no-one is waiting for the
967 * If resync/recovery is interrupted, returns -EINTR;
974 spin_lock_irq(&conf->resync_lock); in raise_barrier()
977 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
978 !atomic_read(&conf->nr_waiting[idx]), in raise_barrier()
979 conf->resync_lock); in raise_barrier()
982 atomic_inc(&conf->barrier[idx]); in raise_barrier()
984 * In raise_barrier() we firstly increase conf->barrier[idx] then in raise_barrier()
985 * check conf->nr_pending[idx]. In _wait_barrier() we firstly in raise_barrier()
986 * increase conf->nr_pending[idx] then check conf->barrier[idx]. in raise_barrier()
987 * A memory barrier here to make sure conf->nr_pending[idx] won't in raise_barrier()
988 * be fetched before conf->barrier[idx] is increased. Otherwise in raise_barrier()
995 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O in raise_barrier()
997 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches in raise_barrier()
1000 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
1001 (!conf->array_frozen && in raise_barrier()
1002 !atomic_read(&conf->nr_pending[idx]) && in raise_barrier()
1003 atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || in raise_barrier()
1004 test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), in raise_barrier()
1005 conf->resync_lock); in raise_barrier()
1007 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { in raise_barrier()
1008 atomic_dec(&conf->barrier[idx]); in raise_barrier()
1009 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
1010 wake_up(&conf->wait_barrier); in raise_barrier()
1011 return -EINTR; in raise_barrier()
1014 atomic_inc(&conf->nr_sync_pending); in raise_barrier()
1015 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
1024 BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); in lower_barrier()
1026 atomic_dec(&conf->barrier[idx]); in lower_barrier()
1027 atomic_dec(&conf->nr_sync_pending); in lower_barrier()
1028 wake_up(&conf->wait_barrier); in lower_barrier()
1036 * We need to increase conf->nr_pending[idx] very early here, in _wait_barrier()
1038 * conf->nr_pending[idx] to be 0. Then we can avoid holding in _wait_barrier()
1039 * conf->resync_lock when there is no barrier raised in same in _wait_barrier()
1043 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
1045 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then in _wait_barrier()
1046 * check conf->barrier[idx]. In raise_barrier() we firstly increase in _wait_barrier()
1047 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory in _wait_barrier()
1048 * barrier is necessary here to make sure conf->barrier[idx] won't be in _wait_barrier()
1049 * fetched before conf->nr_pending[idx] is increased. Otherwise there in _wait_barrier()
1056 * here. If during we check conf->barrier[idx], the array is in _wait_barrier()
1057 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is in _wait_barrier()
1063 if (!READ_ONCE(conf->array_frozen) && in _wait_barrier()
1064 !atomic_read(&conf->barrier[idx])) in _wait_barrier()
1068 * After holding conf->resync_lock, conf->nr_pending[idx] in _wait_barrier()
1071 * raise_barrer() might be waiting for conf->nr_pending[idx] in _wait_barrier()
1074 spin_lock_irq(&conf->resync_lock); in _wait_barrier()
1075 atomic_inc(&conf->nr_waiting[idx]); in _wait_barrier()
1076 atomic_dec(&conf->nr_pending[idx]); in _wait_barrier()
1088 wait_event_lock_irq(conf->wait_barrier, in _wait_barrier()
1089 !conf->array_frozen && in _wait_barrier()
1090 !atomic_read(&conf->barrier[idx]), in _wait_barrier()
1091 conf->resync_lock); in _wait_barrier()
1092 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
1095 atomic_dec(&conf->nr_waiting[idx]); in _wait_barrier()
1096 spin_unlock_irq(&conf->resync_lock); in _wait_barrier()
1110 * conf->barrier[idx] here, memory barrier is unnecessary as well. in wait_read_barrier()
1112 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1114 if (!READ_ONCE(conf->array_frozen)) in wait_read_barrier()
1117 spin_lock_irq(&conf->resync_lock); in wait_read_barrier()
1118 atomic_inc(&conf->nr_waiting[idx]); in wait_read_barrier()
1119 atomic_dec(&conf->nr_pending[idx]); in wait_read_barrier()
1132 wait_event_lock_irq(conf->wait_barrier, in wait_read_barrier()
1133 !conf->array_frozen, in wait_read_barrier()
1134 conf->resync_lock); in wait_read_barrier()
1135 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1138 atomic_dec(&conf->nr_waiting[idx]); in wait_read_barrier()
1139 spin_unlock_irq(&conf->resync_lock); in wait_read_barrier()
1152 atomic_dec(&conf->nr_pending[idx]); in _allow_barrier()
1163 /* conf->resync_lock should be held */
1168 ret = atomic_read(&conf->nr_sync_pending); in get_unqueued_pending()
1170 ret += atomic_read(&conf->nr_pending[idx]) - in get_unqueued_pending()
1171 atomic_read(&conf->nr_queued[idx]); in get_unqueued_pending()
1189 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the in freeze_array()
1191 * normal I/O are queued, sum of all conf->nr_pending[] will match sum in freeze_array()
1192 * of all conf->nr_queued[]. But normal I/O failure is an exception, in freeze_array()
1201 spin_lock_irq(&conf->resync_lock); in freeze_array()
1202 conf->array_frozen = 1; in freeze_array()
1203 mddev_add_trace_msg(conf->mddev, "raid1 wait freeze"); in freeze_array()
1205 conf->wait_barrier, in freeze_array()
1207 conf->resync_lock, in freeze_array()
1209 spin_unlock_irq(&conf->resync_lock); in freeze_array()
1214 spin_lock_irq(&conf->resync_lock); in unfreeze_array()
1215 conf->array_frozen = 0; in unfreeze_array()
1216 spin_unlock_irq(&conf->resync_lock); in unfreeze_array()
1217 wake_up(&conf->wait_barrier); in unfreeze_array()
1223 int size = bio->bi_iter.bi_size; in alloc_behind_master_bio()
1224 unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; in alloc_behind_master_bio()
1229 &r1_bio->mddev->bio_set); in alloc_behind_master_bio()
1233 behind_bio->bi_iter.bi_size = size; in alloc_behind_master_bio()
1250 size -= len; in alloc_behind_master_bio()
1256 r1_bio->behind_master_bio = behind_bio; in alloc_behind_master_bio()
1257 set_bit(R1BIO_BehindIO, &r1_bio->state); in alloc_behind_master_bio()
1263 bio->bi_iter.bi_size); in alloc_behind_master_bio()
1272 struct mddev *mddev = plug->cb.data; in raid1_unplug()
1273 struct r1conf *conf = mddev->private; in raid1_unplug()
1277 spin_lock_irq(&conf->device_lock); in raid1_unplug()
1278 bio_list_merge(&conf->pending_bio_list, &plug->pending); in raid1_unplug()
1279 spin_unlock_irq(&conf->device_lock); in raid1_unplug()
1281 md_wakeup_thread(mddev->thread); in raid1_unplug()
1286 /* we aren't scheduling, so we can do the write-out directly. */ in raid1_unplug()
1287 bio = bio_list_get(&plug->pending); in raid1_unplug()
1294 r1_bio->master_bio = bio; in init_r1bio()
