Lines Matching +full:data +full:- +full:mirror
1 // SPDX-License-Identifier: GPL-2.0-or-later
9 * RAID-1 management functions.
11 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
19 * - bitmap marked during normal i/o
20 * - bitmap used to skip nondirty blocks during sync
22 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
23 * - persistent bitmap code
38 #include "md-bitmap.h"
50 do { if ((md)->queue) blk_add_trace_msg((md)->queue, "raid1 " fmt, ##args); } while (0)
53 #include "raid1-10.c"
55 #define START(node) ((node)->start)
56 #define LAST(node) ((node)->last)
65 sector_t lo = r1_bio->sector; in check_and_add_serial()
66 sector_t hi = lo + r1_bio->sectors; in check_and_add_serial()
67 struct serial_in_rdev *serial = &rdev->serial[idx]; in check_and_add_serial()
69 spin_lock_irqsave(&serial->serial_lock, flags); in check_and_add_serial()
71 if (raid1_rb_iter_first(&serial->serial_rb, lo, hi)) in check_and_add_serial()
72 ret = -EBUSY; in check_and_add_serial()
74 si->start = lo; in check_and_add_serial()
75 si->last = hi; in check_and_add_serial()
76 raid1_rb_insert(si, &serial->serial_rb); in check_and_add_serial()
78 spin_unlock_irqrestore(&serial->serial_lock, flags); in check_and_add_serial()
85 struct mddev *mddev = rdev->mddev; in wait_for_serialization()
87 int idx = sector_to_idx(r1_bio->sector); in wait_for_serialization()
88 struct serial_in_rdev *serial = &rdev->serial[idx]; in wait_for_serialization()
90 if (WARN_ON(!mddev->serial_info_pool)) in wait_for_serialization()
92 si = mempool_alloc(mddev->serial_info_pool, GFP_NOIO); in wait_for_serialization()
93 wait_event(serial->serial_io_wait, in wait_for_serialization()
102 struct mddev *mddev = rdev->mddev; in remove_serial()
104 struct serial_in_rdev *serial = &rdev->serial[idx]; in remove_serial()
106 spin_lock_irqsave(&serial->serial_lock, flags); in remove_serial()
107 for (si = raid1_rb_iter_first(&serial->serial_rb, lo, hi); in remove_serial()
109 if (si->start == lo && si->last == hi) { in remove_serial()
110 raid1_rb_remove(si, &serial->serial_rb); in remove_serial()
111 mempool_free(si, mddev->serial_info_pool); in remove_serial()
118 spin_unlock_irqrestore(&serial->serial_lock, flags); in remove_serial()
119 wake_up(&serial->serial_io_wait); in remove_serial()
123 * for resync bio, r1bio pointer can be retrieved from the per-bio
128 return get_resync_pages(bio)->raid_bio; in get_resync_r1bio()
131 static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data) in r1bio_pool_alloc() argument
133 struct pool_info *pi = data; in r1bio_pool_alloc()
134 int size = offsetof(struct r1bio, bios[pi->raid_disks]); in r1bio_pool_alloc()
147 static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data) in r1buf_pool_alloc() argument
149 struct pool_info *pi = data; in r1buf_pool_alloc()
160 rps = kmalloc_array(pi->raid_disks, sizeof(struct resync_pages), in r1buf_pool_alloc()
166 * Allocate bios : 1 for reading, n-1 for writing in r1buf_pool_alloc()
168 for (j = pi->raid_disks ; j-- ; ) { in r1buf_pool_alloc()
172 bio_init(bio, NULL, bio->bi_inline_vecs, RESYNC_PAGES, 0); in r1buf_pool_alloc()
173 r1_bio->bios[j] = bio; in r1buf_pool_alloc()
176 * Allocate RESYNC_PAGES data pages and attach them to in r1buf_pool_alloc()
178 * If this is a user-requested check/repair, allocate in r1buf_pool_alloc()
181 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) in r1buf_pool_alloc()
182 need_pages = pi->raid_disks; in r1buf_pool_alloc()
185 for (j = 0; j < pi->raid_disks; j++) { in r1buf_pool_alloc()
188 bio = r1_bio->bios[j]; in r1buf_pool_alloc()
198 rp->raid_bio = r1_bio; in r1buf_pool_alloc()
199 bio->bi_private = rp; in r1buf_pool_alloc()
202 r1_bio->master_bio = NULL; in r1buf_pool_alloc()
207 while (--j >= 0) in r1buf_pool_alloc()
211 while (++j < pi->raid_disks) { in r1buf_pool_alloc()
212 bio_uninit(r1_bio->bios[j]); in r1buf_pool_alloc()
213 kfree(r1_bio->bios[j]); in r1buf_pool_alloc()
218 rbio_pool_free(r1_bio, data); in r1buf_pool_alloc()
222 static void r1buf_pool_free(void *__r1_bio, void *data) in r1buf_pool_free() argument
224 struct pool_info *pi = data; in r1buf_pool_free()
229 for (i = pi->raid_disks; i--; ) { in r1buf_pool_free()
230 rp = get_resync_pages(r1bio->bios[i]); in r1buf_pool_free()
232 bio_uninit(r1bio->bios[i]); in r1buf_pool_free()
233 kfree(r1bio->bios[i]); in r1buf_pool_free()
239 rbio_pool_free(r1bio, data); in r1buf_pool_free()
246 for (i = 0; i < conf->raid_disks * 2; i++) { in put_all_bios()
247 struct bio **bio = r1_bio->bios + i; in put_all_bios()
256 struct r1conf *conf = r1_bio->mddev->private; in free_r1bio()
259 mempool_free(r1_bio, &conf->r1bio_pool); in free_r1bio()
264 struct r1conf *conf = r1_bio->mddev->private; in put_buf()
265 sector_t sect = r1_bio->sector; in put_buf()
268 for (i = 0; i < conf->raid_disks * 2; i++) { in put_buf()
269 struct bio *bio = r1_bio->bios[i]; in put_buf()
270 if (bio->bi_end_io) in put_buf()
271 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev); in put_buf()
274 mempool_free(r1_bio, &conf->r1buf_pool); in put_buf()
282 struct mddev *mddev = r1_bio->mddev; in reschedule_retry()
283 struct r1conf *conf = mddev->private; in reschedule_retry()
286 idx = sector_to_idx(r1_bio->sector); in reschedule_retry()
287 spin_lock_irqsave(&conf->device_lock, flags); in reschedule_retry()
288 list_add(&r1_bio->retry_list, &conf->retry_list); in reschedule_retry()
289 atomic_inc(&conf->nr_queued[idx]); in reschedule_retry()
290 spin_unlock_irqrestore(&conf->device_lock, flags); in reschedule_retry()
292 wake_up(&conf->wait_barrier); in reschedule_retry()
293 md_wakeup_thread(mddev->thread); in reschedule_retry()
303 struct bio *bio = r1_bio->master_bio; in call_bio_endio()
305 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in call_bio_endio()
306 bio->bi_status = BLK_STS_IOERR; in call_bio_endio()
313 struct bio *bio = r1_bio->master_bio; in raid_end_bio_io()
314 struct r1conf *conf = r1_bio->mddev->private; in raid_end_bio_io()
315 sector_t sector = r1_bio->sector; in raid_end_bio_io()
318 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid_end_bio_io()
319 pr_debug("raid1: sync end %s on sectors %llu-%llu\n", in raid_end_bio_io()
321 (unsigned long long) bio->bi_iter.bi_sector, in raid_end_bio_io()
322 (unsigned long long) bio_end_sector(bio) - 1); in raid_end_bio_io()
330 * to go idle. All I/Os, even write-behind writes, are done. in raid_end_bio_io()
340 struct r1conf *conf = r1_bio->mddev->private; in update_head_pos()
342 conf->mirrors[disk].head_position = in update_head_pos()
343 r1_bio->sector + (r1_bio->sectors); in update_head_pos()
351 int mirror; in find_bio_disk() local
352 struct r1conf *conf = r1_bio->mddev->private; in find_bio_disk()
353 int raid_disks = conf->raid_disks; in find_bio_disk()
355 for (mirror = 0; mirror < raid_disks * 2; mirror++) in find_bio_disk()
356 if (r1_bio->bios[mirror] == bio) in find_bio_disk()
359 BUG_ON(mirror == raid_disks * 2); in find_bio_disk()
360 update_head_pos(mirror, r1_bio); in find_bio_disk()
362 return mirror; in find_bio_disk()
367 int uptodate = !bio->bi_status; in raid1_end_read_request()
368 struct r1bio *r1_bio = bio->bi_private; in raid1_end_read_request()
369 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_read_request()
370 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; in raid1_end_read_request()
373 * this branch is our 'one mirror IO has finished' event handler: in raid1_end_read_request()
375 update_head_pos(r1_bio->read_disk, r1_bio); in raid1_end_read_request()
378 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_read_request()
379 else if (test_bit(FailFast, &rdev->flags) && in raid1_end_read_request()
380 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_end_read_request()
381 /* This was a fail-fast read so we definitely in raid1_end_read_request()
390 spin_lock_irqsave(&conf->device_lock, flags); in raid1_end_read_request()
391 if (r1_bio->mddev->degraded == conf->raid_disks || in raid1_end_read_request()
392 (r1_bio->mddev->degraded == conf->raid_disks-1 && in raid1_end_read_request()
393 test_bit(In_sync, &rdev->flags))) in raid1_end_read_request()
395 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_end_read_request()
400 rdev_dec_pending(rdev, conf->mddev); in raid1_end_read_request()
406 mdname(conf->mddev), in raid1_end_read_request()
407 rdev->bdev, in raid1_end_read_request()
408 (unsigned long long)r1_bio->sector); in raid1_end_read_request()
409 set_bit(R1BIO_ReadError, &r1_bio->state); in raid1_end_read_request()
418 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in close_write()
419 bio_free_pages(r1_bio->behind_master_bio); in close_write()
420 bio_put(r1_bio->behind_master_bio); in close_write()
421 r1_bio->behind_master_bio = NULL; in close_write()
424 md_bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector, in close_write()
425 r1_bio->sectors, in close_write()
426 !test_bit(R1BIO_Degraded, &r1_bio->state), in close_write()
427 test_bit(R1BIO_BehindIO, &r1_bio->state)); in close_write()
428 md_write_end(r1_bio->mddev); in close_write()
433 if (!atomic_dec_and_test(&r1_bio->remaining)) in r1_bio_write_done()
436 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in r1_bio_write_done()
440 if (test_bit(R1BIO_MadeGood, &r1_bio->state)) in r1_bio_write_done()
449 struct r1bio *r1_bio = bio->bi_private; in raid1_end_write_request()
450 int behind = test_bit(R1BIO_BehindIO, &r1_bio->state); in raid1_end_write_request()
451 struct r1conf *conf = r1_bio->mddev->private; in raid1_end_write_request()
