aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/extent-tree.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs/extent-tree.c')
-rw-r--r--fs/btrfs/extent-tree.c1674
1 files changed, 553 insertions, 1121 deletions
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index fefe83ad2059..f5e7cae63d80 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -49,17 +49,23 @@ struct pending_extent_op {
49 int del; 49 int del;
50}; 50};
51 51
52static int finish_current_insert(struct btrfs_trans_handle *trans, 52static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
53 struct btrfs_root *extent_root, int all); 53 struct btrfs_root *root, u64 parent,
54static int del_pending_extents(struct btrfs_trans_handle *trans, 54 u64 root_objectid, u64 ref_generation,
55 struct btrfs_root *extent_root, int all); 55 u64 owner, struct btrfs_key *ins,
56static int pin_down_bytes(struct btrfs_trans_handle *trans, 56 int ref_mod);
57 struct btrfs_root *root, 57static int update_reserved_extents(struct btrfs_root *root,
58 u64 bytenr, u64 num_bytes, int is_data); 58 u64 bytenr, u64 num, int reserve);
59static int update_block_group(struct btrfs_trans_handle *trans, 59static int update_block_group(struct btrfs_trans_handle *trans,
60 struct btrfs_root *root, 60 struct btrfs_root *root,
61 u64 bytenr, u64 num_bytes, int alloc, 61 u64 bytenr, u64 num_bytes, int alloc,
62 int mark_free); 62 int mark_free);
63static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans,
64 struct btrfs_root *root,
65 u64 bytenr, u64 num_bytes, u64 parent,
66 u64 root_objectid, u64 ref_generation,
67 u64 owner_objectid, int pin,
68 int ref_to_drop);
63 69
64static int do_chunk_alloc(struct btrfs_trans_handle *trans, 70static int do_chunk_alloc(struct btrfs_trans_handle *trans,
65 struct btrfs_root *extent_root, u64 alloc_bytes, 71 struct btrfs_root *extent_root, u64 alloc_bytes,
@@ -554,262 +560,13 @@ out:
554 return ret; 560 return ret;
555} 561}
556 562
557/*
558 * updates all the backrefs that are pending on update_list for the
559 * extent_root
560 */
561static noinline int update_backrefs(struct btrfs_trans_handle *trans,
562 struct btrfs_root *extent_root,
563 struct btrfs_path *path,
564 struct list_head *update_list)
565{
566 struct btrfs_key key;
567 struct btrfs_extent_ref *ref;
568 struct btrfs_fs_info *info = extent_root->fs_info;
569 struct pending_extent_op *op;
570 struct extent_buffer *leaf;
571 int ret = 0;
572 struct list_head *cur = update_list->next;
573 u64 ref_objectid;
574 u64 ref_root = extent_root->root_key.objectid;
575
576 op = list_entry(cur, struct pending_extent_op, list);
577
578search:
579 key.objectid = op->bytenr;
580 key.type = BTRFS_EXTENT_REF_KEY;
581 key.offset = op->orig_parent;
582
583 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1);
584 BUG_ON(ret);
585
586 leaf = path->nodes[0];
587
588loop:
589 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
590
591 ref_objectid = btrfs_ref_objectid(leaf, ref);
592
593 if (btrfs_ref_root(leaf, ref) != ref_root ||
594 btrfs_ref_generation(leaf, ref) != op->orig_generation ||
595 (ref_objectid != op->level &&
596 ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) {
597 printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, "
598 "root %llu, owner %u\n",
599 (unsigned long long)op->bytenr,
600 (unsigned long long)op->orig_parent,
601 (unsigned long long)ref_root, op->level);
602 btrfs_print_leaf(extent_root, leaf);
603 BUG();
604 }
605
606 key.objectid = op->bytenr;
607 key.offset = op->parent;
608 key.type = BTRFS_EXTENT_REF_KEY;
609 ret = btrfs_set_item_key_safe(trans, extent_root, path, &key);
610 BUG_ON(ret);
611 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
612 btrfs_set_ref_generation(leaf, ref, op->generation);
613
614 cur = cur->next;
615
616 list_del_init(&op->list);
617 unlock_extent(&info->extent_ins, op->bytenr,
618 op->bytenr + op->num_bytes - 1, GFP_NOFS);
619 kfree(op);
620
621 if (cur == update_list) {
622 btrfs_mark_buffer_dirty(path->nodes[0]);
623 btrfs_release_path(extent_root, path);
624 goto out;
625 }
626
627 op = list_entry(cur, struct pending_extent_op, list);
628
629 path->slots[0]++;
630 while (path->slots[0] < btrfs_header_nritems(leaf)) {
631 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
632 if (key.objectid == op->bytenr &&
633 key.type == BTRFS_EXTENT_REF_KEY)
634 goto loop;
635 path->slots[0]++;
636 }
637
638 btrfs_mark_buffer_dirty(path->nodes[0]);
639 btrfs_release_path(extent_root, path);
640 goto search;
641
642out:
643 return 0;
644}
645
646static noinline int insert_extents(struct btrfs_trans_handle *trans,
647 struct btrfs_root *extent_root,
648 struct btrfs_path *path,
649 struct list_head *insert_list, int nr)
650{
651 struct btrfs_key *keys;
652 u32 *data_size;
653 struct pending_extent_op *op;
654 struct extent_buffer *leaf;
655 struct list_head *cur = insert_list->next;
656 struct btrfs_fs_info *info = extent_root->fs_info;
657 u64 ref_root = extent_root->root_key.objectid;
658 int i = 0, last = 0, ret;
659 int total = nr * 2;
660
661 if (!nr)
662 return 0;
663
664 keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS);
665 if (!keys)
666 return -ENOMEM;
667
668 data_size = kzalloc(total * sizeof(u32), GFP_NOFS);
669 if (!data_size) {
670 kfree(keys);
671 return -ENOMEM;
672 }
673
674 list_for_each_entry(op, insert_list, list) {
675 keys[i].objectid = op->bytenr;
676 keys[i].offset = op->num_bytes;
677 keys[i].type = BTRFS_EXTENT_ITEM_KEY;
678 data_size[i] = sizeof(struct btrfs_extent_item);
679 i++;
680
681 keys[i].objectid = op->bytenr;
682 keys[i].offset = op->parent;
683 keys[i].type = BTRFS_EXTENT_REF_KEY;
684 data_size[i] = sizeof(struct btrfs_extent_ref);
685 i++;
686 }
687
688 op = list_entry(cur, struct pending_extent_op, list);
689 i = 0;
690 while (i < total) {
691 int c;
692 ret = btrfs_insert_some_items(trans, extent_root, path,
693 keys+i, data_size+i, total-i);
694 BUG_ON(ret < 0);
695
696 if (last && ret > 1)
697 BUG();
698
699 leaf = path->nodes[0];
700 for (c = 0; c < ret; c++) {
701 int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY;
702
703 /*
704 * if the first item we inserted was a backref, then
705 * the EXTENT_ITEM will be the odd c's, else it will
706 * be the even c's
707 */
708 if ((ref_first && (c % 2)) ||
709 (!ref_first && !(c % 2))) {
710 struct btrfs_extent_item *itm;
711
712 itm = btrfs_item_ptr(leaf, path->slots[0] + c,
713 struct btrfs_extent_item);
714 btrfs_set_extent_refs(path->nodes[0], itm, 1);
715 op->del++;
716 } else {
717 struct btrfs_extent_ref *ref;
718
719 ref = btrfs_item_ptr(leaf, path->slots[0] + c,
720 struct btrfs_extent_ref);
721 btrfs_set_ref_root(leaf, ref, ref_root);
722 btrfs_set_ref_generation(leaf, ref,
723 op->generation);
724 btrfs_set_ref_objectid(leaf, ref, op->level);
725 btrfs_set_ref_num_refs(leaf, ref, 1);
726 op->del++;
727 }
728
729 /*
730 * using del to see when its ok to free up the
731 * pending_extent_op. In the case where we insert the
732 * last item on the list in order to help do batching
733 * we need to not free the extent op until we actually
734 * insert the extent_item
735 */
736 if (op->del == 2) {
737 unlock_extent(&info->extent_ins, op->bytenr,
738 op->bytenr + op->num_bytes - 1,
739 GFP_NOFS);
740 cur = cur->next;
741 list_del_init(&op->list);
742 kfree(op);
743 if (cur != insert_list)
744 op = list_entry(cur,
745 struct pending_extent_op,
746 list);
747 }
748 }
749 btrfs_mark_buffer_dirty(leaf);
750 btrfs_release_path(extent_root, path);
751
752 /*
753 * Ok backref's and items usually go right next to eachother,
754 * but if we could only insert 1 item that means that we
755 * inserted on the end of a leaf, and we have no idea what may
756 * be on the next leaf so we just play it safe. In order to
757 * try and help this case we insert the last thing on our
758 * insert list so hopefully it will end up being the last
759 * thing on the leaf and everything else will be before it,
760 * which will let us insert a whole bunch of items at the same
761 * time.
