2 * Copyright (C) 2009 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
27 #include "transaction.h"
30 #include "btrfs_inode.h"
31 #include "async-thread.h"
32 #include "free-space-cache.h"
33 #include "inode-map.h"
35 #include "print-tree.h"
38 * backref_node, mapping_node and tree_block start with this
41 struct rb_node rb_node;
46 * present a tree block in the backref cache
49 struct rb_node rb_node;
53 /* objectid of tree block owner, can be not uptodate */
55 /* link to pending, changed or detached list */
56 struct list_head list;
57 /* list of upper level blocks reference this block */
58 struct list_head upper;
59 /* list of child blocks in the cache */
60 struct list_head lower;
61 /* NULL if this node is not tree root */
62 struct btrfs_root *root;
63 /* extent buffer got by COW the block */
64 struct extent_buffer *eb;
65 /* level of tree block */
67 /* is the block in non-reference counted tree */
68 unsigned int cowonly:1;
69 /* 1 if no child node in the cache */
70 unsigned int lowest:1;
71 /* is the extent buffer locked */
72 unsigned int locked:1;
73 /* has the block been processed */
74 unsigned int processed:1;
75 /* have backrefs of this block been checked */
76 unsigned int checked:1;
78 * 1 if corresponding block has been cowed but some upper
79 * level block pointers may not point to the new location
81 unsigned int pending:1;
83 * 1 if the backref node isn't connected to any other
86 unsigned int detached:1;
90 * present a block pointer in the backref cache
93 struct list_head list[2];
94 struct backref_node *node[2];
99 #define RELOCATION_RESERVED_NODES 256
101 struct backref_cache {
102 /* red black tree of all backref nodes in the cache */
103 struct rb_root rb_root;
104 /* for passing backref nodes to btrfs_reloc_cow_block */
105 struct backref_node *path[BTRFS_MAX_LEVEL];
107 * list of blocks that have been cowed but some block
108 * pointers in upper level blocks may not reflect the
111 struct list_head pending[BTRFS_MAX_LEVEL];
112 /* list of backref nodes with no child node */
113 struct list_head leaves;
114 /* list of blocks that have been cowed in current transaction */
115 struct list_head changed;
116 /* list of detached backref node. */
117 struct list_head detached;
126 * map address of tree root to tree
128 struct mapping_node {
129 struct rb_node rb_node;
134 struct mapping_tree {
135 struct rb_root rb_root;
140 * present a tree block to process
143 struct rb_node rb_node;
145 struct btrfs_key key;
146 unsigned int level:8;
147 unsigned int key_ready:1;
150 #define MAX_EXTENTS 128
152 struct file_extent_cluster {
155 u64 boundary[MAX_EXTENTS];
159 struct reloc_control {
160 /* block group to relocate */
161 struct btrfs_block_group_cache *block_group;
163 struct btrfs_root *extent_root;
164 /* inode for moving data */
165 struct inode *data_inode;
167 struct btrfs_block_rsv *block_rsv;
169 struct backref_cache backref_cache;
171 struct file_extent_cluster cluster;
172 /* tree blocks have been processed */
173 struct extent_io_tree processed_blocks;
174 /* map start of tree root to corresponding reloc tree */
175 struct mapping_tree reloc_root_tree;
176 /* list of reloc trees */
177 struct list_head reloc_roots;
178 /* size of metadata reservation for merging reloc trees */
179 u64 merging_rsv_size;
180 /* size of relocated tree nodes */
182 /* reserved size for block group relocation*/
188 unsigned int stage:8;
189 unsigned int create_reloc_tree:1;
190 unsigned int merge_reloc_tree:1;
191 unsigned int found_file_extent:1;
194 /* stages of data relocation */
195 #define MOVE_DATA_EXTENTS 0
196 #define UPDATE_DATA_PTRS 1
198 static void remove_backref_node(struct backref_cache *cache,
199 struct backref_node *node);
200 static void __mark_block_processed(struct reloc_control *rc,
201 struct backref_node *node);
203 static void mapping_tree_init(struct mapping_tree *tree)
205 tree->rb_root = RB_ROOT;
206 spin_lock_init(&tree->lock);
209 static void backref_cache_init(struct backref_cache *cache)
212 cache->rb_root = RB_ROOT;
213 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
214 INIT_LIST_HEAD(&cache->pending[i]);
215 INIT_LIST_HEAD(&cache->changed);
216 INIT_LIST_HEAD(&cache->detached);
217 INIT_LIST_HEAD(&cache->leaves);
220 static void backref_cache_cleanup(struct backref_cache *cache)
222 struct backref_node *node;
225 while (!list_empty(&cache->detached)) {
226 node = list_entry(cache->detached.next,
227 struct backref_node, list);
228 remove_backref_node(cache, node);
231 while (!list_empty(&cache->leaves)) {
232 node = list_entry(cache->leaves.next,
233 struct backref_node, lower);
234 remove_backref_node(cache, node);
237 cache->last_trans = 0;
239 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
240 ASSERT(list_empty(&cache->pending[i]));
241 ASSERT(list_empty(&cache->changed));
242 ASSERT(list_empty(&cache->detached));
243 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
244 ASSERT(!cache->nr_nodes);
245 ASSERT(!cache->nr_edges);
248 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
250 struct backref_node *node;
252 node = kzalloc(sizeof(*node), GFP_NOFS);
254 INIT_LIST_HEAD(&node->list);
255 INIT_LIST_HEAD(&node->upper);
256 INIT_LIST_HEAD(&node->lower);
257 RB_CLEAR_NODE(&node->rb_node);
263 static void free_backref_node(struct backref_cache *cache,
264 struct backref_node *node)
272 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
274 struct backref_edge *edge;
276 edge = kzalloc(sizeof(*edge), GFP_NOFS);
282 static void free_backref_edge(struct backref_cache *cache,
283 struct backref_edge *edge)
291 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
292 struct rb_node *node)
294 struct rb_node **p = &root->rb_node;
295 struct rb_node *parent = NULL;
296 struct tree_entry *entry;
300 entry = rb_entry(parent, struct tree_entry, rb_node);
302 if (bytenr < entry->bytenr)
304 else if (bytenr > entry->bytenr)
310 rb_link_node(node, parent, p);
311 rb_insert_color(node, root);
315 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
317 struct rb_node *n = root->rb_node;
318 struct tree_entry *entry;
321 entry = rb_entry(n, struct tree_entry, rb_node);
323 if (bytenr < entry->bytenr)
325 else if (bytenr > entry->bytenr)
333 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
336 struct btrfs_fs_info *fs_info = NULL;
337 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
340 fs_info = bnode->root->fs_info;
341 btrfs_panic(fs_info, errno,
342 "Inconsistency in backref cache found at offset %llu",
347 * walk up backref nodes until reach node presents tree root
349 static struct backref_node *walk_up_backref(struct backref_node *node,
350 struct backref_edge *edges[],
353 struct backref_edge *edge;
356 while (!list_empty(&node->upper)) {
357 edge = list_entry(node->upper.next,
358 struct backref_edge, list[LOWER]);
360 node = edge->node[UPPER];
362 BUG_ON(node->detached);
368 * walk down backref nodes to find start of next reference path
370 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
373 struct backref_edge *edge;
374 struct backref_node *lower;
378 edge = edges[idx - 1];
379 lower = edge->node[LOWER];
380 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
384 edge = list_entry(edge->list[LOWER].next,
385 struct backref_edge, list[LOWER]);
386 edges[idx - 1] = edge;
388 return edge->node[UPPER];
394 static void unlock_node_buffer(struct backref_node *node)
397 btrfs_tree_unlock(node->eb);
402 static void drop_node_buffer(struct backref_node *node)
405 unlock_node_buffer(node);
406 free_extent_buffer(node->eb);
411 static void drop_backref_node(struct backref_cache *tree,
412 struct backref_node *node)
414 BUG_ON(!list_empty(&node->upper));
416 drop_node_buffer(node);
417 list_del(&node->list);
418 list_del(&node->lower);
419 if (!RB_EMPTY_NODE(&node->rb_node))
420 rb_erase(&node->rb_node, &tree->rb_root);
421 free_backref_node(tree, node);
425 * remove a backref node from the backref cache
427 static void remove_backref_node(struct backref_cache *cache,
428 struct backref_node *node)
430 struct backref_node *upper;
431 struct backref_edge *edge;
436 BUG_ON(!node->lowest && !node->detached);
437 while (!list_empty(&node->upper)) {
438 edge = list_entry(node->upper.next, struct backref_edge,
440 upper = edge->node[UPPER];
441 list_del(&edge->list[LOWER]);
442 list_del(&edge->list[UPPER]);
443 free_backref_edge(cache, edge);
445 if (RB_EMPTY_NODE(&upper->rb_node)) {
446 BUG_ON(!list_empty(&node->upper));
447 drop_backref_node(cache, node);
453 * add the node to leaf node list if no other
454 * child block cached.
456 if (list_empty(&upper->lower)) {
457 list_add_tail(&upper->lower, &cache->leaves);
462 drop_backref_node(cache, node);
465 static void update_backref_node(struct backref_cache *cache,
466 struct backref_node *node, u64 bytenr)
468 struct rb_node *rb_node;
469 rb_erase(&node->rb_node, &cache->rb_root);
470 node->bytenr = bytenr;
471 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
473 backref_tree_panic(rb_node, -EEXIST, bytenr);
477 * update backref cache after a transaction commit
479 static int update_backref_cache(struct btrfs_trans_handle *trans,
480 struct backref_cache *cache)
482 struct backref_node *node;
485 if (cache->last_trans == 0) {
486 cache->last_trans = trans->transid;
490 if (cache->last_trans == trans->transid)
494 * detached nodes are used to avoid unnecessary backref
495 * lookup. transaction commit changes the extent tree.
496 * so the detached nodes are no longer useful.
498 while (!list_empty(&cache->detached)) {
499 node = list_entry(cache->detached.next,
500 struct backref_node, list);
501 remove_backref_node(cache, node);
504 while (!list_empty(&cache->changed)) {
505 node = list_entry(cache->changed.next,
506 struct backref_node, list);
507 list_del_init(&node->list);
508 BUG_ON(node->pending);
509 update_backref_node(cache, node, node->new_bytenr);
513 * some nodes can be left in the pending list if there were
514 * errors during processing the pending nodes.
516 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
517 list_for_each_entry(node, &cache->pending[level], list) {
518 BUG_ON(!node->pending);
519 if (node->bytenr == node->new_bytenr)
521 update_backref_node(cache, node, node->new_bytenr);
525 cache->last_trans = 0;
530 static int should_ignore_root(struct btrfs_root *root)
532 struct btrfs_root *reloc_root;
534 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
537 reloc_root = root->reloc_root;
541 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
542 root->fs_info->running_transaction->transid - 1)
545 * if there is reloc tree and it was created in previous
546 * transaction backref lookup can find the reloc tree,
547 * so backref node for the fs tree root is useless for
553 * find reloc tree by address of tree root
555 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
558 struct rb_node *rb_node;
559 struct mapping_node *node;
560 struct btrfs_root *root = NULL;
562 spin_lock(&rc->reloc_root_tree.lock);
563 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
565 node = rb_entry(rb_node, struct mapping_node, rb_node);
566 root = (struct btrfs_root *)node->data;
568 spin_unlock(&rc->reloc_root_tree.lock);
572 static int is_cowonly_root(u64 root_objectid)
574 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
575 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
576 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
577 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
578 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
579 root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
580 root_objectid == BTRFS_UUID_TREE_OBJECTID ||
581 root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
582 root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
587 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
590 struct btrfs_key key;
592 key.objectid = root_objectid;
593 key.type = BTRFS_ROOT_ITEM_KEY;
594 if (is_cowonly_root(root_objectid))
597 key.offset = (u64)-1;
599 return btrfs_get_fs_root(fs_info, &key, false);
602 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
603 static noinline_for_stack
604 struct btrfs_root *find_tree_root(struct reloc_control *rc,
605 struct extent_buffer *leaf,
606 struct btrfs_extent_ref_v0 *ref0)
608 struct btrfs_root *root;
609 u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
610 u64 generation = btrfs_ref_generation_v0(leaf, ref0);
612 BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
614 root = read_fs_root(rc->extent_root->fs_info, root_objectid);
615 BUG_ON(IS_ERR(root));
617 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
618 generation != btrfs_root_generation(&root->root_item))
625 static noinline_for_stack
626 int find_inline_backref(struct extent_buffer *leaf, int slot,
627 unsigned long *ptr, unsigned long *end)
629 struct btrfs_key key;
630 struct btrfs_extent_item *ei;
631 struct btrfs_tree_block_info *bi;
634 btrfs_item_key_to_cpu(leaf, &key, slot);
636 item_size = btrfs_item_size_nr(leaf, slot);
637 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
638 if (item_size < sizeof(*ei)) {
639 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
643 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
644 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
645 BTRFS_EXTENT_FLAG_TREE_BLOCK));
647 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
648 item_size <= sizeof(*ei) + sizeof(*bi)) {
649 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
652 if (key.type == BTRFS_METADATA_ITEM_KEY &&
653 item_size <= sizeof(*ei)) {
654 WARN_ON(item_size < sizeof(*ei));
658 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
659 bi = (struct btrfs_tree_block_info *)(ei + 1);
660 *ptr = (unsigned long)(bi + 1);
662 *ptr = (unsigned long)(ei + 1);
664 *end = (unsigned long)ei + item_size;
669 * build backref tree for a given tree block. root of the backref tree
670 * corresponds the tree block, leaves of the backref tree correspond
671 * roots of b-trees that reference the tree block.