1295 r1_bio->sectors = bio_sectors(bio); in init_r1bio()
1296 r1_bio->state = 0; in init_r1bio()
1297 r1_bio->mddev = mddev; in init_r1bio()
1298 r1_bio->sector = bio->bi_iter.bi_sector; in init_r1bio()
1304 struct r1conf *conf = mddev->private; in alloc_r1bio()
1307 r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); in alloc_r1bio()
1309 memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); in alloc_r1bio()
1317 struct r1conf *conf = mddev->private; in raid1_read_request()
1318 struct raid1_info *mirror; in raid1_read_request() local
1335 if (!wait_read_barrier(conf, bio->bi_iter.bi_sector, in raid1_read_request()
1336 bio->bi_opf & REQ_NOWAIT)) { in raid1_read_request()
1345 r1_bio->sectors = max_read_sectors; in raid1_read_request()
1348 * make_request() can abort the operation when read-ahead is being in raid1_read_request()
1357 conf->mirrors[r1_bio->read_disk].rdev->bdev, in raid1_read_request()
1358 r1_bio->sector); in raid1_read_request()
1362 mirror = conf->mirrors + rdisk; in raid1_read_request()
1365 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n", in raid1_read_request()
1367 (unsigned long long)r1_bio->sector, in raid1_read_request()
1368 mirror->rdev->bdev); in raid1_read_request()
1370 if (test_bit(WriteMostly, &mirror->rdev->flags)) { in raid1_read_request()
1372 * Reading from a write-mostly device must take care not to in raid1_read_request()
1373 * over-take any writes that are 'behind' in raid1_read_request()
1376 mddev->bitmap_ops->wait_behind_writes(mddev); in raid1_read_request()
1381 gfp, &conf->bio_split); in raid1_read_request()
1390 r1_bio->master_bio = bio; in raid1_read_request()
1391 r1_bio->sectors = max_sectors; in raid1_read_request()
1394 r1_bio->read_disk = rdisk; in raid1_read_request()
1397 r1_bio->master_bio = bio; in raid1_read_request()
1399 read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp, in raid1_read_request()
1400 &mddev->bio_set); in raid1_read_request()
1402 r1_bio->bios[rdisk] = read_bio; in raid1_read_request()
1404 read_bio->bi_iter.bi_sector = r1_bio->sector + in raid1_read_request()
1405 mirror->rdev->data_offset; in raid1_read_request()
1406 read_bio->bi_end_io = raid1_end_read_request; in raid1_read_request()
1407 if (test_bit(FailFast, &mirror->rdev->flags) && in raid1_read_request()
1408 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_read_request()
1409 read_bio->bi_opf |= MD_FAILFAST; in raid1_read_request()
1410 read_bio->bi_private = r1_bio; in raid1_read_request()
1411 mddev_trace_remap(mddev, read_bio, r1_bio->sector); in raid1_read_request()
1416 atomic_dec(&mirror->rdev->nr_pending); in raid1_read_request()
1417 bio->bi_status = errno_to_blk_status(error); in raid1_read_request()
1418 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_read_request()
1424 struct r1conf *conf = mddev->private; in wait_blocked_rdev()
1425 int disks = conf->raid_disks * 2; in wait_blocked_rdev()
1430 struct md_rdev *rdev = conf->mirrors[i].rdev; in wait_blocked_rdev()
1436 if (test_bit(WriteErrorSeen, &rdev->flags) && in wait_blocked_rdev()
1437 rdev_has_badblock(rdev, bio->bi_iter.bi_sector, in wait_blocked_rdev()
1439 set_bit(BlockedBadBlocks, &rdev->flags); in wait_blocked_rdev()
1442 if (bio->bi_opf & REQ_NOWAIT) in wait_blocked_rdev()
1445 mddev_add_trace_msg(rdev->mddev, "raid1 wait rdev %d blocked", in wait_blocked_rdev()
1446 rdev->raid_disk); in wait_blocked_rdev()
1447 atomic_inc(&rdev->nr_pending); in wait_blocked_rdev()
1448 md_wait_for_blocked_rdev(rdev, rdev->mddev); in wait_blocked_rdev()
1459 struct r1conf *conf = mddev->private; in raid1_write_request()
1469 mddev->cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1470 bio->bi_iter.bi_sector, bio_end_sector(bio))) { in raid1_write_request()
1473 if (bio->bi_opf & REQ_NOWAIT) { in raid1_write_request()
1478 prepare_to_wait(&conf->wait_barrier, in raid1_write_request()
1480 if (!mddev->cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1481 bio->bi_iter.bi_sector, in raid1_write_request()
1486 finish_wait(&conf->wait_barrier, &w); in raid1_write_request()
1494 if (!wait_barrier(conf, bio->bi_iter.bi_sector, in raid1_write_request()
1495 bio->bi_opf & REQ_NOWAIT)) { in raid1_write_request()
1506 r1_bio->sectors = max_write_sectors; in raid1_write_request()
1519 disks = conf->raid_disks * 2; in raid1_write_request()
1520 max_sectors = r1_bio->sectors; in raid1_write_request()
1522 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_write_request()
1525 * The write-behind io is only attempted on drives marked as in raid1_write_request()
1526 * write-mostly, which means we could allocate write behind in raid1_write_request()
1529 if (!is_discard && rdev && test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1532 r1_bio->bios[i] = NULL; in raid1_write_request()
1533 if (!rdev || test_bit(Faulty, &rdev->flags)) in raid1_write_request()
1536 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1537 if (test_bit(WriteErrorSeen, &rdev->flags)) { in raid1_write_request()
1542 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, in raid1_write_request()
1544 if (is_bad && first_bad <= r1_bio->sector) { in raid1_write_request()
1546 bad_sectors -= (r1_bio->sector - first_bad); in raid1_write_request()
1565 if (bio->bi_opf & REQ_ATOMIC) { in raid1_write_request()
1566 error = -EIO; in raid1_write_request()
1570 good_sectors = first_bad - r1_bio->sector; in raid1_write_request()
1575 r1_bio->bios[i] = bio; in raid1_write_request()
1580 * alloc_behind_master_bio allocates a copy of the data payload a page in raid1_write_request()
1584 if (write_behind && mddev->bitmap) in raid1_write_request()
1589 GFP_NOIO, &conf->bio_split); in raid1_write_request()
1598 r1_bio->master_bio = bio; in raid1_write_request()
1599 r1_bio->sectors = max_sectors; in raid1_write_request()
1603 r1_bio->master_bio = bio; in raid1_write_request()
1604 atomic_set(&r1_bio->remaining, 1); in raid1_write_request()
1605 atomic_set(&r1_bio->behind_remaining, 0); in raid1_write_request()
1611 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_write_request()
1612 if (!r1_bio->bios[i]) in raid1_write_request()
1617 mddev->bitmap_info.max_write_behind; in raid1_write_request()
1625 err = mddev->bitmap_ops->get_stats(mddev->bitmap, &stats); in raid1_write_request()
1630 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) in raid1_write_request()
1631 mddev->bitmap_ops->start_behind_write(mddev); in raid1_write_request()
1635 if (r1_bio->behind_master_bio) { in raid1_write_request()
1636 mbio = bio_alloc_clone(rdev->bdev, in raid1_write_request()
1637 r1_bio->behind_master_bio, in raid1_write_request()
1638 GFP_NOIO, &mddev->bio_set); in raid1_write_request()