453 int mirror = find_bio_disk(r1_bio, bio); in raid1_end_write_request() local
454 struct md_rdev *rdev = conf->mirrors[mirror].rdev; in raid1_end_write_request()
456 sector_t lo = r1_bio->sector; in raid1_end_write_request()
457 sector_t hi = r1_bio->sector + r1_bio->sectors; in raid1_end_write_request()
459 discard_error = bio->bi_status && bio_op(bio) == REQ_OP_DISCARD; in raid1_end_write_request()
462 * 'one mirror IO has finished' event handler: in raid1_end_write_request()
464 if (bio->bi_status && !discard_error) { in raid1_end_write_request()
465 set_bit(WriteErrorSeen, &rdev->flags); in raid1_end_write_request()
466 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in raid1_end_write_request()
468 conf->mddev->recovery); in raid1_end_write_request()
470 if (test_bit(FailFast, &rdev->flags) && in raid1_end_write_request()
471 (bio->bi_opf & MD_FAILFAST) && in raid1_end_write_request()
473 !test_bit(WriteMostly, &rdev->flags)) { in raid1_end_write_request()
474 md_error(r1_bio->mddev, rdev); in raid1_end_write_request()
481 if (!test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
482 set_bit(R1BIO_WriteError, &r1_bio->state); in raid1_end_write_request()
485 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1_end_write_request()
487 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
498 * to user-side. So if something waits for IO, then it in raid1_end_write_request()
504 r1_bio->bios[mirror] = NULL; in raid1_end_write_request()
509 * such device for properly reading the data back (we could in raid1_end_write_request()
511 * before rdev->recovery_offset, but for simplicity we don't in raid1_end_write_request()
514 if (test_bit(In_sync, &rdev->flags) && in raid1_end_write_request()
515 !test_bit(Faulty, &rdev->flags)) in raid1_end_write_request()
516 set_bit(R1BIO_Uptodate, &r1_bio->state); in raid1_end_write_request()
519 if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in raid1_end_write_request()
521 r1_bio->bios[mirror] = IO_MADE_GOOD; in raid1_end_write_request()
522 set_bit(R1BIO_MadeGood, &r1_bio->state); in raid1_end_write_request()
527 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_end_write_request()
529 if (test_bit(WriteMostly, &rdev->flags)) in raid1_end_write_request()
530 atomic_dec(&r1_bio->behind_remaining); in raid1_end_write_request()
534 * has safely reached all non-writemostly in raid1_end_write_request()
536 * gets done only once -- we don't ever want to return in raid1_end_write_request()
537 * -EIO here, instead we'll wait in raid1_end_write_request()
539 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) && in raid1_end_write_request()
540 test_bit(R1BIO_Uptodate, &r1_bio->state)) { in raid1_end_write_request()
542 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) { in raid1_end_write_request()
543 struct bio *mbio = r1_bio->master_bio; in raid1_end_write_request()
545 " %llu-%llu\n", in raid1_end_write_request()
546 (unsigned long long) mbio->bi_iter.bi_sector, in raid1_end_write_request()
547 (unsigned long long) bio_end_sector(mbio) - 1); in raid1_end_write_request()
551 } else if (rdev->mddev->serialize_policy) in raid1_end_write_request()
553 if (r1_bio->bios[mirror] == NULL) in raid1_end_write_request()
554 rdev_dec_pending(rdev, conf->mddev); in raid1_end_write_request()
576 len = round_up(start_sector + 1, BARRIER_UNIT_SECTOR_SIZE) - in align_to_barrier_unit_end()
587 * be done. There is a per-array 'next expected sequential IO' sector
588 * number - if this matches on the next IO then we use the last disk.
589 * There is also a per-disk 'last know head position' sector that is
595 * because position is mirror, not device based.
601 const sector_t this_sector = r1_bio->sector; in read_balance()
619 sectors = r1_bio->sectors; in read_balance()
620 best_disk = -1; in read_balance()
621 best_dist_disk = -1; in read_balance()
623 best_pending_disk = -1; in read_balance()
628 clear_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
630 if ((conf->mddev->recovery_cp < this_sector + sectors) || in read_balance()
631 (mddev_is_clustered(conf->mddev) && in read_balance()
632 md_cluster_ops->area_resyncing(conf->mddev, READ, this_sector, in read_balance()
638 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) { in read_balance()
645 rdev = conf->mirrors[disk].rdev; in read_balance()
646 if (r1_bio->bios[disk] == IO_BLOCKED in read_balance()
648 || test_bit(Faulty, &rdev->flags)) in read_balance()
650 if (!test_bit(In_sync, &rdev->flags) && in read_balance()
651 rdev->recovery_offset < this_sector + sectors) in read_balance()
653 if (test_bit(WriteMostly, &rdev->flags)) { in read_balance()
654 /* Don't balance among write-mostly, just in read_balance()
662 best_good_sectors = first_bad - this_sector; in read_balance()
683 bad_sectors -= (this_sector - first_bad); in read_balance()
690 sector_t good_sectors = first_bad - this_sector; in read_balance()
701 best_disk = -1; in read_balance()
707 set_bit(R1BIO_FailFast, &r1_bio->state); in read_balance()
709 nonrot = bdev_nonrot(rdev->bdev); in read_balance()
711 pending = atomic_read(&rdev->nr_pending); in read_balance()
712 dist = abs(this_sector - conf->mirrors[disk].head_position); in read_balance()
718 if (conf->mirrors[disk].next_seq_sect == this_sector in read_balance()
720 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9; in read_balance()
721 struct raid1_info *mirror = &conf->mirrors[disk]; in read_balance() local
738 mirror->seq_start != MaxSector && in read_balance()
739 mirror->next_seq_sect > opt_iosize && in read_balance()
740 mirror->next_seq_sect - opt_iosize >= in read_balance()
741 mirror->seq_start) { in read_balance()
764 * non-rotational, choose the disk with less pending request even the in read_balance()
766 * mixed ratation/non-rotational disks depending on workload. in read_balance()
768 if (best_disk == -1) { in read_balance()
776 rdev = conf->mirrors[best_disk].rdev; in read_balance()
779 atomic_inc(&rdev->nr_pending); in read_balance()
782 if (conf->mirrors[best_disk].next_seq_sect != this_sector) in read_balance()
783 conf->mirrors[best_disk].seq_start = this_sector; in read_balance()
785 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors; in read_balance()
794 if (wq_has_sleeper(&conf->wait_barrier)) in wake_up_barrier()
795 wake_up(&conf->wait_barrier); in wake_up_barrier()
801 raid1_prepare_flush_writes(conf->mddev->bitmap); in flush_bio_list()
805 struct bio *next = bio->bi_next; in flush_bio_list()
818 spin_lock_irq(&conf->device_lock); in flush_pending_writes()
820 if (conf->pending_bio_list.head) { in flush_pending_writes()
824 bio = bio_list_get(&conf->pending_bio_list); in flush_pending_writes()
825 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
829 * current->state might be TASK_UNINTERRUPTIBLE which will in flush_pending_writes()
833 * is a false-positive. Silence the warning by resetting in flush_pending_writes()
841 spin_unlock_irq(&conf->device_lock); in flush_pending_writes()
853 * We choose only to raise the barrier if no-one is waiting for the
865 * If resync/recovery is interrupted, returns -EINTR;
872 spin_lock_irq(&conf->resync_lock); in raise_barrier()
875 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
876 !atomic_read(&conf->nr_waiting[idx]), in raise_barrier()
877 conf->resync_lock); in raise_barrier()
880 atomic_inc(&conf->barrier[idx]); in raise_barrier()
882 * In raise_barrier() we firstly increase conf->barrier[idx] then in raise_barrier()
883 * check conf->nr_pending[idx]. In _wait_barrier() we firstly in raise_barrier()
884 * increase conf->nr_pending[idx] then check conf->barrier[idx]. in raise_barrier()
885 * A memory barrier here to make sure conf->nr_pending[idx] won't in raise_barrier()
886 * be fetched before conf->barrier[idx] is increased. Otherwise in raise_barrier()
893 * B: while conf->nr_pending[idx] is not 0, meaning regular I/O in raise_barrier()
895 * C: while conf->barrier[idx] >= RESYNC_DEPTH, meaning reaches in raise_barrier()
898 wait_event_lock_irq(conf->wait_barrier, in raise_barrier()
899 (!conf->array_frozen && in raise_barrier()
900 !atomic_read(&conf->nr_pending[idx]) && in raise_barrier()
901 atomic_read(&conf->barrier[idx]) < RESYNC_DEPTH) || in raise_barrier()
902 test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery), in raise_barrier()
903 conf->resync_lock); in raise_barrier()
905 if (test_bit(MD_RECOVERY_INTR, &conf->mddev->recovery)) { in raise_barrier()
906 atomic_dec(&conf->barrier[idx]); in raise_barrier()
907 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
908 wake_up(&conf->wait_barrier); in raise_barrier()
909 return -EINTR; in raise_barrier()
912 atomic_inc(&conf->nr_sync_pending); in raise_barrier()
913 spin_unlock_irq(&conf->resync_lock); in raise_barrier()
922 BUG_ON(atomic_read(&conf->barrier[idx]) <= 0); in lower_barrier()
924 atomic_dec(&conf->barrier[idx]); in lower_barrier()
925 atomic_dec(&conf->nr_sync_pending); in lower_barrier()
926 wake_up(&conf->wait_barrier); in lower_barrier()
934 * We need to increase conf->nr_pending[idx] very early here, in _wait_barrier()
936 * conf->nr_pending[idx] to be 0. Then we can avoid holding in _wait_barrier()
937 * conf->resync_lock when there is no barrier raised in same in _wait_barrier()
941 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
943 * In _wait_barrier() we firstly increase conf->nr_pending[idx], then in _wait_barrier()
944 * check conf->barrier[idx]. In raise_barrier() we firstly increase in _wait_barrier()