673 * the basic idea of this function is check backrefs of a given block
674 * to find upper level blocks that reference the block, and then check
675 * backrefs of these upper level blocks recursively. the recursion stop
676 * when tree root is reached or backrefs for the block is cached.
678 * NOTE: if we find backrefs for a block are cached, we know backrefs
679 * for all upper level blocks that directly/indirectly reference the
680 * block are also cached.
682 static noinline_for_stack
683 struct backref_node *build_backref_tree(struct reloc_control *rc,
684 struct btrfs_key *node_key,
685 int level, u64 bytenr)
687 struct backref_cache *cache = &rc->backref_cache;
688 struct btrfs_path *path1;
689 struct btrfs_path *path2;
690 struct extent_buffer *eb;
691 struct btrfs_root *root;
692 struct backref_node *cur;
693 struct backref_node *upper;
694 struct backref_node *lower;
695 struct backref_node *node = NULL;
696 struct backref_node *exist = NULL;
697 struct backref_edge *edge;
698 struct rb_node *rb_node;
699 struct btrfs_key key;
707 bool need_check = true;
709 path1 = btrfs_alloc_path();
710 path2 = btrfs_alloc_path();
711 if (!path1 || !path2) {
715 path1->reada = READA_FORWARD;
716 path2->reada = READA_FORWARD;
718 node = alloc_backref_node(cache);
724 node->bytenr = bytenr;
731 key.objectid = cur->bytenr;
732 key.type = BTRFS_METADATA_ITEM_KEY;
733 key.offset = (u64)-1;
735 path1->search_commit_root = 1;
736 path1->skip_locking = 1;
737 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
744 ASSERT(path1->slots[0]);
748 WARN_ON(cur->checked);
749 if (!list_empty(&cur->upper)) {
751 * the backref was added previously when processing
752 * backref of type BTRFS_TREE_BLOCK_REF_KEY
754 ASSERT(list_is_singular(&cur->upper));
755 edge = list_entry(cur->upper.next, struct backref_edge,
757 ASSERT(list_empty(&edge->list[UPPER]));
758 exist = edge->node[UPPER];
760 * add the upper level block to pending list if we need
764 list_add_tail(&edge->list[UPPER], &list);
771 eb = path1->nodes[0];
774 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
775 ret = btrfs_next_leaf(rc->extent_root, path1);
782 eb = path1->nodes[0];
785 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
786 if (key.objectid != cur->bytenr) {
791 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
792 key.type == BTRFS_METADATA_ITEM_KEY) {
793 ret = find_inline_backref(eb, path1->slots[0],
801 /* update key for inline back ref */
802 struct btrfs_extent_inline_ref *iref;
804 iref = (struct btrfs_extent_inline_ref *)ptr;
805 type = btrfs_get_extent_inline_ref_type(eb, iref,
806 BTRFS_REF_TYPE_BLOCK);
807 if (type == BTRFS_REF_TYPE_INVALID) {
812 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
814 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
815 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
819 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
820 exist->owner == key.offset) ||
821 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
822 exist->bytenr == key.offset))) {
827 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
828 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
829 key.type == BTRFS_EXTENT_REF_V0_KEY) {
830 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
831 struct btrfs_extent_ref_v0 *ref0;
832 ref0 = btrfs_item_ptr(eb, path1->slots[0],
833 struct btrfs_extent_ref_v0);
834 if (key.objectid == key.offset) {
835 root = find_tree_root(rc, eb, ref0);
836 if (root && !should_ignore_root(root))
839 list_add(&cur->list, &useless);
842 if (is_cowonly_root(btrfs_ref_root_v0(eb,
847 ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY);
848 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
850 if (key.objectid == key.offset) {
852 * only root blocks of reloc trees use
853 * backref of this type.
855 root = find_reloc_root(rc, cur->bytenr);
861 edge = alloc_backref_edge(cache);
866 rb_node = tree_search(&cache->rb_root, key.offset);
868 upper = alloc_backref_node(cache);
870 free_backref_edge(cache, edge);
874 upper->bytenr = key.offset;
875 upper->level = cur->level + 1;
877 * backrefs for the upper level block isn't
878 * cached, add the block to pending list
880 list_add_tail(&edge->list[UPPER], &list);
882 upper = rb_entry(rb_node, struct backref_node,
884 ASSERT(upper->checked);
885 INIT_LIST_HEAD(&edge->list[UPPER]);
887 list_add_tail(&edge->list[LOWER], &cur->upper);
888 edge->node[LOWER] = cur;
889 edge->node[UPPER] = upper;
892 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
896 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
897 root = read_fs_root(rc->extent_root->fs_info, key.offset);
903 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
906 if (btrfs_root_level(&root->root_item) == cur->level) {
908 ASSERT(btrfs_root_bytenr(&root->root_item) ==
910 if (should_ignore_root(root))
911 list_add(&cur->list, &useless);
917 level = cur->level + 1;
920 * searching the tree to find upper level blocks
921 * reference the block.
923 path2->search_commit_root = 1;
924 path2->skip_locking = 1;
925 path2->lowest_level = level;
926 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
927 path2->lowest_level = 0;
932 if (ret > 0 && path2->slots[level] > 0)
933 path2->slots[level]--;
935 eb = path2->nodes[level];
936 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
938 btrfs_err(root->fs_info,
939 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
940 cur->bytenr, level - 1, root->objectid,
941 node_key->objectid, node_key->type,
948 for (; level < BTRFS_MAX_LEVEL; level++) {
949 if (!path2->nodes[level]) {
950 ASSERT(btrfs_root_bytenr(&root->root_item) ==
952 if (should_ignore_root(root))
953 list_add(&lower->list, &useless);
959 edge = alloc_backref_edge(cache);
965 eb = path2->nodes[level];
966 rb_node = tree_search(&cache->rb_root, eb->start);
968 upper = alloc_backref_node(cache);
970 free_backref_edge(cache, edge);
974 upper->bytenr = eb->start;
975 upper->owner = btrfs_header_owner(eb);
976 upper->level = lower->level + 1;
977 if (!test_bit(BTRFS_ROOT_REF_COWS,
982 * if we know the block isn't shared
983 * we can void checking its backrefs.
985 if (btrfs_block_can_be_shared(root, eb))
991 * add the block to pending list if we
992 * need check its backrefs, we only do this once
993 * while walking up a tree as we will catch
994 * anything else later on.
996 if (!upper->checked && need_check) {
998 list_add_tail(&edge->list[UPPER],
1003 INIT_LIST_HEAD(&edge->list[UPPER]);
1006 upper = rb_entry(rb_node, struct backref_node,
1008 ASSERT(upper->checked);
1009 INIT_LIST_HEAD(&edge->list[UPPER]);
1011 upper->owner = btrfs_header_owner(eb);
1013 list_add_tail(&edge->list[LOWER], &lower->upper);
1014 edge->node[LOWER] = lower;
1015 edge->node[UPPER] = upper;
1022 btrfs_release_path(path2);
1025 ptr += btrfs_extent_inline_ref_size(key.type);
1035 btrfs_release_path(path1);
1040 /* the pending list isn't empty, take the first block to process */
1041 if (!list_empty(&list)) {
1042 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1043 list_del_init(&edge->list[UPPER]);
1044 cur = edge->node[UPPER];
1049 * everything goes well, connect backref nodes and insert backref nodes
1052 ASSERT(node->checked);
1053 cowonly = node->cowonly;
1055 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1058 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1059 list_add_tail(&node->lower, &cache->leaves);
1062 list_for_each_entry(edge, &node->upper, list[LOWER])
1063 list_add_tail(&edge->list[UPPER], &list);
1065 while (!list_empty(&list)) {
1066 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1067 list_del_init(&edge->list[UPPER]);
1068 upper = edge->node[UPPER];
1069 if (upper->detached) {
1070 list_del(&edge->list[LOWER]);
1071 lower = edge->node[LOWER];
1072 free_backref_edge(cache, edge);
1073 if (list_empty(&lower->upper))
1074 list_add(&lower->list, &useless);
1078 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1079 if (upper->lowest) {
1080 list_del_init(&upper->lower);
1084 list_add_tail(&edge->list[UPPER], &upper->lower);
1088 if (!upper->checked) {
1090 * Still want to blow up for developers since this is a
1097 if (cowonly != upper->cowonly) {
1104 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1107 backref_tree_panic(rb_node, -EEXIST,
1111 list_add_tail(&edge->list[UPPER], &upper->lower);
1113 list_for_each_entry(edge, &upper->upper, list[LOWER])
1114 list_add_tail(&edge->list[UPPER], &list);
1117 * process useless backref nodes. backref nodes for tree leaves
1118 * are deleted from the cache. backref nodes for upper level
1119 * tree blocks are left in the cache to avoid unnecessary backref
1122 while (!list_empty(&useless)) {
1123 upper = list_entry(useless.next, struct backref_node, list);
1124 list_del_init(&upper->list);
1125 ASSERT(list_empty(&upper->upper));
1128 if (upper->lowest) {
1129 list_del_init(&upper->lower);
1132 while (!list_empty(&upper->lower)) {
1133 edge = list_entry(upper->lower.next,
1134 struct backref_edge, list[UPPER]);
1135 list_del(&edge->list[UPPER]);
1136 list_del(&edge->list[LOWER]);
1137 lower = edge->node[LOWER];
1138 free_backref_edge(cache, edge);
1140 if (list_empty(&lower->upper))
1141 list_add(&lower->list, &useless);
1143 __mark_block_processed(rc, upper);
1144 if (upper->level > 0) {
1145 list_add(&upper->list, &cache->detached);
1146 upper->detached = 1;
1148 rb_erase(&upper->rb_node, &cache->rb_root);
1149 free_backref_node(cache, upper);
1153 btrfs_free_path(path1);
1154 btrfs_free_path(path2);
1156 while (!list_empty(&useless)) {
1157 lower = list_entry(useless.next,
1158 struct backref_node, list);
1159 list_del_init(&lower->list);
1161 while (!list_empty(&list)) {
1162 edge = list_first_entry(&list, struct backref_edge,
1164 list_del(&edge->list[UPPER]);
1165 list_del(&edge->list[LOWER]);
1166 lower = edge->node[LOWER];
1167 upper = edge->node[UPPER];
1168 free_backref_edge(cache, edge);
1171 * Lower is no longer linked to any upper backref nodes
1172 * and isn't in the cache, we can free it ourselves.
1174 if (list_empty(&lower->upper) &&
1175 RB_EMPTY_NODE(&lower->rb_node))
1176 list_add(&lower->list, &useless);
1178 if (!RB_EMPTY_NODE(&upper->rb_node))
1181 /* Add this guy's upper edges to the list to process */
1182 list_for_each_entry(edge, &upper->upper, list[LOWER])
1183 list_add_tail(&edge->list[UPPER], &list);
1184 if (list_empty(&upper->upper))
1185 list_add(&upper->list, &useless);
1188 while (!list_empty(&useless)) {
1189 lower = list_entry(useless.next,
1190 struct backref_node, list);
1191 list_del_init(&lower->list);
1194 free_backref_node(cache, lower);
1197 free_backref_node(cache, node);
1198 return ERR_PTR(err);
1200 ASSERT(!node || !node->detached);
1205 * helper to add backref node for the newly created snapshot.