1639 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_write_request()
1641 if (test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1642 atomic_inc(&r1_bio->behind_remaining); in raid1_write_request()
1644 mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, in raid1_write_request()
1645 &mddev->bio_set); in raid1_write_request()
1647 if (mddev->serialize_policy) in raid1_write_request()
1651 r1_bio->bios[i] = mbio; in raid1_write_request()
1653 mbio->bi_iter.bi_sector = (r1_bio->sector + rdev->data_offset); in raid1_write_request()
1654 mbio->bi_end_io = raid1_end_write_request; in raid1_write_request()
1655 if (test_bit(FailFast, &rdev->flags) && in raid1_write_request()
1656 !test_bit(WriteMostly, &rdev->flags) && in raid1_write_request()
1657 conf->raid_disks - mddev->degraded > 1) in raid1_write_request()
1658 mbio->bi_opf |= MD_FAILFAST; in raid1_write_request()
1659 mbio->bi_private = r1_bio; in raid1_write_request()
1661 atomic_inc(&r1_bio->remaining); in raid1_write_request()
1662 mddev_trace_remap(mddev, mbio, r1_bio->sector); in raid1_write_request()
1664 mbio->bi_bdev = (void *)rdev; in raid1_write_request()
1666 spin_lock_irqsave(&conf->device_lock, flags); in raid1_write_request()
1667 bio_list_add(&conf->pending_bio_list, mbio); in raid1_write_request()
1668 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_write_request()
1669 md_wakeup_thread(mddev->thread); in raid1_write_request()
1680 if (r1_bio->bios[k]) { in raid1_write_request()
1681 rdev_dec_pending(conf->mirrors[k].rdev, mddev); in raid1_write_request()
1682 r1_bio->bios[k] = NULL; in raid1_write_request()
1686 bio->bi_status = errno_to_blk_status(error); in raid1_write_request()
1687 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_write_request()
1695 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid1_make_request()
1707 bio->bi_iter.bi_sector, bio_sectors(bio)); in raid1_make_request()
1720 struct r1conf *conf = mddev->private; in raid1_status()
1723 lockdep_assert_held(&mddev->lock); in raid1_status()
1725 seq_printf(seq, " [%d/%d] [", conf->raid_disks, in raid1_status()
1726 conf->raid_disks - mddev->degraded); in raid1_status()
1727 for (i = 0; i < conf->raid_disks; i++) { in raid1_status()
1728 struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev); in raid1_status()
1731 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid1_status()
1737 * raid1_error() - RAID1 error handler.
1743 * - &MD_BROKEN flag is set in &mddev->flags.
1744 * - recovery is disabled.
1746 * - recovery is interrupted.
1747 * - &mddev->degraded is bumped.
1750 * &mddev->fail_last_dev is off.
1754 struct r1conf *conf = mddev->private; in raid1_error()
1757 spin_lock_irqsave(&conf->device_lock, flags); in raid1_error()
1759 if (test_bit(In_sync, &rdev->flags) && in raid1_error()
1760 (conf->raid_disks - mddev->degraded) == 1) { in raid1_error()
1761 set_bit(MD_BROKEN, &mddev->flags); in raid1_error()
1763 if (!mddev->fail_last_dev) { in raid1_error()
1764 conf->recovery_disabled = mddev->recovery_disabled; in raid1_error()
1765 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1769 set_bit(Blocked, &rdev->flags); in raid1_error()
1770 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid1_error()
1771 mddev->degraded++; in raid1_error()
1772 set_bit(Faulty, &rdev->flags); in raid1_error()
1773 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1777 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_error()
1778 set_mask_bits(&mddev->sb_flags, 0, in raid1_error()
1782 mdname(mddev), rdev->bdev, in raid1_error()
1783 mdname(mddev), conf->raid_disks - mddev->degraded); in raid1_error()
1795 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, in print_conf()
1796 conf->raid_disks); in print_conf()
1798 lockdep_assert_held(&conf->mddev->reconfig_mutex); in print_conf()
1799 for (i = 0; i < conf->raid_disks; i++) { in print_conf()
1800 struct md_rdev *rdev = conf->mirrors[i].rdev; in print_conf()
1803 i, !test_bit(In_sync, &rdev->flags), in print_conf()
1804 !test_bit(Faulty, &rdev->flags), in print_conf()
1805 rdev->bdev); in print_conf()
1818 mempool_exit(&conf->r1buf_pool); in close_sync()
1824 struct r1conf *conf = mddev->private; in raid1_spare_active()
1833 * which expects 'In_sync' flags and ->degraded to be consistent. in raid1_spare_active()
1835 spin_lock_irqsave(&conf->device_lock, flags); in raid1_spare_active()
1836 for (i = 0; i < conf->raid_disks; i++) { in raid1_spare_active()
1837 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_spare_active()
1838 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; in raid1_spare_active()
1840 && !test_bit(Candidate, &repl->flags) in raid1_spare_active()
1841 && repl->recovery_offset == MaxSector in raid1_spare_active()
1842 && !test_bit(Faulty, &repl->flags) in raid1_spare_active()
1843 && !test_and_set_bit(In_sync, &repl->flags)) { in raid1_spare_active()
1846 !test_and_clear_bit(In_sync, &rdev->flags)) in raid1_spare_active()
1851 * it gets removed and never re-added in raid1_spare_active()
1853 set_bit(Faulty, &rdev->flags); in raid1_spare_active()
1855 rdev->sysfs_state); in raid1_spare_active()
1859 && rdev->recovery_offset == MaxSector in raid1_spare_active()
1860 && !test_bit(Faulty, &rdev->flags) in raid1_spare_active()
1861 && !test_and_set_bit(In_sync, &rdev->flags)) { in raid1_spare_active()
1863 sysfs_notify_dirent_safe(rdev->sysfs_state); in raid1_spare_active()
1866 mddev->degraded -= count; in raid1_spare_active()
1867 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_spare_active()
1876 struct raid1_info *info = conf->mirrors + disk; in raid1_add_conf()
1879 info += conf->raid_disks; in raid1_add_conf()
1881 if (info->rdev) in raid1_add_conf()
1884 if (bdev_nonrot(rdev->bdev)) { in raid1_add_conf()
1885 set_bit(Nonrot, &rdev->flags); in raid1_add_conf()
1886 WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks + 1); in raid1_add_conf()
1889 rdev->raid_disk = disk; in raid1_add_conf()
1890 info->head_position = 0; in raid1_add_conf()
1891 info->seq_start = MaxSector; in raid1_add_conf()
1892 WRITE_ONCE(info->rdev, rdev); in raid1_add_conf()
1899 struct raid1_info *info = conf->mirrors + disk; in raid1_remove_conf()
1900 struct md_rdev *rdev = info->rdev; in raid1_remove_conf()
1902 if (!rdev || test_bit(In_sync, &rdev->flags) || in raid1_remove_conf()
1903 atomic_read(&rdev->nr_pending)) in raid1_remove_conf()
1906 /* Only remove non-faulty devices if recovery is not possible. */ in raid1_remove_conf()
1907 if (!test_bit(Faulty, &rdev->flags) && in raid1_remove_conf()
1908 rdev->mddev->recovery_disabled != conf->recovery_disabled && in raid1_remove_conf()
1909 rdev->mddev->degraded < conf->raid_disks) in raid1_remove_conf()
1912 if (test_and_clear_bit(Nonrot, &rdev->flags)) in raid1_remove_conf()