762 */
763 if (ret == 1 && !last && (i + ret < total)) {
764 /*
765 * last: where we will pick up the next time around
766 * i: our current key to insert, will be total - 1
767 * cur: the current op we are screwing with
768 * op: duh
769 */
770 last = i + ret;
771 i = total - 1;
772 cur = insert_list->prev;
773 op = list_entry(cur, struct pending_extent_op, list);
774 } else if (last) {
775 /*
776 * ok we successfully inserted the last item on the
777 * list, lets reset everything
778 *
779 * i: our current key to insert, so where we left off
780 * last time
781 * last: done with this
782 * cur: the op we are messing with
783 * op: duh
784 * total: since we inserted the last key, we need to
785 * decrement total so we dont overflow
786 */
787 i = last;
788 last = 0;
789 total--;
790 if (i < total) {
791 cur = insert_list->next;
792 op = list_entry(cur, struct pending_extent_op,
793 list);
794 }
795 } else {
796 i += ret;
797 }
798
799 cond_resched();
800 }
801 ret = 0;
802 kfree(keys);
803 kfree(data_size);
804 return ret;
805}
806
807static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, 563static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
808 struct btrfs_root *root, 564 struct btrfs_root *root,
809 struct btrfs_path *path, 565 struct btrfs_path *path,
810 u64 bytenr, u64 parent, 566 u64 bytenr, u64 parent,
811 u64 ref_root, u64 ref_generation, 567 u64 ref_root, u64 ref_generation,
812 u64 owner_objectid) 568 u64 owner_objectid,
569 int refs_to_add)
813{ 570{
814 struct btrfs_key key; 571 struct btrfs_key key;
815 struct extent_buffer *leaf; 572 struct extent_buffer *leaf;
@@ -829,9 +586,10 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
829 btrfs_set_ref_root(leaf, ref, ref_root); 586 btrfs_set_ref_root(leaf, ref, ref_root);
830 btrfs_set_ref_generation(leaf, ref, ref_generation); 587 btrfs_set_ref_generation(leaf, ref, ref_generation);
831 btrfs_set_ref_objectid(leaf, ref, owner_objectid); 588 btrfs_set_ref_objectid(leaf, ref, owner_objectid);
832 btrfs_set_ref_num_refs(leaf, ref, 1); 589 btrfs_set_ref_num_refs(leaf, ref, refs_to_add);
833 } else if (ret == -EEXIST) { 590 } else if (ret == -EEXIST) {
834 u64 existing_owner; 591 u64 existing_owner;
592
835 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID); 593 BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID);
836 leaf = path->nodes[0]; 594 leaf = path->nodes[0];
837 ref = btrfs_item_ptr(leaf, path->slots[0], 595 ref = btrfs_item_ptr(leaf, path->slots[0],
@@ -845,7 +603,7 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
845 603
846 num_refs = btrfs_ref_num_refs(leaf, ref); 604 num_refs = btrfs_ref_num_refs(leaf, ref);
847 BUG_ON(num_refs == 0); 605 BUG_ON(num_refs == 0);
848 btrfs_set_ref_num_refs(leaf, ref, num_refs + 1); 606 btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add);
849 607
850 existing_owner = btrfs_ref_objectid(leaf, ref); 608 existing_owner = btrfs_ref_objectid(leaf, ref);
851 if (existing_owner != owner_objectid && 609 if (existing_owner != owner_objectid &&
@@ -857,6 +615,7 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans,
857 } else { 615 } else {
858 goto out; 616 goto out;
859 } 617 }
618 btrfs_unlock_up_safe(path, 1);
860 btrfs_mark_buffer_dirty(path->nodes[0]); 619 btrfs_mark_buffer_dirty(path->nodes[0]);
861out: 620out:
862 btrfs_release_path(root, path); 621 btrfs_release_path(root, path);
@@ -865,7 +624,8 @@ out:
865 624
866static noinline int remove_extent_backref(struct btrfs_trans_handle *trans, 625static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
867 struct btrfs_root *root, 626 struct btrfs_root *root,
868 struct btrfs_path *path) 627 struct btrfs_path *path,
628 int refs_to_drop)
869{ 629{
870 struct extent_buffer *leaf; 630 struct extent_buffer *leaf;
871 struct btrfs_extent_ref *ref; 631 struct btrfs_extent_ref *ref;
@@ -875,8 +635,8 @@ static noinline int remove_extent_backref(struct btrfs_trans_handle *trans,
875 leaf = path->nodes[0]; 635 leaf = path->nodes[0];
876 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); 636 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref);
877 num_refs = btrfs_ref_num_refs(leaf, ref); 637 num_refs = btrfs_ref_num_refs(leaf, ref);
878 BUG_ON(num_refs == 0); 638 BUG_ON(num_refs < refs_to_drop);
879 num_refs -= 1; 639 num_refs -= refs_to_drop;
880 if (num_refs == 0) { 640 if (num_refs == 0) {
881 ret = btrfs_del_item(trans, root, path); 641 ret = btrfs_del_item(trans, root, path);
882 } else { 642 } else {
@@ -927,332 +687,28 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
927#endif 687#endif
928} 688}
929 689
930static noinline int free_extents(struct btrfs_trans_handle *trans,
931 struct btrfs_root *extent_root,
932 struct list_head *del_list)
933{
934 struct btrfs_fs_info *info = extent_root->fs_info;
935 struct btrfs_path *path;
936 struct btrfs_key key, found_key;
937 struct extent_buffer *leaf;
938 struct list_head *cur;
939 struct pending_extent_op *op;
940 struct btrfs_extent_item *ei;
941 int ret, num_to_del, extent_slot = 0, found_extent = 0;
942 u32 refs;
943 u64 bytes_freed = 0;
944
945 path = btrfs_alloc_path();
946 if (!path)
947 return -ENOMEM;
948 path->reada = 1;
949
950search:
951 /* search for the backref for the current ref we want to delete */
952 cur = del_list->next;
953 op = list_entry(cur, struct pending_extent_op, list);
954 ret = lookup_extent_backref(trans, extent_root, path, op->bytenr,
955 op->orig_parent,
956 extent_root->root_key.objectid,
957 op->orig_generation, op->level, 1);
958 if (ret) {
959 printk(KERN_ERR "btrfs unable to find backref byte nr %llu "
960 "root %llu gen %llu owner %u\n",
961 (unsigned long long)op->bytenr,
962 (unsigned long long)extent_root->root_key.objectid,
963 (unsigned long long)op->orig_generation, op->level);
964 btrfs_print_leaf(extent_root, path->nodes[0]);
965 WARN_ON(1);
966 goto out;
967 }
968
969 extent_slot = path->slots[0];
970 num_to_del = 1;
971 found_extent = 0;
972
973 /*
974 * if we aren't the first item on the leaf we can move back one and see
975 * if our ref is right next to our extent item
976 */
977 if (likely(extent_slot)) {
978 extent_slot--;
979 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
980 extent_slot);
981 if (found_key.objectid == op->bytenr &&
982 found_key.type == BTRFS_EXTENT_ITEM_KEY &&
983 found_key.offset == op->num_bytes) {
984 num_to_del++;
985 found_extent = 1;
986 }
987 }
988
989 /*
990 * if we didn't find the extent we need to delete the backref and then
991 * search for the extent item key so we can update its ref count
992 */
993 if (!found_extent) {
994 key.objectid = op->bytenr;
995 key.type = BTRFS_EXTENT_ITEM_KEY;
996 key.offset = op->num_bytes;
997
998 ret = remove_extent_backref(trans, extent_root, path);
999 BUG_ON(ret);
1000 btrfs_release_path(extent_root, path);
1001 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1002 BUG_ON(ret);
1003 extent_slot = path->slots[0];
1004 }
1005
1006 /* this is where we update the ref count for the extent */
1007 leaf = path->nodes[0];
1008 ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item);
1009 refs = btrfs_extent_refs(leaf, ei);
1010 BUG_ON(refs == 0);
1011 refs--;
1012 btrfs_set_extent_refs(leaf, ei, refs);
1013
1014 btrfs_mark_buffer_dirty(leaf);
1015
1016 /*
1017 * This extent needs deleting. The reason cur_slot is extent_slot +
1018 * num_to_del is because extent_slot points to the slot where the extent
1019 * is, and if the backref was not right next to the extent we will be
1020 * deleting at least 1 item, and will want to start searching at the
1021 * slot directly next to extent_slot. However if we did find the
1022 * backref next to the extent item them we will be deleting at least 2
1023 * items and will want to start searching directly after the ref slot
1024 */
1025 if (!refs) {
1026 struct list_head *pos, *n, *end;
1027 int cur_slot = extent_slot+num_to_del;
1028 u64 super_used;
1029 u64 root_used;
1030
1031 path->slots[0] = extent_slot;
1032 bytes_freed = op->num_bytes;
1033
1034 mutex_lock(&info->pinned_mutex);
1035 ret = pin_down_bytes(trans, extent_root, op->bytenr,
1036 op->num_bytes, op->level >=
1037 BTRFS_FIRST_FREE_OBJECTID);
1038 mutex_unlock(&info->pinned_mutex);
1039 BUG_ON(ret < 0);
1040 op->del = ret;
1041
1042 /*
1043 * we need to see if we can delete multiple things at once, so
1044 * start looping through the list of extents we are wanting to
1045 * delete and see if their extent/backref's are right next to
1046 * eachother and the extents only have 1 ref
1047 */
1048 for (pos = cur->next; pos != del_list; pos = pos->next) {
1049 struct pending_extent_op *tmp;
1050
1051 tmp = list_entry(pos, struct pending_extent_op, list);
1052
1053 /* we only want to delete extent+ref at this stage */
1054 if (cur_slot >= btrfs_header_nritems(leaf) - 1)
1055 break;
1056
1057 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot);
1058 if (found_key.objectid != tmp->bytenr ||
1059 found_key.type != BTRFS_EXTENT_ITEM_KEY ||
1060 found_key.offset != tmp->num_bytes)
1061 break;
1062
1063 /* check to make sure this extent only has one ref */
1064 ei = btrfs_item_ptr(leaf, cur_slot,
1065 struct btrfs_extent_item);
1066 if (btrfs_extent_refs(leaf, ei) != 1)
1067 break;
1068
1069 btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1);
1070 if (found_key.objectid != tmp->bytenr ||
1071 found_key.type != BTRFS_EXTENT_REF_KEY ||
1072 found_key.offset != tmp->orig_parent)
1073 break;
1074
1075 /*
1076 * the ref is right next to the extent, we can set the
1077 * ref count to 0 since we will delete them both now
1078 */
1079 btrfs_set_extent_refs(leaf, ei, 0);
1080
1081 /* pin down the bytes for this extent */
1082 mutex_lock(&info->pinned_mutex);
1083 ret = pin_down_bytes(trans, extent_root, tmp->bytenr,
1084 tmp->num_bytes, tmp->level >=
1085 BTRFS_FIRST_FREE_OBJECTID);
1086 mutex_unlock(&info->pinned_mutex);
1087 BUG_ON(ret < 0);
1088
1089 /*
1090 * use the del field to tell if we need to go ahead and
1091 * free up the extent when we delete the item or not.
1092 */
1093 tmp->del = ret;
1094 bytes_freed += tmp->num_bytes;
1095
1096 num_to_del += 2;
1097 cur_slot += 2;
1098 }
1099 end = pos;
1100
1101 /* update the free space counters */
1102 spin_lock(&info->delalloc_lock);
1103 super_used = btrfs_super_bytes_used(&info->super_copy);
1104 btrfs_set_super_bytes_used(&info->super_copy,
1105 super_used - bytes_freed);
1106
1107 root_used = btrfs_root_used(&extent_root->root_item);
1108 btrfs_set_root_used(&extent_root->root_item,
1109 root_used - bytes_freed);
1110 spin_unlock(&info->delalloc_lock);
1111
1112 /* delete the items */
1113 ret = btrfs_del_items(trans, extent_root, path,
1114 path->slots[0], num_to_del);
1115 BUG_ON(ret);
1116
1117 /*
1118 * loop through the extents we deleted and do the cleanup work
1119 * on them
1120 */
1121 for (pos = cur, n = pos->next; pos != end;
1122 pos = n, n = pos->next) {
1123 struct pending_extent_op *tmp;
1124 tmp = list_entry(pos, struct pending_extent_op, list);
1125
1126 /*
1127 * remember tmp->del tells us wether or not we pinned
1128 * down the extent
1129 */
1130 ret = update_block_group(trans, extent_root,
1131 tmp->bytenr, tmp->num_bytes, 0,
1132 tmp->del);
1133 BUG_ON(ret);
1134
1135 list_del_init(&tmp->list);
1136 unlock_extent(&info->extent_ins, tmp->bytenr,
1137 tmp->bytenr + tmp->num_bytes - 1,
1138 GFP_NOFS);
1139 kfree(tmp);
1140 }
1141 } else if (refs && found_extent) {
1142 /*
1143 * the ref and extent were right next to eachother, but the
1144 * extent still has a ref, so just free the backref and keep
1145 * going
1146 */
1147 ret = remove_extent_backref(trans, extent_root, path);
1148 BUG_ON(ret);
1149
1150 list_del_init(&op->list);
1151 unlock_extent(&info->extent_ins, op->bytenr,
1152 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1153 kfree(op);
1154 } else {
1155 /*
1156 * the extent has multiple refs and the backref we were looking
1157 * for was not right next to it, so just unlock and go next,
1158 * we're good to go
1159 */
1160 list_del_init(&op->list);
1161 unlock_extent(&info->extent_ins, op->bytenr,