1206 * the backref node is created by cloning backref node that
1207 * corresponds to root of source tree
1209 static int clone_backref_node(struct btrfs_trans_handle *trans,
1210 struct reloc_control *rc,
1211 struct btrfs_root *src,
1212 struct btrfs_root *dest)
1214 struct btrfs_root *reloc_root = src->reloc_root;
1215 struct backref_cache *cache = &rc->backref_cache;
1216 struct backref_node *node = NULL;
1217 struct backref_node *new_node;
1218 struct backref_edge *edge;
1219 struct backref_edge *new_edge;
1220 struct rb_node *rb_node;
1222 if (cache->last_trans > 0)
1223 update_backref_cache(trans, cache);
1225 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1227 node = rb_entry(rb_node, struct backref_node, rb_node);
1231 BUG_ON(node->new_bytenr != reloc_root->node->start);
1235 rb_node = tree_search(&cache->rb_root,
1236 reloc_root->commit_root->start);
1238 node = rb_entry(rb_node, struct backref_node,
1240 BUG_ON(node->detached);
1247 new_node = alloc_backref_node(cache);
1251 new_node->bytenr = dest->node->start;
1252 new_node->level = node->level;
1253 new_node->lowest = node->lowest;
1254 new_node->checked = 1;
1255 new_node->root = dest;
1257 if (!node->lowest) {
1258 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1259 new_edge = alloc_backref_edge(cache);
1263 new_edge->node[UPPER] = new_node;
1264 new_edge->node[LOWER] = edge->node[LOWER];
1265 list_add_tail(&new_edge->list[UPPER],
1269 list_add_tail(&new_node->lower, &cache->leaves);
1272 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1273 &new_node->rb_node);
1275 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1277 if (!new_node->lowest) {
1278 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1279 list_add_tail(&new_edge->list[LOWER],
1280 &new_edge->node[LOWER]->upper);
1285 while (!list_empty(&new_node->lower)) {
1286 new_edge = list_entry(new_node->lower.next,
1287 struct backref_edge, list[UPPER]);
1288 list_del(&new_edge->list[UPPER]);
1289 free_backref_edge(cache, new_edge);
1291 free_backref_node(cache, new_node);
1296 * helper to add 'address of tree root -> reloc tree' mapping
1298 static int __must_check __add_reloc_root(struct btrfs_root *root)
1300 struct btrfs_fs_info *fs_info = root->fs_info;
1301 struct rb_node *rb_node;
1302 struct mapping_node *node;
1303 struct reloc_control *rc = fs_info->reloc_ctl;
1305 node = kmalloc(sizeof(*node), GFP_NOFS);
1309 node->bytenr = root->node->start;
1312 spin_lock(&rc->reloc_root_tree.lock);
1313 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1314 node->bytenr, &node->rb_node);
1315 spin_unlock(&rc->reloc_root_tree.lock);
1317 btrfs_panic(fs_info, -EEXIST,
1318 "Duplicate root found for start=%llu while inserting into relocation tree",
1322 list_add_tail(&root->root_list, &rc->reloc_roots);
1327 * helper to delete the 'address of tree root -> reloc tree'
1330 static void __del_reloc_root(struct btrfs_root *root)
1332 struct btrfs_fs_info *fs_info = root->fs_info;
1333 struct rb_node *rb_node;
1334 struct mapping_node *node = NULL;
1335 struct reloc_control *rc = fs_info->reloc_ctl;
1337 spin_lock(&rc->reloc_root_tree.lock);
1338 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1341 node = rb_entry(rb_node, struct mapping_node, rb_node);
1342 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1344 spin_unlock(&rc->reloc_root_tree.lock);
1348 BUG_ON((struct btrfs_root *)node->data != root);
1350 spin_lock(&fs_info->trans_lock);
1351 list_del_init(&root->root_list);
1352 spin_unlock(&fs_info->trans_lock);
1357 * helper to update the 'address of tree root -> reloc tree'
1360 static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1362 struct btrfs_fs_info *fs_info = root->fs_info;
1363 struct rb_node *rb_node;
1364 struct mapping_node *node = NULL;
1365 struct reloc_control *rc = fs_info->reloc_ctl;
1367 spin_lock(&rc->reloc_root_tree.lock);
1368 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1371 node = rb_entry(rb_node, struct mapping_node, rb_node);
1372 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1374 spin_unlock(&rc->reloc_root_tree.lock);
1378 BUG_ON((struct btrfs_root *)node->data != root);
1380 spin_lock(&rc->reloc_root_tree.lock);
1381 node->bytenr = new_bytenr;
1382 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1383 node->bytenr, &node->rb_node);
1384 spin_unlock(&rc->reloc_root_tree.lock);
1386 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1390 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1391 struct btrfs_root *root, u64 objectid)
1393 struct btrfs_fs_info *fs_info = root->fs_info;
1394 struct btrfs_root *reloc_root;
1395 struct extent_buffer *eb;
1396 struct btrfs_root_item *root_item;
1397 struct btrfs_key root_key;
1400 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1403 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1404 root_key.type = BTRFS_ROOT_ITEM_KEY;
1405 root_key.offset = objectid;
1407 if (root->root_key.objectid == objectid) {
1408 u64 commit_root_gen;
1410 /* called by btrfs_init_reloc_root */
1411 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1412 BTRFS_TREE_RELOC_OBJECTID);
1415 * Set the last_snapshot field to the generation of the commit
1416 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1417 * correctly (returns true) when the relocation root is created
1418 * either inside the critical section of a transaction commit
1419 * (through transaction.c:qgroup_account_snapshot()) and when
1420 * it's created before the transaction commit is started.
1422 commit_root_gen = btrfs_header_generation(root->commit_root);
1423 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1426 * called by btrfs_reloc_post_snapshot_hook.
1427 * the source tree is a reloc tree, all tree blocks
1428 * modified after it was created have RELOC flag
1429 * set in their headers. so it's OK to not update
1430 * the 'last_snapshot'.
1432 ret = btrfs_copy_root(trans, root, root->node, &eb,
1433 BTRFS_TREE_RELOC_OBJECTID);
1437 memcpy(root_item, &root->root_item, sizeof(*root_item));
1438 btrfs_set_root_bytenr(root_item, eb->start);
1439 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1440 btrfs_set_root_generation(root_item, trans->transid);
1442 if (root->root_key.objectid == objectid) {
1443 btrfs_set_root_refs(root_item, 0);
1444 memset(&root_item->drop_progress, 0,
1445 sizeof(struct btrfs_disk_key));
1446 root_item->drop_level = 0;
1449 btrfs_tree_unlock(eb);
1450 free_extent_buffer(eb);
1452 ret = btrfs_insert_root(trans, fs_info->tree_root,
1453 &root_key, root_item);
1457 reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1458 BUG_ON(IS_ERR(reloc_root));
1459 reloc_root->last_trans = trans->transid;
1464 * create reloc tree for a given fs tree. reloc tree is just a
1465 * snapshot of the fs tree with special root objectid.
1467 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *root)
1470 struct btrfs_fs_info *fs_info = root->fs_info;
1471 struct btrfs_root *reloc_root;
1472 struct reloc_control *rc = fs_info->reloc_ctl;
1473 struct btrfs_block_rsv *rsv;
1477 if (root->reloc_root) {
1478 reloc_root = root->reloc_root;
1479 reloc_root->last_trans = trans->transid;
1483 if (!rc || !rc->create_reloc_tree ||
1484 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1487 if (!trans->reloc_reserved) {
1488 rsv = trans->block_rsv;
1489 trans->block_rsv = rc->block_rsv;
1492 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1494 trans->block_rsv = rsv;
1496 ret = __add_reloc_root(reloc_root);
1498 root->reloc_root = reloc_root;
1503 * update root item of reloc tree
1505 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *root)
1508 struct btrfs_fs_info *fs_info = root->fs_info;
1509 struct btrfs_root *reloc_root;
1510 struct btrfs_root_item *root_item;
1513 if (!root->reloc_root)
1516 reloc_root = root->reloc_root;
1517 root_item = &reloc_root->root_item;
1519 if (fs_info->reloc_ctl->merge_reloc_tree &&
1520 btrfs_root_refs(root_item) == 0) {
1521 root->reloc_root = NULL;
1522 __del_reloc_root(reloc_root);
1525 if (reloc_root->commit_root != reloc_root->node) {
1526 btrfs_set_root_node(root_item, reloc_root->node);
1527 free_extent_buffer(reloc_root->commit_root);
1528 reloc_root->commit_root = btrfs_root_node(reloc_root);
1531 ret = btrfs_update_root(trans, fs_info->tree_root,
1532 &reloc_root->root_key, root_item);
1540 * helper to find first cached inode with inode number >= objectid
1543 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1545 struct rb_node *node;
1546 struct rb_node *prev;
1547 struct btrfs_inode *entry;
1548 struct inode *inode;
1550 spin_lock(&root->inode_lock);
1552 node = root->inode_tree.rb_node;
1556 entry = rb_entry(node, struct btrfs_inode, rb_node);
1558 if (objectid < btrfs_ino(entry))
1559 node = node->rb_left;
1560 else if (objectid > btrfs_ino(entry))
1561 node = node->rb_right;
1567 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1568 if (objectid <= btrfs_ino(entry)) {
1572 prev = rb_next(prev);
1576 entry = rb_entry(node, struct btrfs_inode, rb_node);
1577 inode = igrab(&entry->vfs_inode);
1579 spin_unlock(&root->inode_lock);
1583 objectid = btrfs_ino(entry) + 1;
1584 if (cond_resched_lock(&root->inode_lock))
1587 node = rb_next(node);
1589 spin_unlock(&root->inode_lock);
1593 static int in_block_group(u64 bytenr,
1594 struct btrfs_block_group_cache *block_group)
1596 if (bytenr >= block_group->key.objectid &&
1597 bytenr < block_group->key.objectid + block_group->key.offset)
1603 * get new location of data
1605 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1606 u64 bytenr, u64 num_bytes)
1608 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1609 struct btrfs_path *path;
1610 struct btrfs_file_extent_item *fi;
1611 struct extent_buffer *leaf;
1614 path = btrfs_alloc_path();
1618 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1619 ret = btrfs_lookup_file_extent(NULL, root, path,
1620 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1628 leaf = path->nodes[0];
1629 fi = btrfs_item_ptr(leaf, path->slots[0],
1630 struct btrfs_file_extent_item);
1632 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1633 btrfs_file_extent_compression(leaf, fi) ||
1634 btrfs_file_extent_encryption(leaf, fi) ||
1635 btrfs_file_extent_other_encoding(leaf, fi));
1637 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1642 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1645 btrfs_free_path(path);
1650 * update file extent items in the tree leaf to point to
1651 * the new locations.
1653 static noinline_for_stack
1654 int replace_file_extents(struct btrfs_trans_handle *trans,
1655 struct reloc_control *rc,
1656 struct btrfs_root *root,
1657 struct extent_buffer *leaf)
1659 struct btrfs_fs_info *fs_info = root->fs_info;
1660 struct btrfs_key key;
1661 struct btrfs_file_extent_item *fi;
1662 struct inode *inode = NULL;
1674 if (rc->stage != UPDATE_DATA_PTRS)
1677 /* reloc trees always use full backref */
1678 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1679 parent = leaf->start;
1683 nritems = btrfs_header_nritems(leaf);
1684 for (i = 0; i < nritems; i++) {
1686 btrfs_item_key_to_cpu(leaf, &key, i);
1687 if (key.type != BTRFS_EXTENT_DATA_KEY)
1689 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1690 if (btrfs_file_extent_type(leaf, fi) ==
1691 BTRFS_FILE_EXTENT_INLINE)
1693 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1694 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1697 if (!in_block_group(bytenr, rc->block_group))
1701 * if we are modifying block in fs tree, wait for readpage
1702 * to complete and drop the extent cache
1704 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1706 inode = find_next_inode(root, key.objectid);
1708 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1709 btrfs_add_delayed_iput(inode);
1710 inode = find_next_inode(root, key.objectid);
1712 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1714 btrfs_file_extent_num_bytes(leaf, fi);
1715 WARN_ON(!IS_ALIGNED(key.offset,
1716 fs_info->sectorsize));
1717 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1719 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1724 btrfs_drop_extent_cache(BTRFS_I(inode),
1725 key.offset, end, 1);
1726 unlock_extent(&BTRFS_I(inode)->io_tree,
1731 ret = get_new_location(rc->data_inode, &new_bytenr,
1735 * Don't have to abort since we've not changed anything
1736 * in the file extent yet.