1913 WRITE_ONCE(conf->nonrot_disks, conf->nonrot_disks - 1); in raid1_remove_conf()
1915 WRITE_ONCE(info->rdev, NULL); in raid1_remove_conf()
1921 struct r1conf *conf = mddev->private; in raid1_add_disk()
1922 int err = -EEXIST; in raid1_add_disk()
1923 int mirror = 0, repl_slot = -1; in raid1_add_disk() local
1926 int last = conf->raid_disks - 1; in raid1_add_disk()
1928 if (mddev->recovery_disabled == conf->recovery_disabled) in raid1_add_disk()
1929 return -EBUSY; in raid1_add_disk()
1931 if (rdev->raid_disk >= 0) in raid1_add_disk()
1932 first = last = rdev->raid_disk; in raid1_add_disk()
1935 * find the disk ... but prefer rdev->saved_raid_disk in raid1_add_disk()
1938 if (rdev->saved_raid_disk >= 0 && in raid1_add_disk()
1939 rdev->saved_raid_disk >= first && in raid1_add_disk()
1940 rdev->saved_raid_disk < conf->raid_disks && in raid1_add_disk()
1941 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid1_add_disk()
1942 first = last = rdev->saved_raid_disk; in raid1_add_disk()
1944 for (mirror = first; mirror <= last; mirror++) { in raid1_add_disk()
1945 p = conf->mirrors + mirror; in raid1_add_disk()
1946 if (!p->rdev) { in raid1_add_disk()
1951 raid1_add_conf(conf, rdev, mirror, false); in raid1_add_disk()
1955 if (rdev->saved_raid_disk < 0) in raid1_add_disk()
1956 conf->fullsync = 1; in raid1_add_disk()
1959 if (test_bit(WantReplacement, &p->rdev->flags) && in raid1_add_disk()
1960 p[conf->raid_disks].rdev == NULL && repl_slot < 0) in raid1_add_disk()
1961 repl_slot = mirror; in raid1_add_disk()
1966 clear_bit(In_sync, &rdev->flags); in raid1_add_disk()
1967 set_bit(Replacement, &rdev->flags); in raid1_add_disk()
1970 conf->fullsync = 1; in raid1_add_disk()
1979 struct r1conf *conf = mddev->private; in raid1_remove_disk()
1981 int number = rdev->raid_disk; in raid1_remove_disk()
1982 struct raid1_info *p = conf->mirrors + number; in raid1_remove_disk()
1984 if (unlikely(number >= conf->raid_disks)) in raid1_remove_disk()
1987 if (rdev != p->rdev) { in raid1_remove_disk()
1988 number += conf->raid_disks; in raid1_remove_disk()
1989 p = conf->mirrors + number; in raid1_remove_disk()
1993 if (rdev == p->rdev) { in raid1_remove_disk()
1995 err = -EBUSY; in raid1_remove_disk()
1999 if (number < conf->raid_disks && in raid1_remove_disk()
2000 conf->mirrors[conf->raid_disks + number].rdev) { in raid1_remove_disk()
2006 conf->mirrors[conf->raid_disks + number].rdev; in raid1_remove_disk()
2008 if (atomic_read(&repl->nr_pending)) { in raid1_remove_disk()
2015 err = -EBUSY; in raid1_remove_disk()
2019 clear_bit(Replacement, &repl->flags); in raid1_remove_disk()
2020 WRITE_ONCE(p->rdev, repl); in raid1_remove_disk()
2021 conf->mirrors[conf->raid_disks + number].rdev = NULL; in raid1_remove_disk()
2025 clear_bit(WantReplacement, &rdev->flags); in raid1_remove_disk()
2038 update_head_pos(r1_bio->read_disk, r1_bio); in end_sync_read()
2041 * we have read a block, now it needs to be re-written, in end_sync_read()
2042 * or re-read if the read failed. in end_sync_read()
2045 if (!bio->bi_status) in end_sync_read()
2046 set_bit(R1BIO_Uptodate, &r1_bio->state); in end_sync_read()
2048 if (atomic_dec_and_test(&r1_bio->remaining)) in end_sync_read()
2055 sector_t s = r1_bio->sector; in abort_sync_write()
2056 long sectors_to_go = r1_bio->sectors; in abort_sync_write()
2060 mddev->bitmap_ops->end_sync(mddev, s, &sync_blocks); in abort_sync_write()
2062 sectors_to_go -= sync_blocks; in abort_sync_write()
2068 if (atomic_dec_and_test(&r1_bio->remaining)) { in put_sync_write_buf()
2069 struct mddev *mddev = r1_bio->mddev; in put_sync_write_buf()
2070 int s = r1_bio->sectors; in put_sync_write_buf()
2072 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in put_sync_write_buf()
2073 test_bit(R1BIO_WriteError, &r1_bio->state)) in put_sync_write_buf()
2084 int uptodate = !bio->bi_status; in end_sync_write()
2086 struct mddev *mddev = r1_bio->mddev; in end_sync_write()
2087 struct r1conf *conf = mddev->private; in end_sync_write()
2088 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; in end_sync_write()
2092 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
2093 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
2095 mddev->recovery); in end_sync_write()
2096 set_bit(R1BIO_WriteError, &r1_bio->state); in end_sync_write()
2097 } else if (rdev_has_badblock(rdev, r1_bio->sector, r1_bio->sectors) && in end_sync_write()
2098 !rdev_has_badblock(conf->mirrors[r1_bio->read_disk].rdev, in end_sync_write()
2099 r1_bio->sector, r1_bio->sectors)) { in end_sync_write()
2100 set_bit(R1BIO_MadeGood, &r1_bio->state); in end_sync_write()
2113 set_bit(WriteErrorSeen, &rdev->flags); in r1_sync_page_io()
2115 &rdev->flags)) in r1_sync_page_io()
2117 rdev->mddev->recovery); in r1_sync_page_io()
2119 /* need to record an error - either for the block or the device */ in r1_sync_page_io()
2121 md_error(rdev->mddev, rdev); in r1_sync_page_io()
2128 * good data, much like with normal read errors. Only in fix_sync_read_error()
2130 * need to re-issue the read request. in fix_sync_read_error()
2138 struct mddev *mddev = r1_bio->mddev; in fix_sync_read_error()
2139 struct r1conf *conf = mddev->private; in fix_sync_read_error()
2140 struct bio *bio = r1_bio->bios[r1_bio->read_disk]; in fix_sync_read_error()
2141 struct page **pages = get_resync_pages(bio)->pages; in fix_sync_read_error()
2142 sector_t sect = r1_bio->sector; in fix_sync_read_error()
2143 int sectors = r1_bio->sectors; in fix_sync_read_error()
2147 rdev = conf->mirrors[r1_bio->read_disk].rdev; in fix_sync_read_error()
2148 if (test_bit(FailFast, &rdev->flags)) { in fix_sync_read_error()
2149 /* Don't try recovering from here - just fail it in fix_sync_read_error()
2152 if (test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2156 bio->bi_end_io = end_sync_write; in fix_sync_read_error()
2161 int d = r1_bio->read_disk; in fix_sync_read_error()
2168 if (r1_bio->bios[d]->bi_end_io == end_sync_read) { in fix_sync_read_error()
2173 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2182 if (d == conf->raid_disks * 2) in fix_sync_read_error()
2184 } while (!success && d != r1_bio->read_disk); in fix_sync_read_error()
2194 mdname(mddev), bio->bi_bdev, in fix_sync_read_error()
2195 (unsigned long long)r1_bio->sector); in fix_sync_read_error()
2196 for (d = 0; d < conf->raid_disks * 2; d++) { in fix_sync_read_error()
2197 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2198 if (!rdev || test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2207 sectors -= s; in fix_sync_read_error()
2214 /* write it back and re-read */ in fix_sync_read_error()
2215 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2217 d = conf->raid_disks * 2; in fix_sync_read_error()
