1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2009 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
14 #include "transaction.h"
17 #include "btrfs_inode.h"
18 #include "async-thread.h"
19 #include "free-space-cache.h"
20 #include "inode-map.h"
22 #include "print-tree.h"
25 * backref_node, mapping_node and tree_block start with this
28 struct rb_node rb_node;
33 * present a tree block in the backref cache
36 struct rb_node rb_node;
40 /* objectid of tree block owner, can be not uptodate */
42 /* link to pending, changed or detached list */
43 struct list_head list;
44 /* list of upper level blocks reference this block */
45 struct list_head upper;
46 /* list of child blocks in the cache */
47 struct list_head lower;
48 /* NULL if this node is not tree root */
49 struct btrfs_root *root;
50 /* extent buffer got by COW the block */
51 struct extent_buffer *eb;
52 /* level of tree block */
54 /* is the block in non-reference counted tree */
55 unsigned int cowonly:1;
56 /* 1 if no child node in the cache */
57 unsigned int lowest:1;
58 /* is the extent buffer locked */
59 unsigned int locked:1;
60 /* has the block been processed */
61 unsigned int processed:1;
62 /* have backrefs of this block been checked */
63 unsigned int checked:1;
65 * 1 if corresponding block has been cowed but some upper
66 * level block pointers may not point to the new location
68 unsigned int pending:1;
70 * 1 if the backref node isn't connected to any other
73 unsigned int detached:1;
77 * present a block pointer in the backref cache
80 struct list_head list[2];
81 struct backref_node *node[2];
86 #define RELOCATION_RESERVED_NODES 256
88 struct backref_cache {
89 /* red black tree of all backref nodes in the cache */
90 struct rb_root rb_root;
91 /* for passing backref nodes to btrfs_reloc_cow_block */
92 struct backref_node *path[BTRFS_MAX_LEVEL];
94 * list of blocks that have been cowed but some block
95 * pointers in upper level blocks may not reflect the
98 struct list_head pending[BTRFS_MAX_LEVEL];
99 /* list of backref nodes with no child node */
100 struct list_head leaves;
101 /* list of blocks that have been cowed in current transaction */
102 struct list_head changed;
103 /* list of detached backref node. */
104 struct list_head detached;
113 * map address of tree root to tree
115 struct mapping_node {
116 struct rb_node rb_node;
121 struct mapping_tree {
122 struct rb_root rb_root;
127 * present a tree block to process
130 struct rb_node rb_node;
132 struct btrfs_key key;
133 unsigned int level:8;
134 unsigned int key_ready:1;
137 #define MAX_EXTENTS 128
139 struct file_extent_cluster {
142 u64 boundary[MAX_EXTENTS];
146 struct reloc_control {
147 /* block group to relocate */
148 struct btrfs_block_group_cache *block_group;
150 struct btrfs_root *extent_root;
151 /* inode for moving data */
152 struct inode *data_inode;
154 struct btrfs_block_rsv *block_rsv;
156 struct backref_cache backref_cache;
158 struct file_extent_cluster cluster;
159 /* tree blocks have been processed */
160 struct extent_io_tree processed_blocks;
161 /* map start of tree root to corresponding reloc tree */
162 struct mapping_tree reloc_root_tree;
163 /* list of reloc trees */
164 struct list_head reloc_roots;
165 /* size of metadata reservation for merging reloc trees */
166 u64 merging_rsv_size;
167 /* size of relocated tree nodes */
169 /* reserved size for block group relocation*/
175 unsigned int stage:8;
176 unsigned int create_reloc_tree:1;
177 unsigned int merge_reloc_tree:1;
178 unsigned int found_file_extent:1;
181 /* stages of data relocation */
182 #define MOVE_DATA_EXTENTS 0
183 #define UPDATE_DATA_PTRS 1
185 static void remove_backref_node(struct backref_cache *cache,
186 struct backref_node *node);
187 static void __mark_block_processed(struct reloc_control *rc,
188 struct backref_node *node);
190 static void mapping_tree_init(struct mapping_tree *tree)
192 tree->rb_root = RB_ROOT;
193 spin_lock_init(&tree->lock);
196 static void backref_cache_init(struct backref_cache *cache)
199 cache->rb_root = RB_ROOT;
200 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
201 INIT_LIST_HEAD(&cache->pending[i]);
202 INIT_LIST_HEAD(&cache->changed);
203 INIT_LIST_HEAD(&cache->detached);
204 INIT_LIST_HEAD(&cache->leaves);
207 static void backref_cache_cleanup(struct backref_cache *cache)
209 struct backref_node *node;
212 while (!list_empty(&cache->detached)) {
213 node = list_entry(cache->detached.next,
214 struct backref_node, list);
215 remove_backref_node(cache, node);
218 while (!list_empty(&cache->leaves)) {
219 node = list_entry(cache->leaves.next,
220 struct backref_node, lower);
221 remove_backref_node(cache, node);
224 cache->last_trans = 0;
226 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
227 ASSERT(list_empty(&cache->pending[i]));
228 ASSERT(list_empty(&cache->changed));
229 ASSERT(list_empty(&cache->detached));
230 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
231 ASSERT(!cache->nr_nodes);
232 ASSERT(!cache->nr_edges);
235 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
237 struct backref_node *node;
239 node = kzalloc(sizeof(*node), GFP_NOFS);
241 INIT_LIST_HEAD(&node->list);
242 INIT_LIST_HEAD(&node->upper);
243 INIT_LIST_HEAD(&node->lower);
244 RB_CLEAR_NODE(&node->rb_node);
250 static void free_backref_node(struct backref_cache *cache,
251 struct backref_node *node)
259 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
261 struct backref_edge *edge;
263 edge = kzalloc(sizeof(*edge), GFP_NOFS);
269 static void free_backref_edge(struct backref_cache *cache,
270 struct backref_edge *edge)
278 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
279 struct rb_node *node)
281 struct rb_node **p = &root->rb_node;
282 struct rb_node *parent = NULL;
283 struct tree_entry *entry;
287 entry = rb_entry(parent, struct tree_entry, rb_node);
289 if (bytenr < entry->bytenr)
291 else if (bytenr > entry->bytenr)
297 rb_link_node(node, parent, p);
298 rb_insert_color(node, root);
302 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
304 struct rb_node *n = root->rb_node;
305 struct tree_entry *entry;
308 entry = rb_entry(n, struct tree_entry, rb_node);
310 if (bytenr < entry->bytenr)
312 else if (bytenr > entry->bytenr)
320 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
323 struct btrfs_fs_info *fs_info = NULL;
324 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
327 fs_info = bnode->root->fs_info;
328 btrfs_panic(fs_info, errno,
329 "Inconsistency in backref cache found at offset %llu",
334 * walk up backref nodes until reach node presents tree root
336 static struct backref_node *walk_up_backref(struct backref_node *node,
337 struct backref_edge *edges[],