1741 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1744 key.offset -= btrfs_file_extent_offset(leaf, fi);
1745 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1747 btrfs_header_owner(leaf),
1748 key.objectid, key.offset);
1750 btrfs_abort_transaction(trans, ret);
1754 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1755 parent, btrfs_header_owner(leaf),
1756 key.objectid, key.offset);
1758 btrfs_abort_transaction(trans, ret);
1763 btrfs_mark_buffer_dirty(leaf);
1765 btrfs_add_delayed_iput(inode);
1769 static noinline_for_stack
1770 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1771 struct btrfs_path *path, int level)
1773 struct btrfs_disk_key key1;
1774 struct btrfs_disk_key key2;
1775 btrfs_node_key(eb, &key1, slot);
1776 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1777 return memcmp(&key1, &key2, sizeof(key1));
1781 * try to replace tree blocks in fs tree with the new blocks
1782 * in reloc tree. tree blocks haven't been modified since the
1783 * reloc tree was create can be replaced.
1785 * if a block was replaced, level of the block + 1 is returned.
1786 * if no block got replaced, 0 is returned. if there are other
1787 * errors, a negative error number is returned.
1789 static noinline_for_stack
1790 int replace_path(struct btrfs_trans_handle *trans,
1791 struct btrfs_root *dest, struct btrfs_root *src,
1792 struct btrfs_path *path, struct btrfs_key *next_key,
1793 int lowest_level, int max_level)
1795 struct btrfs_fs_info *fs_info = dest->fs_info;
1796 struct extent_buffer *eb;
1797 struct extent_buffer *parent;
1798 struct btrfs_key key;
1810 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1811 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1813 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1815 slot = path->slots[lowest_level];
1816 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1818 eb = btrfs_lock_root_node(dest);
1819 btrfs_set_lock_blocking(eb);
1820 level = btrfs_header_level(eb);
1822 if (level < lowest_level) {
1823 btrfs_tree_unlock(eb);
1824 free_extent_buffer(eb);
1829 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1832 btrfs_set_lock_blocking(eb);
1835 next_key->objectid = (u64)-1;
1836 next_key->type = (u8)-1;
1837 next_key->offset = (u64)-1;
1842 level = btrfs_header_level(parent);
1843 BUG_ON(level < lowest_level);
1845 ret = btrfs_bin_search(parent, &key, level, &slot);
1846 if (ret && slot > 0)
1849 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1850 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1852 old_bytenr = btrfs_node_blockptr(parent, slot);
1853 blocksize = fs_info->nodesize;
1854 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1856 if (level <= max_level) {
1857 eb = path->nodes[level];
1858 new_bytenr = btrfs_node_blockptr(eb,
1859 path->slots[level]);
1860 new_ptr_gen = btrfs_node_ptr_generation(eb,
1861 path->slots[level]);
1867 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1872 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1873 memcmp_node_keys(parent, slot, path, level)) {
1874 if (level <= lowest_level) {
1879 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen);
1883 } else if (!extent_buffer_uptodate(eb)) {
1885 free_extent_buffer(eb);
1888 btrfs_tree_lock(eb);
1890 ret = btrfs_cow_block(trans, dest, eb, parent,
1894 btrfs_set_lock_blocking(eb);
1896 btrfs_tree_unlock(parent);
1897 free_extent_buffer(parent);
1904 btrfs_tree_unlock(parent);
1905 free_extent_buffer(parent);
1910 btrfs_node_key_to_cpu(path->nodes[level], &key,
1911 path->slots[level]);
1912 btrfs_release_path(path);
1914 path->lowest_level = level;
1915 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1916 path->lowest_level = 0;
1920 * Info qgroup to trace both subtrees.
1922 * We must trace both trees.
1923 * 1) Tree reloc subtree
1924 * If not traced, we will leak data numbers
1926 * If not traced, we will double count old data
1927 * and tree block numbers, if current trans doesn't free
1928 * data reloc tree inode.
1930 ret = btrfs_qgroup_trace_subtree(trans, src, parent,
1931 btrfs_header_generation(parent),
1932 btrfs_header_level(parent));
1935 ret = btrfs_qgroup_trace_subtree(trans, dest,
1937 btrfs_header_generation(path->nodes[level]),
1938 btrfs_header_level(path->nodes[level]));
1943 * swap blocks in fs tree and reloc tree.
1945 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1946 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1947 btrfs_mark_buffer_dirty(parent);
1949 btrfs_set_node_blockptr(path->nodes[level],
1950 path->slots[level], old_bytenr);
1951 btrfs_set_node_ptr_generation(path->nodes[level],
1952 path->slots[level], old_ptr_gen);
1953 btrfs_mark_buffer_dirty(path->nodes[level]);
1955 ret = btrfs_inc_extent_ref(trans, src, old_bytenr,
1956 blocksize, path->nodes[level]->start,
1957 src->root_key.objectid, level - 1, 0);
1959 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr,
1960 blocksize, 0, dest->root_key.objectid,
1964 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1965 path->nodes[level]->start,
1966 src->root_key.objectid, level - 1, 0);
1969 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1970 0, dest->root_key.objectid, level - 1,
1974 btrfs_unlock_up_safe(path, 0);
1979 btrfs_tree_unlock(parent);
1980 free_extent_buffer(parent);
1985 * helper to find next relocated block in reloc tree
1987 static noinline_for_stack
1988 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1991 struct extent_buffer *eb;
1996 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1998 for (i = 0; i < *level; i++) {
1999 free_extent_buffer(path->nodes[i]);
2000 path->nodes[i] = NULL;
2003 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2004 eb = path->nodes[i];
2005 nritems = btrfs_header_nritems(eb);
2006 while (path->slots[i] + 1 < nritems) {
2008 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
2015 free_extent_buffer(path->nodes[i]);
2016 path->nodes[i] = NULL;
2022 * walk down reloc tree to find relocated block of lowest level
2024 static noinline_for_stack
2025 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2028 struct btrfs_fs_info *fs_info = root->fs_info;
2029 struct extent_buffer *eb = NULL;
2036 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2038 for (i = *level; i > 0; i--) {
2039 eb = path->nodes[i];
2040 nritems = btrfs_header_nritems(eb);
2041 while (path->slots[i] < nritems) {
2042 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2043 if (ptr_gen > last_snapshot)
2047 if (path->slots[i] >= nritems) {
2058 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2059 eb = read_tree_block(fs_info, bytenr, ptr_gen);
2062 } else if (!extent_buffer_uptodate(eb)) {
2063 free_extent_buffer(eb);
2066 BUG_ON(btrfs_header_level(eb) != i - 1);
2067 path->nodes[i - 1] = eb;
2068 path->slots[i - 1] = 0;
2074 * invalidate extent cache for file extents whose key in range of
2075 * [min_key, max_key)
2077 static int invalidate_extent_cache(struct btrfs_root *root,
2078 struct btrfs_key *min_key,
2079 struct btrfs_key *max_key)
2081 struct btrfs_fs_info *fs_info = root->fs_info;
2082 struct inode *inode = NULL;
2087 objectid = min_key->objectid;
2092 if (objectid > max_key->objectid)
2095 inode = find_next_inode(root, objectid);
2098 ino = btrfs_ino(BTRFS_I(inode));
2100 if (ino > max_key->objectid) {
2106 if (!S_ISREG(inode->i_mode))
2109 if (unlikely(min_key->objectid == ino)) {
2110 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2112 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2115 start = min_key->offset;
2116 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2122 if (unlikely(max_key->objectid == ino)) {
2123 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2125 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2128 if (max_key->offset == 0)
2130 end = max_key->offset;
2131 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2138 /* the lock_extent waits for readpage to complete */
2139 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2140 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2141 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2146 static int find_next_key(struct btrfs_path *path, int level,
2147 struct btrfs_key *key)
2150 while (level < BTRFS_MAX_LEVEL) {
2151 if (!path->nodes[level])
2153 if (path->slots[level] + 1 <
2154 btrfs_header_nritems(path->nodes[level])) {
2155 btrfs_node_key_to_cpu(path->nodes[level], key,
2156 path->slots[level] + 1);
2165 * merge the relocated tree blocks in reloc tree with corresponding
2168 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2169 struct btrfs_root *root)
2171 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2172 LIST_HEAD(inode_list);
2173 struct btrfs_key key;
2174 struct btrfs_key next_key;
2175 struct btrfs_trans_handle *trans = NULL;
2176 struct btrfs_root *reloc_root;
2177 struct btrfs_root_item *root_item;
2178 struct btrfs_path *path;
2179 struct extent_buffer *leaf;
2187 path = btrfs_alloc_path();
2190 path->reada = READA_FORWARD;
2192 reloc_root = root->reloc_root;
2193 root_item = &reloc_root->root_item;
2195 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2196 level = btrfs_root_level(root_item);
2197 extent_buffer_get(reloc_root->node);
2198 path->nodes[level] = reloc_root->node;
2199 path->slots[level] = 0;
2201 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2203 level = root_item->drop_level;
2205 path->lowest_level = level;
2206 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2207 path->lowest_level = 0;
2209 btrfs_free_path(path);
2213 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2214 path->slots[level]);
2215 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2217 btrfs_unlock_up_safe(path, 0);
2220 min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2221 memset(&next_key, 0, sizeof(next_key));
2224 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2225 BTRFS_RESERVE_FLUSH_ALL);
2230 trans = btrfs_start_transaction(root, 0);
2231 if (IS_ERR(trans)) {
2232 err = PTR_ERR(trans);
2236 trans->block_rsv = rc->block_rsv;
2241 ret = walk_down_reloc_tree(reloc_root, path, &level);
2249 if (!find_next_key(path, level, &key) &&
2250 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2253 ret = replace_path(trans, root, reloc_root, path,
2254 &next_key, level, max_level);
2263 btrfs_node_key_to_cpu(path->nodes[level], &key,
2264 path->slots[level]);
2268 ret = walk_up_reloc_tree(reloc_root, path, &level);
2274 * save the merging progress in the drop_progress.
2275 * this is OK since root refs == 1 in this case.
2277 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2278 path->slots[level]);
2279 root_item->drop_level = level;
2281 btrfs_end_transaction_throttle(trans);
2284 btrfs_btree_balance_dirty(fs_info);
2286 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2287 invalidate_extent_cache(root, &key, &next_key);
2291 * handle the case only one block in the fs tree need to be
2292 * relocated and the block is tree root.