2218 d--; in fix_sync_read_error()
2219 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2221 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2225 r1_bio->bios[d]->bi_end_io = NULL; in fix_sync_read_error()
2230 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2232 d = conf->raid_disks * 2; in fix_sync_read_error()
2233 d--; in fix_sync_read_error()
2234 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2236 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2240 atomic_add(s, &rdev->corrected_errors); in fix_sync_read_error()
2242 sectors -= s; in fix_sync_read_error()
2246 set_bit(R1BIO_Uptodate, &r1_bio->state); in fix_sync_read_error()
2247 bio->bi_status = 0; in fix_sync_read_error()
2258 * attempt an over-write in process_checks()
2260 struct mddev *mddev = r1_bio->mddev; in process_checks()
2261 struct r1conf *conf = mddev->private; in process_checks()
2267 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); in process_checks()
2268 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2270 struct bio *b = r1_bio->bios[i]; in process_checks()
2272 if (b->bi_end_io != end_sync_read) in process_checks()
2275 status = b->bi_status; in process_checks()
2276 bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ); in process_checks()
2277 b->bi_status = status; in process_checks()
2278 b->bi_iter.bi_sector = r1_bio->sector + in process_checks()
2279 conf->mirrors[i].rdev->data_offset; in process_checks()
2280 b->bi_end_io = end_sync_read; in process_checks()
2281 rp->raid_bio = r1_bio; in process_checks()
2282 b->bi_private = rp; in process_checks()
2285 md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); in process_checks()
2287 for (primary = 0; primary < conf->raid_disks * 2; primary++) in process_checks()
2288 if (r1_bio->bios[primary]->bi_end_io == end_sync_read && in process_checks()
2289 !r1_bio->bios[primary]->bi_status) { in process_checks()
2290 r1_bio->bios[primary]->bi_end_io = NULL; in process_checks()
2291 rdev_dec_pending(conf->mirrors[primary].rdev, mddev); in process_checks()
2294 r1_bio->read_disk = primary; in process_checks()
2295 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2297 struct bio *pbio = r1_bio->bios[primary]; in process_checks()
2298 struct bio *sbio = r1_bio->bios[i]; in process_checks()
2299 blk_status_t status = sbio->bi_status; in process_checks()
2300 struct page **ppages = get_resync_pages(pbio)->pages; in process_checks()
2301 struct page **spages = get_resync_pages(sbio)->pages; in process_checks()
2306 if (sbio->bi_end_io != end_sync_read) in process_checks()
2309 sbio->bi_status = 0; in process_checks()
2312 page_len[j++] = bi->bv_len; in process_checks()
2315 for (j = vcnt; j-- ; ) { in process_checks()
2324 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); in process_checks()
2325 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) in process_checks()
2328 sbio->bi_end_io = NULL; in process_checks()
2329 rdev_dec_pending(conf->mirrors[i].rdev, mddev); in process_checks()
2339 struct r1conf *conf = mddev->private; in sync_request_write()
2341 int disks = conf->raid_disks * 2; in sync_request_write()
2344 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) { in sync_request_write()
2346 * ouch - failed to read all of that. in sync_request_write()
2350 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) || in sync_request_write()
2352 conf->recovery_disabled = mddev->recovery_disabled; in sync_request_write()
2353 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in sync_request_write()
2354 md_done_sync(mddev, r1_bio->sectors, 0); in sync_request_write()
2360 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in sync_request_write()
2366 atomic_set(&r1_bio->remaining, 1); in sync_request_write()
2368 wbio = r1_bio->bios[i]; in sync_request_write()
2369 if (wbio->bi_end_io == NULL || in sync_request_write()
2370 (wbio->bi_end_io == end_sync_read && in sync_request_write()
2371 (i == r1_bio->read_disk || in sync_request_write()
2372 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) in sync_request_write()
2374 if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { in sync_request_write()
2379 wbio->bi_opf = REQ_OP_WRITE; in sync_request_write()
2380 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) in sync_request_write()
2381 wbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2383 wbio->bi_end_io = end_sync_write; in sync_request_write()
2384 atomic_inc(&r1_bio->remaining); in sync_request_write()
2385 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); in sync_request_write()
2403 sector_t sect = r1_bio->sector; in fix_read_error()
2404 int sectors = r1_bio->sectors; in fix_read_error()
2405 int read_disk = r1_bio->read_disk; in fix_read_error()
2406 struct mddev *mddev = conf->mddev; in fix_read_error()
2407 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; in fix_read_error()
2410 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; in fix_read_error()
2424 rdev = conf->mirrors[d].rdev; in fix_read_error()
2426 (test_bit(In_sync, &rdev->flags) || in fix_read_error()
2427 (!test_bit(Faulty, &rdev->flags) && in fix_read_error()
2428 rdev->recovery_offset >= sect + s)) && in fix_read_error()
2430 atomic_inc(&rdev->nr_pending); in fix_read_error()
2432 conf->tmppage, REQ_OP_READ, false)) in fix_read_error()
2440 if (d == conf->raid_disks * 2) in fix_read_error()
2445 /* Cannot read from anywhere - mark it bad */ in fix_read_error()
2446 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; in fix_read_error()
2451 /* write it back and re-read */ in fix_read_error()
2455 d = conf->raid_disks * 2; in fix_read_error()
2456 d--; in fix_read_error()
2457 rdev = conf->mirrors[d].rdev; in fix_read_error()
2459 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2460 atomic_inc(&rdev->nr_pending); in fix_read_error()
2462 conf->tmppage, REQ_OP_WRITE); in fix_read_error()
2469 d = conf->raid_disks * 2; in fix_read_error()
2470 d--; in fix_read_error()
2471 rdev = conf->mirrors[d].rdev; in fix_read_error()
2473 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2474 atomic_inc(&rdev->nr_pending); in fix_read_error()
2476 conf->tmppage, REQ_OP_READ)) { in fix_read_error()
2477 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2481 rdev->data_offset), in fix_read_error()
2482 rdev->bdev); in fix_read_error()
2487 sectors -= s; in fix_read_error()
2494 struct mddev *mddev = r1_bio->mddev; in narrow_write_error()
2495 struct r1conf *conf = mddev->private; in narrow_write_error()
2496 struct md_rdev *rdev = conf->mirrors[i].rdev; in narrow_write_error()
2498 /* bio has the data to be written to device 'i' where in narrow_write_error()
2512 int sect_to_write = r1_bio->sectors; in narrow_write_error()