1162 op->bytenr + op->num_bytes - 1, GFP_NOFS);
1163 kfree(op);
1164 }
1165
1166 btrfs_release_path(extent_root, path);
1167 if (!list_empty(del_list))
1168 goto search;
1169
1170out:
1171 btrfs_free_path(path);
1172 return ret;
1173}
1174
1175static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, 690static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1176 struct btrfs_root *root, u64 bytenr, 691 struct btrfs_root *root, u64 bytenr,
692 u64 num_bytes,
1177 u64 orig_parent, u64 parent, 693 u64 orig_parent, u64 parent,
1178 u64 orig_root, u64 ref_root, 694 u64 orig_root, u64 ref_root,
1179 u64 orig_generation, u64 ref_generation, 695 u64 orig_generation, u64 ref_generation,
1180 u64 owner_objectid) 696 u64 owner_objectid)
1181{ 697{
1182 int ret; 698 int ret;
1183 struct btrfs_root *extent_root = root->fs_info->extent_root; 699 int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID;
1184 struct btrfs_path *path;
1185
1186 if (root == root->fs_info->extent_root) {
1187 struct pending_extent_op *extent_op;
1188 u64 num_bytes;
1189
1190 BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL);
1191 num_bytes = btrfs_level_size(root, (int)owner_objectid);
1192 mutex_lock(&root->fs_info->extent_ins_mutex);
1193 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
1194 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
1195 u64 priv;
1196 ret = get_state_private(&root->fs_info->extent_ins,
1197 bytenr, &priv);
1198 BUG_ON(ret);
1199 extent_op = (struct pending_extent_op *)
1200 (unsigned long)priv;
1201 BUG_ON(extent_op->parent != orig_parent);
1202 BUG_ON(extent_op->generation != orig_generation);
1203 700
1204 extent_op->parent = parent; 701 ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes,
1205 extent_op->generation = ref_generation; 702 orig_parent, parent, orig_root,
1206 } else { 703 ref_root, orig_generation,
1207 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); 704 ref_generation, owner_objectid, pin);
1208 BUG_ON(!extent_op);
1209
1210 extent_op->type = PENDING_BACKREF_UPDATE;
1211 extent_op->bytenr = bytenr;
1212 extent_op->num_bytes = num_bytes;
1213 extent_op->parent = parent;
1214 extent_op->orig_parent = orig_parent;
1215 extent_op->generation = ref_generation;
1216 extent_op->orig_generation = orig_generation;
1217 extent_op->level = (int)owner_objectid;
1218 INIT_LIST_HEAD(&extent_op->list);
1219 extent_op->del = 0;
1220
1221 set_extent_bits(&root->fs_info->extent_ins,
1222 bytenr, bytenr + num_bytes - 1,
1223 EXTENT_WRITEBACK, GFP_NOFS);
1224 set_state_private(&root->fs_info->extent_ins,
1225 bytenr, (unsigned long)extent_op);
1226 }
1227 mutex_unlock(&root->fs_info->extent_ins_mutex);
1228 return 0;
1229 }
1230
1231 path = btrfs_alloc_path();
1232 if (!path)
1233 return -ENOMEM;
1234 ret = lookup_extent_backref(trans, extent_root, path,
1235 bytenr, orig_parent, orig_root,
1236 orig_generation, owner_objectid, 1);
1237 if (ret)
1238 goto out;
1239 ret = remove_extent_backref(trans, extent_root, path);
1240 if (ret)
1241 goto out;
1242 ret = insert_extent_backref(trans, extent_root, path, bytenr,
1243 parent, ref_root, ref_generation,
1244 owner_objectid);
1245 BUG_ON(ret); 705 BUG_ON(ret);
1246 finish_current_insert(trans, extent_root, 0);
1247 del_pending_extents(trans, extent_root, 0);
1248out:
1249 btrfs_free_path(path);
1250 return ret; 706 return ret;
1251} 707}
1252 708
1253int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, 709int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1254 struct btrfs_root *root, u64 bytenr, 710 struct btrfs_root *root, u64 bytenr,
1255 u64 orig_parent, u64 parent, 711 u64 num_bytes, u64 orig_parent, u64 parent,
1256 u64 ref_root, u64 ref_generation, 712 u64 ref_root, u64 ref_generation,
1257 u64 owner_objectid) 713 u64 owner_objectid)
1258{ 714{
@@ -1260,20 +716,36 @@ int btrfs_update_extent_ref(struct btrfs_trans_handle *trans,
1260 if (ref_root == BTRFS_TREE_LOG_OBJECTID && 716 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1261 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) 717 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1262 return 0; 718 return 0;
1263 ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent, 719
1264 parent, ref_root, ref_root, 720 ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes,
1265 ref_generation, ref_generation, 721 orig_parent, parent, ref_root,
1266 owner_objectid); 722 ref_root, ref_generation,
723 ref_generation, owner_objectid);
1267 return ret; 724 return ret;
1268} 725}
1269
1270static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, 726static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1271 struct btrfs_root *root, u64 bytenr, 727 struct btrfs_root *root, u64 bytenr,
728 u64 num_bytes,
1272 u64 orig_parent, u64 parent, 729 u64 orig_parent, u64 parent,
1273 u64 orig_root, u64 ref_root, 730 u64 orig_root, u64 ref_root,
1274 u64 orig_generation, u64 ref_generation, 731 u64 orig_generation, u64 ref_generation,
1275 u64 owner_objectid) 732 u64 owner_objectid)
1276{ 733{
734 int ret;
735
736 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root,
737 ref_generation, owner_objectid,
738 BTRFS_ADD_DELAYED_REF, 0);
739 BUG_ON(ret);
740 return ret;
741}
742
743static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root, u64 bytenr,
745 u64 num_bytes, u64 parent, u64 ref_root,
746 u64 ref_generation, u64 owner_objectid,
747 int refs_to_add)
748{
1277 struct btrfs_path *path; 749 struct btrfs_path *path;
1278 int ret; 750 int ret;
1279 struct btrfs_key key; 751 struct btrfs_key key;
@@ -1286,17 +758,24 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1286 return -ENOMEM; 758 return -ENOMEM;
1287 759
1288 path->reada = 1; 760 path->reada = 1;
761 path->leave_spinning = 1;
1289 key.objectid = bytenr; 762 key.objectid = bytenr;
1290 key.type = BTRFS_EXTENT_ITEM_KEY; 763 key.type = BTRFS_EXTENT_ITEM_KEY;
1291 key.offset = (u64)-1; 764 key.offset = num_bytes;
1292 765
1293 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, 766 /* first find the extent item and update its reference count */
1294 0, 1); 767 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
1295 if (ret < 0) 768 path, 0, 1);
769 if (ret < 0) {
770 btrfs_set_path_blocking(path);
1296 return ret; 771 return ret;
1297 BUG_ON(ret == 0 || path->slots[0] == 0); 772 }
1298 773
1299 path->slots[0]--; 774 if (ret > 0) {
775 WARN_ON(1);
776 btrfs_free_path(path);
777 return -EIO;
778 }
1300 l = path->nodes[0]; 779 l = path->nodes[0];
1301 780
1302 btrfs_item_key_to_cpu(l, &key, path->slots[0]); 781 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
@@ -1310,21 +789,24 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1310 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); 789 BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY);
1311 790
1312 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); 791 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
792
1313 refs = btrfs_extent_refs(l, item); 793 refs = btrfs_extent_refs(l, item);
1314 btrfs_set_extent_refs(l, item, refs + 1); 794 btrfs_set_extent_refs(l, item, refs + refs_to_add);
795 btrfs_unlock_up_safe(path, 1);
796
1315 btrfs_mark_buffer_dirty(path->nodes[0]); 797 btrfs_mark_buffer_dirty(path->nodes[0]);
1316 798
1317 btrfs_release_path(root->fs_info->extent_root, path); 799 btrfs_release_path(root->fs_info->extent_root, path);
1318 800
1319 path->reada = 1; 801 path->reada = 1;
802 path->leave_spinning = 1;
803
804 /* now insert the actual backref */
1320 ret = insert_extent_backref(trans, root->fs_info->extent_root, 805 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1321 path, bytenr, parent, 806 path, bytenr, parent,
1322 ref_root, ref_generation, 807 ref_root, ref_generation,
1323 owner_objectid); 808 owner_objectid, refs_to_add);
1324 BUG_ON(ret); 809 BUG_ON(ret);
1325 finish_current_insert(trans, root->fs_info->extent_root, 0);
1326 del_pending_extents(trans, root->fs_info->extent_root, 0);
1327
1328 btrfs_free_path(path); 810 btrfs_free_path(path);
1329 return 0; 811 return 0;
1330} 812}
@@ -1339,68 +821,278 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1339 if (ref_root == BTRFS_TREE_LOG_OBJECTID && 821 if (ref_root == BTRFS_TREE_LOG_OBJECTID &&
1340 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) 822 owner_objectid < BTRFS_FIRST_FREE_OBJECTID)
1341 return 0; 823 return 0;
1342 ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent, 824
825 ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent,
1343 0, ref_root, 0, ref_generation, 826 0, ref_root, 0, ref_generation,
1344 owner_objectid); 827 owner_objectid);
1345 return ret; 828 return ret;
1346} 829}
1347 830
1348int btrfs_extent_post_op(struct btrfs_trans_handle *trans, 831static int drop_delayed_ref(struct btrfs_trans_handle *trans,
1349 struct btrfs_root *root) 832 struct btrfs_root *root,
833 struct btrfs_delayed_ref_node *node)
834{
835 int ret = 0;
836 struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node);
837
838 BUG_ON(node->ref_mod == 0);
839 ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes,
840 node->parent, ref->root, ref->generation,
841 ref->owner_objectid, ref->pin, node->ref_mod);
842
843 return ret;
844}
845
846/* helper function to actually process a single delayed ref entry */
847static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans,
848 struct btrfs_root *root,
849 struct btrfs_delayed_ref_node *node,
850 int insert_reserved)
1350{ 851{
1351 u64 start;
1352 u64 end;
1353 int ret; 852 int ret;
853 struct btrfs_delayed_ref *ref;
854
855 if (node->parent == (u64)-1) {
856 struct btrfs_delayed_ref_head *head;
857 /*
858 * we've hit the end of the chain and we were supposed
859 * to insert this extent into the tree. But, it got
860 * deleted before we ever needed to insert it, so all
861 * we have to do is clean up the accounting
862 */
863 if (insert_reserved) {
864 update_reserved_extents(root, node->bytenr,
865 node->num_bytes, 0);
866 }
867 head = btrfs_delayed_node_to_head(node);
868 mutex_unlock(&head->mutex);
869 return 0;
870 }
1354 871
1355 while(1) { 872 ref = btrfs_delayed_node_to_ref(node);
1356 finish_current_insert(trans, root->fs_info->extent_root, 1); 873 if (ref->action == BTRFS_ADD_DELAYED_REF) {
1357 del_pending_extents(trans, root->fs_info->extent_root, 1); 874 if (insert_reserved) {
875 struct btrfs_key ins;
1358 876
1359 /* is there more work to do? */ 877 ins.objectid = node->bytenr;
1360 ret = find_first_extent_bit(&root->fs_info->pending_del, 878 ins.offset = node->num_bytes;
1361 0, &start, &end, EXTENT_WRITEBACK); 879 ins.type = BTRFS_EXTENT_ITEM_KEY;
1362 if (!ret) 880
1363 continue; 881 /* record the full extent allocation */
1364 ret = find_first_extent_bit(&root->fs_info->extent_ins, 882 ret = __btrfs_alloc_reserved_extent(trans, root,
1365 0, &start, &end, EXTENT_WRITEBACK); 883 node->parent, ref->root,
1366 if (!ret) 884 ref->generation, ref->owner_objectid,
1367 continue; 885 &ins, node->ref_mod);
1368 break; 886 update_reserved_extents(root, node->bytenr,
887 node->num_bytes, 0);
888 } else {
889 /* just add one backref */
890 ret = add_extent_ref(trans, root, node->bytenr,
891 node->num_bytes,
892 node->parent, ref->root, ref->generation,
893 ref->owner_objectid, node->ref_mod);
894 }
895 BUG_ON(ret);
896 } else if (ref->action == BTRFS_DROP_DELAYED_REF) {
897 WARN_ON(insert_reserved);
898 ret = drop_delayed_ref(trans, root, node);
1369 } 899 }
1370 return 0; 900 return 0;
1371} 901}
1372 902
1373int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans, 903static noinline struct btrfs_delayed_ref_node *
1374 struct btrfs_root *root, u64 bytenr, 904select_delayed_ref(struct btrfs_delayed_ref_head *head)
1375 u64 num_bytes, u32 *refs)
1376{ 905{
1377 struct btrfs_path *path; 906 struct rb_node *node;
907 struct btrfs_delayed_ref_node *ref;
908 int action = BTRFS_ADD_DELAYED_REF;
909again:
910 /*
911 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
912 * this prevents ref count from going down to zero when
913 * there still are pending delayed ref.