340 struct backref_edge *edge;
343 while (!list_empty(&node->upper)) {
344 edge = list_entry(node->upper.next,
345 struct backref_edge, list[LOWER]);
347 node = edge->node[UPPER];
349 BUG_ON(node->detached);
355 * walk down backref nodes to find start of next reference path
357 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
360 struct backref_edge *edge;
361 struct backref_node *lower;
365 edge = edges[idx - 1];
366 lower = edge->node[LOWER];
367 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
371 edge = list_entry(edge->list[LOWER].next,
372 struct backref_edge, list[LOWER]);
373 edges[idx - 1] = edge;
375 return edge->node[UPPER];
381 static void unlock_node_buffer(struct backref_node *node)
384 btrfs_tree_unlock(node->eb);
389 static void drop_node_buffer(struct backref_node *node)
392 unlock_node_buffer(node);
393 free_extent_buffer(node->eb);
398 static void drop_backref_node(struct backref_cache *tree,
399 struct backref_node *node)
401 BUG_ON(!list_empty(&node->upper));
403 drop_node_buffer(node);
404 list_del(&node->list);
405 list_del(&node->lower);
406 if (!RB_EMPTY_NODE(&node->rb_node))
407 rb_erase(&node->rb_node, &tree->rb_root);
408 free_backref_node(tree, node);
412 * remove a backref node from the backref cache
414 static void remove_backref_node(struct backref_cache *cache,
415 struct backref_node *node)
417 struct backref_node *upper;
418 struct backref_edge *edge;
423 BUG_ON(!node->lowest && !node->detached);
424 while (!list_empty(&node->upper)) {
425 edge = list_entry(node->upper.next, struct backref_edge,
427 upper = edge->node[UPPER];
428 list_del(&edge->list[LOWER]);
429 list_del(&edge->list[UPPER]);
430 free_backref_edge(cache, edge);
432 if (RB_EMPTY_NODE(&upper->rb_node)) {
433 BUG_ON(!list_empty(&node->upper));
434 drop_backref_node(cache, node);
440 * add the node to leaf node list if no other
441 * child block cached.
443 if (list_empty(&upper->lower)) {
444 list_add_tail(&upper->lower, &cache->leaves);
449 drop_backref_node(cache, node);
452 static void update_backref_node(struct backref_cache *cache,
453 struct backref_node *node, u64 bytenr)
455 struct rb_node *rb_node;
456 rb_erase(&node->rb_node, &cache->rb_root);
457 node->bytenr = bytenr;
458 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
460 backref_tree_panic(rb_node, -EEXIST, bytenr);
464 * update backref cache after a transaction commit
466 static int update_backref_cache(struct btrfs_trans_handle *trans,
467 struct backref_cache *cache)
469 struct backref_node *node;
472 if (cache->last_trans == 0) {
473 cache->last_trans = trans->transid;
477 if (cache->last_trans == trans->transid)
481 * detached nodes are used to avoid unnecessary backref
482 * lookup. transaction commit changes the extent tree.
483 * so the detached nodes are no longer useful.
485 while (!list_empty(&cache->detached)) {
486 node = list_entry(cache->detached.next,
487 struct backref_node, list);
488 remove_backref_node(cache, node);
491 while (!list_empty(&cache->changed)) {
492 node = list_entry(cache->changed.next,
493 struct backref_node, list);
494 list_del_init(&node->list);
495 BUG_ON(node->pending);
496 update_backref_node(cache, node, node->new_bytenr);
500 * some nodes can be left in the pending list if there were
501 * errors during processing the pending nodes.
503 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
504 list_for_each_entry(node, &cache->pending[level], list) {
505 BUG_ON(!node->pending);
506 if (node->bytenr == node->new_bytenr)
508 update_backref_node(cache, node, node->new_bytenr);
512 cache->last_trans = 0;
517 static int should_ignore_root(struct btrfs_root *root)
519 struct btrfs_root *reloc_root;
521 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
524 reloc_root = root->reloc_root;
528 if (btrfs_header_generation(reloc_root->commit_root) ==
529 root->fs_info->running_transaction->transid)
532 * if there is reloc tree and it was created in previous
533 * transaction backref lookup can find the reloc tree,
534 * so backref node for the fs tree root is useless for
540 * find reloc tree by address of tree root
542 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
545 struct rb_node *rb_node;
546 struct mapping_node *node;
547 struct btrfs_root *root = NULL;
549 spin_lock(&rc->reloc_root_tree.lock);
550 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
552 node = rb_entry(rb_node, struct mapping_node, rb_node);
553 root = (struct btrfs_root *)node->data;
555 spin_unlock(&rc->reloc_root_tree.lock);
559 static int is_cowonly_root(u64 root_objectid)
561 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
562 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
563 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
564 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
565 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
566 root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
567 root_objectid == BTRFS_UUID_TREE_OBJECTID ||
568 root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
569 root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
574 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
577 struct btrfs_key key;
579 key.objectid = root_objectid;
580 key.type = BTRFS_ROOT_ITEM_KEY;
581 if (is_cowonly_root(root_objectid))
584 key.offset = (u64)-1;
586 return btrfs_get_fs_root(fs_info, &key, false);
589 static noinline_for_stack
590 int find_inline_backref(struct extent_buffer *leaf, int slot,
591 unsigned long *ptr, unsigned long *end)
593 struct btrfs_key key;
594 struct btrfs_extent_item *ei;
595 struct btrfs_tree_block_info *bi;
598 btrfs_item_key_to_cpu(leaf, &key, slot);
600 item_size = btrfs_item_size_nr(leaf, slot);
601 if (item_size < sizeof(*ei)) {
602 btrfs_print_v0_err(leaf->fs_info);
603 btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL);
606 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
607 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
608 BTRFS_EXTENT_FLAG_TREE_BLOCK));
610 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
611 item_size <= sizeof(*ei) + sizeof(*bi)) {
612 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
615 if (key.type == BTRFS_METADATA_ITEM_KEY &&
616 item_size <= sizeof(*ei)) {
617 WARN_ON(item_size < sizeof(*ei));
621 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
622 bi = (struct btrfs_tree_block_info *)(ei + 1);
623 *ptr = (unsigned long)(bi + 1);
625 *ptr = (unsigned long)(ei + 1);
627 *end = (unsigned long)ei + item_size;
632 * build backref tree for a given tree block. root of the backref tree
633 * corresponds the tree block, leaves of the backref tree correspond
634 * roots of b-trees that reference the tree block.