2294 leaf = btrfs_lock_root_node(root);
2295 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2296 btrfs_tree_unlock(leaf);
2297 free_extent_buffer(leaf);
2301 btrfs_free_path(path);
2304 memset(&root_item->drop_progress, 0,
2305 sizeof(root_item->drop_progress));
2306 root_item->drop_level = 0;
2307 btrfs_set_root_refs(root_item, 0);
2308 btrfs_update_reloc_root(trans, root);
2312 btrfs_end_transaction_throttle(trans);
2314 btrfs_btree_balance_dirty(fs_info);
2316 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2317 invalidate_extent_cache(root, &key, &next_key);
2322 static noinline_for_stack
2323 int prepare_to_merge(struct reloc_control *rc, int err)
2325 struct btrfs_root *root = rc->extent_root;
2326 struct btrfs_fs_info *fs_info = root->fs_info;
2327 struct btrfs_root *reloc_root;
2328 struct btrfs_trans_handle *trans;
2329 LIST_HEAD(reloc_roots);
2333 mutex_lock(&fs_info->reloc_mutex);
2334 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2335 rc->merging_rsv_size += rc->nodes_relocated * 2;
2336 mutex_unlock(&fs_info->reloc_mutex);
2340 num_bytes = rc->merging_rsv_size;
2341 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2342 BTRFS_RESERVE_FLUSH_ALL);
2347 trans = btrfs_join_transaction(rc->extent_root);
2348 if (IS_ERR(trans)) {
2350 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2352 return PTR_ERR(trans);
2356 if (num_bytes != rc->merging_rsv_size) {
2357 btrfs_end_transaction(trans);
2358 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2364 rc->merge_reloc_tree = 1;
2366 while (!list_empty(&rc->reloc_roots)) {
2367 reloc_root = list_entry(rc->reloc_roots.next,
2368 struct btrfs_root, root_list);
2369 list_del_init(&reloc_root->root_list);
2371 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2372 BUG_ON(IS_ERR(root));
2373 BUG_ON(root->reloc_root != reloc_root);
2376 * set reference count to 1, so btrfs_recover_relocation
2377 * knows it should resumes merging
2380 btrfs_set_root_refs(&reloc_root->root_item, 1);
2381 btrfs_update_reloc_root(trans, root);
2383 list_add(&reloc_root->root_list, &reloc_roots);
2386 list_splice(&reloc_roots, &rc->reloc_roots);
2389 btrfs_commit_transaction(trans);
2391 btrfs_end_transaction(trans);
2395 static noinline_for_stack
2396 void free_reloc_roots(struct list_head *list)
2398 struct btrfs_root *reloc_root;
2400 while (!list_empty(list)) {
2401 reloc_root = list_entry(list->next, struct btrfs_root,
2403 __del_reloc_root(reloc_root);
2404 free_extent_buffer(reloc_root->node);
2405 free_extent_buffer(reloc_root->commit_root);
2406 reloc_root->node = NULL;
2407 reloc_root->commit_root = NULL;
2411 static noinline_for_stack
2412 void merge_reloc_roots(struct reloc_control *rc)
2414 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2415 struct btrfs_root *root;
2416 struct btrfs_root *reloc_root;
2417 LIST_HEAD(reloc_roots);
2421 root = rc->extent_root;
2424 * this serializes us with btrfs_record_root_in_transaction,
2425 * we have to make sure nobody is in the middle of
2426 * adding their roots to the list while we are
2429 mutex_lock(&fs_info->reloc_mutex);
2430 list_splice_init(&rc->reloc_roots, &reloc_roots);
2431 mutex_unlock(&fs_info->reloc_mutex);
2433 while (!list_empty(&reloc_roots)) {
2435 reloc_root = list_entry(reloc_roots.next,
2436 struct btrfs_root, root_list);
2438 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2439 root = read_fs_root(fs_info,
2440 reloc_root->root_key.offset);
2441 BUG_ON(IS_ERR(root));
2442 BUG_ON(root->reloc_root != reloc_root);
2444 ret = merge_reloc_root(rc, root);
2446 if (list_empty(&reloc_root->root_list))
2447 list_add_tail(&reloc_root->root_list,
2452 list_del_init(&reloc_root->root_list);
2455 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2457 if (list_empty(&reloc_root->root_list))
2458 list_add_tail(&reloc_root->root_list,
2470 btrfs_handle_fs_error(fs_info, ret, NULL);
2471 if (!list_empty(&reloc_roots))
2472 free_reloc_roots(&reloc_roots);
2474 /* new reloc root may be added */
2475 mutex_lock(&fs_info->reloc_mutex);
2476 list_splice_init(&rc->reloc_roots, &reloc_roots);
2477 mutex_unlock(&fs_info->reloc_mutex);
2478 if (!list_empty(&reloc_roots))
2479 free_reloc_roots(&reloc_roots);
2482 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2485 static void free_block_list(struct rb_root *blocks)
2487 struct tree_block *block;
2488 struct rb_node *rb_node;
2489 while ((rb_node = rb_first(blocks))) {
2490 block = rb_entry(rb_node, struct tree_block, rb_node);
2491 rb_erase(rb_node, blocks);
2496 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2497 struct btrfs_root *reloc_root)
2499 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2500 struct btrfs_root *root;
2502 if (reloc_root->last_trans == trans->transid)
2505 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2506 BUG_ON(IS_ERR(root));
2507 BUG_ON(root->reloc_root != reloc_root);
2509 return btrfs_record_root_in_trans(trans, root);
2512 static noinline_for_stack
2513 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2514 struct reloc_control *rc,
2515 struct backref_node *node,
2516 struct backref_edge *edges[])
2518 struct backref_node *next;
2519 struct btrfs_root *root;
2525 next = walk_up_backref(next, edges, &index);
2528 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2530 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2531 record_reloc_root_in_trans(trans, root);
2535 btrfs_record_root_in_trans(trans, root);
2536 root = root->reloc_root;
2538 if (next->new_bytenr != root->node->start) {
2539 BUG_ON(next->new_bytenr);
2540 BUG_ON(!list_empty(&next->list));
2541 next->new_bytenr = root->node->start;
2543 list_add_tail(&next->list,
2544 &rc->backref_cache.changed);
2545 __mark_block_processed(rc, next);
2551 next = walk_down_backref(edges, &index);
2552 if (!next || next->level <= node->level)
2559 /* setup backref node path for btrfs_reloc_cow_block */
2561 rc->backref_cache.path[next->level] = next;
2564 next = edges[index]->node[UPPER];
2570 * select a tree root for relocation. return NULL if the block
2571 * is reference counted. we should use do_relocation() in this
2572 * case. return a tree root pointer if the block isn't reference
2573 * counted. return -ENOENT if the block is root of reloc tree.
2575 static noinline_for_stack
2576 struct btrfs_root *select_one_root(struct backref_node *node)
2578 struct backref_node *next;
2579 struct btrfs_root *root;
2580 struct btrfs_root *fs_root = NULL;
2581 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2587 next = walk_up_backref(next, edges, &index);
2591 /* no other choice for non-references counted tree */
2592 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2595 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2601 next = walk_down_backref(edges, &index);
2602 if (!next || next->level <= node->level)
2607 return ERR_PTR(-ENOENT);
2611 static noinline_for_stack
2612 u64 calcu_metadata_size(struct reloc_control *rc,
2613 struct backref_node *node, int reserve)
2615 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2616 struct backref_node *next = node;
2617 struct backref_edge *edge;
2618 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2622 BUG_ON(reserve && node->processed);
2627 if (next->processed && (reserve || next != node))
2630 num_bytes += fs_info->nodesize;
2632 if (list_empty(&next->upper))
2635 edge = list_entry(next->upper.next,
2636 struct backref_edge, list[LOWER]);
2637 edges[index++] = edge;
2638 next = edge->node[UPPER];
2640 next = walk_down_backref(edges, &index);
2645 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2646 struct reloc_control *rc,
2647 struct backref_node *node)
2649 struct btrfs_root *root = rc->extent_root;
2650 struct btrfs_fs_info *fs_info = root->fs_info;
2655 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2657 trans->block_rsv = rc->block_rsv;
2658 rc->reserved_bytes += num_bytes;
2661 * We are under a transaction here so we can only do limited flushing.
2662 * If we get an enospc just kick back -EAGAIN so we know to drop the
2663 * transaction and try to refill when we can flush all the things.
2665 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2666 BTRFS_RESERVE_FLUSH_LIMIT);
2668 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2669 while (tmp <= rc->reserved_bytes)
2672 * only one thread can access block_rsv at this point,
2673 * so we don't need hold lock to protect block_rsv.
2674 * we expand more reservation size here to allow enough
2675 * space for relocation and we will return eailer in
2678 rc->block_rsv->size = tmp + fs_info->nodesize *
2679 RELOCATION_RESERVED_NODES;
2687 * relocate a block tree, and then update pointers in upper level
2688 * blocks that reference the block to point to the new location.
2690 * if called by link_to_upper, the block has already been relocated.
2691 * in that case this function just updates pointers.
2693 static int do_relocation(struct btrfs_trans_handle *trans,
2694 struct reloc_control *rc,
2695 struct backref_node *node,
2696 struct btrfs_key *key,
2697 struct btrfs_path *path, int lowest)
2699 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2700 struct backref_node *upper;
2701 struct backref_edge *edge;
2702 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2703 struct btrfs_root *root;
2704 struct extent_buffer *eb;
2712 BUG_ON(lowest && node->eb);
2714 path->lowest_level = node->level + 1;
2715 rc->backref_cache.path[node->level] = node;
2716 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2719 upper = edge->node[UPPER];
2720 root = select_reloc_root(trans, rc, upper, edges);
2723 if (upper->eb && !upper->locked) {
2725 ret = btrfs_bin_search(upper->eb, key,
2726 upper->level, &slot);
2728 bytenr = btrfs_node_blockptr(upper->eb, slot);
2729 if (node->eb->start == bytenr)
2732 drop_node_buffer(upper);
2736 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2743 btrfs_release_path(path);
2748 upper->eb = path->nodes[upper->level];
2749 path->nodes[upper->level] = NULL;
2751 BUG_ON(upper->eb != path->nodes[upper->level]);
2755 path->locks[upper->level] = 0;
2757 slot = path->slots[upper->level];
2758 btrfs_release_path(path);
2760 ret = btrfs_bin_search(upper->eb, key, upper->level,
2765 bytenr = btrfs_node_blockptr(upper->eb, slot);
2767 if (bytenr != node->bytenr) {
2768 btrfs_err(root->fs_info,
2769 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2770 bytenr, node->bytenr, slot,
2776 if (node->eb->start == bytenr)
2780 blocksize = root->fs_info->nodesize;
2781 generation = btrfs_node_ptr_generation(upper->eb, slot);
2782 eb = read_tree_block(fs_info, bytenr, generation);
2786 } else if (!extent_buffer_uptodate(eb)) {
2787 free_extent_buffer(eb);
2791 btrfs_tree_lock(eb);
2792 btrfs_set_lock_blocking(eb);
2795 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2797 btrfs_tree_unlock(eb);
2798 free_extent_buffer(eb);
2803 BUG_ON(node->eb != eb);
2805 btrfs_set_node_blockptr(upper->eb, slot,
2807 btrfs_set_node_ptr_generation(upper->eb, slot,
2809 btrfs_mark_buffer_dirty(upper->eb);
2811 ret = btrfs_inc_extent_ref(trans, root,
2812 node->eb->start, blocksize,
2814 btrfs_header_owner(upper->eb),
2818 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2822 if (!upper->pending)
2823 drop_node_buffer(upper);
2825 unlock_node_buffer(upper);
2830 if (!err && node->pending) {
2831 drop_node_buffer(node);
2832 list_move_tail(&node->list, &rc->backref_cache.changed);
2836 path->lowest_level = 0;
2837 BUG_ON(err == -ENOSPC);
2841 static int link_to_upper(struct btrfs_trans_handle *trans,
2842 struct reloc_control *rc,
2843 struct backref_node *node,
2844 struct btrfs_path *path)
2846 struct btrfs_key key;
2848 btrfs_node_key_to_cpu(node->eb, &key, 0);
2849 return do_relocation(trans, rc, node, &key, path, 0);
2852 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2853 struct reloc_control *rc,
2854 struct btrfs_path *path, int err)
2857 struct backref_cache *cache = &rc->backref_cache;
2858 struct backref_node *node;
2862 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2863 while (!list_empty(&cache->pending[level])) {
2864 node = list_entry(cache->pending[level].next,
2865 struct backref_node, list);
2866 list_move_tail(&node->list, &list);
2867 BUG_ON(!node->pending);
2870 ret = link_to_upper(trans, rc, node, path);
2875 list_splice_init(&list, &cache->pending[level]);
2880 static void mark_block_processed(struct reloc_control *rc,
2881 u64 bytenr, u32 blocksize)
2883 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2887 static void __mark_block_processed(struct reloc_control *rc,
2888 struct backref_node *node)
2891 if (node->level == 0 ||
2892 in_block_group(node->bytenr, rc->block_group)) {
2893 blocksize = rc->extent_root->fs_info->nodesize;
2894 mark_block_processed(rc, node->bytenr, blocksize);
2896 node->processed = 1;
2900 * mark a block and all blocks directly/indirectly reference the block
2903 static void update_processed_blocks(struct reloc_control *rc,
2904 struct backref_node *node)
2906 struct backref_node *next = node;
2907 struct backref_edge *edge;
2908 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2914 if (next->processed)
2917 __mark_block_processed(rc, next);
2919 if (list_empty(&next->upper))
2922 edge = list_entry(next->upper.next,
2923 struct backref_edge, list[LOWER]);
2924 edges[index++] = edge;
2925 next = edge->node[UPPER];
2927 next = walk_down_backref(edges, &index);
2931 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2933 u32 blocksize = rc->extent_root->fs_info->nodesize;
2935 if (test_range_bit(&rc->processed_blocks, bytenr,