2515 if (rdev->badblocks.shift < 0) in narrow_write_error()
2518 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2519 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2520 sector = r1_bio->sector; in narrow_write_error()
2522 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2523 - sector; in narrow_write_error()
2531 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in narrow_write_error()
2532 wbio = bio_alloc_clone(rdev->bdev, in narrow_write_error()
2533 r1_bio->behind_master_bio, in narrow_write_error()
2534 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2536 wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio, in narrow_write_error()
2537 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2540 wbio->bi_opf = REQ_OP_WRITE; in narrow_write_error()
2541 wbio->bi_iter.bi_sector = r1_bio->sector; in narrow_write_error()
2542 wbio->bi_iter.bi_size = r1_bio->sectors << 9; in narrow_write_error()
2544 bio_trim(wbio, sector - r1_bio->sector, sectors); in narrow_write_error()
2545 wbio->bi_iter.bi_sector += rdev->data_offset; in narrow_write_error()
2554 sect_to_write -= sectors; in narrow_write_error()
2564 int s = r1_bio->sectors; in handle_sync_write_finished()
2565 for (m = 0; m < conf->raid_disks * 2 ; m++) { in handle_sync_write_finished()
2566 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_sync_write_finished()
2567 struct bio *bio = r1_bio->bios[m]; in handle_sync_write_finished()
2568 if (bio->bi_end_io == NULL) in handle_sync_write_finished()
2570 if (!bio->bi_status && in handle_sync_write_finished()
2571 test_bit(R1BIO_MadeGood, &r1_bio->state)) { in handle_sync_write_finished()
2572 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); in handle_sync_write_finished()
2574 if (bio->bi_status && in handle_sync_write_finished()
2575 test_bit(R1BIO_WriteError, &r1_bio->state)) { in handle_sync_write_finished()
2576 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) in handle_sync_write_finished()
2577 md_error(conf->mddev, rdev); in handle_sync_write_finished()
2581 md_done_sync(conf->mddev, s, 1); in handle_sync_write_finished()
2589 for (m = 0; m < conf->raid_disks * 2 ; m++) in handle_write_finished()
2590 if (r1_bio->bios[m] == IO_MADE_GOOD) { in handle_write_finished()
2591 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_write_finished()
2593 r1_bio->sector, in handle_write_finished()
2594 r1_bio->sectors, 0); in handle_write_finished()
2595 rdev_dec_pending(rdev, conf->mddev); in handle_write_finished()
2596 } else if (r1_bio->bios[m] != NULL) { in handle_write_finished()
2603 md_error(conf->mddev, in handle_write_finished()
2604 conf->mirrors[m].rdev); in handle_write_finished()
2606 rdev_dec_pending(conf->mirrors[m].rdev, in handle_write_finished()
2607 conf->mddev); in handle_write_finished()
2610 spin_lock_irq(&conf->device_lock); in handle_write_finished()
2611 list_add(&r1_bio->retry_list, &conf->bio_end_io_list); in handle_write_finished()
2612 idx = sector_to_idx(r1_bio->sector); in handle_write_finished()
2613 atomic_inc(&conf->nr_queued[idx]); in handle_write_finished()
2614 spin_unlock_irq(&conf->device_lock); in handle_write_finished()
2619 wake_up(&conf->wait_barrier); in handle_write_finished()
2620 md_wakeup_thread(conf->mddev->thread); in handle_write_finished()
2622 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in handle_write_finished()
2630 struct mddev *mddev = conf->mddev; in handle_read_error()
2635 clear_bit(R1BIO_ReadError, &r1_bio->state); in handle_read_error()
2639 * other devices. When we find one, we re-write in handle_read_error()
2645 bio = r1_bio->bios[r1_bio->read_disk]; in handle_read_error()
2647 r1_bio->bios[r1_bio->read_disk] = NULL; in handle_read_error()
2649 rdev = conf->mirrors[r1_bio->read_disk].rdev; in handle_read_error()
2650 if (mddev->ro == 0 in handle_read_error()
2651 && !test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2655 } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2658 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; in handle_read_error()
2661 rdev_dec_pending(rdev, conf->mddev); in handle_read_error()
2662 sector = r1_bio->sector; in handle_read_error()
2663 bio = r1_bio->master_bio; in handle_read_error()
2666 r1_bio->state = 0; in handle_read_error()
2667 raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); in handle_read_error()
2673 struct mddev *mddev = thread->mddev; in raid1d()
2676 struct r1conf *conf = mddev->private; in raid1d()
2677 struct list_head *head = &conf->retry_list; in raid1d()
2683 if (!list_empty_careful(&conf->bio_end_io_list) && in raid1d()
2684 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid1d()
2686 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2687 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) in raid1d()
2688 list_splice_init(&conf->bio_end_io_list, &tmp); in raid1d()
2689 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2693 list_del(&r1_bio->retry_list); in raid1d()
2694 idx = sector_to_idx(r1_bio->sector); in raid1d()
2695 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2696 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2707 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2709 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2712 r1_bio = list_entry(head->prev, struct r1bio, retry_list); in raid1d()
2713 list_del(head->prev); in raid1d()
2714 idx = sector_to_idx(r1_bio->sector); in raid1d()
2715 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2716 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2718 mddev = r1_bio->mddev; in raid1d()
2719 conf = mddev->private; in raid1d()
2720 if (test_bit(R1BIO_IsSync, &r1_bio->state)) { in raid1d()
2721 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2722 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2726 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2727 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2729 else if (test_bit(R1BIO_ReadError, &r1_bio->state)) in raid1d()
2735 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid1d()
2746 BUG_ON(mempool_initialized(&conf->r1buf_pool)); in init_resync()
2748 return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, in init_resync()
2749 r1buf_pool_free, conf->poolinfo); in init_resync()
2754 struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); in raid1_alloc_init_r1buf()
2759 for (i = conf->poolinfo->raid_disks; i--; ) { in raid1_alloc_init_r1buf()
2760 bio = r1bio->bios[i]; in raid1_alloc_init_r1buf()
2761 rps = bio->bi_private; in raid1_alloc_init_r1buf()
2763 bio->bi_private = rps; in raid1_alloc_init_r1buf()
2765 r1bio->master_bio = NULL; in raid1_alloc_init_r1buf()
2772 * We need to make sure that no normal I/O request - particularly write
2773 * requests - conflict with active sync requests.