636 * the basic idea of this function is check backrefs of a given block
637 * to find upper level blocks that reference the block, and then check
638 * backrefs of these upper level blocks recursively. the recursion stop
639 * when tree root is reached or backrefs for the block is cached.
641 * NOTE: if we find backrefs for a block are cached, we know backrefs
642 * for all upper level blocks that directly/indirectly reference the
643 * block are also cached.
645 static noinline_for_stack
646 struct backref_node *build_backref_tree(struct reloc_control *rc,
647 struct btrfs_key *node_key,
648 int level, u64 bytenr)
650 struct backref_cache *cache = &rc->backref_cache;
651 struct btrfs_path *path1;
652 struct btrfs_path *path2;
653 struct extent_buffer *eb;
654 struct btrfs_root *root;
655 struct backref_node *cur;
656 struct backref_node *upper;
657 struct backref_node *lower;
658 struct backref_node *node = NULL;
659 struct backref_node *exist = NULL;
660 struct backref_edge *edge;
661 struct rb_node *rb_node;
662 struct btrfs_key key;
670 bool need_check = true;
672 path1 = btrfs_alloc_path();
673 path2 = btrfs_alloc_path();
674 if (!path1 || !path2) {
678 path1->reada = READA_FORWARD;
679 path2->reada = READA_FORWARD;
681 node = alloc_backref_node(cache);
687 node->bytenr = bytenr;
694 key.objectid = cur->bytenr;
695 key.type = BTRFS_METADATA_ITEM_KEY;
696 key.offset = (u64)-1;
698 path1->search_commit_root = 1;
699 path1->skip_locking = 1;
700 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
707 ASSERT(path1->slots[0]);
711 WARN_ON(cur->checked);
712 if (!list_empty(&cur->upper)) {
714 * the backref was added previously when processing
715 * backref of type BTRFS_TREE_BLOCK_REF_KEY
717 ASSERT(list_is_singular(&cur->upper));
718 edge = list_entry(cur->upper.next, struct backref_edge,
720 ASSERT(list_empty(&edge->list[UPPER]));
721 exist = edge->node[UPPER];
723 * add the upper level block to pending list if we need
727 list_add_tail(&edge->list[UPPER], &list);
734 eb = path1->nodes[0];
737 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
738 ret = btrfs_next_leaf(rc->extent_root, path1);
745 eb = path1->nodes[0];
748 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
749 if (key.objectid != cur->bytenr) {
754 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
755 key.type == BTRFS_METADATA_ITEM_KEY) {
756 ret = find_inline_backref(eb, path1->slots[0],
764 /* update key for inline back ref */
765 struct btrfs_extent_inline_ref *iref;
767 iref = (struct btrfs_extent_inline_ref *)ptr;
768 type = btrfs_get_extent_inline_ref_type(eb, iref,
769 BTRFS_REF_TYPE_BLOCK);
770 if (type == BTRFS_REF_TYPE_INVALID) {
775 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
777 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
778 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
782 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
783 exist->owner == key.offset) ||
784 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
785 exist->bytenr == key.offset))) {
790 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
791 if (key.objectid == key.offset) {
793 * only root blocks of reloc trees use
794 * backref of this type.
796 root = find_reloc_root(rc, cur->bytenr);
802 edge = alloc_backref_edge(cache);
807 rb_node = tree_search(&cache->rb_root, key.offset);
809 upper = alloc_backref_node(cache);
811 free_backref_edge(cache, edge);
815 upper->bytenr = key.offset;
816 upper->level = cur->level + 1;
818 * backrefs for the upper level block isn't
819 * cached, add the block to pending list
821 list_add_tail(&edge->list[UPPER], &list);
823 upper = rb_entry(rb_node, struct backref_node,
825 ASSERT(upper->checked);
826 INIT_LIST_HEAD(&edge->list[UPPER]);
828 list_add_tail(&edge->list[LOWER], &cur->upper);
829 edge->node[LOWER] = cur;
830 edge->node[UPPER] = upper;
833 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
835 btrfs_print_v0_err(rc->extent_root->fs_info);
836 btrfs_handle_fs_error(rc->extent_root->fs_info, err,
839 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
843 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
844 root = read_fs_root(rc->extent_root->fs_info, key.offset);
850 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
853 if (btrfs_root_level(&root->root_item) == cur->level) {
855 ASSERT(btrfs_root_bytenr(&root->root_item) ==
857 if (should_ignore_root(root))
858 list_add(&cur->list, &useless);
864 level = cur->level + 1;
867 * searching the tree to find upper level blocks
868 * reference the block.