945 * conf->barrier[idx], then check conf->nr_pending[idx]. A memory in _wait_barrier()
946 * barrier is necessary here to make sure conf->barrier[idx] won't be in _wait_barrier()
947 * fetched before conf->nr_pending[idx] is increased. Otherwise there in _wait_barrier()
954 * here. If during we check conf->barrier[idx], the array is in _wait_barrier()
955 * frozen (conf->array_frozen is 1), and chonf->barrier[idx] is in _wait_barrier()
961 if (!READ_ONCE(conf->array_frozen) && in _wait_barrier()
962 !atomic_read(&conf->barrier[idx])) in _wait_barrier()
966 * After holding conf->resync_lock, conf->nr_pending[idx] in _wait_barrier()
969 * raise_barrer() might be waiting for conf->nr_pending[idx] in _wait_barrier()
972 spin_lock_irq(&conf->resync_lock); in _wait_barrier()
973 atomic_inc(&conf->nr_waiting[idx]); in _wait_barrier()
974 atomic_dec(&conf->nr_pending[idx]); in _wait_barrier()
986 wait_event_lock_irq(conf->wait_barrier, in _wait_barrier()
987 !conf->array_frozen && in _wait_barrier()
988 !atomic_read(&conf->barrier[idx]), in _wait_barrier()
989 conf->resync_lock); in _wait_barrier()
990 atomic_inc(&conf->nr_pending[idx]); in _wait_barrier()
993 atomic_dec(&conf->nr_waiting[idx]); in _wait_barrier()
994 spin_unlock_irq(&conf->resync_lock); in _wait_barrier()
1008 * conf->barrier[idx] here, memory barrier is unnecessary as well. in wait_read_barrier()
1010 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1012 if (!READ_ONCE(conf->array_frozen)) in wait_read_barrier()
1015 spin_lock_irq(&conf->resync_lock); in wait_read_barrier()
1016 atomic_inc(&conf->nr_waiting[idx]); in wait_read_barrier()
1017 atomic_dec(&conf->nr_pending[idx]); in wait_read_barrier()
1030 wait_event_lock_irq(conf->wait_barrier, in wait_read_barrier()
1031 !conf->array_frozen, in wait_read_barrier()
1032 conf->resync_lock); in wait_read_barrier()
1033 atomic_inc(&conf->nr_pending[idx]); in wait_read_barrier()
1036 atomic_dec(&conf->nr_waiting[idx]); in wait_read_barrier()
1037 spin_unlock_irq(&conf->resync_lock); in wait_read_barrier()
1050 atomic_dec(&conf->nr_pending[idx]); in _allow_barrier()
1061 /* conf->resync_lock should be held */
1066 ret = atomic_read(&conf->nr_sync_pending); in get_unqueued_pending()
1068 ret += atomic_read(&conf->nr_pending[idx]) - in get_unqueued_pending()
1069 atomic_read(&conf->nr_queued[idx]); in get_unqueued_pending()
1087 * Every flying I/O contributes to a conf->nr_pending[idx], idx is the in freeze_array()
1089 * normal I/O are queued, sum of all conf->nr_pending[] will match sum in freeze_array()
1090 * of all conf->nr_queued[]. But normal I/O failure is an exception, in freeze_array()
1099 spin_lock_irq(&conf->resync_lock); in freeze_array()
1100 conf->array_frozen = 1; in freeze_array()
1101 raid1_log(conf->mddev, "wait freeze"); in freeze_array()
1103 conf->wait_barrier, in freeze_array()
1105 conf->resync_lock, in freeze_array()
1107 spin_unlock_irq(&conf->resync_lock); in freeze_array()
1112 spin_lock_irq(&conf->resync_lock); in unfreeze_array()
1113 conf->array_frozen = 0; in unfreeze_array()
1114 spin_unlock_irq(&conf->resync_lock); in unfreeze_array()
1115 wake_up(&conf->wait_barrier); in unfreeze_array()
1121 int size = bio->bi_iter.bi_size; in alloc_behind_master_bio()
1122 unsigned vcnt = (size + PAGE_SIZE - 1) >> PAGE_SHIFT; in alloc_behind_master_bio()
1127 &r1_bio->mddev->bio_set); in alloc_behind_master_bio()
1131 behind_bio->bi_iter.bi_size = size; in alloc_behind_master_bio()
1148 size -= len; in alloc_behind_master_bio()
1154 r1_bio->behind_master_bio = behind_bio; in alloc_behind_master_bio()
1155 set_bit(R1BIO_BehindIO, &r1_bio->state); in alloc_behind_master_bio()
1161 bio->bi_iter.bi_size); in alloc_behind_master_bio()
1170 struct mddev *mddev = plug->cb.data; in raid1_unplug()
1171 struct r1conf *conf = mddev->private; in raid1_unplug()
1175 spin_lock_irq(&conf->device_lock); in raid1_unplug()
1176 bio_list_merge(&conf->pending_bio_list, &plug->pending); in raid1_unplug()
1177 spin_unlock_irq(&conf->device_lock); in raid1_unplug()
1179 md_wakeup_thread(mddev->thread); in raid1_unplug()
1184 /* we aren't scheduling, so we can do the write-out directly. */ in raid1_unplug()
1185 bio = bio_list_get(&plug->pending); in raid1_unplug()
1192 r1_bio->master_bio = bio; in init_r1bio()
1193 r1_bio->sectors = bio_sectors(bio); in init_r1bio()
1194 r1_bio->state = 0; in init_r1bio()
1195 r1_bio->mddev = mddev; in init_r1bio()
1196 r1_bio->sector = bio->bi_iter.bi_sector; in init_r1bio()
1202 struct r1conf *conf = mddev->private; in alloc_r1bio()
1205 r1_bio = mempool_alloc(&conf->r1bio_pool, GFP_NOIO); in alloc_r1bio()
1207 memset(r1_bio->bios, 0, conf->raid_disks * sizeof(r1_bio->bios[0])); in alloc_r1bio()
1215 struct r1conf *conf = mddev->private; in raid1_read_request()
1216 struct raid1_info *mirror; in raid1_read_request() local
1218 struct bitmap *bitmap = mddev->bitmap; in raid1_read_request()
1220 const blk_opf_t do_sync = bio->bi_opf & REQ_SYNC; in raid1_read_request()
1235 struct md_rdev *rdev = conf->mirrors[r1_bio->read_disk].rdev; in raid1_read_request()
1238 snprintf(b, sizeof(b), "%pg", rdev->bdev); in raid1_read_request()
1247 if (!wait_read_barrier(conf, bio->bi_iter.bi_sector, in raid1_read_request()
1248 bio->bi_opf & REQ_NOWAIT)) { in raid1_read_request()
1257 r1_bio->sectors = max_read_sectors; in raid1_read_request()
1260 * make_request() can abort the operation when read-ahead is being in raid1_read_request()
1271 (unsigned long long)r1_bio->sector); in raid1_read_request()
1276 mirror = conf->mirrors + rdisk; in raid1_read_request()
1279 pr_info_ratelimited("md/raid1:%s: redirecting sector %llu to other mirror: %pg\n", in raid1_read_request()
1281 (unsigned long long)r1_bio->sector, in raid1_read_request()
1282 mirror->rdev->bdev); in raid1_read_request()
1284 if (test_bit(WriteMostly, &mirror->rdev->flags) && in raid1_read_request()
1287 * Reading from a write-mostly device must take care not to in raid1_read_request()
1288 * over-take any writes that are 'behind' in raid1_read_request()
1291 wait_event(bitmap->behind_wait, in raid1_read_request()
1292 atomic_read(&bitmap->behind_writes) == 0); in raid1_read_request()
1297 gfp, &conf->bio_split); in raid1_read_request()
1301 r1_bio->master_bio = bio; in raid1_read_request()
1302 r1_bio->sectors = max_sectors; in raid1_read_request()
1305 r1_bio->read_disk = rdisk; in raid1_read_request()
1308 r1_bio->master_bio = bio; in raid1_read_request()
1310 read_bio = bio_alloc_clone(mirror->rdev->bdev, bio, gfp, in raid1_read_request()
1311 &mddev->bio_set); in raid1_read_request()
1313 r1_bio->bios[rdisk] = read_bio; in raid1_read_request()
1315 read_bio->bi_iter.bi_sector = r1_bio->sector + in raid1_read_request()
1316 mirror->rdev->data_offset; in raid1_read_request()
1317 read_bio->bi_end_io = raid1_end_read_request; in raid1_read_request()
1318 read_bio->bi_opf = op | do_sync; in raid1_read_request()
1319 if (test_bit(FailFast, &mirror->rdev->flags) && in raid1_read_request()
1320 test_bit(R1BIO_FailFast, &r1_bio->state)) in raid1_read_request()
1321 read_bio->bi_opf |= MD_FAILFAST; in raid1_read_request()
1322 read_bio->bi_private = r1_bio; in raid1_read_request()
1324 if (mddev->gendisk) in raid1_read_request()
1325 trace_block_bio_remap(read_bio, disk_devt(mddev->gendisk), in raid1_read_request()
1326 r1_bio->sector); in raid1_read_request()
1334 struct r1conf *conf = mddev->private; in raid1_write_request()
1337 struct bitmap *bitmap = mddev->bitmap; in raid1_write_request()
1346 md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1347 bio->bi_iter.bi_sector, bio_end_sector(bio))) { in raid1_write_request()
1350 if (bio->bi_opf & REQ_NOWAIT) { in raid1_write_request()
1355 prepare_to_wait(&conf->wait_barrier, in raid1_write_request()
1357 if (!md_cluster_ops->area_resyncing(mddev, WRITE, in raid1_write_request()
1358 bio->bi_iter.bi_sector, in raid1_write_request()
1363 finish_wait(&conf->wait_barrier, &w); in raid1_write_request()
1371 if (!wait_barrier(conf, bio->bi_iter.bi_sector, in raid1_write_request()
1372 bio->bi_opf & REQ_NOWAIT)) { in raid1_write_request()
1379 r1_bio->sectors = max_write_sectors; in raid1_write_request()
1392 disks = conf->raid_disks * 2; in raid1_write_request()
1394 max_sectors = r1_bio->sectors; in raid1_write_request()
1396 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_write_request()
1399 * The write-behind io is only attempted on drives marked as in raid1_write_request()
1400 * write-mostly, which means we could allocate write behind in raid1_write_request()
1403 if (!is_discard && rdev && test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1406 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) { in raid1_write_request()
1407 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1411 r1_bio->bios[i] = NULL; in raid1_write_request()
1412 if (!rdev || test_bit(Faulty, &rdev->flags)) { in raid1_write_request()
1413 if (i < conf->raid_disks) in raid1_write_request()
1414 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1_write_request()
1418 atomic_inc(&rdev->nr_pending); in raid1_write_request()
1419 if (test_bit(WriteErrorSeen, &rdev->flags)) { in raid1_write_request()
1424 is_bad = is_badblock(rdev, r1_bio->sector, max_sectors, in raid1_write_request()
1429 set_bit(BlockedBadBlocks, &rdev->flags); in raid1_write_request()
1433 if (is_bad && first_bad <= r1_bio->sector) { in raid1_write_request()
1435 bad_sectors -= (r1_bio->sector - first_bad); in raid1_write_request()
1444 * missing, so it might be re-added, in raid1_write_request()
