diff options
-rw-r--r-- | drivers/md/dm-snap.c | 295 |
1 files changed, 174 insertions, 121 deletions
diff --git a/drivers/md/dm-snap.c b/drivers/md/dm-snap.c index 475514bda9d0..14bd1a1815b1 100644 --- a/drivers/md/dm-snap.c +++ b/drivers/md/dm-snap.c | |||
@@ -392,6 +392,8 @@ static void read_snapshot_metadata(struct dm_snapshot *s) | |||
392 | down_write(&s->lock); | 392 | down_write(&s->lock); |
393 | s->valid = 0; | 393 | s->valid = 0; |
394 | up_write(&s->lock); | 394 | up_write(&s->lock); |
395 | |||
396 | dm_table_event(s->table); | ||
395 | } | 397 | } |
396 | } | 398 | } |
397 | 399 | ||
@@ -601,6 +603,11 @@ static void error_bios(struct bio *bio) | |||
601 | } | 603 | } |
602 | } | 604 | } |
603 | 605 | ||
606 | static inline void error_snapshot_bios(struct pending_exception *pe) | ||
607 | { | ||
608 | error_bios(bio_list_get(&pe->snapshot_bios)); | ||
609 | } | ||
610 | |||
604 | static struct bio *__flush_bios(struct pending_exception *pe) | 611 | static struct bio *__flush_bios(struct pending_exception *pe) |
605 | { | 612 | { |
606 | /* | 613 | /* |
@@ -616,6 +623,28 @@ static struct bio *__flush_bios(struct pending_exception *pe) | |||
616 | return NULL; | 623 | return NULL; |
617 | } | 624 | } |
618 | 625 | ||
626 | static void __invalidate_snapshot(struct dm_snapshot *s, | ||
627 | struct pending_exception *pe, int err) | ||
628 | { | ||
629 | if (!s->valid) | ||
630 | return; | ||
631 | |||
632 | if (err == -EIO) | ||
633 | DMERR("Invalidating snapshot: Error reading/writing."); | ||
634 | else if (err == -ENOMEM) | ||
635 | DMERR("Invalidating snapshot: Unable to allocate exception."); | ||
636 | |||
637 | if (pe) | ||
638 | remove_exception(&pe->e); | ||
639 | |||
640 | if (s->store.drop_snapshot) | ||
641 | s->store.drop_snapshot(&s->store); | ||
642 | |||
643 | s->valid = 0; | ||
644 | |||
645 | dm_table_event(s->table); | ||
646 | } | ||
647 | |||
619 | static void pending_complete(struct pending_exception *pe, int success) | 648 | static void pending_complete(struct pending_exception *pe, int success) |
620 | { | 649 | { |
621 | struct exception *e; | 650 | struct exception *e; |
@@ -623,50 +652,53 @@ static void pending_complete(struct pending_exception *pe, int success) | |||
623 | struct dm_snapshot *s = pe->snap; | 652 | struct dm_snapshot *s = pe->snap; |
624 | struct bio *flush = NULL; | 653 | struct bio *flush = NULL; |
625 | 654 | ||
626 | if (success) { | 655 | if (!success) { |
627 | e = alloc_exception(); | 656 | /* Read/write error - snapshot is unusable */ |
628 | if (!e) { | ||
629 | DMWARN("Unable to allocate exception."); | ||
630 | down_write(&s->lock); | ||
631 | s->store.drop_snapshot(&s->store); | ||
632 | s->valid = 0; | ||
633 | flush = __flush_bios(pe); | ||
634 | up_write(&s->lock); | ||
635 | |||
636 | error_bios(bio_list_get(&pe->snapshot_bios)); | ||
637 | goto out; | ||
638 | } | ||
639 | *e = pe->e; | ||
640 | |||
641 | /* | ||
642 | * Add a proper exception, and remove the | ||
643 | * in-flight exception from the list. | ||
644 | */ | ||
645 | down_write(&s->lock); | 657 | down_write(&s->lock); |
646 | insert_exception(&s->complete, e); | 658 | __invalidate_snapshot(s, pe, -EIO); |
647 | remove_exception(&pe->e); | ||
648 | flush = __flush_bios(pe); | 659 | flush = __flush_bios(pe); |
649 | |||
650 | /* Submit any pending write bios */ | ||
651 | up_write(&s->lock); | 660 | up_write(&s->lock); |
652 | 661 | ||
653 | flush_bios(bio_list_get(&pe->snapshot_bios)); | 662 | error_snapshot_bios(pe); |
654 | } else { | 663 | goto out; |
655 | /* Read/write error - snapshot is unusable */ | 664 | } |
665 | |||