914 */
915 node = rb_prev(&head->node.rb_node);
916 while (1) {
917 if (!node)
918 break;
919 ref = rb_entry(node, struct btrfs_delayed_ref_node,
920 rb_node);
921 if (ref->bytenr != head->node.bytenr)
922 break;
923 if (btrfs_delayed_node_to_ref(ref)->action == action)
924 return ref;
925 node = rb_prev(node);
926 }
927 if (action == BTRFS_ADD_DELAYED_REF) {
928 action = BTRFS_DROP_DELAYED_REF;
929 goto again;
930 }
931 return NULL;
932}
933
934static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
935 struct btrfs_root *root,
936 struct list_head *cluster)
937{
938 struct btrfs_delayed_ref_root *delayed_refs;
939 struct btrfs_delayed_ref_node *ref;
940 struct btrfs_delayed_ref_head *locked_ref = NULL;
1378 int ret; 941 int ret;
1379 struct btrfs_key key; 942 int count = 0;
1380 struct extent_buffer *l; 943 int must_insert_reserved = 0;
1381 struct btrfs_extent_item *item;
1382 944
1383 WARN_ON(num_bytes < root->sectorsize); 945 delayed_refs = &trans->transaction->delayed_refs;
1384 path = btrfs_alloc_path(); 946 while (1) {
1385 path->reada = 1; 947 if (!locked_ref) {
1386 key.objectid = bytenr; 948 /* pick a new head ref from the cluster list */
1387 key.offset = num_bytes; 949 if (list_empty(cluster))
1388 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); 950 break;
1389 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, 951
1390 0, 0); 952 locked_ref = list_entry(cluster->next,
1391 if (ret < 0) 953 struct btrfs_delayed_ref_head, cluster);
1392 goto out; 954
1393 if (ret != 0) { 955 /* grab the lock that says we are going to process
1394 btrfs_print_leaf(root, path->nodes[0]); 956 * all the refs for this head */
1395 printk(KERN_INFO "btrfs failed to find block number %llu\n", 957 ret = btrfs_delayed_ref_lock(trans, locked_ref);
1396 (unsigned long long)bytenr); 958
1397 BUG(); 959 /*
960 * we may have dropped the spin lock to get the head
961 * mutex lock, and that might have given someone else
962 * time to free the head. If that's true, it has been
963 * removed from our list and we can move on.
964 */
965 if (ret == -EAGAIN) {
966 locked_ref = NULL;
967 count++;
968 continue;
969 }
970 }
971
972 /*
973 * record the must insert reserved flag before we
974 * drop the spin lock.
975 */
976 must_insert_reserved = locked_ref->must_insert_reserved;
977 locked_ref->must_insert_reserved = 0;
978
979 /*
980 * locked_ref is the head node, so we have to go one
981 * node back for any delayed ref updates
982 */
983 ref = select_delayed_ref(locked_ref);
984 if (!ref) {
985 /* All delayed refs have been processed, Go ahead
986 * and send the head node to run_one_delayed_ref,
987 * so that any accounting fixes can happen
988 */
989 ref = &locked_ref->node;
990 list_del_init(&locked_ref->cluster);
991 locked_ref = NULL;
992 }
993
994 ref->in_tree = 0;
995 rb_erase(&ref->rb_node, &delayed_refs->root);
996 delayed_refs->num_entries--;
997 spin_unlock(&delayed_refs->lock);
998
999 ret = run_one_delayed_ref(trans, root, ref,
1000 must_insert_reserved);
1001 BUG_ON(ret);
1002 btrfs_put_delayed_ref(ref);
1003
1004 count++;
1005 cond_resched();
1006 spin_lock(&delayed_refs->lock);
1007 }
1008 return count;
1009}
1010
1011/*
1012 * this starts processing the delayed reference count updates and
1013 * extent insertions we have queued up so far. count can be
1014 * 0, which means to process everything in the tree at the start
1015 * of the run (but not newly added entries), or it can be some target
1016 * number you'd like to process.
1017 */
1018int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
1019 struct btrfs_root *root, unsigned long count)
1020{
1021 struct rb_node *node;
1022 struct btrfs_delayed_ref_root *delayed_refs;
1023 struct btrfs_delayed_ref_node *ref;
1024 struct list_head cluster;
1025 int ret;
1026 int run_all = count == (unsigned long)-1;
1027 int run_most = 0;
1028
1029 if (root == root->fs_info->extent_root)
1030 root = root->fs_info->tree_root;
1031
1032 delayed_refs = &trans->transaction->delayed_refs;
1033 INIT_LIST_HEAD(&cluster);
1034again:
1035 spin_lock(&delayed_refs->lock);
1036 if (count == 0) {
1037 count = delayed_refs->num_entries * 2;
1038 run_most = 1;
1039 }
1040 while (1) {
1041 if (!(run_all || run_most) &&
1042 delayed_refs->num_heads_ready < 64)
1043 break;
1044
1045 /*
1046 * go find something we can process in the rbtree. We start at
1047 * the beginning of the tree, and then build a cluster
1048 * of refs to process starting at the first one we are able to
1049 * lock
1050 */
1051 ret = btrfs_find_ref_cluster(trans, &cluster,
1052 delayed_refs->run_delayed_start);
1053 if (ret)
1054 break;
1055
1056 ret = run_clustered_refs(trans, root, &cluster);
1057 BUG_ON(ret < 0);
1058
1059 count -= min_t(unsigned long, ret, count);
1060
1061 if (count == 0)
1062 break;
1063 }
1064
1065 if (run_all) {
1066 node = rb_first(&delayed_refs->root);
1067 if (!node)
1068 goto out;
1069 count = (unsigned long)-1;
1070
1071 while (node) {
1072 ref = rb_entry(node, struct btrfs_delayed_ref_node,
1073 rb_node);
1074 if (btrfs_delayed_ref_is_head(ref)) {
1075 struct btrfs_delayed_ref_head *head;
1076
1077 head = btrfs_delayed_node_to_head(ref);
1078 atomic_inc(&ref->refs);
1079
1080 spin_unlock(&delayed_refs->lock);
1081 mutex_lock(&head->mutex);
1082 mutex_unlock(&head->mutex);
1083
1084 btrfs_put_delayed_ref(ref);
1085 cond_resched();
1086 goto again;
1087 }
1088 node = rb_next(node);
1089 }
1090 spin_unlock(&delayed_refs->lock);
1091 schedule_timeout(1);
1092 goto again;
1398 } 1093 }
1399 l = path->nodes[0];
1400 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1401 *refs = btrfs_extent_refs(l, item);
1402out: 1094out:
1403 btrfs_free_path(path); 1095 spin_unlock(&delayed_refs->lock);
1404 return 0; 1096 return 0;
1405} 1097}
1406 1098
@@ -1624,7 +1316,7 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1624 int refi = 0; 1316 int refi = 0;
1625 int slot; 1317 int slot;
1626 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, 1318 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
1627 u64, u64, u64, u64, u64, u64, u64, u64); 1319 u64, u64, u64, u64, u64, u64, u64, u64, u64);
1628 1320
1629 ref_root = btrfs_header_owner(buf); 1321 ref_root = btrfs_header_owner(buf);
1630 ref_generation = btrfs_header_generation(buf); 1322 ref_generation = btrfs_header_generation(buf);
@@ -1696,12 +1388,19 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1696 1388
1697 if (level == 0) { 1389 if (level == 0) {
1698 btrfs_item_key_to_cpu(buf, &key, slot); 1390 btrfs_item_key_to_cpu(buf, &key, slot);
1391 fi = btrfs_item_ptr(buf, slot,
1392 struct btrfs_file_extent_item);
1393
1394 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1395 if (bytenr == 0)
1396 continue;
1699 1397
1700 ret = process_func(trans, root, bytenr, 1398 ret = process_func(trans, root, bytenr,
1701 orig_buf->start, buf->start, 1399 btrfs_file_extent_disk_num_bytes(buf, fi),
1702 orig_root, ref_root, 1400 orig_buf->start, buf->start,
1703 orig_generation, ref_generation, 1401 orig_root, ref_root,
1704 key.objectid); 1402 orig_generation, ref_generation,
1403 key.objectid);
1705 1404
1706 if (ret) { 1405 if (ret) {
1707 faili = slot; 1406 faili = slot;
@@ -1709,7 +1408,7 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans,
1709 goto fail; 1408 goto fail;
1710 } 1409 }
1711 } else { 1410 } else {
1712 ret = process_func(trans, root, bytenr, 1411 ret = process_func(trans, root, bytenr, buf->len,
1713 orig_buf->start, buf->start, 1412 orig_buf->start, buf->start,
1714 orig_root, ref_root, 1413 orig_root, ref_root,
1715 orig_generation, ref_generation, 1414 orig_generation, ref_generation,
@@ -1786,17 +1485,17 @@ int btrfs_update_ref(struct btrfs_trans_handle *trans,
1786 if (bytenr == 0) 1485 if (bytenr == 0)
1787 continue; 1486 continue;
1788 ret = __btrfs_update_extent_ref(trans, root, bytenr, 1487 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1789 orig_buf->start, buf->start, 1488 btrfs_file_extent_disk_num_bytes(buf, fi),
1790 orig_root, ref_root, 1489 orig_buf->start, buf->start,
1791 orig_generation, ref_generation, 1490 orig_root, ref_root, orig_generation,
1792 key.objectid); 1491 ref_generation, key.objectid);
1793 if (ret) 1492 if (ret)
1794 goto fail; 1493 goto fail;
1795 } else { 1494 } else {
1796 bytenr = btrfs_node_blockptr(buf, slot); 1495 bytenr = btrfs_node_blockptr(buf, slot);
1797 ret = __btrfs_update_extent_ref(trans, root, bytenr, 1496 ret = __btrfs_update_extent_ref(trans, root, bytenr,
1798 orig_buf->start, buf->start, 1497 buf->len, orig_buf->start,
1799 orig_root, ref_root, 1498 buf->start, orig_root, ref_root,
1800 orig_generation, ref_generation, 1499 orig_generation, ref_generation,
1801 level - 1); 1500 level - 1);
1802 if (ret) 1501 if (ret)
@@ -1815,7 +1514,6 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
1815 struct btrfs_block_group_cache *cache) 1514 struct btrfs_block_group_cache *cache)
1816{ 1515{
1817 int ret; 1516 int ret;
1818 int pending_ret;
1819 struct btrfs_root *extent_root = root->fs_info->extent_root; 1517 struct btrfs_root *extent_root = root->fs_info->extent_root;
1820 unsigned long bi; 1518 unsigned long bi;
1821 struct extent_buffer *leaf; 1519 struct extent_buffer *leaf;
@@ -1831,12 +1529,8 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
1831 btrfs_mark_buffer_dirty(leaf); 1529 btrfs_mark_buffer_dirty(leaf);
1832 btrfs_release_path(extent_root, path); 1530 btrfs_release_path(extent_root, path);
1833fail: 1531fail:
1834 finish_current_insert(trans, extent_root, 0);
1835 pending_ret = del_pending_extents(trans, extent_root, 0);
1836 if (ret) 1532 if (ret)
1837 return ret; 1533 return ret;
1838 if (pending_ret)
1839 return pending_ret;
1840 return 0; 1534 return 0;
1841 1535
1842} 1536}
@@ -2361,6 +2055,8 @@ int btrfs_update_pinned_extents(struct btrfs_root *root,
2361 clear_extent_dirty(&fs_info->pinned_extents, 2055 clear_extent_dirty(&fs_info->pinned_extents,
2362 bytenr, bytenr + num - 1, GFP_NOFS); 2056 bytenr, bytenr + num - 1, GFP_NOFS);
2363 } 2057 }
2058 mutex_unlock(&root->fs_info->pinned_mutex);
2059
2364 while (num > 0) { 2060 while (num > 0) {
2365 cache = btrfs_lookup_block_group(fs_info, bytenr); 2061 cache = btrfs_lookup_block_group(fs_info, bytenr);
2366 BUG_ON(!cache); 2062 BUG_ON(!cache);
@@ -2452,8 +2148,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2452 u64 end; 2148 u64 end;