1449 * in-sync is recorded in the bad in raid1_write_request()
1455 int good_sectors = first_bad - r1_bio->sector; in raid1_write_request()
1460 r1_bio->bios[i] = bio; in raid1_write_request()
1468 if (r1_bio->bios[j]) in raid1_write_request()
1469 rdev_dec_pending(conf->mirrors[j].rdev, mddev); in raid1_write_request()
1471 allow_barrier(conf, bio->bi_iter.bi_sector); in raid1_write_request()
1473 if (bio->bi_opf & REQ_NOWAIT) { in raid1_write_request()
1477 raid1_log(mddev, "wait rdev %d blocked", blocked_rdev->raid_disk); in raid1_write_request()
1479 wait_barrier(conf, bio->bi_iter.bi_sector, false); in raid1_write_request()
1485 * alloc_behind_master_bio allocates a copy of the data payload a page in raid1_write_request()
1494 GFP_NOIO, &conf->bio_split); in raid1_write_request()
1498 r1_bio->master_bio = bio; in raid1_write_request()
1499 r1_bio->sectors = max_sectors; in raid1_write_request()
1503 r1_bio->master_bio = bio; in raid1_write_request()
1504 atomic_set(&r1_bio->remaining, 1); in raid1_write_request()
1505 atomic_set(&r1_bio->behind_remaining, 0); in raid1_write_request()
1511 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_write_request()
1512 if (!r1_bio->bios[i]) in raid1_write_request()
1521 (atomic_read(&bitmap->behind_writes) in raid1_write_request()
1522 < mddev->bitmap_info.max_write_behind) && in raid1_write_request()
1523 !waitqueue_active(&bitmap->behind_wait)) { in raid1_write_request()
1527 md_bitmap_startwrite(bitmap, r1_bio->sector, r1_bio->sectors, in raid1_write_request()
1528 test_bit(R1BIO_BehindIO, &r1_bio->state)); in raid1_write_request()
1532 if (r1_bio->behind_master_bio) { in raid1_write_request()
1533 mbio = bio_alloc_clone(rdev->bdev, in raid1_write_request()
1534 r1_bio->behind_master_bio, in raid1_write_request()
1535 GFP_NOIO, &mddev->bio_set); in raid1_write_request()
1536 if (test_bit(CollisionCheck, &rdev->flags)) in raid1_write_request()
1538 if (test_bit(WriteMostly, &rdev->flags)) in raid1_write_request()
1539 atomic_inc(&r1_bio->behind_remaining); in raid1_write_request()
1541 mbio = bio_alloc_clone(rdev->bdev, bio, GFP_NOIO, in raid1_write_request()
1542 &mddev->bio_set); in raid1_write_request()
1544 if (mddev->serialize_policy) in raid1_write_request()
1548 r1_bio->bios[i] = mbio; in raid1_write_request()
1550 mbio->bi_iter.bi_sector = (r1_bio->sector + rdev->data_offset); in raid1_write_request()
1551 mbio->bi_end_io = raid1_end_write_request; in raid1_write_request()
1552 mbio->bi_opf = bio_op(bio) | (bio->bi_opf & (REQ_SYNC | REQ_FUA)); in raid1_write_request()
1553 if (test_bit(FailFast, &rdev->flags) && in raid1_write_request()
1554 !test_bit(WriteMostly, &rdev->flags) && in raid1_write_request()
1555 conf->raid_disks - mddev->degraded > 1) in raid1_write_request()
1556 mbio->bi_opf |= MD_FAILFAST; in raid1_write_request()
1557 mbio->bi_private = r1_bio; in raid1_write_request()
1559 atomic_inc(&r1_bio->remaining); in raid1_write_request()
1561 if (mddev->gendisk) in raid1_write_request()
1562 trace_block_bio_remap(mbio, disk_devt(mddev->gendisk), in raid1_write_request()
1563 r1_bio->sector); in raid1_write_request()
1565 mbio->bi_bdev = (void *)rdev; in raid1_write_request()
1567 spin_lock_irqsave(&conf->device_lock, flags); in raid1_write_request()
1568 bio_list_add(&conf->pending_bio_list, mbio); in raid1_write_request()
1569 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_write_request()
1570 md_wakeup_thread(mddev->thread); in raid1_write_request()
1584 if (unlikely(bio->bi_opf & REQ_PREFLUSH) in raid1_make_request()
1596 bio->bi_iter.bi_sector, bio_sectors(bio)); in raid1_make_request()
1610 struct r1conf *conf = mddev->private; in raid1_status()
1613 lockdep_assert_held(&mddev->lock); in raid1_status()
1615 seq_printf(seq, " [%d/%d] [", conf->raid_disks, in raid1_status()
1616 conf->raid_disks - mddev->degraded); in raid1_status()
1617 for (i = 0; i < conf->raid_disks; i++) { in raid1_status()
1618 struct md_rdev *rdev = READ_ONCE(conf->mirrors[i].rdev); in raid1_status()
1621 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_"); in raid1_status()
1627 * raid1_error() - RAID1 error handler.
1633 * - &MD_BROKEN flag is set in &mddev->flags.
1634 * - recovery is disabled.
1636 * - recovery is interrupted.
1637 * - &mddev->degraded is bumped.
1640 * &mddev->fail_last_dev is off.
1644 struct r1conf *conf = mddev->private; in raid1_error()
1647 spin_lock_irqsave(&conf->device_lock, flags); in raid1_error()
1649 if (test_bit(In_sync, &rdev->flags) && in raid1_error()
1650 (conf->raid_disks - mddev->degraded) == 1) { in raid1_error()
1651 set_bit(MD_BROKEN, &mddev->flags); in raid1_error()
1653 if (!mddev->fail_last_dev) { in raid1_error()
1654 conf->recovery_disabled = mddev->recovery_disabled; in raid1_error()
1655 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1659 set_bit(Blocked, &rdev->flags); in raid1_error()
1660 if (test_and_clear_bit(In_sync, &rdev->flags)) in raid1_error()
1661 mddev->degraded++; in raid1_error()
1662 set_bit(Faulty, &rdev->flags); in raid1_error()
1663 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_error()
1667 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_error()
1668 set_mask_bits(&mddev->sb_flags, 0, in raid1_error()
1672 mdname(mddev), rdev->bdev, in raid1_error()
1673 mdname(mddev), conf->raid_disks - mddev->degraded); in raid1_error()
1685 pr_debug(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded, in print_conf()
1686 conf->raid_disks); in print_conf()
1688 lockdep_assert_held(&conf->mddev->reconfig_mutex); in print_conf()
1689 for (i = 0; i < conf->raid_disks; i++) { in print_conf()
1690 struct md_rdev *rdev = conf->mirrors[i].rdev; in print_conf()
1693 i, !test_bit(In_sync, &rdev->flags), in print_conf()
1694 !test_bit(Faulty, &rdev->flags), in print_conf()
1695 rdev->bdev); in print_conf()
1708 mempool_exit(&conf->r1buf_pool); in close_sync()
1714 struct r1conf *conf = mddev->private; in raid1_spare_active()
1723 * which expects 'In_sync' flags and ->degraded to be consistent. in raid1_spare_active()
1725 spin_lock_irqsave(&conf->device_lock, flags); in raid1_spare_active()
1726 for (i = 0; i < conf->raid_disks; i++) { in raid1_spare_active()
1727 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_spare_active()
1728 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev; in raid1_spare_active()
1730 && !test_bit(Candidate, &repl->flags) in raid1_spare_active()
1731 && repl->recovery_offset == MaxSector in raid1_spare_active()
1732 && !test_bit(Faulty, &repl->flags) in raid1_spare_active()
1733 && !test_and_set_bit(In_sync, &repl->flags)) { in raid1_spare_active()
1736 !test_and_clear_bit(In_sync, &rdev->flags)) in raid1_spare_active()
1741 * it gets removed and never re-added in raid1_spare_active()
1743 set_bit(Faulty, &rdev->flags); in raid1_spare_active()
1745 rdev->sysfs_state); in raid1_spare_active()
1749 && rdev->recovery_offset == MaxSector in raid1_spare_active()
1750 && !test_bit(Faulty, &rdev->flags) in raid1_spare_active()
1751 && !test_and_set_bit(In_sync, &rdev->flags)) { in raid1_spare_active()
1753 sysfs_notify_dirent_safe(rdev->sysfs_state); in raid1_spare_active()
1756 mddev->degraded -= count; in raid1_spare_active()
1757 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_spare_active()
1765 struct r1conf *conf = mddev->private; in raid1_add_disk()
1766 int err = -EEXIST; in raid1_add_disk()
1767 int mirror = 0, repl_slot = -1; in raid1_add_disk() local
1770 int last = conf->raid_disks - 1; in raid1_add_disk()
1772 if (mddev->recovery_disabled == conf->recovery_disabled) in raid1_add_disk()
1773 return -EBUSY; in raid1_add_disk()
1776 return -ENXIO; in raid1_add_disk()
1778 if (rdev->raid_disk >= 0) in raid1_add_disk()
1779 first = last = rdev->raid_disk; in raid1_add_disk()
1782 * find the disk ... but prefer rdev->saved_raid_disk in raid1_add_disk()
1785 if (rdev->saved_raid_disk >= 0 && in raid1_add_disk()
1786 rdev->saved_raid_disk >= first && in raid1_add_disk()
1787 rdev->saved_raid_disk < conf->raid_disks && in raid1_add_disk()
1788 conf->mirrors[rdev->saved_raid_disk].rdev == NULL) in raid1_add_disk()
1789 first = last = rdev->saved_raid_disk; in raid1_add_disk()
1791 for (mirror = first; mirror <= last; mirror++) { in raid1_add_disk()
1792 p = conf->mirrors + mirror; in raid1_add_disk()
1793 if (!p->rdev) { in raid1_add_disk()
1794 if (mddev->gendisk) in raid1_add_disk()
1795 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_add_disk()
1796 rdev->data_offset << 9); in raid1_add_disk()
1798 p->head_position = 0; in raid1_add_disk()
1799 rdev->raid_disk = mirror; in raid1_add_disk()
1804 if (rdev->saved_raid_disk < 0) in raid1_add_disk()
1805 conf->fullsync = 1; in raid1_add_disk()
1806 WRITE_ONCE(p->rdev, rdev); in raid1_add_disk()
1809 if (test_bit(WantReplacement, &p->rdev->flags) && in raid1_add_disk()
1810 p[conf->raid_disks].rdev == NULL && repl_slot < 0) in raid1_add_disk()
1811 repl_slot = mirror; in raid1_add_disk()
1816 p = conf->mirrors + repl_slot; in raid1_add_disk()
1817 clear_bit(In_sync, &rdev->flags); in raid1_add_disk()
1818 set_bit(Replacement, &rdev->flags); in raid1_add_disk()
1819 rdev->raid_disk = repl_slot; in raid1_add_disk()
1821 conf->fullsync = 1; in raid1_add_disk()
1822 WRITE_ONCE(p[conf->raid_disks].rdev, rdev); in raid1_add_disk()
1831 struct r1conf *conf = mddev->private; in raid1_remove_disk()
1833 int number = rdev->raid_disk; in raid1_remove_disk()
1834 struct raid1_info *p = conf->mirrors + number; in raid1_remove_disk()