2936 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2941 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2942 struct tree_block *block)
2944 struct extent_buffer *eb;
2946 BUG_ON(block->key_ready);
2947 eb = read_tree_block(fs_info, block->bytenr, block->key.offset);
2950 } else if (!extent_buffer_uptodate(eb)) {
2951 free_extent_buffer(eb);
2954 WARN_ON(btrfs_header_level(eb) != block->level);
2955 if (block->level == 0)
2956 btrfs_item_key_to_cpu(eb, &block->key, 0);
2958 btrfs_node_key_to_cpu(eb, &block->key, 0);
2959 free_extent_buffer(eb);
2960 block->key_ready = 1;
2965 * helper function to relocate a tree block
2967 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2968 struct reloc_control *rc,
2969 struct backref_node *node,
2970 struct btrfs_key *key,
2971 struct btrfs_path *path)
2973 struct btrfs_root *root;
2979 BUG_ON(node->processed);
2980 root = select_one_root(node);
2981 if (root == ERR_PTR(-ENOENT)) {
2982 update_processed_blocks(rc, node);
2986 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2987 ret = reserve_metadata_space(trans, rc, node);
2993 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2994 BUG_ON(node->new_bytenr);
2995 BUG_ON(!list_empty(&node->list));
2996 btrfs_record_root_in_trans(trans, root);
2997 root = root->reloc_root;
2998 node->new_bytenr = root->node->start;
3000 list_add_tail(&node->list, &rc->backref_cache.changed);
3002 path->lowest_level = node->level;
3003 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
3004 btrfs_release_path(path);
3009 update_processed_blocks(rc, node);
3011 ret = do_relocation(trans, rc, node, key, path, 1);
3014 if (ret || node->level == 0 || node->cowonly)
3015 remove_backref_node(&rc->backref_cache, node);
3020 * relocate a list of blocks
3022 static noinline_for_stack
3023 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3024 struct reloc_control *rc, struct rb_root *blocks)
3026 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3027 struct backref_node *node;
3028 struct btrfs_path *path;
3029 struct tree_block *block;
3030 struct rb_node *rb_node;
3034 path = btrfs_alloc_path();
3037 goto out_free_blocks;
3040 rb_node = rb_first(blocks);
3042 block = rb_entry(rb_node, struct tree_block, rb_node);
3043 if (!block->key_ready)
3044 readahead_tree_block(fs_info, block->bytenr);
3045 rb_node = rb_next(rb_node);
3048 rb_node = rb_first(blocks);
3050 block = rb_entry(rb_node, struct tree_block, rb_node);
3051 if (!block->key_ready) {
3052 err = get_tree_block_key(fs_info, block);
3056 rb_node = rb_next(rb_node);
3059 rb_node = rb_first(blocks);
3061 block = rb_entry(rb_node, struct tree_block, rb_node);
3063 node = build_backref_tree(rc, &block->key,
3064 block->level, block->bytenr);
3066 err = PTR_ERR(node);
3070 ret = relocate_tree_block(trans, rc, node, &block->key,
3073 if (ret != -EAGAIN || rb_node == rb_first(blocks))
3077 rb_node = rb_next(rb_node);
3080 err = finish_pending_nodes(trans, rc, path, err);
3083 btrfs_free_path(path);
3085 free_block_list(blocks);
3089 static noinline_for_stack
3090 int prealloc_file_extent_cluster(struct inode *inode,
3091 struct file_extent_cluster *cluster)
3096 u64 offset = BTRFS_I(inode)->index_cnt;
3100 u64 prealloc_start = cluster->start - offset;
3101 u64 prealloc_end = cluster->end - offset;
3103 struct extent_changeset *data_reserved = NULL;
3105 BUG_ON(cluster->start != cluster->boundary[0]);
3108 ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3109 prealloc_end + 1 - prealloc_start);
3113 cur_offset = prealloc_start;
3114 while (nr < cluster->nr) {
3115 start = cluster->boundary[nr] - offset;
3116 if (nr + 1 < cluster->nr)
3117 end = cluster->boundary[nr + 1] - 1 - offset;
3119 end = cluster->end - offset;
3121 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3122 num_bytes = end + 1 - start;
3123 if (cur_offset < start)
3124 btrfs_free_reserved_data_space(inode, data_reserved,
3125 cur_offset, start - cur_offset);
3126 ret = btrfs_prealloc_file_range(inode, 0, start,
3127 num_bytes, num_bytes,
3128 end + 1, &alloc_hint);
3129 cur_offset = end + 1;
3130 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3135 if (cur_offset < prealloc_end)
3136 btrfs_free_reserved_data_space(inode, data_reserved,
3137 cur_offset, prealloc_end + 1 - cur_offset);
3139 inode_unlock(inode);
3140 extent_changeset_free(data_reserved);
3144 static noinline_for_stack
3145 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3148 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3149 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3150 struct extent_map *em;
3153 em = alloc_extent_map();
3158 em->len = end + 1 - start;
3159 em->block_len = em->len;
3160 em->block_start = block_start;
3161 em->bdev = fs_info->fs_devices->latest_bdev;
3162 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3164 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3166 write_lock(&em_tree->lock);
3167 ret = add_extent_mapping(em_tree, em, 0);
3168 write_unlock(&em_tree->lock);
3169 if (ret != -EEXIST) {
3170 free_extent_map(em);
3173 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3175 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3179 static int relocate_file_extent_cluster(struct inode *inode,
3180 struct file_extent_cluster *cluster)
3182 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3185 u64 offset = BTRFS_I(inode)->index_cnt;
3186 unsigned long index;
3187 unsigned long last_index;
3189 struct file_ra_state *ra;
3190 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3197 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3201 ret = prealloc_file_extent_cluster(inode, cluster);
3205 file_ra_state_init(ra, inode->i_mapping);
3207 ret = setup_extent_mapping(inode, cluster->start - offset,
3208 cluster->end - offset, cluster->start);
3212 index = (cluster->start - offset) >> PAGE_SHIFT;
3213 last_index = (cluster->end - offset) >> PAGE_SHIFT;
3214 while (index <= last_index) {
3215 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3220 page = find_lock_page(inode->i_mapping, index);
3222 page_cache_sync_readahead(inode->i_mapping,
3224 last_index + 1 - index);
3225 page = find_or_create_page(inode->i_mapping, index,
3228 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3235 if (PageReadahead(page)) {
3236 page_cache_async_readahead(inode->i_mapping,
3237 ra, NULL, page, index,
3238 last_index + 1 - index);
3241 if (!PageUptodate(page)) {
3242 btrfs_readpage(NULL, page);
3244 if (!PageUptodate(page)) {
3247 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3249 btrfs_delalloc_release_extents(BTRFS_I(inode),
3256 page_start = page_offset(page);
3257 page_end = page_start + PAGE_SIZE - 1;
3259 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3261 set_page_extent_mapped(page);
3263 if (nr < cluster->nr &&
3264 page_start + offset == cluster->boundary[nr]) {
3265 set_extent_bits(&BTRFS_I(inode)->io_tree,
3266 page_start, page_end,
3271 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL, 0);
3272 set_page_dirty(page);
3274 unlock_extent(&BTRFS_I(inode)->io_tree,
3275 page_start, page_end);
3280 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
3281 balance_dirty_pages_ratelimited(inode->i_mapping);
3282 btrfs_throttle(fs_info);
3284 WARN_ON(nr != cluster->nr);
3290 static noinline_for_stack
3291 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3292 struct file_extent_cluster *cluster)
3296 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3297 ret = relocate_file_extent_cluster(inode, cluster);
3304 cluster->start = extent_key->objectid;
3306 BUG_ON(cluster->nr >= MAX_EXTENTS);
3307 cluster->end = extent_key->objectid + extent_key->offset - 1;
3308 cluster->boundary[cluster->nr] = extent_key->objectid;
3311 if (cluster->nr >= MAX_EXTENTS) {
3312 ret = relocate_file_extent_cluster(inode, cluster);
3320 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3321 static int get_ref_objectid_v0(struct reloc_control *rc,
3322 struct btrfs_path *path,
3323 struct btrfs_key *extent_key,
3324 u64 *ref_objectid, int *path_change)
3326 struct btrfs_key key;
3327 struct extent_buffer *leaf;
3328 struct btrfs_extent_ref_v0 *ref0;
3332 leaf = path->nodes[0];
3333 slot = path->slots[0];
3335 if (slot >= btrfs_header_nritems(leaf)) {
3336 ret = btrfs_next_leaf(rc->extent_root, path);
3340 leaf = path->nodes[0];
3341 slot = path->slots[0];
3345 btrfs_item_key_to_cpu(leaf, &key, slot);
3346 if (key.objectid != extent_key->objectid)
3349 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3353 ref0 = btrfs_item_ptr(leaf, slot,
3354 struct btrfs_extent_ref_v0);
3355 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3363 * helper to add a tree block to the list.
3364 * the major work is getting the generation and level of the block
3366 static int add_tree_block(struct reloc_control *rc,
3367 struct btrfs_key *extent_key,
3368 struct btrfs_path *path,
3369 struct rb_root *blocks)
3371 struct extent_buffer *eb;
3372 struct btrfs_extent_item *ei;
3373 struct btrfs_tree_block_info *bi;
3374 struct tree_block *block;
3375 struct rb_node *rb_node;
3380 eb = path->nodes[0];
3381 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3383 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3384 item_size >= sizeof(*ei) + sizeof(*bi)) {
3385 ei = btrfs_item_ptr(eb, path->slots[0],
3386 struct btrfs_extent_item);
3387 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3388 bi = (struct btrfs_tree_block_info *)(ei + 1);
3389 level = btrfs_tree_block_level(eb, bi);
3391 level = (int)extent_key->offset;
3393 generation = btrfs_extent_generation(eb, ei);
3395 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3399 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3400 ret = get_ref_objectid_v0(rc, path, extent_key,
3404 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3405 level = (int)ref_owner;
3406 /* FIXME: get real generation */
3413 btrfs_release_path(path);
3415 BUG_ON(level == -1);
3417 block = kmalloc(sizeof(*block), GFP_NOFS);
3421 block->bytenr = extent_key->objectid;
3422 block->key.objectid = rc->extent_root->fs_info->nodesize;
3423 block->key.offset = generation;
3424 block->level = level;
3425 block->key_ready = 0;
3427 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3429 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3435 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3437 static int __add_tree_block(struct reloc_control *rc,
3438 u64 bytenr, u32 blocksize,
3439 struct rb_root *blocks)
3441 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3442 struct btrfs_path *path;
3443 struct btrfs_key key;
3445 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3447 if (tree_block_processed(bytenr, rc))
3450 if (tree_search(blocks, bytenr))
3453 path = btrfs_alloc_path();
3457 key.objectid = bytenr;
3459 key.type = BTRFS_METADATA_ITEM_KEY;
3460 key.offset = (u64)-1;
3462 key.type = BTRFS_EXTENT_ITEM_KEY;
3463 key.offset = blocksize;
3466 path->search_commit_root = 1;
3467 path->skip_locking = 1;
3468 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3472 if (ret > 0 && skinny) {
3473 if (path->slots[0]) {
3475 btrfs_item_key_to_cpu(path->nodes[0], &key,
3477 if (key.objectid == bytenr &&
3478 (key.type == BTRFS_METADATA_ITEM_KEY ||
3479 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3480 key.offset == blocksize)))
3486 btrfs_release_path(path);
3492 btrfs_print_leaf(path->nodes[0]);
3494 "tree block extent item (%llu) is not found in extent tree",
3501 ret = add_tree_block(rc, &key, path, blocks);
3503 btrfs_free_path(path);
3508 * helper to check if the block use full backrefs for pointers in it
3510 static int block_use_full_backref(struct reloc_control *rc,
3511 struct extent_buffer *eb)
3516 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3517 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3520 ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3521 eb->start, btrfs_header_level(eb), 1,
3525 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3532 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3533 struct btrfs_block_group_cache *block_group,
3534 struct inode *inode,
3537 struct btrfs_key key;
3538 struct btrfs_root *root = fs_info->tree_root;
3539 struct btrfs_trans_handle *trans;
3546 key.type = BTRFS_INODE_ITEM_KEY;
3549 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3550 if (IS_ERR(inode) || is_bad_inode(inode)) {
3557 ret = btrfs_check_trunc_cache_free_space(fs_info,
3558 &fs_info->global_block_rsv);
3562 trans = btrfs_join_transaction(root);
3563 if (IS_ERR(trans)) {
3564 ret = PTR_ERR(trans);
3568 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3570 btrfs_end_transaction(trans);
3571 btrfs_btree_balance_dirty(fs_info);
3578 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3579 * this function scans fs tree to find blocks reference the data extent
3581 static int find_data_references(struct reloc_control *rc,
3582 struct btrfs_key *extent_key,
3583 struct extent_buffer *leaf,
3584 struct btrfs_extent_data_ref *ref,
3585 struct rb_root *blocks)
3587 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3588 struct btrfs_path *path;
3589 struct tree_block *block;
3590 struct btrfs_root *root;
3591 struct btrfs_file_extent_item *fi;
3592 struct rb_node *rb_node;
3593 struct btrfs_key key;
3604 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3605 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3606 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3607 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3610 * This is an extent belonging to the free space cache, lets just delete
3611 * it and redo the search.