2453 int ret; 2149 int ret;
2454 2150
2455 mutex_lock(&root->fs_info->pinned_mutex);
2456 while (1) { 2151 while (1) {
2152 mutex_lock(&root->fs_info->pinned_mutex);
2457 ret = find_first_extent_bit(unpin, 0, &start, &end, 2153 ret = find_first_extent_bit(unpin, 0, &start, &end,
2458 EXTENT_DIRTY); 2154 EXTENT_DIRTY);
2459 if (ret) 2155 if (ret)
@@ -2461,209 +2157,21 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2461 2157
2462 ret = btrfs_discard_extent(root, start, end + 1 - start); 2158 ret = btrfs_discard_extent(root, start, end + 1 - start);
2463 2159
2160 /* unlocks the pinned mutex */
2464 btrfs_update_pinned_extents(root, start, end + 1 - start, 0); 2161 btrfs_update_pinned_extents(root, start, end + 1 - start, 0);
2465 clear_extent_dirty(unpin, start, end, GFP_NOFS); 2162 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2466 2163
2467 if (need_resched()) { 2164 cond_resched();
2468 mutex_unlock(&root->fs_info->pinned_mutex);
2469 cond_resched();
2470 mutex_lock(&root->fs_info->pinned_mutex);
2471 }
2472 } 2165 }
2473 mutex_unlock(&root->fs_info->pinned_mutex); 2166 mutex_unlock(&root->fs_info->pinned_mutex);
2474 return ret; 2167 return ret;
2475} 2168}
2476 2169
2477static int finish_current_insert(struct btrfs_trans_handle *trans,
2478 struct btrfs_root *extent_root, int all)
2479{
2480 u64 start;
2481 u64 end;
2482 u64 priv;
2483 u64 search = 0;
2484 struct btrfs_fs_info *info = extent_root->fs_info;
2485 struct btrfs_path *path;
2486 struct pending_extent_op *extent_op, *tmp;
2487 struct list_head insert_list, update_list;
2488 int ret;
2489 int num_inserts = 0, max_inserts, restart = 0;
2490
2491 path = btrfs_alloc_path();
2492 INIT_LIST_HEAD(&insert_list);
2493 INIT_LIST_HEAD(&update_list);
2494
2495 max_inserts = extent_root->leafsize /
2496 (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) +
2497 sizeof(struct btrfs_extent_ref) +
2498 sizeof(struct btrfs_extent_item));
2499again:
2500 mutex_lock(&info->extent_ins_mutex);
2501 while (1) {
2502 ret = find_first_extent_bit(&info->extent_ins, search, &start,
2503 &end, EXTENT_WRITEBACK);
2504 if (ret) {
2505 if (restart && !num_inserts &&
2506 list_empty(&update_list)) {
2507 restart = 0;
2508 search = 0;
2509 continue;
2510 }
2511 break;
2512 }
2513
2514 ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS);
2515 if (!ret) {
2516 if (all)
2517 restart = 1;
2518 search = end + 1;
2519 if (need_resched()) {
2520 mutex_unlock(&info->extent_ins_mutex);
2521 cond_resched();
2522 mutex_lock(&info->extent_ins_mutex);
2523 }
2524 continue;
2525 }
2526
2527 ret = get_state_private(&info->extent_ins, start, &priv);
2528 BUG_ON(ret);
2529 extent_op = (struct pending_extent_op *)(unsigned long) priv;
2530
2531 if (extent_op->type == PENDING_EXTENT_INSERT) {
2532 num_inserts++;
2533 list_add_tail(&extent_op->list, &insert_list);
2534 search = end + 1;
2535 if (num_inserts == max_inserts) {
2536 restart = 1;
2537 break;
2538 }
2539 } else if (extent_op->type == PENDING_BACKREF_UPDATE) {
2540 list_add_tail(&extent_op->list, &update_list);
2541 search = end + 1;
2542 } else {
2543 BUG();
2544 }
2545 }
2546
2547 /*
2548 * process the update list, clear the writeback bit for it, and if
2549 * somebody marked this thing for deletion then just unlock it and be
2550 * done, the free_extents will handle it
2551 */
2552 list_for_each_entry_safe(extent_op, tmp, &update_list, list) {
2553 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2554 extent_op->bytenr + extent_op->num_bytes - 1,
2555 EXTENT_WRITEBACK, GFP_NOFS);
2556 if (extent_op->del) {
2557 list_del_init(&extent_op->list);
2558 unlock_extent(&info->extent_ins, extent_op->bytenr,
2559 extent_op->bytenr + extent_op->num_bytes
2560 - 1, GFP_NOFS);
2561 kfree(extent_op);
2562 }
2563 }
2564 mutex_unlock(&info->extent_ins_mutex);
2565
2566 /*
2567 * still have things left on the update list, go ahead an update
2568 * everything
2569 */
2570 if (!list_empty(&update_list)) {
2571 ret = update_backrefs(trans, extent_root, path, &update_list);
2572 BUG_ON(ret);
2573
2574 /* we may have COW'ed new blocks, so lets start over */
2575 if (all)
2576 restart = 1;
2577 }
2578
2579 /*
2580 * if no inserts need to be done, but we skipped some extents and we
2581 * need to make sure everything is cleaned then reset everything and
2582 * go back to the beginning
2583 */
2584 if (!num_inserts && restart) {
2585 search = 0;
2586 restart = 0;
2587 INIT_LIST_HEAD(&update_list);
2588 INIT_LIST_HEAD(&insert_list);
2589 goto again;
2590 } else if (!num_inserts) {
2591 goto out;
2592 }
2593
2594 /*
2595 * process the insert extents list. Again if we are deleting this
2596 * extent, then just unlock it, pin down the bytes if need be, and be
2597 * done with it. Saves us from having to actually insert the extent
2598 * into the tree and then subsequently come along and delete it
2599 */
2600 mutex_lock(&info->extent_ins_mutex);
2601 list_for_each_entry_safe(extent_op, tmp, &insert_list, list) {
2602 clear_extent_bits(&info->extent_ins, extent_op->bytenr,
2603 extent_op->bytenr + extent_op->num_bytes - 1,
2604 EXTENT_WRITEBACK, GFP_NOFS);
2605 if (extent_op->del) {
2606 u64 used;
2607 list_del_init(&extent_op->list);
2608 unlock_extent(&info->extent_ins, extent_op->bytenr,
2609 extent_op->bytenr + extent_op->num_bytes
2610 - 1, GFP_NOFS);
2611
2612 mutex_lock(&extent_root->fs_info->pinned_mutex);
2613 ret = pin_down_bytes(trans, extent_root,
2614 extent_op->bytenr,
2615 extent_op->num_bytes, 0);
2616 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2617
2618 spin_lock(&info->delalloc_lock);
2619 used = btrfs_super_bytes_used(&info->super_copy);
2620 btrfs_set_super_bytes_used(&info->super_copy,
2621 used - extent_op->num_bytes);
2622 used = btrfs_root_used(&extent_root->root_item);
2623 btrfs_set_root_used(&extent_root->root_item,
2624 used - extent_op->num_bytes);
2625 spin_unlock(&info->delalloc_lock);
2626
2627 ret = update_block_group(trans, extent_root,
2628 extent_op->bytenr,
2629 extent_op->num_bytes,
2630 0, ret > 0);
2631 BUG_ON(ret);
2632 kfree(extent_op);
2633 num_inserts--;
2634 }
2635 }
2636 mutex_unlock(&info->extent_ins_mutex);
2637
2638 ret = insert_extents(trans, extent_root, path, &insert_list,
2639 num_inserts);
2640 BUG_ON(ret);
2641
2642 /*
2643 * if restart is set for whatever reason we need to go back and start
2644 * searching through the pending list again.
2645 *
2646 * We just inserted some extents, which could have resulted in new
2647 * blocks being allocated, which would result in new blocks needing
2648 * updates, so if all is set we _must_ restart to get the updated
2649 * blocks.
2650 */
2651 if (restart || all) {
2652 INIT_LIST_HEAD(&insert_list);
2653 INIT_LIST_HEAD(&update_list);
2654 search = 0;
2655 restart = 0;
2656 num_inserts = 0;
2657 goto again;
2658 }
2659out:
2660 btrfs_free_path(path);
2661 return 0;
2662}
2663
2664static int pin_down_bytes(struct btrfs_trans_handle *trans, 2170static int pin_down_bytes(struct btrfs_trans_handle *trans,
2665 struct btrfs_root *root, 2171 struct btrfs_root *root,
2666 u64 bytenr, u64 num_bytes, int is_data) 2172 struct btrfs_path *path,
2173 u64 bytenr, u64 num_bytes, int is_data,
2174 struct extent_buffer **must_clean)
2667{ 2175{
2668 int err = 0; 2176 int err = 0;
2669 struct extent_buffer *buf; 2177 struct extent_buffer *buf;
@@ -2686,17 +2194,19 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans,
2686 u64 header_transid = btrfs_header_generation(buf); 2194 u64 header_transid = btrfs_header_generation(buf);
2687 if (header_owner != BTRFS_TREE_LOG_OBJECTID && 2195 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2688 header_owner != BTRFS_TREE_RELOC_OBJECTID && 2196 header_owner != BTRFS_TREE_RELOC_OBJECTID &&
2197 header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2689 header_transid == trans->transid && 2198 header_transid == trans->transid &&
2690 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { 2199 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2691 clean_tree_block(NULL, root, buf); 2200 *must_clean = buf;
2692 btrfs_tree_unlock(buf);
2693 free_extent_buffer(buf);
2694 return 1; 2201 return 1;
2695 } 2202 }
2696 btrfs_tree_unlock(buf); 2203 btrfs_tree_unlock(buf);
2697 } 2204 }
2698 free_extent_buffer(buf); 2205 free_extent_buffer(buf);
2699pinit: 2206pinit:
2207 btrfs_set_path_blocking(path);
2208 mutex_lock(&root->fs_info->pinned_mutex);
2209 /* unlocks the pinned mutex */
2700 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); 2210 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2701 2211
2702 BUG_ON(err < 0); 2212 BUG_ON(err < 0);
@@ -2710,7 +2220,8 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2710 struct btrfs_root *root, 2220 struct btrfs_root *root,
2711 u64 bytenr, u64 num_bytes, u64 parent, 2221 u64 bytenr, u64 num_bytes, u64 parent,
2712 u64 root_objectid, u64 ref_generation, 2222 u64 root_objectid, u64 ref_generation,
2713 u64 owner_objectid, int pin, int mark_free) 2223 u64 owner_objectid, int pin, int mark_free,
2224 int refs_to_drop)
2714{ 2225{
2715 struct btrfs_path *path; 2226 struct btrfs_path *path;
2716 struct btrfs_key key; 2227 struct btrfs_key key;
@@ -2732,6 +2243,7 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2732 return -ENOMEM; 2243 return -ENOMEM;
2733 2244
2734 path->reada = 1; 2245 path->reada = 1;
2246 path->leave_spinning = 1;
2735 ret = lookup_extent_backref(trans, extent_root, path, 2247 ret = lookup_extent_backref(trans, extent_root, path,
2736 bytenr, parent, root_objectid, 2248 bytenr, parent, root_objectid,
2737 ref_generation, owner_objectid, 1); 2249 ref_generation, owner_objectid, 1);
@@ -2753,9 +2265,11 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2753 break; 2265 break;
2754 } 2266 }
2755 if (!found_extent) { 2267 if (!found_extent) {
2756 ret = remove_extent_backref(trans, extent_root, path); 2268 ret = remove_extent_backref(trans, extent_root, path,
2269 refs_to_drop);
2757 BUG_ON(ret); 2270 BUG_ON(ret);
2758 btrfs_release_path(extent_root, path); 2271 btrfs_release_path(extent_root, path);
2272 path->leave_spinning = 1;
2759 ret = btrfs_search_slot(trans, extent_root, 2273 ret = btrfs_search_slot(trans, extent_root,
2760 &key, path, -1, 1); 2274 &key, path, -1, 1);
2761 if (ret) { 2275 if (ret) {