1836 if (unlikely(number >= conf->raid_disks)) in raid1_remove_disk()
1839 if (rdev != p->rdev) in raid1_remove_disk()
1840 p = conf->mirrors + conf->raid_disks + number; in raid1_remove_disk()
1843 if (rdev == p->rdev) { in raid1_remove_disk()
1844 if (test_bit(In_sync, &rdev->flags) || in raid1_remove_disk()
1845 atomic_read(&rdev->nr_pending)) { in raid1_remove_disk()
1846 err = -EBUSY; in raid1_remove_disk()
1849 /* Only remove non-faulty devices if recovery in raid1_remove_disk()
1852 if (!test_bit(Faulty, &rdev->flags) && in raid1_remove_disk()
1853 mddev->recovery_disabled != conf->recovery_disabled && in raid1_remove_disk()
1854 mddev->degraded < conf->raid_disks) { in raid1_remove_disk()
1855 err = -EBUSY; in raid1_remove_disk()
1858 WRITE_ONCE(p->rdev, NULL); in raid1_remove_disk()
1859 if (conf->mirrors[conf->raid_disks + number].rdev) { in raid1_remove_disk()
1865 conf->mirrors[conf->raid_disks + number].rdev; in raid1_remove_disk()
1867 if (atomic_read(&repl->nr_pending)) { in raid1_remove_disk()
1874 err = -EBUSY; in raid1_remove_disk()
1878 clear_bit(Replacement, &repl->flags); in raid1_remove_disk()
1879 WRITE_ONCE(p->rdev, repl); in raid1_remove_disk()
1880 conf->mirrors[conf->raid_disks + number].rdev = NULL; in raid1_remove_disk()
1884 clear_bit(WantReplacement, &rdev->flags); in raid1_remove_disk()
1897 update_head_pos(r1_bio->read_disk, r1_bio); in end_sync_read()
1900 * we have read a block, now it needs to be re-written, in end_sync_read()
1901 * or re-read if the read failed. in end_sync_read()
1904 if (!bio->bi_status) in end_sync_read()
1905 set_bit(R1BIO_Uptodate, &r1_bio->state); in end_sync_read()
1907 if (atomic_dec_and_test(&r1_bio->remaining)) in end_sync_read()
1914 sector_t s = r1_bio->sector; in abort_sync_write()
1915 long sectors_to_go = r1_bio->sectors; in abort_sync_write()
1919 md_bitmap_end_sync(mddev->bitmap, s, &sync_blocks, 1); in abort_sync_write()
1921 sectors_to_go -= sync_blocks; in abort_sync_write()
1927 if (atomic_dec_and_test(&r1_bio->remaining)) { in put_sync_write_buf()
1928 struct mddev *mddev = r1_bio->mddev; in put_sync_write_buf()
1929 int s = r1_bio->sectors; in put_sync_write_buf()
1931 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in put_sync_write_buf()
1932 test_bit(R1BIO_WriteError, &r1_bio->state)) in put_sync_write_buf()
1943 int uptodate = !bio->bi_status; in end_sync_write()
1945 struct mddev *mddev = r1_bio->mddev; in end_sync_write()
1946 struct r1conf *conf = mddev->private; in end_sync_write()
1949 struct md_rdev *rdev = conf->mirrors[find_bio_disk(r1_bio, bio)].rdev; in end_sync_write()
1953 set_bit(WriteErrorSeen, &rdev->flags); in end_sync_write()
1954 if (!test_and_set_bit(WantReplacement, &rdev->flags)) in end_sync_write()
1956 mddev->recovery); in end_sync_write()
1957 set_bit(R1BIO_WriteError, &r1_bio->state); in end_sync_write()
1958 } else if (is_badblock(rdev, r1_bio->sector, r1_bio->sectors, in end_sync_write()
1960 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev, in end_sync_write()
1961 r1_bio->sector, in end_sync_write()
1962 r1_bio->sectors, in end_sync_write()
1965 set_bit(R1BIO_MadeGood, &r1_bio->state); in end_sync_write()
1977 set_bit(WriteErrorSeen, &rdev->flags); in r1_sync_page_io()
1979 &rdev->flags)) in r1_sync_page_io()
1981 rdev->mddev->recovery); in r1_sync_page_io()
1983 /* need to record an error - either for the block or the device */ in r1_sync_page_io()
1985 md_error(rdev->mddev, rdev); in r1_sync_page_io()
1992 * good data, much like with normal read errors. Only in fix_sync_read_error()
1994 * need to re-issue the read request. in fix_sync_read_error()
2002 struct mddev *mddev = r1_bio->mddev; in fix_sync_read_error()
2003 struct r1conf *conf = mddev->private; in fix_sync_read_error()
2004 struct bio *bio = r1_bio->bios[r1_bio->read_disk]; in fix_sync_read_error()
2005 struct page **pages = get_resync_pages(bio)->pages; in fix_sync_read_error()
2006 sector_t sect = r1_bio->sector; in fix_sync_read_error()
2007 int sectors = r1_bio->sectors; in fix_sync_read_error()
2011 rdev = conf->mirrors[r1_bio->read_disk].rdev; in fix_sync_read_error()
2012 if (test_bit(FailFast, &rdev->flags)) { in fix_sync_read_error()
2013 /* Don't try recovering from here - just fail it in fix_sync_read_error()
2016 if (test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2020 bio->bi_end_io = end_sync_write; in fix_sync_read_error()
2025 int d = r1_bio->read_disk; in fix_sync_read_error()
2032 if (r1_bio->bios[d]->bi_end_io == end_sync_read) { in fix_sync_read_error()
2037 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2046 if (d == conf->raid_disks * 2) in fix_sync_read_error()
2048 } while (!success && d != r1_bio->read_disk); in fix_sync_read_error()
2058 mdname(mddev), bio->bi_bdev, in fix_sync_read_error()
2059 (unsigned long long)r1_bio->sector); in fix_sync_read_error()
2060 for (d = 0; d < conf->raid_disks * 2; d++) { in fix_sync_read_error()
2061 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2062 if (!rdev || test_bit(Faulty, &rdev->flags)) in fix_sync_read_error()
2068 conf->recovery_disabled = in fix_sync_read_error()
2069 mddev->recovery_disabled; in fix_sync_read_error()
2070 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in fix_sync_read_error()
2071 md_done_sync(mddev, r1_bio->sectors, 0); in fix_sync_read_error()
2076 sectors -= s; in fix_sync_read_error()
2083 /* write it back and re-read */ in fix_sync_read_error()
2084 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2086 d = conf->raid_disks * 2; in fix_sync_read_error()
2087 d--; in fix_sync_read_error()
2088 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2090 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2094 r1_bio->bios[d]->bi_end_io = NULL; in fix_sync_read_error()
2099 while (d != r1_bio->read_disk) { in fix_sync_read_error()
2101 d = conf->raid_disks * 2; in fix_sync_read_error()
2102 d--; in fix_sync_read_error()
2103 if (r1_bio->bios[d]->bi_end_io != end_sync_read) in fix_sync_read_error()
2105 rdev = conf->mirrors[d].rdev; in fix_sync_read_error()
2109 atomic_add(s, &rdev->corrected_errors); in fix_sync_read_error()
2111 sectors -= s; in fix_sync_read_error()
2115 set_bit(R1BIO_Uptodate, &r1_bio->state); in fix_sync_read_error()
2116 bio->bi_status = 0; in fix_sync_read_error()
2127 * attempt an over-write in process_checks()
2129 struct mddev *mddev = r1_bio->mddev; in process_checks()
2130 struct r1conf *conf = mddev->private; in process_checks()
2136 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9); in process_checks()
2137 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2139 struct bio *b = r1_bio->bios[i]; in process_checks()
2141 if (b->bi_end_io != end_sync_read) in process_checks()
2144 status = b->bi_status; in process_checks()
2145 bio_reset(b, conf->mirrors[i].rdev->bdev, REQ_OP_READ); in process_checks()
2146 b->bi_status = status; in process_checks()
2147 b->bi_iter.bi_sector = r1_bio->sector + in process_checks()
2148 conf->mirrors[i].rdev->data_offset; in process_checks()
2149 b->bi_end_io = end_sync_read; in process_checks()
2150 rp->raid_bio = r1_bio; in process_checks()
2151 b->bi_private = rp; in process_checks()
2154 md_bio_reset_resync_pages(b, rp, r1_bio->sectors << 9); in process_checks()
2156 for (primary = 0; primary < conf->raid_disks * 2; primary++) in process_checks()
2157 if (r1_bio->bios[primary]->bi_end_io == end_sync_read && in process_checks()
2158 !r1_bio->bios[primary]->bi_status) { in process_checks()
2159 r1_bio->bios[primary]->bi_end_io = NULL; in process_checks()
2160 rdev_dec_pending(conf->mirrors[primary].rdev, mddev); in process_checks()
2163 r1_bio->read_disk = primary; in process_checks()
2164 for (i = 0; i < conf->raid_disks * 2; i++) { in process_checks()
2166 struct bio *pbio = r1_bio->bios[primary]; in process_checks()
2167 struct bio *sbio = r1_bio->bios[i]; in process_checks()
2168 blk_status_t status = sbio->bi_status; in process_checks()
2169 struct page **ppages = get_resync_pages(pbio)->pages; in process_checks()
2170 struct page **spages = get_resync_pages(sbio)->pages; in process_checks()
2175 if (sbio->bi_end_io != end_sync_read) in process_checks()
2178 sbio->bi_status = 0; in process_checks()
2181 page_len[j++] = bi->bv_len; in process_checks()
2184 for (j = vcnt; j-- ; ) { in process_checks()
2193 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches); in process_checks()
2194 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery) in process_checks()
2197 sbio->bi_end_io = NULL; in process_checks()
2198 rdev_dec_pending(conf->mirrors[i].rdev, mddev); in process_checks()
2208 struct r1conf *conf = mddev->private; in sync_request_write()
2210 int disks = conf->raid_disks * 2; in sync_request_write()
2213 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) in sync_request_write()
2214 /* ouch - failed to read all of that. */ in sync_request_write()
2218 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in sync_request_write()
2224 atomic_set(&r1_bio->remaining, 1); in sync_request_write()
2226 wbio = r1_bio->bios[i]; in sync_request_write()
2227 if (wbio->bi_end_io == NULL || in sync_request_write()
2228 (wbio->bi_end_io == end_sync_read && in sync_request_write()
2229 (i == r1_bio->read_disk || in sync_request_write()
2230 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery)))) in sync_request_write()
2232 if (test_bit(Faulty, &conf->mirrors[i].rdev->flags)) { in sync_request_write()
2237 wbio->bi_opf = REQ_OP_WRITE; in sync_request_write()
2238 if (test_bit(FailFast, &conf->mirrors[i].rdev->flags)) in sync_request_write()