870 path2->search_commit_root = 1;
871 path2->skip_locking = 1;
872 path2->lowest_level = level;
873 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
874 path2->lowest_level = 0;
879 if (ret > 0 && path2->slots[level] > 0)
880 path2->slots[level]--;
882 eb = path2->nodes[level];
883 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
885 btrfs_err(root->fs_info,
886 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
887 cur->bytenr, level - 1, root->objectid,
888 node_key->objectid, node_key->type,
895 for (; level < BTRFS_MAX_LEVEL; level++) {
896 if (!path2->nodes[level]) {
897 ASSERT(btrfs_root_bytenr(&root->root_item) ==
899 if (should_ignore_root(root))
900 list_add(&lower->list, &useless);
906 edge = alloc_backref_edge(cache);
912 eb = path2->nodes[level];
913 rb_node = tree_search(&cache->rb_root, eb->start);
915 upper = alloc_backref_node(cache);
917 free_backref_edge(cache, edge);
921 upper->bytenr = eb->start;
922 upper->owner = btrfs_header_owner(eb);
923 upper->level = lower->level + 1;
924 if (!test_bit(BTRFS_ROOT_REF_COWS,
929 * if we know the block isn't shared
930 * we can void checking its backrefs.
932 if (btrfs_block_can_be_shared(root, eb))
938 * add the block to pending list if we
939 * need check its backrefs, we only do this once
940 * while walking up a tree as we will catch
941 * anything else later on.
943 if (!upper->checked && need_check) {
945 list_add_tail(&edge->list[UPPER],
950 INIT_LIST_HEAD(&edge->list[UPPER]);
953 upper = rb_entry(rb_node, struct backref_node,
955 ASSERT(upper->checked);
956 INIT_LIST_HEAD(&edge->list[UPPER]);
958 upper->owner = btrfs_header_owner(eb);
960 list_add_tail(&edge->list[LOWER], &lower->upper);
961 edge->node[LOWER] = lower;
962 edge->node[UPPER] = upper;
969 btrfs_release_path(path2);
972 ptr += btrfs_extent_inline_ref_size(key.type);
982 btrfs_release_path(path1);
987 /* the pending list isn't empty, take the first block to process */
988 if (!list_empty(&list)) {
989 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
990 list_del_init(&edge->list[UPPER]);
991 cur = edge->node[UPPER];
996 * everything goes well, connect backref nodes and insert backref nodes
999 ASSERT(node->checked);
1000 cowonly = node->cowonly;
1002 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1005 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1006 list_add_tail(&node->lower, &cache->leaves);
1009 list_for_each_entry(edge, &node->upper, list[LOWER])
1010 list_add_tail(&edge->list[UPPER], &list);
1012 while (!list_empty(&list)) {
1013 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1014 list_del_init(&edge->list[UPPER]);
1015 upper = edge->node[UPPER];
1016 if (upper->detached) {
1017 list_del(&edge->list[LOWER]);
1018 lower = edge->node[LOWER];
1019 free_backref_edge(cache, edge);
1020 if (list_empty(&lower->upper))
1021 list_add(&lower->list, &useless);
1025 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1026 if (upper->lowest) {
1027 list_del_init(&upper->lower);
1031 list_add_tail(&edge->list[UPPER], &upper->lower);
1035 if (!upper->checked) {
1037 * Still want to blow up for developers since this is a
1044 if (cowonly != upper->cowonly) {
1051 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1054 backref_tree_panic(rb_node, -EEXIST,
1058 list_add_tail(&edge->list[UPPER], &upper->lower);
1060 list_for_each_entry(edge, &upper->upper, list[LOWER])
1061 list_add_tail(&edge->list[UPPER], &list);
1064 * process useless backref nodes. backref nodes for tree leaves
1065 * are deleted from the cache. backref nodes for upper level
1066 * tree blocks are left in the cache to avoid unnecessary backref
1069 while (!list_empty(&useless)) {
1070 upper = list_entry(useless.next, struct backref_node, list);
1071 list_del_init(&upper->list);
1072 ASSERT(list_empty(&upper->upper));
1075 if (upper->lowest) {
1076 list_del_init(&upper->lower);
1079 while (!list_empty(&upper->lower)) {
1080 edge = list_entry(upper->lower.next,
1081 struct backref_edge, list[UPPER]);
1082 list_del(&edge->list[UPPER]);
1083 list_del(&edge->list[LOWER]);
1084 lower = edge->node[LOWER];
1085 free_backref_edge(cache, edge);
1087 if (list_empty(&lower->upper))
1088 list_add(&lower->list, &useless);
1090 __mark_block_processed(rc, upper);
1091 if (upper->level > 0) {
1092 list_add(&upper->list, &cache->detached);
1093 upper->detached = 1;
1095 rb_erase(&upper->rb_node, &cache->rb_root);
1096 free_backref_node(cache, upper);
1100 btrfs_free_path(path1);
1101 btrfs_free_path(path2);
1103 while (!list_empty(&useless)) {
1104 lower = list_entry(useless.next,
1105 struct backref_node, list);
1106 list_del_init(&lower->list);
1108 while (!list_empty(&list)) {
1109 edge = list_first_entry(&list, struct backref_edge,
1111 list_del(&edge->list[UPPER]);
1112 list_del(&edge->list[LOWER]);
1113 lower = edge->node[LOWER];
1114 upper = edge->node[UPPER];
1115 free_backref_edge(cache, edge);
1118 * Lower is no longer linked to any upper backref nodes
1119 * and isn't in the cache, we can free it ourselves.
1121 if (list_empty(&lower->upper) &&
1122 RB_EMPTY_NODE(&lower->rb_node))
1123 list_add(&lower->list, &useless);
1125 if (!RB_EMPTY_NODE(&upper->rb_node))
1128 /* Add this guy's upper edges to the list to process */
1129 list_for_each_entry(edge, &upper->upper, list[LOWER])
1130 list_add_tail(&edge->list[UPPER], &list);
1131 if (list_empty(&upper->upper))
1132 list_add(&upper->list, &useless);
1135 while (!list_empty(&useless)) {
1136 lower = list_entry(useless.next,
1137 struct backref_node, list);
1138 list_del_init(&lower->list);
1141 free_backref_node(cache, lower);
1144 remove_backref_node(cache, node);
1145 return ERR_PTR(err);
1147 ASSERT(!node || !node->detached);
1152 * helper to add backref node for the newly created snapshot.