2782 struct r1conf *conf = mddev->private; in raid1_sync_request()
2786 int disk = -1; in raid1_sync_request()
2788 int wonly = -1; in raid1_sync_request()
2797 if (!mempool_initialized(&conf->r1buf_pool)) in raid1_sync_request()
2805 * We can find the current addess in mddev->curr_resync in raid1_sync_request()
2807 if (mddev->curr_resync < max_sector) /* aborted */ in raid1_sync_request()
2808 mddev->bitmap_ops->end_sync(mddev, mddev->curr_resync, in raid1_sync_request()
2811 conf->fullsync = 0; in raid1_sync_request()
2813 mddev->bitmap_ops->close_sync(mddev); in raid1_sync_request()
2817 conf->cluster_sync_low = 0; in raid1_sync_request()
2818 conf->cluster_sync_high = 0; in raid1_sync_request()
2823 if (mddev->bitmap == NULL && in raid1_sync_request()
2824 mddev->recovery_cp == MaxSector && in raid1_sync_request()
2825 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid1_sync_request()
2826 conf->fullsync == 0) { in raid1_sync_request()
2828 return max_sector - sector_nr; in raid1_sync_request()
2833 if (!mddev->bitmap_ops->start_sync(mddev, sector_nr, &sync_blocks, true) && in raid1_sync_request()
2834 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2841 * If there is non-resync activity waiting for a turn, then let it in raid1_sync_request()
2844 if (atomic_read(&conf->nr_waiting[idx])) in raid1_sync_request()
2851 mddev->bitmap_ops->cond_end_sync(mddev, sector_nr, in raid1_sync_request()
2853 (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid1_sync_request()
2864 * and any others (which will be non-In_sync devices) for WRITE. in raid1_sync_request()
2869 r1_bio->mddev = mddev; in raid1_sync_request()
2870 r1_bio->sector = sector_nr; in raid1_sync_request()
2871 r1_bio->state = 0; in raid1_sync_request()
2872 set_bit(R1BIO_IsSync, &r1_bio->state); in raid1_sync_request()
2876 for (i = 0; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2878 bio = r1_bio->bios[i]; in raid1_sync_request()
2880 rdev = conf->mirrors[i].rdev; in raid1_sync_request()
2882 test_bit(Faulty, &rdev->flags)) { in raid1_sync_request()
2883 if (i < conf->raid_disks) in raid1_sync_request()
2885 } else if (!test_bit(In_sync, &rdev->flags)) { in raid1_sync_request()
2886 bio->bi_opf = REQ_OP_WRITE; in raid1_sync_request()
2887 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2897 good_sectors = first_bad - sector_nr; in raid1_sync_request()
2899 bad_sectors -= (sector_nr - first_bad); in raid1_sync_request()
2906 if (test_bit(WriteMostly, &rdev->flags)) { in raid1_sync_request()
2913 bio->bi_opf = REQ_OP_READ; in raid1_sync_request()
2914 bio->bi_end_io = end_sync_read; in raid1_sync_request()
2916 } else if (!test_bit(WriteErrorSeen, &rdev->flags) && in raid1_sync_request()
2917 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid1_sync_request()
2918 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { in raid1_sync_request()
2925 bio->bi_opf = REQ_OP_WRITE; in raid1_sync_request()
2926 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2930 if (rdev && bio->bi_end_io) { in raid1_sync_request()
2931 atomic_inc(&rdev->nr_pending); in raid1_sync_request()
2932 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; in raid1_sync_request()
2933 bio_set_dev(bio, rdev->bdev); in raid1_sync_request()
2934 if (test_bit(FailFast, &rdev->flags)) in raid1_sync_request()
2935 bio->bi_opf |= MD_FAILFAST; in raid1_sync_request()
2940 r1_bio->read_disk = disk; in raid1_sync_request()
2947 for (i = 0 ; i < conf->raid_disks * 2 ; i++) in raid1_sync_request()
2948 if (r1_bio->bios[i]->bi_end_io == end_sync_write) { in raid1_sync_request()
2949 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_sync_request()
2954 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid1_sync_request()
2964 conf->recovery_disabled = mddev->recovery_disabled; in raid1_sync_request()
2965 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_sync_request()
2972 /* only resync enough to reach the next bad->good in raid1_sync_request()
2977 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) in raid1_sync_request()
2979 write_targets += read_targets-1; in raid1_sync_request()
2982 /* There is nowhere to write, so all non-sync in raid1_sync_request()
2983 * drives must be failed - so we are finished in raid1_sync_request()
2988 rv = max_sector - sector_nr; in raid1_sync_request()
2994 if (max_sector > mddev->resync_max) in raid1_sync_request()
2995 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid1_sync_request()
3004 len = (max_sector - sector_nr) << 9; in raid1_sync_request()
3008 if (!mddev->bitmap_ops->start_sync(mddev, sector_nr, in raid1_sync_request()
3010 !conf->fullsync && in raid1_sync_request()
3011 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in raid1_sync_request()
3017 for (i = 0 ; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
3020 bio = r1_bio->bios[i]; in raid1_sync_request()
3022 if (bio->bi_end_io) { in raid1_sync_request()
3034 sync_blocks -= (len>>9); in raid1_sync_request()
3037 r1_bio->sectors = nr_sectors; in raid1_sync_request()
3040 conf->cluster_sync_high < sector_nr + nr_sectors) { in raid1_sync_request()
3041 conf->cluster_sync_low = mddev->curr_resync_completed; in raid1_sync_request()
3042 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; in raid1_sync_request()
3044 mddev->cluster_ops->resync_info_update(mddev, in raid1_sync_request()
3045 conf->cluster_sync_low, in raid1_sync_request()
3046 conf->cluster_sync_high); in raid1_sync_request()
3049 /* For a user-requested sync, we read all readable devices and do a in raid1_sync_request()
3052 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
3053 atomic_set(&r1_bio->remaining, read_targets); in raid1_sync_request()
3054 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { in raid1_sync_request()
3055 bio = r1_bio->bios[i]; in raid1_sync_request()
3056 if (bio->bi_end_io == end_sync_read) { in raid1_sync_request()
3057 read_targets--; in raid1_sync_request()
3060 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
3065 atomic_set(&r1_bio->remaining, 1); in raid1_sync_request()
3066 bio = r1_bio->bios[r1_bio->read_disk]; in raid1_sync_request()
3069 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
3080 return mddev->dev_sectors; in raid1_size()
3089 int err = -ENOMEM; in setup_conf()
3095 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
3097 if (!conf->nr_pending) in setup_conf()
3100 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
3102 if (!conf->nr_waiting) in setup_conf()
3105 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
3107 if (!conf->nr_queued) in setup_conf()
3110 conf->barrier = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
3112 if (!conf->barrier) in setup_conf()
3115 conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), in setup_conf()
3116 mddev->raid_disks, 2), in setup_conf()
3118 if (!conf->mirrors) in setup_conf()
3121 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
3122 if (!conf->tmppage) in setup_conf()
3125 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); in setup_conf()
3126 if (!conf->poolinfo) in setup_conf()
3128 conf->poolinfo->raid_disks = mddev->raid_disks * 2; in setup_conf()
3129 err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, in setup_conf()
3130 rbio_pool_free, conf->poolinfo); in setup_conf()
3134 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
3138 conf->poolinfo->mddev = mddev; in setup_conf()
3140 err = -EINVAL; in setup_conf()
3141 spin_lock_init(&conf->device_lock); in setup_conf()