@@ -2771,8 +2285,9 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2771 btrfs_print_leaf(extent_root, path->nodes[0]); 2285 btrfs_print_leaf(extent_root, path->nodes[0]);
2772 WARN_ON(1); 2286 WARN_ON(1);
2773 printk(KERN_ERR "btrfs unable to find ref byte nr %llu " 2287 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2774 "root %llu gen %llu owner %llu\n", 2288 "parent %llu root %llu gen %llu owner %llu\n",
2775 (unsigned long long)bytenr, 2289 (unsigned long long)bytenr,
2290 (unsigned long long)parent,
2776 (unsigned long long)root_objectid, 2291 (unsigned long long)root_objectid,
2777 (unsigned long long)ref_generation, 2292 (unsigned long long)ref_generation,
2778 (unsigned long long)owner_objectid); 2293 (unsigned long long)owner_objectid);
@@ -2782,17 +2297,23 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2782 ei = btrfs_item_ptr(leaf, extent_slot, 2297 ei = btrfs_item_ptr(leaf, extent_slot,
2783 struct btrfs_extent_item); 2298 struct btrfs_extent_item);
2784 refs = btrfs_extent_refs(leaf, ei); 2299 refs = btrfs_extent_refs(leaf, ei);
2785 BUG_ON(refs == 0);
2786 refs -= 1;
2787 btrfs_set_extent_refs(leaf, ei, refs);
2788 2300
2301 /*
2302 * we're not allowed to delete the extent item if there
2303 * are other delayed ref updates pending
2304 */
2305
2306 BUG_ON(refs < refs_to_drop);
2307 refs -= refs_to_drop;
2308 btrfs_set_extent_refs(leaf, ei, refs);
2789 btrfs_mark_buffer_dirty(leaf); 2309 btrfs_mark_buffer_dirty(leaf);
2790 2310
2791 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { 2311 if (refs == 0 && found_extent &&
2312 path->slots[0] == extent_slot + 1) {
2792 struct btrfs_extent_ref *ref; 2313 struct btrfs_extent_ref *ref;
2793 ref = btrfs_item_ptr(leaf, path->slots[0], 2314 ref = btrfs_item_ptr(leaf, path->slots[0],
2794 struct btrfs_extent_ref); 2315 struct btrfs_extent_ref);
2795 BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1); 2316 BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop);
2796 /* if the back ref and the extent are next to each other 2317 /* if the back ref and the extent are next to each other
2797 * they get deleted below in one shot 2318 * they get deleted below in one shot
2798 */ 2319 */
@@ -2800,11 +2321,13 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2800 num_to_del = 2; 2321 num_to_del = 2;
2801 } else if (found_extent) { 2322 } else if (found_extent) {
2802 /* otherwise delete the extent back ref */ 2323 /* otherwise delete the extent back ref */
2803 ret = remove_extent_backref(trans, extent_root, path); 2324 ret = remove_extent_backref(trans, extent_root, path,
2325 refs_to_drop);
2804 BUG_ON(ret); 2326 BUG_ON(ret);
2805 /* if refs are 0, we need to setup the path for deletion */ 2327 /* if refs are 0, we need to setup the path for deletion */
2806 if (refs == 0) { 2328 if (refs == 0) {
2807 btrfs_release_path(extent_root, path); 2329 btrfs_release_path(extent_root, path);
2330 path->leave_spinning = 1;
2808 ret = btrfs_search_slot(trans, extent_root, &key, path, 2331 ret = btrfs_search_slot(trans, extent_root, &key, path,
2809 -1, 1); 2332 -1, 1);
2810 BUG_ON(ret); 2333 BUG_ON(ret);
@@ -2814,16 +2337,18 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2814 if (refs == 0) { 2337 if (refs == 0) {
2815 u64 super_used; 2338 u64 super_used;
2816 u64 root_used; 2339 u64 root_used;
2340 struct extent_buffer *must_clean = NULL;
2817 2341
2818 if (pin) { 2342 if (pin) {
2819 mutex_lock(&root->fs_info->pinned_mutex); 2343 ret = pin_down_bytes(trans, root, path,
2820 ret = pin_down_bytes(trans, root, bytenr, num_bytes, 2344 bytenr, num_bytes,
2821 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); 2345 owner_objectid >= BTRFS_FIRST_FREE_OBJECTID,
2822 mutex_unlock(&root->fs_info->pinned_mutex); 2346 &must_clean);
2823 if (ret > 0) 2347 if (ret > 0)
2824 mark_free = 1; 2348 mark_free = 1;
2825 BUG_ON(ret < 0); 2349 BUG_ON(ret < 0);
2826 } 2350 }
2351
2827 /* block accounting for super block */ 2352 /* block accounting for super block */
2828 spin_lock(&info->delalloc_lock); 2353 spin_lock(&info->delalloc_lock);
2829 super_used = btrfs_super_bytes_used(&info->super_copy); 2354 super_used = btrfs_super_bytes_used(&info->super_copy);
@@ -2835,14 +2360,34 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2835 btrfs_set_root_used(&root->root_item, 2360 btrfs_set_root_used(&root->root_item,
2836 root_used - num_bytes); 2361 root_used - num_bytes);
2837 spin_unlock(&info->delalloc_lock); 2362 spin_unlock(&info->delalloc_lock);
2363
2364 /*
2365 * it is going to be very rare for someone to be waiting
2366 * on the block we're freeing. del_items might need to
2367 * schedule, so rather than get fancy, just force it
2368 * to blocking here
2369 */
2370 if (must_clean)
2371 btrfs_set_lock_blocking(must_clean);
2372
2838 ret = btrfs_del_items(trans, extent_root, path, path->slots[0], 2373 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2839 num_to_del); 2374 num_to_del);
2840 BUG_ON(ret); 2375 BUG_ON(ret);
2841 btrfs_release_path(extent_root, path); 2376 btrfs_release_path(extent_root, path);
2842 2377
2378 if (must_clean) {
2379 clean_tree_block(NULL, root, must_clean);
2380 btrfs_tree_unlock(must_clean);
2381 free_extent_buffer(must_clean);
2382 }
2383
2843 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { 2384 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2844 ret = btrfs_del_csums(trans, root, bytenr, num_bytes); 2385 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2845 BUG_ON(ret); 2386 BUG_ON(ret);
2387 } else {
2388 invalidate_mapping_pages(info->btree_inode->i_mapping,
2389 bytenr >> PAGE_CACHE_SHIFT,
2390 (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
2846 } 2391 }
2847 2392
2848 ret = update_block_group(trans, root, bytenr, num_bytes, 0, 2393 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
@@ -2850,218 +2395,103 @@ static int __free_extent(struct btrfs_trans_handle *trans,
2850 BUG_ON(ret); 2395 BUG_ON(ret);
2851 } 2396 }
2852 btrfs_free_path(path); 2397 btrfs_free_path(path);
2853 finish_current_insert(trans, extent_root, 0);
2854 return ret; 2398 return ret;
2855} 2399}
2856 2400
2857/* 2401/*
2858 * find all the blocks marked as pending in the radix tree and remove 2402 * remove an extent from the root, returns 0 on success
2859 * them from the extent map
2860 */ 2403 */
2861static int del_pending_extents(struct btrfs_trans_handle *trans, 2404static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
2862 struct btrfs_root *extent_root, int all) 2405 struct btrfs_root *root,
2406 u64 bytenr, u64 num_bytes, u64 parent,
2407 u64 root_objectid, u64 ref_generation,
2408 u64 owner_objectid, int pin,
2409 int refs_to_drop)
2863{ 2410{
2864 int ret; 2411 WARN_ON(num_bytes < root->sectorsize);
2865 int err = 0;
2866 u64 start;
2867 u64 end;
2868 u64 priv;
2869 u64 search = 0;
2870 int nr = 0, skipped = 0;
2871 struct extent_io_tree *pending_del;
2872 struct extent_io_tree *extent_ins;
2873 struct pending_extent_op *extent_op;
2874 struct btrfs_fs_info *info = extent_root->fs_info;
2875 struct list_head delete_list;
2876
2877 INIT_LIST_HEAD(&delete_list);
2878 extent_ins = &extent_root->fs_info->extent_ins;
2879 pending_del = &extent_root->fs_info->pending_del;
2880
2881again:
2882 mutex_lock(&info->extent_ins_mutex);
2883 while (1) {
2884 ret = find_first_extent_bit(pending_del, search, &start, &end,
2885 EXTENT_WRITEBACK);
2886 if (ret) {
2887 if (all && skipped && !nr) {
2888 search = 0;
2889 skipped = 0;
2890 continue;
2891 }
2892 mutex_unlock(&info->extent_ins_mutex);
2893 break;
2894 }
2895
2896 ret = try_lock_extent(extent_ins, start, end, GFP_NOFS);
2897 if (!ret) {
2898 search = end+1;
2899 skipped = 1;
2900
2901 if (need_resched()) {
2902 mutex_unlock(&info->extent_ins_mutex);
2903 cond_resched();
2904 mutex_lock(&info->extent_ins_mutex);
2905 }
2906
2907 continue;
2908 }
2909 BUG_ON(ret < 0);
2910
2911 ret = get_state_private(pending_del, start, &priv);
2912 BUG_ON(ret);
2913 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2914
2915 clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK,
2916 GFP_NOFS);
2917 if (!test_range_bit(extent_ins, start, end,
2918 EXTENT_WRITEBACK, 0)) {
2919 list_add_tail(&extent_op->list, &delete_list);
2920 nr++;
2921 } else {
2922 kfree(extent_op);
2923
2924 ret = get_state_private(&info->extent_ins, start,
2925 &priv);
2926 BUG_ON(ret);
2927 extent_op = (struct pending_extent_op *)
2928 (unsigned long)priv;
2929
2930 clear_extent_bits(&info->extent_ins, start, end,
2931 EXTENT_WRITEBACK, GFP_NOFS);
2932
2933 if (extent_op->type == PENDING_BACKREF_UPDATE) {
2934 list_add_tail(&extent_op->list, &delete_list);
2935 search = end + 1;
2936 nr++;
2937 continue;
2938 }
2939
2940 mutex_lock(&extent_root->fs_info->pinned_mutex);
2941 ret = pin_down_bytes(trans, extent_root, start,
2942 end + 1 - start, 0);
2943 mutex_unlock(&extent_root->fs_info->pinned_mutex);
2944
2945 ret = update_block_group(trans, extent_root, start,
2946 end + 1 - start, 0, ret > 0);
2947
2948 unlock_extent(extent_ins, start, end, GFP_NOFS);
2949 BUG_ON(ret);
2950 kfree(extent_op);
2951 }
2952 if (ret)
2953 err = ret;
2954
2955 search = end + 1;
2956
2957 if (need_resched()) {
2958 mutex_unlock(&info->extent_ins_mutex);
2959 cond_resched();
2960 mutex_lock(&info->extent_ins_mutex);
2961 }
2962 }
2963 2412
2964 if (nr) { 2413 /*
2965 ret = free_extents(trans, extent_root, &delete_list); 2414 * if metadata always pin
2966 BUG_ON(ret); 2415 * if data pin when any transaction has committed this
2967 } 2416 */
2417 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID ||
2418 ref_generation != trans->transid)
2419 pin = 1;
2968 2420
2969 if (all && skipped) { 2421 if (ref_generation != trans->transid)
2970 INIT_LIST_HEAD(&delete_list); 2422 pin = 1;
2971 search = 0;
2972 nr = 0;
2973 goto again;
2974 }
2975 2423
2976 if (!err) 2424 return __free_extent(trans, root, bytenr, num_bytes, parent,
2977 finish_current_insert(trans, extent_root, 0); 2425 root_objectid, ref_generation,
2978 return err; 2426 owner_objectid, pin, pin == 0, refs_to_drop);
2979} 2427}
2980 2428
2981/* 2429/*
2982 * remove an extent from the root, returns 0 on success 2430 * when we free an extent, it is possible (and likely) that we free the last
2431 * delayed ref for that extent as well. This searches the delayed ref tree for
2432 * a given extent, and if there are no other delayed refs to be processed, it
2433 * removes it from the tree.