1153 * the backref node is created by cloning backref node that
1154 * corresponds to root of source tree
1156 static int clone_backref_node(struct btrfs_trans_handle *trans,
1157 struct reloc_control *rc,
1158 struct btrfs_root *src,
1159 struct btrfs_root *dest)
1161 struct btrfs_root *reloc_root = src->reloc_root;
1162 struct backref_cache *cache = &rc->backref_cache;
1163 struct backref_node *node = NULL;
1164 struct backref_node *new_node;
1165 struct backref_edge *edge;
1166 struct backref_edge *new_edge;
1167 struct rb_node *rb_node;
1169 if (cache->last_trans > 0)
1170 update_backref_cache(trans, cache);
1172 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1174 node = rb_entry(rb_node, struct backref_node, rb_node);
1178 BUG_ON(node->new_bytenr != reloc_root->node->start);
1182 rb_node = tree_search(&cache->rb_root,
1183 reloc_root->commit_root->start);
1185 node = rb_entry(rb_node, struct backref_node,
1187 BUG_ON(node->detached);
1194 new_node = alloc_backref_node(cache);
1198 new_node->bytenr = dest->node->start;
1199 new_node->level = node->level;
1200 new_node->lowest = node->lowest;
1201 new_node->checked = 1;
1202 new_node->root = dest;
1204 if (!node->lowest) {
1205 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1206 new_edge = alloc_backref_edge(cache);
1210 new_edge->node[UPPER] = new_node;
1211 new_edge->node[LOWER] = edge->node[LOWER];
1212 list_add_tail(&new_edge->list[UPPER],
1216 list_add_tail(&new_node->lower, &cache->leaves);
1219 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1220 &new_node->rb_node);
1222 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1224 if (!new_node->lowest) {
1225 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1226 list_add_tail(&new_edge->list[LOWER],
1227 &new_edge->node[LOWER]->upper);
1232 while (!list_empty(&new_node->lower)) {
1233 new_edge = list_entry(new_node->lower.next,
1234 struct backref_edge, list[UPPER]);
1235 list_del(&new_edge->list[UPPER]);
1236 free_backref_edge(cache, new_edge);
1238 free_backref_node(cache, new_node);
1243 * helper to add 'address of tree root -> reloc tree' mapping
1245 static int __must_check __add_reloc_root(struct btrfs_root *root)
1247 struct btrfs_fs_info *fs_info = root->fs_info;
1248 struct rb_node *rb_node;
1249 struct mapping_node *node;
1250 struct reloc_control *rc = fs_info->reloc_ctl;
1252 node = kmalloc(sizeof(*node), GFP_NOFS);
1256 node->bytenr = root->commit_root->start;
1259 spin_lock(&rc->reloc_root_tree.lock);
1260 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1261 node->bytenr, &node->rb_node);
1262 spin_unlock(&rc->reloc_root_tree.lock);
1264 btrfs_panic(fs_info, -EEXIST,
1265 "Duplicate root found for start=%llu while inserting into relocation tree",
1269 list_add_tail(&root->root_list, &rc->reloc_roots);
1274 * helper to delete the 'address of tree root -> reloc tree'
1277 static void __del_reloc_root(struct btrfs_root *root)
1279 struct btrfs_fs_info *fs_info = root->fs_info;
1280 struct rb_node *rb_node;
1281 struct mapping_node *node = NULL;
1282 struct reloc_control *rc = fs_info->reloc_ctl;
1284 if (rc && root->node) {
1285 spin_lock(&rc->reloc_root_tree.lock);
1286 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1287 root->commit_root->start);
1289 node = rb_entry(rb_node, struct mapping_node, rb_node);
1290 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1291 RB_CLEAR_NODE(&node->rb_node);
1293 spin_unlock(&rc->reloc_root_tree.lock);
1294 ASSERT(!node || (struct btrfs_root *)node->data == root);
1297 spin_lock(&fs_info->trans_lock);
1298 list_del_init(&root->root_list);
1299 spin_unlock(&fs_info->trans_lock);
1304 * helper to update the 'address of tree root -> reloc tree'
1307 static int __update_reloc_root(struct btrfs_root *root)
1309 struct btrfs_fs_info *fs_info = root->fs_info;
1310 struct rb_node *rb_node;
1311 struct mapping_node *node = NULL;
1312 struct reloc_control *rc = fs_info->reloc_ctl;
1314 spin_lock(&rc->reloc_root_tree.lock);
1315 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1316 root->commit_root->start);
1318 node = rb_entry(rb_node, struct mapping_node, rb_node);
1319 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1321 spin_unlock(&rc->reloc_root_tree.lock);
1325 BUG_ON((struct btrfs_root *)node->data != root);
1327 spin_lock(&rc->reloc_root_tree.lock);
1328 node->bytenr = root->node->start;
1329 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1330 node->bytenr, &node->rb_node);
1331 spin_unlock(&rc->reloc_root_tree.lock);
1333 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1337 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1338 struct btrfs_root *root, u64 objectid)
1340 struct btrfs_fs_info *fs_info = root->fs_info;
1341 struct btrfs_root *reloc_root;
1342 struct extent_buffer *eb;
1343 struct btrfs_root_item *root_item;
1344 struct btrfs_key root_key;
1347 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1350 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1351 root_key.type = BTRFS_ROOT_ITEM_KEY;
1352 root_key.offset = objectid;
1354 if (root->root_key.objectid == objectid) {
1355 u64 commit_root_gen;
1357 /* called by btrfs_init_reloc_root */
1358 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1359 BTRFS_TREE_RELOC_OBJECTID);
1362 * Set the last_snapshot field to the generation of the commit
1363 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1364 * correctly (returns true) when the relocation root is created
1365 * either inside the critical section of a transaction commit
1366 * (through transaction.c:qgroup_account_snapshot()) and when
1367 * it's created before the transaction commit is started.
1369 commit_root_gen = btrfs_header_generation(root->commit_root);
1370 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1373 * called by btrfs_reloc_post_snapshot_hook.
1374 * the source tree is a reloc tree, all tree blocks
1375 * modified after it was created have RELOC flag
1376 * set in their headers. so it's OK to not update
1377 * the 'last_snapshot'.