3613 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3614 ret = delete_block_group_cache(fs_info, rc->block_group,
3615 NULL, ref_objectid);
3621 path = btrfs_alloc_path();
3624 path->reada = READA_FORWARD;
3626 root = read_fs_root(fs_info, ref_root);
3628 err = PTR_ERR(root);
3632 key.objectid = ref_objectid;
3633 key.type = BTRFS_EXTENT_DATA_KEY;
3634 if (ref_offset > ((u64)-1 << 32))
3637 key.offset = ref_offset;
3639 path->search_commit_root = 1;
3640 path->skip_locking = 1;
3641 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3647 leaf = path->nodes[0];
3648 nritems = btrfs_header_nritems(leaf);
3650 * the references in tree blocks that use full backrefs
3651 * are not counted in
3653 if (block_use_full_backref(rc, leaf))
3657 rb_node = tree_search(blocks, leaf->start);
3662 path->slots[0] = nritems;
3665 while (ref_count > 0) {
3666 while (path->slots[0] >= nritems) {
3667 ret = btrfs_next_leaf(root, path);
3672 if (WARN_ON(ret > 0))
3675 leaf = path->nodes[0];
3676 nritems = btrfs_header_nritems(leaf);
3679 if (block_use_full_backref(rc, leaf))
3683 rb_node = tree_search(blocks, leaf->start);
3688 path->slots[0] = nritems;
3692 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3693 if (WARN_ON(key.objectid != ref_objectid ||
3694 key.type != BTRFS_EXTENT_DATA_KEY))
3697 fi = btrfs_item_ptr(leaf, path->slots[0],
3698 struct btrfs_file_extent_item);
3700 if (btrfs_file_extent_type(leaf, fi) ==
3701 BTRFS_FILE_EXTENT_INLINE)
3704 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3705 extent_key->objectid)
3708 key.offset -= btrfs_file_extent_offset(leaf, fi);
3709 if (key.offset != ref_offset)
3717 if (!tree_block_processed(leaf->start, rc)) {
3718 block = kmalloc(sizeof(*block), GFP_NOFS);
3723 block->bytenr = leaf->start;
3724 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3726 block->key_ready = 1;
3727 rb_node = tree_insert(blocks, block->bytenr,
3730 backref_tree_panic(rb_node, -EEXIST,
3736 path->slots[0] = nritems;
3742 btrfs_free_path(path);
3747 * helper to find all tree blocks that reference a given data extent
3749 static noinline_for_stack
3750 int add_data_references(struct reloc_control *rc,
3751 struct btrfs_key *extent_key,
3752 struct btrfs_path *path,
3753 struct rb_root *blocks)
3755 struct btrfs_key key;
3756 struct extent_buffer *eb;
3757 struct btrfs_extent_data_ref *dref;
3758 struct btrfs_extent_inline_ref *iref;
3761 u32 blocksize = rc->extent_root->fs_info->nodesize;
3765 eb = path->nodes[0];
3766 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3767 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3768 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3769 if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3773 ptr += sizeof(struct btrfs_extent_item);
3776 iref = (struct btrfs_extent_inline_ref *)ptr;
3777 key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3778 BTRFS_REF_TYPE_DATA);
3779 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3780 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3781 ret = __add_tree_block(rc, key.offset, blocksize,
3783 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3784 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3785 ret = find_data_references(rc, extent_key,
3789 btrfs_err(rc->extent_root->fs_info,
3790 "extent %llu slot %d has an invalid inline ref type",
3791 eb->start, path->slots[0]);
3797 ptr += btrfs_extent_inline_ref_size(key.type);
3803 eb = path->nodes[0];
3804 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3805 ret = btrfs_next_leaf(rc->extent_root, path);
3812 eb = path->nodes[0];
3815 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3816 if (key.objectid != extent_key->objectid)
3819 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3820 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3821 key.type == BTRFS_EXTENT_REF_V0_KEY) {
3823 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3824 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3826 ret = __add_tree_block(rc, key.offset, blocksize,
3828 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3829 dref = btrfs_item_ptr(eb, path->slots[0],
3830 struct btrfs_extent_data_ref);
3831 ret = find_data_references(rc, extent_key,
3843 btrfs_release_path(path);
3845 free_block_list(blocks);
3850 * helper to find next unprocessed extent
3852 static noinline_for_stack
3853 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3854 struct btrfs_key *extent_key)
3856 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3857 struct btrfs_key key;
3858 struct extent_buffer *leaf;
3859 u64 start, end, last;
3862 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3865 if (rc->search_start >= last) {
3870 key.objectid = rc->search_start;
3871 key.type = BTRFS_EXTENT_ITEM_KEY;
3874 path->search_commit_root = 1;
3875 path->skip_locking = 1;
3876 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3881 leaf = path->nodes[0];
3882 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3883 ret = btrfs_next_leaf(rc->extent_root, path);
3886 leaf = path->nodes[0];
3889 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3890 if (key.objectid >= last) {
3895 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3896 key.type != BTRFS_METADATA_ITEM_KEY) {
3901 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3902 key.objectid + key.offset <= rc->search_start) {
3907 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3908 key.objectid + fs_info->nodesize <=
3914 ret = find_first_extent_bit(&rc->processed_blocks,
3915 key.objectid, &start, &end,
3916 EXTENT_DIRTY, NULL);
3918 if (ret == 0 && start <= key.objectid) {
3919 btrfs_release_path(path);
3920 rc->search_start = end + 1;
3922 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3923 rc->search_start = key.objectid + key.offset;
3925 rc->search_start = key.objectid +
3927 memcpy(extent_key, &key, sizeof(key));
3931 btrfs_release_path(path);
3935 static void set_reloc_control(struct reloc_control *rc)
3937 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3939 mutex_lock(&fs_info->reloc_mutex);
3940 fs_info->reloc_ctl = rc;
3941 mutex_unlock(&fs_info->reloc_mutex);
3944 static void unset_reloc_control(struct reloc_control *rc)
3946 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3948 mutex_lock(&fs_info->reloc_mutex);
3949 fs_info->reloc_ctl = NULL;
3950 mutex_unlock(&fs_info->reloc_mutex);
3953 static int check_extent_flags(u64 flags)
3955 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3956 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3958 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3959 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3961 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3962 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3967 static noinline_for_stack
3968 int prepare_to_relocate(struct reloc_control *rc)
3970 struct btrfs_trans_handle *trans;
3973 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3974 BTRFS_BLOCK_RSV_TEMP);
3978 memset(&rc->cluster, 0, sizeof(rc->cluster));
3979 rc->search_start = rc->block_group->key.objectid;
3980 rc->extents_found = 0;
3981 rc->nodes_relocated = 0;
3982 rc->merging_rsv_size = 0;
3983 rc->reserved_bytes = 0;
3984 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3985 RELOCATION_RESERVED_NODES;
3986 ret = btrfs_block_rsv_refill(rc->extent_root,
3987 rc->block_rsv, rc->block_rsv->size,
3988 BTRFS_RESERVE_FLUSH_ALL);
3992 rc->create_reloc_tree = 1;
3993 set_reloc_control(rc);
3995 trans = btrfs_join_transaction(rc->extent_root);
3996 if (IS_ERR(trans)) {
3997 unset_reloc_control(rc);
3999 * extent tree is not a ref_cow tree and has no reloc_root to
4000 * cleanup. And callers are responsible to free the above
4003 return PTR_ERR(trans);
4005 btrfs_commit_transaction(trans);
4009 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
4011 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
4012 struct rb_root blocks = RB_ROOT;
4013 struct btrfs_key key;
4014 struct btrfs_trans_handle *trans = NULL;
4015 struct btrfs_path *path;
4016 struct btrfs_extent_item *ei;
4023 path = btrfs_alloc_path();
4026 path->reada = READA_FORWARD;
4028 ret = prepare_to_relocate(rc);
4035 rc->reserved_bytes = 0;
4036 ret = btrfs_block_rsv_refill(rc->extent_root,
4037 rc->block_rsv, rc->block_rsv->size,
4038 BTRFS_RESERVE_FLUSH_ALL);
4044 trans = btrfs_start_transaction(rc->extent_root, 0);
4045 if (IS_ERR(trans)) {
4046 err = PTR_ERR(trans);
4051 if (update_backref_cache(trans, &rc->backref_cache)) {
4052 btrfs_end_transaction(trans);
4056 ret = find_next_extent(rc, path, &key);
4062 rc->extents_found++;
4064 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4065 struct btrfs_extent_item);
4066 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4067 if (item_size >= sizeof(*ei)) {
4068 flags = btrfs_extent_flags(path->nodes[0], ei);
4069 ret = check_extent_flags(flags);
4073 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4075 int path_change = 0;
4078 sizeof(struct btrfs_extent_item_v0));
4079 ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
4085 if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
4086 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
4088 flags = BTRFS_EXTENT_FLAG_DATA;
4091 btrfs_release_path(path);
4093 path->search_commit_root = 1;
4094 path->skip_locking = 1;
4095 ret = btrfs_search_slot(NULL, rc->extent_root,
4108 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4109 ret = add_tree_block(rc, &key, path, &blocks);
4110 } else if (rc->stage == UPDATE_DATA_PTRS &&
4111 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4112 ret = add_data_references(rc, &key, path, &blocks);
4114 btrfs_release_path(path);
4122 if (!RB_EMPTY_ROOT(&blocks)) {
4123 ret = relocate_tree_blocks(trans, rc, &blocks);
4126 * if we fail to relocate tree blocks, force to update
4127 * backref cache when committing transaction.
4129 rc->backref_cache.last_trans = trans->transid - 1;
4131 if (ret != -EAGAIN) {
4135 rc->extents_found--;
4136 rc->search_start = key.objectid;
4140 btrfs_end_transaction_throttle(trans);
4141 btrfs_btree_balance_dirty(fs_info);
4144 if (rc->stage == MOVE_DATA_EXTENTS &&
4145 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4146 rc->found_file_extent = 1;
4147 ret = relocate_data_extent(rc->data_inode,
4148 &key, &rc->cluster);
4155 if (trans && progress && err == -ENOSPC) {
4156 ret = btrfs_force_chunk_alloc(trans, fs_info,
4157 rc->block_group->flags);
4165 btrfs_release_path(path);
4166 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4169 btrfs_end_transaction_throttle(trans);
4170 btrfs_btree_balance_dirty(fs_info);
4174 ret = relocate_file_extent_cluster(rc->data_inode,
4180 rc->create_reloc_tree = 0;
4181 set_reloc_control(rc);
4183 backref_cache_cleanup(&rc->backref_cache);
4184 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4186 err = prepare_to_merge(rc, err);
4188 merge_reloc_roots(rc);
4190 rc->merge_reloc_tree = 0;
4191 unset_reloc_control(rc);
4192 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4194 /* get rid of pinned extents */
4195 trans = btrfs_join_transaction(rc->extent_root);
4196 if (IS_ERR(trans)) {
4197 err = PTR_ERR(trans);
4200 btrfs_commit_transaction(trans);
4202 btrfs_free_block_rsv(fs_info, rc->block_rsv);
4203 btrfs_free_path(path);
4207 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root, u64 objectid)
4210 struct btrfs_path *path;
4211 struct btrfs_inode_item *item;
4212 struct extent_buffer *leaf;
4215 path = btrfs_alloc_path();
4219 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4223 leaf = path->nodes[0];
4224 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4225 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4226 btrfs_set_inode_generation(leaf, item, 1);
4227 btrfs_set_inode_size(leaf, item, 0);
4228 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4229 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4230 BTRFS_INODE_PREALLOC);
4231 btrfs_mark_buffer_dirty(leaf);
4233 btrfs_free_path(path);
4238 * helper to create inode for data relocation.