2983 */ 2434 */
2984static int __btrfs_free_extent(struct btrfs_trans_handle *trans, 2435static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
2985 struct btrfs_root *root, 2436 struct btrfs_root *root, u64 bytenr)
2986 u64 bytenr, u64 num_bytes, u64 parent,
2987 u64 root_objectid, u64 ref_generation,
2988 u64 owner_objectid, int pin)
2989{ 2437{
2990 struct btrfs_root *extent_root = root->fs_info->extent_root; 2438 struct btrfs_delayed_ref_head *head;
2991 int pending_ret; 2439 struct btrfs_delayed_ref_root *delayed_refs;
2440 struct btrfs_delayed_ref_node *ref;
2441 struct rb_node *node;
2992 int ret; 2442 int ret;
2993 2443
2994 WARN_ON(num_bytes < root->sectorsize); 2444 delayed_refs = &trans->transaction->delayed_refs;
2995 if (root == extent_root) { 2445 spin_lock(&delayed_refs->lock);
2996 struct pending_extent_op *extent_op = NULL; 2446 head = btrfs_find_delayed_ref_head(trans, bytenr);
2997 2447 if (!head)
2998 mutex_lock(&root->fs_info->extent_ins_mutex); 2448 goto out;
2999 if (test_range_bit(&root->fs_info->extent_ins, bytenr,
3000 bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) {
3001 u64 priv;
3002 ret = get_state_private(&root->fs_info->extent_ins,
3003 bytenr, &priv);
3004 BUG_ON(ret);
3005 extent_op = (struct pending_extent_op *)
3006 (unsigned long)priv;
3007 2449
3008 extent_op->del = 1; 2450 node = rb_prev(&head->node.rb_node);
3009 if (extent_op->type == PENDING_EXTENT_INSERT) { 2451 if (!node)
3010 mutex_unlock(&root->fs_info->extent_ins_mutex); 2452 goto out;
3011 return 0;
3012 }
3013 }
3014 2453
3015 if (extent_op) { 2454 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3016 ref_generation = extent_op->orig_generation;
3017 parent = extent_op->orig_parent;
3018 }
3019 2455
3020 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); 2456 /* there are still entries for this ref, we can't drop it */
3021 BUG_ON(!extent_op); 2457 if (ref->bytenr == bytenr)
3022 2458 goto out;
3023 extent_op->type = PENDING_EXTENT_DELETE;
3024 extent_op->bytenr = bytenr;
3025 extent_op->num_bytes = num_bytes;
3026 extent_op->parent = parent;
3027 extent_op->orig_parent = parent;
3028 extent_op->generation = ref_generation;
3029 extent_op->orig_generation = ref_generation;
3030 extent_op->level = (int)owner_objectid;
3031 INIT_LIST_HEAD(&extent_op->list);
3032 extent_op->del = 0;
3033
3034 set_extent_bits(&root->fs_info->pending_del,
3035 bytenr, bytenr + num_bytes - 1,
3036 EXTENT_WRITEBACK, GFP_NOFS);
3037 set_state_private(&root->fs_info->pending_del,
3038 bytenr, (unsigned long)extent_op);
3039 mutex_unlock(&root->fs_info->extent_ins_mutex);
3040 return 0;
3041 }
3042 /* if metadata always pin */
3043 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
3044 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
3045 mutex_lock(&root->fs_info->pinned_mutex);
3046 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
3047 mutex_unlock(&root->fs_info->pinned_mutex);
3048 update_reserved_extents(root, bytenr, num_bytes, 0);
3049 return 0;
3050 }
3051 pin = 1;
3052 }
3053 2459
3054 /* if data pin when any transaction has committed this */ 2460 /*
3055 if (ref_generation != trans->transid) 2461 * waiting for the lock here would deadlock. If someone else has it
3056 pin = 1; 2462 * locked they are already in the process of dropping it anyway
2463 */
2464 if (!mutex_trylock(&head->mutex))
2465 goto out;
3057 2466
3058 ret = __free_extent(trans, root, bytenr, num_bytes, parent, 2467 /*
3059 root_objectid, ref_generation, 2468 * at this point we have a head with no other entries. Go
3060 owner_objectid, pin, pin == 0); 2469 * ahead and process it.
2470 */
2471 head->node.in_tree = 0;
2472 rb_erase(&head->node.rb_node, &delayed_refs->root);
3061 2473
3062 finish_current_insert(trans, root->fs_info->extent_root, 0); 2474 delayed_refs->num_entries--;
3063 pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0); 2475
3064 return ret ? ret : pending_ret; 2476 /*
2477 * we don't take a ref on the node because we're removing it from the
2478 * tree, so we just steal the ref the tree was holding.
2479 */
2480 delayed_refs->num_heads--;
2481 if (list_empty(&head->cluster))
2482 delayed_refs->num_heads_ready--;
2483
2484 list_del_init(&head->cluster);
2485 spin_unlock(&delayed_refs->lock);
2486
2487 ret = run_one_delayed_ref(trans, root->fs_info->tree_root,
2488 &head->node, head->must_insert_reserved);
2489 BUG_ON(ret);
2490 btrfs_put_delayed_ref(&head->node);
2491 return 0;
2492out:
2493 spin_unlock(&delayed_refs->lock);
2494 return 0;
3065} 2495}
3066 2496
3067int btrfs_free_extent(struct btrfs_trans_handle *trans, 2497int btrfs_free_extent(struct btrfs_trans_handle *trans,
@@ -3072,9 +2502,30 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
3072{ 2502{
3073 int ret; 2503 int ret;
3074 2504
3075 ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent, 2505 /*
3076 root_objectid, ref_generation, 2506 * tree log blocks never actually go into the extent allocation
3077 owner_objectid, pin); 2507 * tree, just update pinning info and exit early.
2508 *
2509 * data extents referenced by the tree log do need to have
2510 * their reference counts bumped.
2511 */
2512 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID &&
2513 owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2514 mutex_lock(&root->fs_info->pinned_mutex);
2515
2516 /* unlocks the pinned mutex */
2517 btrfs_update_pinned_extents(root, bytenr, num_bytes, 1);
2518 update_reserved_extents(root, bytenr, num_bytes, 0);
2519 ret = 0;
2520 } else {
2521 ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent,
2522 root_objectid, ref_generation,
2523 owner_objectid,
2524 BTRFS_DROP_DELAYED_REF, 1);
2525 BUG_ON(ret);
2526 ret = check_ref_cleanup(trans, root, bytenr);
2527 BUG_ON(ret);
2528 }
3078 return ret; 2529 return ret;
3079} 2530}
3080 2531
@@ -3475,10 +2926,10 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
3475static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, 2926static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3476 struct btrfs_root *root, u64 parent, 2927 struct btrfs_root *root, u64 parent,
3477 u64 root_objectid, u64 ref_generation, 2928 u64 root_objectid, u64 ref_generation,
3478 u64 owner, struct btrfs_key *ins) 2929 u64 owner, struct btrfs_key *ins,
2930 int ref_mod)
3479{ 2931{
3480 int ret; 2932 int ret;
3481 int pending_ret;
3482 u64 super_used; 2933 u64 super_used;
3483 u64 root_used; 2934 u64 root_used;
3484 u64 num_bytes = ins->offset; 2935 u64 num_bytes = ins->offset;
@@ -3503,33 +2954,6 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3503 btrfs_set_root_used(&root->root_item, root_used + num_bytes); 2954 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
3504 spin_unlock(&info->delalloc_lock); 2955 spin_unlock(&info->delalloc_lock);
3505 2956
3506 if (root == extent_root) {
3507 struct pending_extent_op *extent_op;
3508
3509 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
3510 BUG_ON(!extent_op);
3511
3512 extent_op->type = PENDING_EXTENT_INSERT;
3513 extent_op->bytenr = ins->objectid;
3514 extent_op->num_bytes = ins->offset;
3515 extent_op->parent = parent;
3516 extent_op->orig_parent = 0;
3517 extent_op->generation = ref_generation;
3518 extent_op->orig_generation = 0;
3519 extent_op->level = (int)owner;
3520 INIT_LIST_HEAD(&extent_op->list);
3521 extent_op->del = 0;
3522
3523 mutex_lock(&root->fs_info->extent_ins_mutex);
3524 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
3525 ins->objectid + ins->offset - 1,
3526 EXTENT_WRITEBACK, GFP_NOFS);
3527 set_state_private(&root->fs_info->extent_ins,
3528 ins->objectid, (unsigned long)extent_op);
3529 mutex_unlock(&root->fs_info->extent_ins_mutex);
3530 goto update_block;
3531 }
3532
3533 memcpy(&keys[0], ins, sizeof(*ins)); 2957 memcpy(&keys[0], ins, sizeof(*ins));
3534 keys[1].objectid = ins->objectid; 2958 keys[1].objectid = ins->objectid;
3535 keys[1].type = BTRFS_EXTENT_REF_KEY; 2959 keys[1].type = BTRFS_EXTENT_REF_KEY;
@@ -3540,37 +2964,31 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3540 path = btrfs_alloc_path(); 2964 path = btrfs_alloc_path();
3541 BUG_ON(!path); 2965 BUG_ON(!path);
3542 2966
2967 path->leave_spinning = 1;
3543 ret = btrfs_insert_empty_items(trans, extent_root, path, keys, 2968 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
3544 sizes, 2); 2969 sizes, 2);
3545 BUG_ON(ret); 2970 BUG_ON(ret);
3546 2971
3547 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], 2972 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3548 struct btrfs_extent_item); 2973 struct btrfs_extent_item);
3549 btrfs_set_extent_refs(path->nodes[0], extent_item, 1); 2974 btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod);
3550 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, 2975 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3551 struct btrfs_extent_ref); 2976 struct btrfs_extent_ref);
3552 2977
3553 btrfs_set_ref_root(path->nodes[0], ref, root_objectid); 2978 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
3554 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); 2979 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
3555 btrfs_set_ref_objectid(path->nodes[0], ref, owner); 2980 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
3556 btrfs_set_ref_num_refs(path->nodes[0], ref, 1); 2981 btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod);
3557 2982
3558 btrfs_mark_buffer_dirty(path->nodes[0]); 2983 btrfs_mark_buffer_dirty(path->nodes[0]);
3559 2984
3560 trans->alloc_exclude_start = 0; 2985 trans->alloc_exclude_start = 0;
3561 trans->alloc_exclude_nr = 0; 2986 trans->alloc_exclude_nr = 0;
3562 btrfs_free_path(path); 2987 btrfs_free_path(path);
3563 finish_current_insert(trans, extent_root, 0);
3564 pending_ret = del_pending_extents(trans, extent_root, 0);
3565 2988
3566 if (ret) 2989 if (ret)
3567 goto out; 2990 goto out;
3568 if (pending_ret) {
3569 ret = pending_ret;
3570 goto out;
3571 }
3572 2991
3573update_block:
3574 ret = update_block_group(trans, root, ins->objectid, 2992 ret = update_block_group(trans, root, ins->objectid,
3575 ins->offset, 1, 0); 2993 ins->offset, 1, 0);
3576 if (ret) { 2994 if (ret) {
@@ -3592,9 +3010,12 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans,
3592 3010
3593 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) 3011 if (root_objectid == BTRFS_TREE_LOG_OBJECTID)
3594 return 0; 3012 return 0;
3595 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, 3013
3596 ref_generation, owner, ins); 3014 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3597 update_reserved_extents(root, ins->objectid, ins->offset, 0); 3015 ins->offset, parent, root_objectid,
3016 ref_generation, owner,
3017 BTRFS_ADD_DELAYED_EXTENT, 0);
3018 BUG_ON(ret);
3598 return ret; 3019 return ret;
3599} 3020}
3600 3021
@@ -3621,7 +3042,7 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans,
3621 BUG_ON(ret); 3042 BUG_ON(ret);