1379 ret = btrfs_copy_root(trans, root, root->node, &eb,
1380 BTRFS_TREE_RELOC_OBJECTID);
1384 memcpy(root_item, &root->root_item, sizeof(*root_item));
1385 btrfs_set_root_bytenr(root_item, eb->start);
1386 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1387 btrfs_set_root_generation(root_item, trans->transid);
1389 if (root->root_key.objectid == objectid) {
1390 btrfs_set_root_refs(root_item, 0);
1391 memset(&root_item->drop_progress, 0,
1392 sizeof(struct btrfs_disk_key));
1393 root_item->drop_level = 0;
1396 btrfs_tree_unlock(eb);
1397 free_extent_buffer(eb);
1399 ret = btrfs_insert_root(trans, fs_info->tree_root,
1400 &root_key, root_item);
1404 reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1405 BUG_ON(IS_ERR(reloc_root));
1406 reloc_root->last_trans = trans->transid;
1411 * create reloc tree for a given fs tree. reloc tree is just a
1412 * snapshot of the fs tree with special root objectid.
1414 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1415 struct btrfs_root *root)
1417 struct btrfs_fs_info *fs_info = root->fs_info;
1418 struct btrfs_root *reloc_root;
1419 struct reloc_control *rc = fs_info->reloc_ctl;
1420 struct btrfs_block_rsv *rsv;
1424 if (root->reloc_root) {
1425 reloc_root = root->reloc_root;
1426 reloc_root->last_trans = trans->transid;
1430 if (!rc || !rc->create_reloc_tree ||
1431 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1434 if (!trans->reloc_reserved) {
1435 rsv = trans->block_rsv;
1436 trans->block_rsv = rc->block_rsv;
1439 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1441 trans->block_rsv = rsv;
1443 ret = __add_reloc_root(reloc_root);
1445 root->reloc_root = reloc_root;
1450 * update root item of reloc tree
1452 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1453 struct btrfs_root *root)
1455 struct btrfs_fs_info *fs_info = root->fs_info;
1456 struct btrfs_root *reloc_root;
1457 struct btrfs_root_item *root_item;
1460 if (!root->reloc_root)
1463 reloc_root = root->reloc_root;
1464 root_item = &reloc_root->root_item;
1466 if (fs_info->reloc_ctl->merge_reloc_tree &&
1467 btrfs_root_refs(root_item) == 0) {
1468 root->reloc_root = NULL;
1469 __del_reloc_root(reloc_root);
1472 if (reloc_root->commit_root != reloc_root->node) {
1473 __update_reloc_root(reloc_root);
1474 btrfs_set_root_node(root_item, reloc_root->node);
1475 free_extent_buffer(reloc_root->commit_root);
1476 reloc_root->commit_root = btrfs_root_node(reloc_root);
1479 ret = btrfs_update_root(trans, fs_info->tree_root,
1480 &reloc_root->root_key, root_item);
1488 * helper to find first cached inode with inode number >= objectid
1491 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1493 struct rb_node *node;
1494 struct rb_node *prev;
1495 struct btrfs_inode *entry;
1496 struct inode *inode;
1498 spin_lock(&root->inode_lock);
1500 node = root->inode_tree.rb_node;
1504 entry = rb_entry(node, struct btrfs_inode, rb_node);
1506 if (objectid < btrfs_ino(entry))
1507 node = node->rb_left;
1508 else if (objectid > btrfs_ino(entry))
1509 node = node->rb_right;
1515 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1516 if (objectid <= btrfs_ino(entry)) {
1520 prev = rb_next(prev);
1524 entry = rb_entry(node, struct btrfs_inode, rb_node);
1525 inode = igrab(&entry->vfs_inode);
1527 spin_unlock(&root->inode_lock);
1531 objectid = btrfs_ino(entry) + 1;
1532 if (cond_resched_lock(&root->inode_lock))
1535 node = rb_next(node);
1537 spin_unlock(&root->inode_lock);
1541 static int in_block_group(u64 bytenr,
1542 struct btrfs_block_group_cache *block_group)
1544 if (bytenr >= block_group->key.objectid &&
1545 bytenr < block_group->key.objectid + block_group->key.offset)
1551 * get new location of data
1553 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1554 u64 bytenr, u64 num_bytes)
1556 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1557 struct btrfs_path *path;
1558 struct btrfs_file_extent_item *fi;
1559 struct extent_buffer *leaf;
1562 path = btrfs_alloc_path();
1566 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1567 ret = btrfs_lookup_file_extent(NULL, root, path,
1568 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1576 leaf = path->nodes[0];
1577 fi = btrfs_item_ptr(leaf, path->slots[0],
1578 struct btrfs_file_extent_item);
1580 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1581 btrfs_file_extent_compression(leaf, fi) ||
1582 btrfs_file_extent_encryption(leaf, fi) ||
1583 btrfs_file_extent_other_encoding(leaf, fi));
1585 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1590 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1593 btrfs_free_path(path);
1598 * update file extent items in the tree leaf to point to
1599 * the new locations.
1601 static noinline_for_stack
1602 int replace_file_extents(struct btrfs_trans_handle *trans,
1603 struct reloc_control *rc,
1604 struct btrfs_root *root,
1605 struct extent_buffer *leaf)
1607 struct btrfs_fs_info *fs_info = root->fs_info;
1608 struct btrfs_key key;
1609 struct btrfs_file_extent_item *fi;
1610 struct inode *inode = NULL;
1622 if (rc->stage != UPDATE_DATA_PTRS)
1625 /* reloc trees always use full backref */
1626 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1627 parent = leaf->start;
1631 nritems = btrfs_header_nritems(leaf);
1632 for (i = 0; i < nritems; i++) {
1634 btrfs_item_key_to_cpu(leaf, &key, i);
1635 if (key.type != BTRFS_EXTENT_DATA_KEY)
1637 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1638 if (btrfs_file_extent_type(leaf, fi) ==
1639 BTRFS_FILE_EXTENT_INLINE)
1641 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1642 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1645 if (!in_block_group(bytenr, rc->block_group))
1649 * if we are modifying block in fs tree, wait for readpage
1650 * to complete and drop the extent cache
1652 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1654 inode = find_next_inode(root, key.objectid);
1656 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1657 btrfs_add_delayed_iput(inode);
1658 inode = find_next_inode(root, key.objectid);
1660 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1662 btrfs_file_extent_num_bytes(leaf, fi);
1663 WARN_ON(!IS_ALIGNED(key.offset,
1664 fs_info->sectorsize));
1665 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1667 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1672 btrfs_drop_extent_cache(BTRFS_I(inode),
1673 key.offset, end, 1);
1674 unlock_extent(&BTRFS_I(inode)->io_tree,
1679 ret = get_new_location(rc->data_inode, &new_bytenr,
1683 * Don't have to abort since we've not changed anything
1684 * in the file extent yet.