2239 wbio->bi_opf |= MD_FAILFAST; in sync_request_write()
2241 wbio->bi_end_io = end_sync_write; in sync_request_write()
2242 atomic_inc(&r1_bio->remaining); in sync_request_write()
2243 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio)); in sync_request_write()
2261 sector_t sect = r1_bio->sector; in fix_read_error()
2262 int sectors = r1_bio->sectors; in fix_read_error()
2263 int read_disk = r1_bio->read_disk; in fix_read_error()
2264 struct mddev *mddev = conf->mddev; in fix_read_error()
2265 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; in fix_read_error()
2268 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; in fix_read_error()
2285 rdev = conf->mirrors[d].rdev; in fix_read_error()
2287 (test_bit(In_sync, &rdev->flags) || in fix_read_error()
2288 (!test_bit(Faulty, &rdev->flags) && in fix_read_error()
2289 rdev->recovery_offset >= sect + s)) && in fix_read_error()
2292 atomic_inc(&rdev->nr_pending); in fix_read_error()
2294 conf->tmppage, REQ_OP_READ, false)) in fix_read_error()
2302 if (d == conf->raid_disks * 2) in fix_read_error()
2307 /* Cannot read from anywhere - mark it bad */ in fix_read_error()
2308 struct md_rdev *rdev = conf->mirrors[read_disk].rdev; in fix_read_error()
2313 /* write it back and re-read */ in fix_read_error()
2317 d = conf->raid_disks * 2; in fix_read_error()
2318 d--; in fix_read_error()
2319 rdev = conf->mirrors[d].rdev; in fix_read_error()
2321 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2322 atomic_inc(&rdev->nr_pending); in fix_read_error()
2324 conf->tmppage, REQ_OP_WRITE); in fix_read_error()
2331 d = conf->raid_disks * 2; in fix_read_error()
2332 d--; in fix_read_error()
2333 rdev = conf->mirrors[d].rdev; in fix_read_error()
2335 !test_bit(Faulty, &rdev->flags)) { in fix_read_error()
2336 atomic_inc(&rdev->nr_pending); in fix_read_error()
2338 conf->tmppage, REQ_OP_READ)) { in fix_read_error()
2339 atomic_add(s, &rdev->corrected_errors); in fix_read_error()
2343 rdev->data_offset), in fix_read_error()
2344 rdev->bdev); in fix_read_error()
2349 sectors -= s; in fix_read_error()
2356 struct mddev *mddev = r1_bio->mddev; in narrow_write_error()
2357 struct r1conf *conf = mddev->private; in narrow_write_error()
2358 struct md_rdev *rdev = conf->mirrors[i].rdev; in narrow_write_error()
2360 /* bio has the data to be written to device 'i' where in narrow_write_error()
2374 int sect_to_write = r1_bio->sectors; in narrow_write_error()
2377 if (rdev->badblocks.shift < 0) in narrow_write_error()
2380 block_sectors = roundup(1 << rdev->badblocks.shift, in narrow_write_error()
2381 bdev_logical_block_size(rdev->bdev) >> 9); in narrow_write_error()
2382 sector = r1_bio->sector; in narrow_write_error()
2384 & ~(sector_t)(block_sectors - 1)) in narrow_write_error()
2385 - sector; in narrow_write_error()
2393 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) { in narrow_write_error()
2394 wbio = bio_alloc_clone(rdev->bdev, in narrow_write_error()
2395 r1_bio->behind_master_bio, in narrow_write_error()
2396 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2398 wbio = bio_alloc_clone(rdev->bdev, r1_bio->master_bio, in narrow_write_error()
2399 GFP_NOIO, &mddev->bio_set); in narrow_write_error()
2402 wbio->bi_opf = REQ_OP_WRITE; in narrow_write_error()
2403 wbio->bi_iter.bi_sector = r1_bio->sector; in narrow_write_error()
2404 wbio->bi_iter.bi_size = r1_bio->sectors << 9; in narrow_write_error()
2406 bio_trim(wbio, sector - r1_bio->sector, sectors); in narrow_write_error()
2407 wbio->bi_iter.bi_sector += rdev->data_offset; in narrow_write_error()
2416 sect_to_write -= sectors; in narrow_write_error()
2426 int s = r1_bio->sectors; in handle_sync_write_finished()
2427 for (m = 0; m < conf->raid_disks * 2 ; m++) { in handle_sync_write_finished()
2428 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_sync_write_finished()
2429 struct bio *bio = r1_bio->bios[m]; in handle_sync_write_finished()
2430 if (bio->bi_end_io == NULL) in handle_sync_write_finished()
2432 if (!bio->bi_status && in handle_sync_write_finished()
2433 test_bit(R1BIO_MadeGood, &r1_bio->state)) { in handle_sync_write_finished()
2434 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0); in handle_sync_write_finished()
2436 if (bio->bi_status && in handle_sync_write_finished()
2437 test_bit(R1BIO_WriteError, &r1_bio->state)) { in handle_sync_write_finished()
2438 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0)) in handle_sync_write_finished()
2439 md_error(conf->mddev, rdev); in handle_sync_write_finished()
2443 md_done_sync(conf->mddev, s, 1); in handle_sync_write_finished()
2451 for (m = 0; m < conf->raid_disks * 2 ; m++) in handle_write_finished()
2452 if (r1_bio->bios[m] == IO_MADE_GOOD) { in handle_write_finished()
2453 struct md_rdev *rdev = conf->mirrors[m].rdev; in handle_write_finished()
2455 r1_bio->sector, in handle_write_finished()
2456 r1_bio->sectors, 0); in handle_write_finished()
2457 rdev_dec_pending(rdev, conf->mddev); in handle_write_finished()
2458 } else if (r1_bio->bios[m] != NULL) { in handle_write_finished()
2465 md_error(conf->mddev, in handle_write_finished()
2466 conf->mirrors[m].rdev); in handle_write_finished()
2468 set_bit(R1BIO_Degraded, &r1_bio->state); in handle_write_finished()
2470 rdev_dec_pending(conf->mirrors[m].rdev, in handle_write_finished()
2471 conf->mddev); in handle_write_finished()
2474 spin_lock_irq(&conf->device_lock); in handle_write_finished()
2475 list_add(&r1_bio->retry_list, &conf->bio_end_io_list); in handle_write_finished()
2476 idx = sector_to_idx(r1_bio->sector); in handle_write_finished()
2477 atomic_inc(&conf->nr_queued[idx]); in handle_write_finished()
2478 spin_unlock_irq(&conf->device_lock); in handle_write_finished()
2483 wake_up(&conf->wait_barrier); in handle_write_finished()
2484 md_wakeup_thread(conf->mddev->thread); in handle_write_finished()
2486 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in handle_write_finished()
2494 struct mddev *mddev = conf->mddev; in handle_read_error()
2499 clear_bit(R1BIO_ReadError, &r1_bio->state); in handle_read_error()
2503 * other devices. When we find one, we re-write in handle_read_error()
2509 bio = r1_bio->bios[r1_bio->read_disk]; in handle_read_error()
2511 r1_bio->bios[r1_bio->read_disk] = NULL; in handle_read_error()
2513 rdev = conf->mirrors[r1_bio->read_disk].rdev; in handle_read_error()
2514 if (mddev->ro == 0 in handle_read_error()
2515 && !test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2519 } else if (mddev->ro == 0 && test_bit(FailFast, &rdev->flags)) { in handle_read_error()
2522 r1_bio->bios[r1_bio->read_disk] = IO_BLOCKED; in handle_read_error()
2525 rdev_dec_pending(rdev, conf->mddev); in handle_read_error()
2526 sector = r1_bio->sector; in handle_read_error()
2527 bio = r1_bio->master_bio; in handle_read_error()
2530 r1_bio->state = 0; in handle_read_error()
2531 raid1_read_request(mddev, bio, r1_bio->sectors, r1_bio); in handle_read_error()
2537 struct mddev *mddev = thread->mddev; in raid1d()
2540 struct r1conf *conf = mddev->private; in raid1d()
2541 struct list_head *head = &conf->retry_list; in raid1d()
2547 if (!list_empty_careful(&conf->bio_end_io_list) && in raid1d()
2548 !test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) { in raid1d()
2550 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2551 if (!test_bit(MD_SB_CHANGE_PENDING, &mddev->sb_flags)) in raid1d()
2552 list_splice_init(&conf->bio_end_io_list, &tmp); in raid1d()
2553 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2557 list_del(&r1_bio->retry_list); in raid1d()
2558 idx = sector_to_idx(r1_bio->sector); in raid1d()
2559 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2560 if (mddev->degraded) in raid1d()
2561 set_bit(R1BIO_Degraded, &r1_bio->state); in raid1d()
2562 if (test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2573 spin_lock_irqsave(&conf->device_lock, flags); in raid1d()
2575 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2578 r1_bio = list_entry(head->prev, struct r1bio, retry_list); in raid1d()
2579 list_del(head->prev); in raid1d()
2580 idx = sector_to_idx(r1_bio->sector); in raid1d()
2581 atomic_dec(&conf->nr_queued[idx]); in raid1d()
2582 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1d()
2584 mddev = r1_bio->mddev; in raid1d()
2585 conf = mddev->private; in raid1d()
2586 if (test_bit(R1BIO_IsSync, &r1_bio->state)) { in raid1d()
2587 if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2588 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2592 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) || in raid1d()
2593 test_bit(R1BIO_WriteError, &r1_bio->state)) in raid1d()
2595 else if (test_bit(R1BIO_ReadError, &r1_bio->state)) in raid1d()
2601 if (mddev->sb_flags & ~(1<<MD_SB_CHANGE_PENDING)) in raid1d()
2612 BUG_ON(mempool_initialized(&conf->r1buf_pool)); in init_resync()
2614 return mempool_init(&conf->r1buf_pool, buffs, r1buf_pool_alloc, in init_resync()
2615 r1buf_pool_free, conf->poolinfo); in init_resync()
2620 struct r1bio *r1bio = mempool_alloc(&conf->r1buf_pool, GFP_NOIO); in raid1_alloc_init_r1buf()
2625 for (i = conf->poolinfo->raid_disks; i--; ) { in raid1_alloc_init_r1buf()
2626 bio = r1bio->bios[i]; in raid1_alloc_init_r1buf()
2627 rps = bio->bi_private; in raid1_alloc_init_r1buf()
2629 bio->bi_private = rps; in raid1_alloc_init_r1buf()
2631 r1bio->master_bio = NULL; in raid1_alloc_init_r1buf()
2638 * We need to make sure that no normal I/O request - particularly write
2639 * requests - conflict with active sync requests.