666 | e = alloc_exception(); | ||
667 | if (!e) { | ||
656 | down_write(&s->lock); | 668 | down_write(&s->lock); |
657 | if (s->valid) | 669 | __invalidate_snapshot(s, pe, -ENOMEM); |
658 | DMERR("Error reading/writing snapshot"); | ||
659 | s->store.drop_snapshot(&s->store); | ||
660 | s->valid = 0; | ||
661 | remove_exception(&pe->e); | ||
662 | flush = __flush_bios(pe); | 670 | flush = __flush_bios(pe); |
663 | up_write(&s->lock); | 671 | up_write(&s->lock); |
664 | 672 | ||
665 | error_bios(bio_list_get(&pe->snapshot_bios)); | 673 | error_snapshot_bios(pe); |
674 | goto out; | ||
675 | } | ||
676 | *e = pe->e; | ||
666 | 677 | ||
667 | dm_table_event(s->table); | 678 | /* |
679 | * Add a proper exception, and remove the | ||
680 | * in-flight exception from the list. | ||
681 | */ | ||
682 | down_write(&s->lock); | ||
683 | if (!s->valid) { | ||
684 | flush = __flush_bios(pe); | ||
685 | up_write(&s->lock); | ||
686 | |||
687 | free_exception(e); | ||
688 | |||
689 | error_snapshot_bios(pe); | ||
690 | goto out; | ||
668 | } | 691 | } |
669 | 692 | ||
693 | insert_exception(&s->complete, e); | ||
694 | remove_exception(&pe->e); | ||
695 | flush = __flush_bios(pe); | ||
696 | |||
697 | up_write(&s->lock); | ||
698 | |||
699 | /* Submit any pending write bios */ | ||
700 | flush_bios(bio_list_get(&pe->snapshot_bios)); | ||
701 | |||
670 | out: | 702 | out: |
671 | primary_pe = pe->primary_pe; | 703 | primary_pe = pe->primary_pe; |
672 | 704 | ||
@@ -758,39 +790,45 @@ __find_pending_exception(struct dm_snapshot *s, struct bio *bio) | |||
758 | if (e) { | 790 | if (e) { |
759 | /* cast the exception to a pending exception */ | 791 | /* cast the exception to a pending exception */ |
760 | pe = container_of(e, struct pending_exception, e); | 792 | pe = container_of(e, struct pending_exception, e); |
793 | goto out; | ||
794 | } | ||
761 | 795 | ||
762 | } else { | 796 | /* |
763 | /* | 797 | * Create a new pending exception, we don't want |
764 | * Create a new pending exception, we don't want | 798 | * to hold the lock while we do this. |
765 | * to hold the lock while we do this. | 799 | */ |
766 | */ | 800 | up_write(&s->lock); |
767 | up_write(&s->lock); | 801 | pe = alloc_pending_exception(); |
768 | pe = alloc_pending_exception(); | 802 | down_write(&s->lock); |
769 | down_write(&s->lock); | ||
770 | 803 | ||
771 | e = lookup_exception(&s->pending, chunk); | 804 | if (!s->valid) { |
772 | if (e) { | 805 | free_pending_exception(pe); |
773 | free_pending_exception(pe); | 806 | return NULL; |
774 | pe = container_of(e, struct pending_exception, e); | 807 | } |
775 | } else { | ||
776 | pe->e.old_chunk = chunk; | ||
777 | bio_list_init(&pe->origin_bios); | ||
778 | bio_list_init(&pe->snapshot_bios); | ||
779 | pe->primary_pe = NULL; | ||
780 | atomic_set(&pe->sibling_count, 1); | ||
781 | pe->snap = s; | ||
782 | pe->started = 0; | ||
783 | |||
784 | if (s->store.prepare_exception(&s->store, &pe->e)) { | ||
785 | free_pending_exception(pe); | ||
786 | s->valid = 0; | ||
787 | return NULL; | ||
788 | } | ||
789 | 808 | ||
790 | insert_exception(&s->pending, &pe->e); | 809 | e = lookup_exception(&s->pending, chunk); |
791 | } | 810 | if (e) { |
811 | free_pending_exception(pe); | ||
812 | pe = container_of(e, struct pending_exception, e); | ||
813 | goto out; | ||
792 | } | 814 | } |
793 | 815 | ||
816 | pe->e.old_chunk = chunk; | ||
817 | bio_list_init(&pe->origin_bios); | ||
818 | bio_list_init(&pe->snapshot_bios); | ||
819 | pe->primary_pe = NULL; | ||
820 | atomic_set(&pe->sibling_count, 1); | ||
821 | pe->snap = s; | ||
822 | pe->started = 0; | ||