1689 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1692 key.offset -= btrfs_file_extent_offset(leaf, fi);
1693 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1695 btrfs_header_owner(leaf),
1696 key.objectid, key.offset);
1698 btrfs_abort_transaction(trans, ret);
1702 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1703 parent, btrfs_header_owner(leaf),
1704 key.objectid, key.offset);
1706 btrfs_abort_transaction(trans, ret);
1711 btrfs_mark_buffer_dirty(leaf);
1713 btrfs_add_delayed_iput(inode);
1717 static noinline_for_stack
1718 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1719 struct btrfs_path *path, int level)
1721 struct btrfs_disk_key key1;
1722 struct btrfs_disk_key key2;
1723 btrfs_node_key(eb, &key1, slot);
1724 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1725 return memcmp(&key1, &key2, sizeof(key1));
1729 * try to replace tree blocks in fs tree with the new blocks
1730 * in reloc tree. tree blocks haven't been modified since the
1731 * reloc tree was create can be replaced.
1733 * if a block was replaced, level of the block + 1 is returned.
1734 * if no block got replaced, 0 is returned. if there are other
1735 * errors, a negative error number is returned.
1737 static noinline_for_stack
1738 int replace_path(struct btrfs_trans_handle *trans,
1739 struct btrfs_root *dest, struct btrfs_root *src,
1740 struct btrfs_path *path, struct btrfs_key *next_key,
1741 int lowest_level, int max_level)
1743 struct btrfs_fs_info *fs_info = dest->fs_info;
1744 struct extent_buffer *eb;
1745 struct extent_buffer *parent;
1746 struct btrfs_key key;
1758 ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1759 ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1761 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1763 slot = path->slots[lowest_level];
1764 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1766 eb = btrfs_lock_root_node(dest);
1767 btrfs_set_lock_blocking(eb);
1768 level = btrfs_header_level(eb);
1770 if (level < lowest_level) {
1771 btrfs_tree_unlock(eb);
1772 free_extent_buffer(eb);
1777 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1780 btrfs_set_lock_blocking(eb);
1783 next_key->objectid = (u64)-1;
1784 next_key->type = (u8)-1;
1785 next_key->offset = (u64)-1;
1790 struct btrfs_key first_key;
1792 level = btrfs_header_level(parent);
1793 ASSERT(level >= lowest_level);
1795 ret = btrfs_bin_search(parent, &key, level, &slot);
1796 if (ret && slot > 0)
1799 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1800 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1802 old_bytenr = btrfs_node_blockptr(parent, slot);
1803 blocksize = fs_info->nodesize;
1804 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1805 btrfs_node_key_to_cpu(parent, &first_key, slot);
1807 if (level <= max_level) {
1808 eb = path->nodes[level];
1809 new_bytenr = btrfs_node_blockptr(eb,
1810 path->slots[level]);
1811 new_ptr_gen = btrfs_node_ptr_generation(eb,
1812 path->slots[level]);
1818 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1823 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1824 memcmp_node_keys(parent, slot, path, level)) {
1825 if (level <= lowest_level) {
1830 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1831 level - 1, &first_key);
1835 } else if (!extent_buffer_uptodate(eb)) {
1837 free_extent_buffer(eb);
1840 btrfs_tree_lock(eb);
1842 ret = btrfs_cow_block(trans, dest, eb, parent,
1846 btrfs_set_lock_blocking(eb);
1848 btrfs_tree_unlock(parent);
1849 free_extent_buffer(parent);
1856 btrfs_tree_unlock(parent);
1857 free_extent_buffer(parent);
1862 btrfs_node_key_to_cpu(path->nodes[level], &key,
1863 path->slots[level]);
1864 btrfs_release_path(path);
1866 path->lowest_level = level;
1867 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1868 path->lowest_level = 0;
1872 * Info qgroup to trace both subtrees.
1874 * We must trace both trees.
1875 * 1) Tree reloc subtree
1876 * If not traced, we will leak data numbers
1878 * If not traced, we will double count old data
1879 * and tree block numbers, if current trans doesn't free
1880 * data reloc tree inode.
1882 ret = btrfs_qgroup_trace_subtree(trans, parent,
1883 btrfs_header_generation(parent),
1884 btrfs_header_level(parent));
1887 ret = btrfs_qgroup_trace_subtree(trans, path->nodes[level],
1888 btrfs_header_generation(path->nodes[level]),
1889 btrfs_header_level(path->nodes[level]));
1894 * swap blocks in fs tree and reloc tree.