2648 struct r1conf *conf = mddev->private; in raid1_sync_request()
2652 int disk = -1; in raid1_sync_request()
2654 int wonly = -1; in raid1_sync_request()
2663 if (!mempool_initialized(&conf->r1buf_pool)) in raid1_sync_request()
2667 max_sector = mddev->dev_sectors; in raid1_sync_request()
2672 * We can find the current addess in mddev->curr_resync in raid1_sync_request()
2674 if (mddev->curr_resync < max_sector) /* aborted */ in raid1_sync_request()
2675 md_bitmap_end_sync(mddev->bitmap, mddev->curr_resync, in raid1_sync_request()
2678 conf->fullsync = 0; in raid1_sync_request()
2680 md_bitmap_close_sync(mddev->bitmap); in raid1_sync_request()
2684 conf->cluster_sync_low = 0; in raid1_sync_request()
2685 conf->cluster_sync_high = 0; in raid1_sync_request()
2690 if (mddev->bitmap == NULL && in raid1_sync_request()
2691 mddev->recovery_cp == MaxSector && in raid1_sync_request()
2692 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) && in raid1_sync_request()
2693 conf->fullsync == 0) { in raid1_sync_request()
2695 return max_sector - sector_nr; in raid1_sync_request()
2700 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) && in raid1_sync_request()
2701 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2708 * If there is non-resync activity waiting for a turn, then let it in raid1_sync_request()
2711 if (atomic_read(&conf->nr_waiting[idx])) in raid1_sync_request()
2718 md_bitmap_cond_end_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2719 mddev_is_clustered(mddev) && (sector_nr + 2 * RESYNC_SECTORS > conf->cluster_sync_high)); in raid1_sync_request()
2731 * and any others (which will be non-In_sync devices) for WRITE. in raid1_sync_request()
2736 r1_bio->mddev = mddev; in raid1_sync_request()
2737 r1_bio->sector = sector_nr; in raid1_sync_request()
2738 r1_bio->state = 0; in raid1_sync_request()
2739 set_bit(R1BIO_IsSync, &r1_bio->state); in raid1_sync_request()
2743 for (i = 0; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2745 bio = r1_bio->bios[i]; in raid1_sync_request()
2747 rdev = conf->mirrors[i].rdev; in raid1_sync_request()
2749 test_bit(Faulty, &rdev->flags)) { in raid1_sync_request()
2750 if (i < conf->raid_disks) in raid1_sync_request()
2752 } else if (!test_bit(In_sync, &rdev->flags)) { in raid1_sync_request()
2753 bio->bi_opf = REQ_OP_WRITE; in raid1_sync_request()
2754 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2764 good_sectors = first_bad - sector_nr; in raid1_sync_request()
2766 bad_sectors -= (sector_nr - first_bad); in raid1_sync_request()
2773 if (test_bit(WriteMostly, &rdev->flags)) { in raid1_sync_request()
2780 bio->bi_opf = REQ_OP_READ; in raid1_sync_request()
2781 bio->bi_end_io = end_sync_read; in raid1_sync_request()
2783 } else if (!test_bit(WriteErrorSeen, &rdev->flags) && in raid1_sync_request()
2784 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && in raid1_sync_request()
2785 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) { in raid1_sync_request()
2792 bio->bi_opf = REQ_OP_WRITE; in raid1_sync_request()
2793 bio->bi_end_io = end_sync_write; in raid1_sync_request()
2797 if (rdev && bio->bi_end_io) { in raid1_sync_request()
2798 atomic_inc(&rdev->nr_pending); in raid1_sync_request()
2799 bio->bi_iter.bi_sector = sector_nr + rdev->data_offset; in raid1_sync_request()
2800 bio_set_dev(bio, rdev->bdev); in raid1_sync_request()
2801 if (test_bit(FailFast, &rdev->flags)) in raid1_sync_request()
2802 bio->bi_opf |= MD_FAILFAST; in raid1_sync_request()
2807 r1_bio->read_disk = disk; in raid1_sync_request()
2814 for (i = 0 ; i < conf->raid_disks * 2 ; i++) in raid1_sync_request()
2815 if (r1_bio->bios[i]->bi_end_io == end_sync_write) { in raid1_sync_request()
2816 struct md_rdev *rdev = conf->mirrors[i].rdev; in raid1_sync_request()
2821 set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); in raid1_sync_request()
2831 conf->recovery_disabled = mddev->recovery_disabled; in raid1_sync_request()
2832 set_bit(MD_RECOVERY_INTR, &mddev->recovery); in raid1_sync_request()
2839 /* only resync enough to reach the next bad->good in raid1_sync_request()
2844 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0) in raid1_sync_request()
2846 write_targets += read_targets-1; in raid1_sync_request()
2849 /* There is nowhere to write, so all non-sync in raid1_sync_request()
2850 * drives must be failed - so we are finished in raid1_sync_request()
2855 rv = max_sector - sector_nr; in raid1_sync_request()
2861 if (max_sector > mddev->resync_max) in raid1_sync_request()
2862 max_sector = mddev->resync_max; /* Don't do IO beyond here */ in raid1_sync_request()
2871 len = (max_sector - sector_nr) << 9; in raid1_sync_request()
2875 if (!md_bitmap_start_sync(mddev->bitmap, sector_nr, in raid1_sync_request()
2877 !conf->fullsync && in raid1_sync_request()
2878 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) in raid1_sync_request()
2884 for (i = 0 ; i < conf->raid_disks * 2; i++) { in raid1_sync_request()
2887 bio = r1_bio->bios[i]; in raid1_sync_request()
2889 if (bio->bi_end_io) { in raid1_sync_request()
2901 sync_blocks -= (len>>9); in raid1_sync_request()
2904 r1_bio->sectors = nr_sectors; in raid1_sync_request()
2907 conf->cluster_sync_high < sector_nr + nr_sectors) { in raid1_sync_request()
2908 conf->cluster_sync_low = mddev->curr_resync_completed; in raid1_sync_request()
2909 conf->cluster_sync_high = conf->cluster_sync_low + CLUSTER_RESYNC_WINDOW_SECTORS; in raid1_sync_request()
2911 md_cluster_ops->resync_info_update(mddev, in raid1_sync_request()
2912 conf->cluster_sync_low, in raid1_sync_request()
2913 conf->cluster_sync_high); in raid1_sync_request()
2916 /* For a user-requested sync, we read all readable devices and do a in raid1_sync_request()
2919 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) { in raid1_sync_request()
2920 atomic_set(&r1_bio->remaining, read_targets); in raid1_sync_request()
2921 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) { in raid1_sync_request()
2922 bio = r1_bio->bios[i]; in raid1_sync_request()
2923 if (bio->bi_end_io == end_sync_read) { in raid1_sync_request()
2924 read_targets--; in raid1_sync_request()
2927 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2932 atomic_set(&r1_bio->remaining, 1); in raid1_sync_request()
2933 bio = r1_bio->bios[r1_bio->read_disk]; in raid1_sync_request()
2936 bio->bi_opf &= ~MD_FAILFAST; in raid1_sync_request()
2947 return mddev->dev_sectors; in raid1_size()
2956 int err = -ENOMEM; in setup_conf()
2962 conf->nr_pending = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2964 if (!conf->nr_pending) in setup_conf()
2967 conf->nr_waiting = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2969 if (!conf->nr_waiting) in setup_conf()
2972 conf->nr_queued = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2974 if (!conf->nr_queued) in setup_conf()
2977 conf->barrier = kcalloc(BARRIER_BUCKETS_NR, in setup_conf()
2979 if (!conf->barrier) in setup_conf()
2982 conf->mirrors = kzalloc(array3_size(sizeof(struct raid1_info), in setup_conf()
2983 mddev->raid_disks, 2), in setup_conf()
2985 if (!conf->mirrors) in setup_conf()
2988 conf->tmppage = alloc_page(GFP_KERNEL); in setup_conf()
2989 if (!conf->tmppage) in setup_conf()
2992 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL); in setup_conf()
2993 if (!conf->poolinfo) in setup_conf()
2995 conf->poolinfo->raid_disks = mddev->raid_disks * 2; in setup_conf()
2996 err = mempool_init(&conf->r1bio_pool, NR_RAID_BIOS, r1bio_pool_alloc, in setup_conf()
2997 rbio_pool_free, conf->poolinfo); in setup_conf()
3001 err = bioset_init(&conf->bio_split, BIO_POOL_SIZE, 0, 0); in setup_conf()
3005 conf->poolinfo->mddev = mddev; in setup_conf()
3007 err = -EINVAL; in setup_conf()
3008 spin_lock_init(&conf->device_lock); in setup_conf()
3010 int disk_idx = rdev->raid_disk; in setup_conf()
3011 if (disk_idx >= mddev->raid_disks in setup_conf()