3622 put_block_group(block_group); 3043 put_block_group(block_group);
3623 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, 3044 ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid,
3624 ref_generation, owner, ins); 3045 ref_generation, owner, ins, 1);
3625 return ret; 3046 return ret;
3626} 3047}
3627 3048
@@ -3640,20 +3061,18 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
3640 u64 search_end, struct btrfs_key *ins, u64 data) 3061 u64 search_end, struct btrfs_key *ins, u64 data)
3641{ 3062{
3642 int ret; 3063 int ret;
3643
3644 ret = __btrfs_reserve_extent(trans, root, num_bytes, 3064 ret = __btrfs_reserve_extent(trans, root, num_bytes,
3645 min_alloc_size, empty_size, hint_byte, 3065 min_alloc_size, empty_size, hint_byte,
3646 search_end, ins, data); 3066 search_end, ins, data);
3647 BUG_ON(ret); 3067 BUG_ON(ret);
3648 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { 3068 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
3649 ret = __btrfs_alloc_reserved_extent(trans, root, parent, 3069 ret = btrfs_add_delayed_ref(trans, ins->objectid,
3650 root_objectid, ref_generation, 3070 ins->offset, parent, root_objectid,
3651 owner_objectid, ins); 3071 ref_generation, owner_objectid,
3072 BTRFS_ADD_DELAYED_EXTENT, 0);
3652 BUG_ON(ret); 3073 BUG_ON(ret);
3653
3654 } else {
3655 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3656 } 3074 }
3075 update_reserved_extents(root, ins->objectid, ins->offset, 1);
3657 return ret; 3076 return ret;
3658} 3077}
3659 3078
@@ -3789,7 +3208,7 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
3789 3208
3790 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); 3209 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
3791 3210
3792 ret = __btrfs_free_extent(trans, root, disk_bytenr, 3211 ret = btrfs_free_extent(trans, root, disk_bytenr,
3793 btrfs_file_extent_disk_num_bytes(leaf, fi), 3212 btrfs_file_extent_disk_num_bytes(leaf, fi),
3794 leaf->start, leaf_owner, leaf_generation, 3213 leaf->start, leaf_owner, leaf_generation,
3795 key.objectid, 0); 3214 key.objectid, 0);
@@ -3829,7 +3248,7 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3829 */ 3248 */
3830 for (i = 0; i < ref->nritems; i++) { 3249 for (i = 0; i < ref->nritems; i++) {
3831 info = ref->extents + sorted[i].slot; 3250 info = ref->extents + sorted[i].slot;
3832 ret = __btrfs_free_extent(trans, root, info->bytenr, 3251 ret = btrfs_free_extent(trans, root, info->bytenr,
3833 info->num_bytes, ref->bytenr, 3252 info->num_bytes, ref->bytenr,
3834 ref->owner, ref->generation, 3253 ref->owner, ref->generation,
3835 info->objectid, 0); 3254 info->objectid, 0);
@@ -3846,12 +3265,13 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
3846 return 0; 3265 return 0;
3847} 3266}
3848 3267
3849static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, 3268static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans,
3269 struct btrfs_root *root, u64 start,
3850 u64 len, u32 *refs) 3270 u64 len, u32 *refs)
3851{ 3271{
3852 int ret; 3272 int ret;
3853 3273
3854 ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs); 3274 ret = btrfs_lookup_extent_ref(trans, root, start, len, refs);
3855 BUG_ON(ret); 3275 BUG_ON(ret);
3856 3276
3857#if 0 /* some debugging code in case we see problems here */ 3277#if 0 /* some debugging code in case we see problems here */
@@ -3959,7 +3379,8 @@ static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
3959 * we just decrement it below and don't update any 3379 * we just decrement it below and don't update any
3960 * of the refs the leaf points to. 3380 * of the refs the leaf points to.
3961 */ 3381 */
3962 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); 3382 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3383 blocksize, &refs);
3963 BUG_ON(ret); 3384 BUG_ON(ret);
3964 if (refs != 1) 3385 if (refs != 1)
3965 continue; 3386 continue;
@@ -4010,7 +3431,7 @@ static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans,
4010 */ 3431 */
4011 for (i = 0; i < refi; i++) { 3432 for (i = 0; i < refi; i++) {
4012 bytenr = sorted[i].bytenr; 3433 bytenr = sorted[i].bytenr;
4013 ret = __btrfs_free_extent(trans, root, bytenr, 3434 ret = btrfs_free_extent(trans, root, bytenr,
4014 blocksize, eb->start, 3435 blocksize, eb->start,
4015 root_owner, root_gen, 0, 1); 3436 root_owner, root_gen, 0, 1);
4016 BUG_ON(ret); 3437 BUG_ON(ret);
@@ -4053,7 +3474,7 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4053 3474
4054 WARN_ON(*level < 0); 3475 WARN_ON(*level < 0);
4055 WARN_ON(*level >= BTRFS_MAX_LEVEL); 3476 WARN_ON(*level >= BTRFS_MAX_LEVEL);
4056 ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, 3477 ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start,
4057 path->nodes[*level]->len, &refs); 3478 path->nodes[*level]->len, &refs);
4058 BUG_ON(ret); 3479 BUG_ON(ret);
4059 if (refs > 1) 3480 if (refs > 1)
@@ -4104,7 +3525,8 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4104 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); 3525 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
4105 blocksize = btrfs_level_size(root, *level - 1); 3526 blocksize = btrfs_level_size(root, *level - 1);
4106 3527
4107 ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); 3528 ret = drop_snap_lookup_refcount(trans, root, bytenr,
3529 blocksize, &refs);
4108 BUG_ON(ret); 3530 BUG_ON(ret);
4109 3531
4110 /* 3532 /*
@@ -4119,7 +3541,7 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
4119 root_gen = btrfs_header_generation(parent); 3541 root_gen = btrfs_header_generation(parent);
4120 path->slots[*level]++; 3542 path->slots[*level]++;
4121 3543
4122 ret = __btrfs_free_extent(trans, root, bytenr, 3544 ret = btrfs_free_extent(trans, root, bytenr,
4123 blocksize, parent->start, 3545 blocksize, parent->start,
4124 root_owner, root_gen, 3546 root_owner, root_gen,
4125 *level - 1, 1); 3547 *level - 1, 1);
@@ -4165,7 +3587,7 @@ out:
4165 * cleanup and free the reference on the last node 3587 * cleanup and free the reference on the last node
4166 * we processed 3588 * we processed
4167 */ 3589 */
4168 ret = __btrfs_free_extent(trans, root, bytenr, blocksize, 3590 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
4169 parent->start, root_owner, root_gen, 3591 parent->start, root_owner, root_gen,
4170 *level, 1); 3592 *level, 1);
4171 free_extent_buffer(path->nodes[*level]); 3593 free_extent_buffer(path->nodes[*level]);
@@ -4354,6 +3776,7 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4354 struct btrfs_path *path; 3776 struct btrfs_path *path;
4355 int i; 3777 int i;
4356 int orig_level; 3778 int orig_level;
3779 int update_count;
4357 struct btrfs_root_item *root_item = &root->root_item; 3780 struct btrfs_root_item *root_item = &root->root_item;
4358 3781
4359 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); 3782 WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex));
@@ -4395,6 +3818,7 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4395 } 3818 }
4396 } 3819 }
4397 while (1) { 3820 while (1) {
3821 unsigned long update;
4398 wret = walk_down_tree(trans, root, path, &level); 3822 wret = walk_down_tree(trans, root, path, &level);
4399 if (wret > 0) 3823 if (wret > 0)
4400 break; 3824 break;
@@ -4407,12 +3831,21 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
4407 break; 3831 break;
4408 if (wret < 0) 3832 if (wret < 0)
4409 ret = wret; 3833 ret = wret;
4410 if (trans->transaction->in_commit) { 3834 if (trans->transaction->in_commit ||
3835 trans->transaction->delayed_refs.flushing) {
4411 ret = -EAGAIN; 3836 ret = -EAGAIN;
4412 break; 3837 break;
4413 } 3838 }
4414 atomic_inc(&root->fs_info->throttle_gen); 3839 atomic_inc(&root->fs_info->throttle_gen);
4415 wake_up(&root->fs_info->transaction_throttle); 3840 wake_up(&root->fs_info->transaction_throttle);
3841 for (update_count = 0; update_count < 16; update_count++) {
3842 update = trans->delayed_ref_updates;
3843 trans->delayed_ref_updates = 0;
3844 if (update)
3845 btrfs_run_delayed_refs(trans, root, update);
3846 else
3847 break;
3848 }
4416 } 3849 }
4417 for (i = 0; i <= orig_level; i++) { 3850 for (i = 0; i <= orig_level; i++) {
4418 if (path->nodes[i]) { 3851 if (path->nodes[i]) {
@@ -5457,6 +4890,7 @@ static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans,
5457 root->root_key.objectid, 4890 root->root_key.objectid,
5458 trans->transid, key.objectid); 4891 trans->transid, key.objectid);
5459 BUG_ON(ret); 4892 BUG_ON(ret);
4893
5460 ret = btrfs_free_extent(trans, root, 4894 ret = btrfs_free_extent(trans, root,
5461 bytenr, num_bytes, leaf->start, 4895 bytenr, num_bytes, leaf->start,
5462 btrfs_header_owner(leaf), 4896 btrfs_header_owner(leaf),
@@ -5768,9 +5202,6 @@ static noinline int relocate_tree_block(struct btrfs_trans_handle *trans,
5768 ref_path, NULL, NULL); 5202 ref_path, NULL, NULL);
5769 BUG_ON(ret); 5203 BUG_ON(ret);
5770 5204
5771 if (root == root->fs_info->extent_root)
5772 btrfs_extent_post_op(trans, root);
5773
5774 return 0; 5205 return 0;
5775} 5206}
5776 5207
@@ -6038,6 +5469,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
6038 if (!path) 5469 if (!path)
6039 return -ENOMEM; 5470 return -ENOMEM;
6040 5471
5472 path->leave_spinning = 1;
6041 ret = btrfs_insert_empty_inode(trans, root, path, objectid); 5473 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
6042 if (ret) 5474 if (ret)
6043 goto out; 5475 goto out;
@@ -6208,6 +5640,9 @@ again:
6208 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); 5640 btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1);
6209 mutex_unlock(&root->fs_info->cleaner_mutex); 5641 mutex_unlock(&root->fs_info->cleaner_mutex);
6210 5642
5643 trans = btrfs_start_transaction(info->tree_root, 1);
5644 btrfs_commit_transaction(trans, info->tree_root);
5645
6211 while (1) { 5646 while (1) {
6212 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 5647 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6213 if (ret < 0) 5648 if (ret < 0)
@@ -6466,7 +5901,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6466 5901
6467 extent_root = root->fs_info->extent_root; 5902 extent_root = root->fs_info->extent_root;
6468 5903
6469 root->fs_info->last_trans_new_blockgroup = trans->transid; 5904 root->fs_info->last_trans_log_full_commit = trans->transid;
6470 5905
6471 cache = kzalloc(sizeof(*cache), GFP_NOFS); 5906 cache = kzalloc(sizeof(*cache), GFP_NOFS);
6472 if (!cache) 5907 if (!cache)
@@ -6500,9 +5935,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
6500 sizeof(cache->item)); 5935 sizeof(cache->item));
6501 BUG_ON(ret); 5936 BUG_ON(ret);
6502 5937
6503 finish_current_insert(trans, extent_root, 0);
6504 ret = del_pending_extents(trans, extent_root, 0);
6505 BUG_ON(ret);
6506 set_avail_alloc_bits(extent_root->fs_info, type); 5938 set_avail_alloc_bits(extent_root->fs_info, type);
6507 5939
6508 return 0; 5940 return 0;
generated by cgit v1.2.2 (git 2.25.0) at 2024-10-08 11:01:50 -0400