4239 * the inode is in data relocation tree and its link count is 0
4241 static noinline_for_stack
4242 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4243 struct btrfs_block_group_cache *group)
4245 struct inode *inode = NULL;
4246 struct btrfs_trans_handle *trans;
4247 struct btrfs_root *root;
4248 struct btrfs_key key;
4252 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4254 return ERR_CAST(root);
4256 trans = btrfs_start_transaction(root, 6);
4258 return ERR_CAST(trans);
4260 err = btrfs_find_free_objectid(root, &objectid);
4264 err = __insert_orphan_inode(trans, root, objectid);
4267 key.objectid = objectid;
4268 key.type = BTRFS_INODE_ITEM_KEY;
4270 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4271 BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
4272 BTRFS_I(inode)->index_cnt = group->key.objectid;
4274 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4276 btrfs_end_transaction(trans);
4277 btrfs_btree_balance_dirty(fs_info);
4281 inode = ERR_PTR(err);
4286 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4288 struct reloc_control *rc;
4290 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4294 INIT_LIST_HEAD(&rc->reloc_roots);
4295 backref_cache_init(&rc->backref_cache);
4296 mapping_tree_init(&rc->reloc_root_tree);
4297 extent_io_tree_init(&rc->processed_blocks, NULL);
4302 * Print the block group being relocated
4304 static void describe_relocation(struct btrfs_fs_info *fs_info,
4305 struct btrfs_block_group_cache *block_group)
4307 char buf[128]; /* prefixed by a '|' that'll be dropped */
4308 u64 flags = block_group->flags;
4310 /* Shouldn't happen */
4312 strcpy(buf, "|NONE");
4316 #define DESCRIBE_FLAG(f, d) \
4317 if (flags & BTRFS_BLOCK_GROUP_##f) { \
4318 bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \
4319 flags &= ~BTRFS_BLOCK_GROUP_##f; \
4321 DESCRIBE_FLAG(DATA, "data");
4322 DESCRIBE_FLAG(SYSTEM, "system");
4323 DESCRIBE_FLAG(METADATA, "metadata");
4324 DESCRIBE_FLAG(RAID0, "raid0");
4325 DESCRIBE_FLAG(RAID1, "raid1");
4326 DESCRIBE_FLAG(DUP, "dup");
4327 DESCRIBE_FLAG(RAID10, "raid10");
4328 DESCRIBE_FLAG(RAID5, "raid5");
4329 DESCRIBE_FLAG(RAID6, "raid6");
4331 snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags);
4332 #undef DESCRIBE_FLAG
4336 "relocating block group %llu flags %s",
4337 block_group->key.objectid, buf + 1);
4341 * function to relocate all extents in a block group.
4343 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4345 struct btrfs_root *extent_root = fs_info->extent_root;
4346 struct reloc_control *rc;
4347 struct inode *inode;
4348 struct btrfs_path *path;
4353 rc = alloc_reloc_control(fs_info);
4357 rc->extent_root = extent_root;
4359 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4360 BUG_ON(!rc->block_group);
4362 ret = btrfs_inc_block_group_ro(fs_info, rc->block_group);
4369 path = btrfs_alloc_path();
4375 inode = lookup_free_space_inode(fs_info, rc->block_group, path);
4376 btrfs_free_path(path);
4379 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4381 ret = PTR_ERR(inode);
4383 if (ret && ret != -ENOENT) {
4388 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4389 if (IS_ERR(rc->data_inode)) {
4390 err = PTR_ERR(rc->data_inode);
4391 rc->data_inode = NULL;
4395 describe_relocation(fs_info, rc->block_group);
4397 btrfs_wait_block_group_reservations(rc->block_group);
4398 btrfs_wait_nocow_writers(rc->block_group);
4399 btrfs_wait_ordered_roots(fs_info, U64_MAX,
4400 rc->block_group->key.objectid,
4401 rc->block_group->key.offset);
4404 mutex_lock(&fs_info->cleaner_mutex);
4405 ret = relocate_block_group(rc);
4406 mutex_unlock(&fs_info->cleaner_mutex);
4412 if (rc->extents_found == 0)
4415 btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4417 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4418 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4424 invalidate_mapping_pages(rc->data_inode->i_mapping,
4426 rc->stage = UPDATE_DATA_PTRS;
4430 WARN_ON(rc->block_group->pinned > 0);
4431 WARN_ON(rc->block_group->reserved > 0);
4432 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4435 btrfs_dec_block_group_ro(rc->block_group);
4436 iput(rc->data_inode);
4437 btrfs_put_block_group(rc->block_group);
4442 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4444 struct btrfs_fs_info *fs_info = root->fs_info;
4445 struct btrfs_trans_handle *trans;
4448 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4450 return PTR_ERR(trans);
4452 memset(&root->root_item.drop_progress, 0,
4453 sizeof(root->root_item.drop_progress));
4454 root->root_item.drop_level = 0;
4455 btrfs_set_root_refs(&root->root_item, 0);
4456 ret = btrfs_update_root(trans, fs_info->tree_root,
4457 &root->root_key, &root->root_item);
4459 err = btrfs_end_transaction(trans);
4466 * recover relocation interrupted by system crash.
4468 * this function resumes merging reloc trees with corresponding fs trees.
4469 * this is important for keeping the sharing of tree blocks
4471 int btrfs_recover_relocation(struct btrfs_root *root)
4473 struct btrfs_fs_info *fs_info = root->fs_info;
4474 LIST_HEAD(reloc_roots);
4475 struct btrfs_key key;
4476 struct btrfs_root *fs_root;
4477 struct btrfs_root *reloc_root;
4478 struct btrfs_path *path;
4479 struct extent_buffer *leaf;
4480 struct reloc_control *rc = NULL;
4481 struct btrfs_trans_handle *trans;
4485 path = btrfs_alloc_path();
4488 path->reada = READA_BACK;
4490 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4491 key.type = BTRFS_ROOT_ITEM_KEY;
4492 key.offset = (u64)-1;
4495 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4502 if (path->slots[0] == 0)
4506 leaf = path->nodes[0];
4507 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4508 btrfs_release_path(path);
4510 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4511 key.type != BTRFS_ROOT_ITEM_KEY)
4514 reloc_root = btrfs_read_fs_root(root, &key);
4515 if (IS_ERR(reloc_root)) {
4516 err = PTR_ERR(reloc_root);
4520 list_add(&reloc_root->root_list, &reloc_roots);
4522 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4523 fs_root = read_fs_root(fs_info,
4524 reloc_root->root_key.offset);
4525 if (IS_ERR(fs_root)) {
4526 ret = PTR_ERR(fs_root);
4527 if (ret != -ENOENT) {
4531 ret = mark_garbage_root(reloc_root);
4539 if (key.offset == 0)
4544 btrfs_release_path(path);
4546 if (list_empty(&reloc_roots))
4549 rc = alloc_reloc_control(fs_info);
4555 rc->extent_root = fs_info->extent_root;
4557 set_reloc_control(rc);
4559 trans = btrfs_join_transaction(rc->extent_root);
4560 if (IS_ERR(trans)) {
4561 unset_reloc_control(rc);
4562 err = PTR_ERR(trans);
4566 rc->merge_reloc_tree = 1;
4568 while (!list_empty(&reloc_roots)) {
4569 reloc_root = list_entry(reloc_roots.next,
4570 struct btrfs_root, root_list);
4571 list_del(&reloc_root->root_list);
4573 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4574 list_add_tail(&reloc_root->root_list,
4579 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
4580 if (IS_ERR(fs_root)) {
4581 err = PTR_ERR(fs_root);
4585 err = __add_reloc_root(reloc_root);
4586 BUG_ON(err < 0); /* -ENOMEM or logic error */
4587 fs_root->reloc_root = reloc_root;
4590 err = btrfs_commit_transaction(trans);
4594 merge_reloc_roots(rc);
4596 unset_reloc_control(rc);
4598 trans = btrfs_join_transaction(rc->extent_root);
4599 if (IS_ERR(trans)) {
4600 err = PTR_ERR(trans);
4603 err = btrfs_commit_transaction(trans);
4607 if (!list_empty(&reloc_roots))
4608 free_reloc_roots(&reloc_roots);
4610 btrfs_free_path(path);
4613 /* cleanup orphan inode in data relocation tree */
4614 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4615 if (IS_ERR(fs_root))
4616 err = PTR_ERR(fs_root);
4618 err = btrfs_orphan_cleanup(fs_root);
4624 * helper to add ordered checksum for data relocation.
4626 * cloning checksum properly handles the nodatasum extents.
4627 * it also saves CPU time to re-calculate the checksum.
4629 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4631 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4632 struct btrfs_ordered_sum *sums;
4633 struct btrfs_ordered_extent *ordered;
4639 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4640 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4642 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4643 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4644 disk_bytenr + len - 1, &list, 0);
4648 while (!list_empty(&list)) {
4649 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4650 list_del_init(&sums->list);
4653 * We need to offset the new_bytenr based on where the csum is.
4654 * We need to do this because we will read in entire prealloc
4655 * extents but we may have written to say the middle of the
4656 * prealloc extent, so we need to make sure the csum goes with
4657 * the right disk offset.
4659 * We can do this because the data reloc inode refers strictly
4660 * to the on disk bytes, so we don't have to worry about
4661 * disk_len vs real len like with real inodes since it's all
4664 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4665 sums->bytenr = new_bytenr;
4667 btrfs_add_ordered_sum(inode, ordered, sums);
4670 btrfs_put_ordered_extent(ordered);
4674 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4675 struct btrfs_root *root, struct extent_buffer *buf,
4676 struct extent_buffer *cow)
4678 struct btrfs_fs_info *fs_info = root->fs_info;
4679 struct reloc_control *rc;
4680 struct backref_node *node;
4685 rc = fs_info->reloc_ctl;
4689 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4690 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4692 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4693 if (buf == root->node)
4694 __update_reloc_root(root, cow->start);
4697 level = btrfs_header_level(buf);
4698 if (btrfs_header_generation(buf) <=
4699 btrfs_root_last_snapshot(&root->root_item))
4702 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4703 rc->create_reloc_tree) {
4704 WARN_ON(!first_cow && level == 0);
4706 node = rc->backref_cache.path[level];
4707 BUG_ON(node->bytenr != buf->start &&
4708 node->new_bytenr != buf->start);
4710 drop_node_buffer(node);
4711 extent_buffer_get(cow);
4713 node->new_bytenr = cow->start;
4715 if (!node->pending) {
4716 list_move_tail(&node->list,
4717 &rc->backref_cache.pending[level]);
4722 __mark_block_processed(rc, node);
4724 if (first_cow && level > 0)
4725 rc->nodes_relocated += buf->len;
4728 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4729 ret = replace_file_extents(trans, rc, root, cow);
4734 * called before creating snapshot. it calculates metadata reservation
4735 * required for relocating tree blocks in the snapshot
4737 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4738 u64 *bytes_to_reserve)
4740 struct btrfs_root *root;
4741 struct reloc_control *rc;
4743 root = pending->root;
4744 if (!root->reloc_root)
4747 rc = root->fs_info->reloc_ctl;
4748 if (!rc->merge_reloc_tree)
4751 root = root->reloc_root;
4752 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4754 * relocation is in the stage of merging trees. the space
4755 * used by merging a reloc tree is twice the size of
4756 * relocated tree nodes in the worst case. half for cowing
4757 * the reloc tree, half for cowing the fs tree. the space
4758 * used by cowing the reloc tree will be freed after the
4759 * tree is dropped. if we create snapshot, cowing the fs
4760 * tree may use more space than it frees. so we need
4761 * reserve extra space.
4763 *bytes_to_reserve += rc->nodes_relocated;
4767 * called after snapshot is created. migrate block reservation
4768 * and create reloc root for the newly created snapshot
4770 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4771 struct btrfs_pending_snapshot *pending)
4773 struct btrfs_root *root = pending->root;
4774 struct btrfs_root *reloc_root;
4775 struct btrfs_root *new_root;
4776 struct reloc_control *rc;
4779 if (!root->reloc_root)
4782 rc = root->fs_info->reloc_ctl;
4783 rc->merging_rsv_size += rc->nodes_relocated;
4785 if (rc->merge_reloc_tree) {
4786 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4788 rc->nodes_relocated, 1);
4793 new_root = pending->snap;
4794 reloc_root = create_reloc_root(trans, root->reloc_root,
4795 new_root->root_key.objectid);
4796 if (IS_ERR(reloc_root))
4797 return PTR_ERR(reloc_root);
4799 ret = __add_reloc_root(reloc_root);
4801 new_root->reloc_root = reloc_root;
4803 if (rc->create_reloc_tree)
4804 ret = clone_backref_node(trans, rc, root, reloc_root);