823 | |||
824 | if (s->store.prepare_exception(&s->store, &pe->e)) { | ||
825 | free_pending_exception(pe); | ||
826 | return NULL; | ||
827 | } | ||
828 | |||
829 | insert_exception(&s->pending, &pe->e); | ||
830 | |||
831 | out: | ||
794 | return pe; | 832 | return pe; |
795 | } | 833 | } |
796 | 834 | ||
@@ -807,13 +845,15 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio, | |||
807 | { | 845 | { |
808 | struct exception *e; | 846 | struct exception *e; |
809 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; | 847 | struct dm_snapshot *s = (struct dm_snapshot *) ti->private; |
848 | int copy_needed = 0; | ||
810 | int r = 1; | 849 | int r = 1; |
811 | chunk_t chunk; | 850 | chunk_t chunk; |
812 | struct pending_exception *pe; | 851 | struct pending_exception *pe = NULL; |
813 | 852 | ||
814 | chunk = sector_to_chunk(s, bio->bi_sector); | 853 | chunk = sector_to_chunk(s, bio->bi_sector); |
815 | 854 | ||
816 | /* Full snapshots are not usable */ | 855 | /* Full snapshots are not usable */ |
856 | /* To get here the table must be live so s->active is always set. */ | ||
817 | if (!s->valid) | 857 | if (!s->valid) |
818 | return -EIO; | 858 | return -EIO; |
819 | 859 | ||
@@ -831,36 +871,41 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio, | |||
831 | * to copy an exception */ | 871 | * to copy an exception */ |
832 | down_write(&s->lock); | 872 | down_write(&s->lock); |
833 | 873 | ||
874 | if (!s->valid) { | ||
875 | r = -EIO; | ||
876 | goto out_unlock; | ||
877 | } | ||
878 | |||
834 | /* If the block is already remapped - use that, else remap it */ | 879 | /* If the block is already remapped - use that, else remap it */ |
835 | e = lookup_exception(&s->complete, chunk); | 880 | e = lookup_exception(&s->complete, chunk); |
836 | if (e) { | 881 | if (e) { |
837 | remap_exception(s, e, bio); | 882 | remap_exception(s, e, bio); |
838 | up_write(&s->lock); | 883 | goto out_unlock; |
839 | 884 | } | |
840 | } else { | 885 | |
841 | pe = __find_pending_exception(s, bio); | 886 | pe = __find_pending_exception(s, bio); |
842 | 887 | if (!pe) { | |
843 | if (!pe) { | 888 | __invalidate_snapshot(s, pe, -ENOMEM); |
844 | if (s->store.drop_snapshot) | 889 | r = -EIO; |
845 | s->store.drop_snapshot(&s->store); | 890 | goto out_unlock; |
846 | s->valid = 0; | ||
847 | r = -EIO; | ||
848 | up_write(&s->lock); | ||
849 | } else { | ||
850 | remap_exception(s, &pe->e, bio); | ||
851 | bio_list_add(&pe->snapshot_bios, bio); | ||
852 | |||
853 | if (!pe->started) { | ||
854 | /* this is protected by snap->lock */ | ||
855 | pe->started = 1; | ||
856 | up_write(&s->lock); | ||
857 | start_copy(pe); | ||
858 | } else | ||
859 | up_write(&s->lock); | ||
860 | r = 0; | ||
861 | } | ||
862 | } | 891 | } |
863 | 892 | ||
893 | remap_exception(s, &pe->e, bio); | ||
894 | bio_list_add(&pe->snapshot_bios, bio); | ||
895 | |||
896 | if (!pe->started) { | ||
897 | /* this is protected by snap->lock */ | ||
898 | pe->started = 1; | ||
899 | copy_needed = 1; | ||
900 | } | ||
901 | |||
902 | r = 0; | ||
903 | |||
904 | out_unlock: | ||
905 | up_write(&s->lock); | ||
906 | |||
907 | if (copy_needed) | ||
908 | start_copy(pe); | ||
864 | } else { | 909 | } else { |
865 | /* | 910 | /* |
866 | * FIXME: this read path scares me because we | 911 | * FIXME: this read path scares me because we |
@@ -872,6 +917,11 @@ static int snapshot_map(struct dm_target *ti, struct bio *bio, | |||
872 | /* Do reads */ | 917 | /* Do reads */ |
873 | down_read(&s->lock); | 918 | down_read(&s->lock); |
874 | 919 | ||
920 | if (!s->valid) { | ||
921 | up_read(&s->lock); | ||
922 | return -EIO; | ||