3014 if (test_bit(Replacement, &rdev->flags)) in setup_conf()
3015 disk = conf->mirrors + mddev->raid_disks + disk_idx; in setup_conf()
3017 disk = conf->mirrors + disk_idx; in setup_conf()
3019 if (disk->rdev) in setup_conf()
3021 disk->rdev = rdev; in setup_conf()
3022 disk->head_position = 0; in setup_conf()
3023 disk->seq_start = MaxSector; in setup_conf()
3025 conf->raid_disks = mddev->raid_disks; in setup_conf()
3026 conf->mddev = mddev; in setup_conf()
3027 INIT_LIST_HEAD(&conf->retry_list); in setup_conf()
3028 INIT_LIST_HEAD(&conf->bio_end_io_list); in setup_conf()
3030 spin_lock_init(&conf->resync_lock); in setup_conf()
3031 init_waitqueue_head(&conf->wait_barrier); in setup_conf()
3033 bio_list_init(&conf->pending_bio_list); in setup_conf()
3034 conf->recovery_disabled = mddev->recovery_disabled - 1; in setup_conf()
3036 err = -EIO; in setup_conf()
3037 for (i = 0; i < conf->raid_disks * 2; i++) { in setup_conf()
3039 disk = conf->mirrors + i; in setup_conf()
3041 if (i < conf->raid_disks && in setup_conf()
3042 disk[conf->raid_disks].rdev) { in setup_conf()
3044 if (!disk->rdev) { in setup_conf()
3048 disk->rdev = in setup_conf()
3049 disk[conf->raid_disks].rdev; in setup_conf()
3050 disk[conf->raid_disks].rdev = NULL; in setup_conf()
3051 } else if (!test_bit(In_sync, &disk->rdev->flags)) in setup_conf()
3052 /* Original is not in_sync - bad */ in setup_conf()
3056 if (!disk->rdev || in setup_conf()
3057 !test_bit(In_sync, &disk->rdev->flags)) { in setup_conf()
3058 disk->head_position = 0; in setup_conf()
3059 if (disk->rdev && in setup_conf()
3060 (disk->rdev->saved_raid_disk < 0)) in setup_conf()
3061 conf->fullsync = 1; in setup_conf()
3065 err = -ENOMEM; in setup_conf()
3066 rcu_assign_pointer(conf->thread, in setup_conf()
3068 if (!conf->thread) in setup_conf()
3075 mempool_exit(&conf->r1bio_pool); in setup_conf()
3076 kfree(conf->mirrors); in setup_conf()
3077 safe_put_page(conf->tmppage); in setup_conf()
3078 kfree(conf->poolinfo); in setup_conf()
3079 kfree(conf->nr_pending); in setup_conf()
3080 kfree(conf->nr_waiting); in setup_conf()
3081 kfree(conf->nr_queued); in setup_conf()
3082 kfree(conf->barrier); in setup_conf()
3083 bioset_exit(&conf->bio_split); in setup_conf()
3097 if (mddev->level != 1) { in raid1_run()
3099 mdname(mddev), mddev->level); in raid1_run()
3100 return -EIO; in raid1_run()
3102 if (mddev->reshape_position != MaxSector) { in raid1_run()
3105 return -EIO; in raid1_run()
3113 if (mddev->private == NULL) in raid1_run()
3116 conf = mddev->private; in raid1_run()
3121 if (mddev->queue) in raid1_run()
3122 blk_queue_max_write_zeroes_sectors(mddev->queue, 0); in raid1_run()
3125 if (!mddev->gendisk) in raid1_run()
3127 disk_stack_limits(mddev->gendisk, rdev->bdev, in raid1_run()
3128 rdev->data_offset << 9); in raid1_run()
3131 mddev->degraded = 0; in raid1_run()
3132 for (i = 0; i < conf->raid_disks; i++) in raid1_run()
3133 if (conf->mirrors[i].rdev == NULL || in raid1_run()
3134 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) || in raid1_run()
3135 test_bit(Faulty, &conf->mirrors[i].rdev->flags)) in raid1_run()
3136 mddev->degraded++; in raid1_run()
3140 if (conf->raid_disks - mddev->degraded < 1) { in raid1_run()
3141 md_unregister_thread(mddev, &conf->thread); in raid1_run()
3142 ret = -EINVAL; in raid1_run()
3146 if (conf->raid_disks - mddev->degraded == 1) in raid1_run()
3147 mddev->recovery_cp = MaxSector; in raid1_run()
3149 if (mddev->recovery_cp != MaxSector) in raid1_run()
3150 pr_info("md/raid1:%s: not clean -- starting background reconstruction\n", in raid1_run()
3153 mdname(mddev), mddev->raid_disks - mddev->degraded, in raid1_run()
3154 mddev->raid_disks); in raid1_run()
3159 rcu_assign_pointer(mddev->thread, conf->thread); in raid1_run()
3160 rcu_assign_pointer(conf->thread, NULL); in raid1_run()
3161 mddev->private = conf; in raid1_run()
3162 set_bit(MD_FAILFAST_SUPPORTED, &mddev->flags); in raid1_run()
3168 md_unregister_thread(mddev, &mddev->thread); in raid1_run()
3182 mempool_exit(&conf->r1bio_pool); in raid1_free()
3183 kfree(conf->mirrors); in raid1_free()
3184 safe_put_page(conf->tmppage); in raid1_free()
3185 kfree(conf->poolinfo); in raid1_free()
3186 kfree(conf->nr_pending); in raid1_free()
3187 kfree(conf->nr_waiting); in raid1_free()
3188 kfree(conf->nr_queued); in raid1_free()
3189 kfree(conf->barrier); in raid1_free()
3190 bioset_exit(&conf->bio_split); in raid1_free()
3204 if (mddev->external_size && in raid1_resize()
3205 mddev->array_sectors > newsize) in raid1_resize()
3206 return -EINVAL; in raid1_resize()
3207 if (mddev->bitmap) { in raid1_resize()
3208 int ret = md_bitmap_resize(mddev->bitmap, newsize, 0, 0); in raid1_resize()
3213 if (sectors > mddev->dev_sectors && in raid1_resize()
3214 mddev->recovery_cp > mddev->dev_sectors) { in raid1_resize()
3215 mddev->recovery_cp = mddev->dev_sectors; in raid1_resize()
3216 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_resize()
3218 mddev->dev_sectors = sectors; in raid1_resize()
3219 mddev->resync_max_sectors = sectors; in raid1_resize()
3227 * 2/ resize conf->mirrors in raid1_reshape()
3231 * Then resize conf->mirrors and swap in the new r1bio pool. in raid1_reshape()
3239 struct r1conf *conf = mddev->private; in raid1_reshape()
3249 if (mddev->chunk_sectors != mddev->new_chunk_sectors || in raid1_reshape()
3250 mddev->layout != mddev->new_layout || in raid1_reshape()
3251 mddev->level != mddev->new_level) { in raid1_reshape()
3252 mddev->new_chunk_sectors = mddev->chunk_sectors; in raid1_reshape()
3253 mddev->new_layout = mddev->layout; in raid1_reshape()
3254 mddev->new_level = mddev->level; in raid1_reshape()
3255 return -EINVAL; in raid1_reshape()
3261 raid_disks = mddev->raid_disks + mddev->delta_disks; in raid1_reshape()
3263 if (raid_disks < conf->raid_disks) { in raid1_reshape()
3265 for (d= 0; d < conf->raid_disks; d++) in raid1_reshape()
3266 if (conf->mirrors[d].rdev) in raid1_reshape()
3269 return -EBUSY; in raid1_reshape()
3274 return -ENOMEM; in raid1_reshape()
3275 newpoolinfo->mddev = mddev; in raid1_reshape()
3276 newpoolinfo->raid_disks = raid_disks * 2; in raid1_reshape()
3290 return -ENOMEM; in raid1_reshape()
3296 oldpool = conf->r1bio_pool; in raid1_reshape()
3297 conf->r1bio_pool = newpool; in raid1_reshape()
3299 for (d = d2 = 0; d < conf->raid_disks; d++) { in raid1_reshape()
3300 struct md_rdev *rdev = conf->mirrors[d].rdev; in raid1_reshape()
3301 if (rdev && rdev->raid_disk != d2) { in raid1_reshape()
3303 rdev->raid_disk = d2; in raid1_reshape()
3307 mdname(mddev), rdev->raid_disk); in raid1_reshape()
3312 kfree(conf->mirrors); in raid1_reshape()
3313 conf->mirrors = newmirrors; in raid1_reshape()
3314 kfree(conf->poolinfo); in raid1_reshape()
3315 conf->poolinfo = newpoolinfo; in raid1_reshape()
3317 spin_lock_irqsave(&conf->device_lock, flags); in raid1_reshape()
3318 mddev->degraded += (raid_disks - conf->raid_disks); in raid1_reshape()
3319 spin_unlock_irqrestore(&conf->device_lock, flags); in raid1_reshape()
3320 conf->raid_disks = mddev->raid_disks = raid_disks; in raid1_reshape()
3321 mddev->delta_disks = 0; in raid1_reshape()
3325 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery); in raid1_reshape()
3326 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery); in raid1_reshape()
3327 md_wakeup_thread(mddev->thread); in raid1_reshape()
3335 struct r1conf *conf = mddev->private; in raid1_quiesce()
3348 if (mddev->level == 5 && mddev->raid_disks == 2) { in raid1_takeover()
3350 mddev->new_level = 1; in raid1_takeover()
3351 mddev->new_layout = 0; in raid1_takeover()
3352 mddev->new_chunk_sectors = 0; in raid1_takeover()
3356 conf->array_frozen = 1; in raid1_takeover()
3362 return ERR_PTR(-EINVAL); in raid1_takeover()
3400 MODULE_ALIAS("md-personality-3"); /* RAID1 */
3401 MODULE_ALIAS("md-raid1");
3402 MODULE_ALIAS("md-level-1");