3142 conf->raid_disks = mddev->raid_disks; in setup_conf()
3144 int disk_idx = rdev->raid_disk; in setup_conf()
3146 if (disk_idx >= conf->raid_disks || disk_idx < 0) in setup_conf()
3150 test_bit(Replacement, &rdev->flags))) in setup_conf()
3153 conf->mddev = mddev; in setup_conf()
3154 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
3155 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
3157 spin_lock_init(&conf->resync_lock); in setup_conf()
3158 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
3160 bio_list_init(&conf->pending_bio_list); in setup_conf()
3161 conf->recovery_disabled = mddev->recovery_disabled - 1; in setup_conf()
3163 err = -EIO; in setup_conf()
3164 for (i = 0; i < conf->raid_disks * 2; i++) { in setup_conf()
3166 disk = conf->mirrors + i; in setup_conf()
3168 if (i < conf->raid_disks && in setup_conf()
3169 disk[conf->raid_disks].rdev) { in setup_conf()
3171 if (!disk->rdev) { in setup_conf()
3175 disk->rdev = in setup_conf()
3176 disk[conf->raid_disks].rdev; in setup_conf()
3177 disk[conf->raid_disks].rdev = NULL; in setup_conf()
3178 } else if (!test_bit(In_sync, &disk->rdev->flags)) in setup_conf()
3179 /* Original is not in_sync - bad */ in setup_conf()
3183 if (!disk->rdev || in setup_conf()
3184 !test_bit(In_sync, &disk->rdev->flags)) { in setup_conf()
3185 disk->head_position = 0; in setup_conf()
3186 if (disk->rdev && in setup_conf()
3187 (disk->rdev->saved_raid_disk < 0)) in setup_conf()
3188 conf->fullsync = 1; in setup_conf()
3192 err = -ENOMEM; in setup_conf()
3193 rcu_assign_pointer(conf->thread, in setup_conf()
3195 if (!conf->thread) in setup_conf()
3202 mempool_exit(&conf->r1bio_pool); in setup_conf()
3203 kfree(conf->mirrors); in setup_conf()
3204 safe_put_page(conf->tmppage); in setup_conf()
3205 kfree(conf->poolinfo); in setup_conf()
3206 kfree(conf->nr_pending); in setup_conf()
3207 kfree(conf->nr_waiting); in setup_conf()
3208 kfree(conf->nr_queued); in setup_conf()
3209 kfree(conf->barrier); in setup_conf()
3210 bioset_exit(&conf->bio_split); in setup_conf()
3227 return queue_limits_set(mddev->gendisk->queue, &lim); in raid1_set_limits()
3236 if (mddev->level != 1) { in raid1_run()
3238 mdname(mddev), mddev->level); in raid1_run()
3239 return -EIO; in raid1_run()
3241 if (mddev->reshape_position != MaxSector) { in raid1_run()
3244 return -EIO; in raid1_run()
3252 if (mddev->private == NULL) in raid1_run()
3255 conf = mddev->private; in raid1_run()
3263 if (!mddev->private) in raid1_run()
3269 mddev->degraded = 0; in raid1_run()
3270 for (i = 0; i < conf->raid_disks; i++) in raid1_run()
3271 if (conf->mirrors[i].rdev == NULL || in raid1_run()
3272 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || in raid1_run()
3273 test_bit(Faulty, &conf->mirrors[i].rdev->flags)) in raid1_run()
3274 mddev->degraded++; in raid1_run()
3278 if (conf->raid_disks - mddev->degraded < 1) { in raid1_run()
3279 md_unregister_thread(mddev, &conf->thread); in raid1_run()
3280 if (!mddev->private) in raid1_run()
3282 return -EINVAL; in raid1_run()
3285 if (conf->raid_disks - mddev->degraded == 1) in raid1_run()
3286 mddev->recovery_cp = MaxSector; in raid1_run()
3288 if (mddev->recovery_cp != MaxSector) in raid1_run()
3289 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", in raid1_run()
3292 mdname(mddev), mddev->raid_disks - mddev->degraded, in raid1_run()
3293 mddev->raid_disks); in raid1_run()
3298 rcu_assign_pointer(mddev->thread, conf->thread); in raid1_run()
3299 rcu_assign_pointer(conf->thread, NULL); in raid1_run()
3300 mddev->private = conf; in raid1_run()
3301 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid1_run()
3307 md_unregister_thread(mddev, &mddev->thread); in raid1_run()
3315 mempool_exit(&conf->r1bio_pool); in raid1_free()
3316 kfree(conf->mirrors); in raid1_free()
3317 safe_put_page(conf->tmppage); in raid1_free()
3318 kfree(conf->poolinfo); in raid1_free()
3319 kfree(conf->nr_pending); in raid1_free()
3320 kfree(conf->nr_waiting); in raid1_free()
3321 kfree(conf->nr_queued); in raid1_free()
3322 kfree(conf->barrier); in raid1_free()
3323 bioset_exit(&conf->bio_split); in raid1_free()
3339 if (mddev->external_size && in raid1_resize()
3340 mddev->array_sectors > newsize) in raid1_resize()
3341 return -EINVAL; in raid1_resize()
3343 ret = mddev->bitmap_ops->resize(mddev, newsize, 0, false); in raid1_resize()
3348 if (sectors > mddev->dev_sectors && in raid1_resize()
3349 mddev->recovery_cp > mddev->dev_sectors) { in raid1_resize()
3350 mddev->recovery_cp = mddev->dev_sectors; in raid1_resize()
3351 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_resize()
3353 mddev->dev_sectors = sectors; in raid1_resize()
3354 mddev->resync_max_sectors = sectors; in raid1_resize()
3362 * 2/ resize conf->mirrors in raid1_reshape()
3366 * Then resize conf->mirrors and swap in the new r1bio pool. in raid1_reshape()
3374 struct r1conf *conf = mddev->private; in raid1_reshape()
3384 if (mddev->chunk_sectors != mddev->new_chunk_sectors || in raid1_reshape()
3385 mddev->layout != mddev->new_layout || in raid1_reshape()
3386 mddev->level != mddev->new_level) { in raid1_reshape()
3387 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid1_reshape()
3388 mddev->new_layout = mddev->layout; in raid1_reshape()
3389 mddev->new_level = mddev->level; in raid1_reshape()
3390 return -EINVAL; in raid1_reshape()
3396 raid_disks = mddev->raid_disks + mddev->delta_disks; in raid1_reshape()
3398 if (raid_disks < conf->raid_disks) { in raid1_reshape()
3400 for (d= 0; d < conf->raid_disks; d++) in raid1_reshape()
3401 if (conf->mirrors[d].rdev) in raid1_reshape()
3404 return -EBUSY; in raid1_reshape()
3409 return -ENOMEM; in raid1_reshape()
3410 newpoolinfo->mddev = mddev; in raid1_reshape()
3411 newpoolinfo->raid_disks = raid_disks * 2; in raid1_reshape()
3425 return -ENOMEM; in raid1_reshape()
3431 oldpool = conf->r1bio_pool; in raid1_reshape()
3432 conf->r1bio_pool = newpool; in raid1_reshape()
3434 for (d = d2 = 0; d < conf->raid_disks; d++) { in raid1_reshape()
3435 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid1_reshape()
3436 if (rdev && rdev->raid_disk != d2) { in raid1_reshape()
3438 rdev->raid_disk = d2; in raid1_reshape()
3442 mdname(mddev), rdev->raid_disk); in raid1_reshape()
3447 kfree(conf->mirrors); in raid1_reshape()
3448 conf->mirrors = newmirrors; in raid1_reshape()
3449 kfree(conf->poolinfo); in raid1_reshape()
3450 conf->poolinfo = newpoolinfo; in raid1_reshape()
3452 spin_lock_irqsave(&conf->device_lock, flags); in raid1_reshape()
3453 mddev->degraded += (raid_disks - conf->raid_disks); in raid1_reshape()
3454 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_reshape()
3455 conf->raid_disks = mddev->raid_disks = raid_disks; in raid1_reshape()
3456 mddev->delta_disks = 0; in raid1_reshape()
3460 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); in raid1_reshape()
3461 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_reshape()
3462 md_wakeup_thread(mddev->thread); in raid1_reshape()
3470 struct r1conf *conf = mddev->private; in raid1_quiesce()
3483 if (mddev->level == 5 && mddev->raid_disks == 2) { in raid1_takeover()
3485 mddev->new_level = 1; in raid1_takeover()
3486 mddev->new_layout = 0; in raid1_takeover()
3487 mddev->new_chunk_sectors = 0; in raid1_takeover()
3491 conf->array_frozen = 1; in raid1_takeover()
3497 return ERR_PTR(-EINVAL); in raid1_takeover()
3539 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3540 MODULE_ALIAS("md-raid1");
3541 MODULE_ALIAS("md-level-1");