923 | } | ||
924 | |||
875 | /* See if it it has been remapped */ | 925 | /* See if it it has been remapped */ |
876 | e = lookup_exception(&s->complete, chunk); | 926 | e = lookup_exception(&s->complete, chunk); |
877 | if (e) | 927 | if (e) |
@@ -948,15 +998,15 @@ static int __origin_write(struct list_head *snapshots, struct bio *bio) | |||
948 | /* Do all the snapshots on this origin */ | 998 | /* Do all the snapshots on this origin */ |
949 | list_for_each_entry (snap, snapshots, list) { | 999 | list_for_each_entry (snap, snapshots, list) { |
950 | 1000 | ||
1001 | down_write(&snap->lock); | ||
1002 | |||
951 | /* Only deal with valid and active snapshots */ | 1003 | /* Only deal with valid and active snapshots */ |
952 | if (!snap->valid || !snap->active) | 1004 | if (!snap->valid || !snap->active) |
953 | continue; | 1005 | goto next_snapshot; |
954 | 1006 | ||
955 | /* Nothing to do if writing beyond end of snapshot */ | 1007 | /* Nothing to do if writing beyond end of snapshot */ |
956 | if (bio->bi_sector >= dm_table_get_size(snap->table)) | 1008 | if (bio->bi_sector >= dm_table_get_size(snap->table)) |
957 | continue; | 1009 | goto next_snapshot; |
958 | |||
959 | down_write(&snap->lock); | ||
960 | 1010 | ||
961 | /* | 1011 | /* |
962 | * Remember, different snapshots can have | 1012 | * Remember, different snapshots can have |
@@ -973,40 +1023,43 @@ static int __origin_write(struct list_head *snapshots, struct bio *bio) | |||
973 | * won't destroy the primary_pe while we're inside this loop. | 1023 | * won't destroy the primary_pe while we're inside this loop. |
974 | */ | 1024 | */ |
975 | e = lookup_exception(&snap->complete, chunk); | 1025 | e = lookup_exception(&snap->complete, chunk); |
976 | if (!e) { | 1026 | if (e) |
977 | pe = __find_pending_exception(snap, bio); | 1027 | goto next_snapshot; |
978 | if (!pe) { | 1028 | |
979 | snap->store.drop_snapshot(&snap->store); | 1029 | pe = __find_pending_exception(snap, bio); |
980 | snap->valid = 0; | 1030 | if (!pe) { |
981 | 1031 | __invalidate_snapshot(snap, pe, ENOMEM); | |
982 | } else { | 1032 | goto next_snapshot; |
983 | if (!primary_pe) { | 1033 | } |
984 | /* | 1034 | |
985 | * Either every pe here has same | 1035 | if (!primary_pe) { |
986 | * primary_pe or none has one yet. | 1036 | /* |
987 | */ | 1037 | * Either every pe here has same |
988 | if (pe->primary_pe) | 1038 | * primary_pe or none has one yet. |
989 | primary_pe = pe->primary_pe; | 1039 | */ |
990 | else { | 1040 | if (pe->primary_pe) |
991 | primary_pe = pe; | 1041 | primary_pe = pe->primary_pe; |
992 | first = 1; | 1042 | else { |
993 | } | 1043 | primary_pe = pe; |
994 | 1044 | first = 1; | |
995 | bio_list_add(&primary_pe->origin_bios, | ||
996 | bio); | ||
997 | r = 0; | ||
998 | } | ||
999 | if (!pe->primary_pe) { | ||
1000 | atomic_inc(&primary_pe->sibling_count); | ||
1001 | pe->primary_pe = primary_pe; | ||
1002 | } | ||
1003 | if (!pe->started) { | ||
1004 | pe->started = 1; | ||
1005 | list_add_tail(&pe->list, &pe_queue); | ||
1006 | } | ||
1007 | } | 1045 | } |
1046 | |||
1047 | bio_list_add(&primary_pe->origin_bios, bio); | ||
1048 | |||
1049 | r = 0; | ||
1050 | } | ||
1051 | |||
1052 | if (!pe->primary_pe) { | ||
1053 | atomic_inc(&primary_pe->sibling_count); | ||
1054 | pe->primary_pe = primary_pe; | ||
1055 | } | ||
1056 | |||
1057 | if (!pe->started) { | ||
1058 | pe->started = 1; | ||
1059 | list_add_tail(&pe->list, &pe_queue); | ||
1008 | } | 1060 | } |
1009 | 1061 | ||
1062 | next_snapshot: | ||
1010 | up_write(&snap->lock); | 1063 | up_write(&snap->lock); |
1011 | } | 1064 | } |
1012 | 1065 | ||