1896 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1897 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1898 btrfs_mark_buffer_dirty(parent);
1900 btrfs_set_node_blockptr(path->nodes[level],
1901 path->slots[level], old_bytenr);
1902 btrfs_set_node_ptr_generation(path->nodes[level],
1903 path->slots[level], old_ptr_gen);
1904 btrfs_mark_buffer_dirty(path->nodes[level]);
1906 ret = btrfs_inc_extent_ref(trans, src, old_bytenr,
1907 blocksize, path->nodes[level]->start,
1908 src->root_key.objectid, level - 1, 0);
1910 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr,
1911 blocksize, 0, dest->root_key.objectid,
1915 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1916 path->nodes[level]->start,
1917 src->root_key.objectid, level - 1, 0);
1920 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1921 0, dest->root_key.objectid, level - 1,
1925 btrfs_unlock_up_safe(path, 0);
1930 btrfs_tree_unlock(parent);
1931 free_extent_buffer(parent);
1936 * helper to find next relocated block in reloc tree
1938 static noinline_for_stack
1939 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1942 struct extent_buffer *eb;
1947 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1949 for (i = 0; i < *level; i++) {
1950 free_extent_buffer(path->nodes[i]);
1951 path->nodes[i] = NULL;
1954 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1955 eb = path->nodes[i];
1956 nritems = btrfs_header_nritems(eb);
1957 while (path->slots[i] + 1 < nritems) {
1959 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1966 free_extent_buffer(path->nodes[i]);
1967 path->nodes[i] = NULL;
1973 * walk down reloc tree to find relocated block of lowest level
1975 static noinline_for_stack
1976 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1979 struct btrfs_fs_info *fs_info = root->fs_info;
1980 struct extent_buffer *eb = NULL;
1987 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1989 for (i = *level; i > 0; i--) {
1990 struct btrfs_key first_key;
1992 eb = path->nodes[i];
1993 nritems = btrfs_header_nritems(eb);
1994 while (path->slots[i] < nritems) {
1995 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1996 if (ptr_gen > last_snapshot)
2000 if (path->slots[i] >= nritems) {
2011 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2012 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2013 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2017 } else if (!extent_buffer_uptodate(eb)) {
2018 free_extent_buffer(eb);
2021 BUG_ON(btrfs_header_level(eb) != i - 1);
2022 path->nodes[i - 1] = eb;
2023 path->slots[i - 1] = 0;
2029 * invalidate extent cache for file extents whose key in range of
2030 * [min_key, max_key)
2032 static int invalidate_extent_cache(struct btrfs_root *root,
2033 struct btrfs_key *min_key,
2034 struct btrfs_key *max_key)
2036 struct btrfs_fs_info *fs_info = root->fs_info;
2037 struct inode *inode = NULL;
2042 objectid = min_key->objectid;
2047 if (objectid > max_key->objectid)
2050 inode = find_next_inode(root, objectid);
2053 ino = btrfs_ino(BTRFS_I(inode));
2055 if (ino > max_key->objectid) {
2061 if (!S_ISREG(inode->i_mode))
2064 if (unlikely(min_key->objectid == ino)) {
2065 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2067 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2070 start = min_key->offset;
2071 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2077 if (unlikely(max_key->objectid == ino)) {
2078 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2080 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2083 if (max_key->offset == 0)
2085 end = max_key->offset;
2086 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2093 /* the lock_extent waits for readpage to complete */
2094 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2095 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2096 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2101 static int find_next_key(struct btrfs_path *path, int level,
2102 struct btrfs_key *key)
2105 while (level < BTRFS_MAX_LEVEL) {
2106 if (!path->nodes[level])
2108 if (path->slots[level] + 1 <
2109 btrfs_header_nritems(path->nodes[level])) {
2110 btrfs_node_key_to_cpu(path->nodes[level], key,
2111 path->slots[level] + 1);
2120 * merge the relocated tree blocks in reloc tree with corresponding
2123 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2124 struct btrfs_root *root)
2126 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2127 LIST_HEAD(inode_list);
2128 struct btrfs_key key;
2129 struct btrfs_key next_key;
2130 struct btrfs_trans_handle *trans = NULL;
2131 struct btrfs_root *reloc_root;
2132 struct btrfs_root_item *root_item;
2133 struct btrfs_path *path;
2134 struct extent_buffer *leaf;
2142 path = btrfs_alloc_path();
2145 path->reada = READA_FORWARD;
2147 reloc_root = root->reloc_root;
2148 root_item = &reloc_root->root_item;
2150 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2151 level = btrfs_root_level(root_item);
2152 extent_buffer_get(reloc_root->node);
2153 path->nodes[level] = reloc_root->node;
2154 path->slots[level] = 0;
2156 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2158 level = root_item->drop_level;
2160 path->lowest_level = level;
2161 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2162 path->lowest_level = 0;
2164 btrfs_free_path(path);
2168 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2169 path->slots[level]);
2170 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2172 btrfs_unlock_up_safe(path, 0);
2175 min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2176 memset(&next_key, 0, sizeof(next_key));
2179 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2180 BTRFS_RESERVE_FLUSH_ALL);
2185 trans = btrfs_start_transaction(root, 0);
2186 if (IS_ERR(trans)) {
2187 err = PTR_ERR(trans);
2191 trans->block_rsv = rc->block_rsv;
2196 ret = walk_down_reloc_tree(reloc_root, path, &level);
2204 if (!find_next_key(path, level, &key) &&
2205 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2208 ret = replace_path(trans, root, reloc_root, path,
2209 &next_key, level, max_level);
2218 btrfs_node_key_to_cpu(path->nodes[level], &key,
2219 path->slots[level]);
2223 ret = walk_up_reloc_tree(reloc_root, path, &level);
2229 * save the merging progress in the drop_progress.
2230 * this is OK since root refs == 1 in this case.
2232 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2233 path->slots[level]);
2234 root_item->drop_level = level;
2236 btrfs_end_transaction_throttle(trans);
2239 btrfs_btree_balance_dirty(fs_info);
2241 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2242 invalidate_extent_cache(root, &key, &next_key);
2246 * handle the case only one block in the fs tree need to be
2247 * relocated and the block is tree root.
2249 leaf = btrfs_lock_root_node(root);
2250 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2251 btrfs_tree_unlock(leaf);
2252 free_extent_buffer(leaf);
2256 btrfs_free_path(path);
2259 memset(&root_item->drop_progress, 0,
2260 sizeof(root_item->drop_progress));
2261 root_item->drop_level = 0;
2262 btrfs_set_root_refs(root_item, 0);
2263 btrfs_update_reloc_root(trans, root);
2267 btrfs_end_transaction_throttle(trans);
2269 btrfs_btree_balance_dirty(fs_info);
2271 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2272 invalidate_extent_cache(root, &key, &next_key);
2277 static noinline_for_stack
2278 int prepare_to_merge(struct reloc_control *rc, int err)
2280 struct btrfs_root *root = rc->extent_root;
2281 struct btrfs_fs_info *fs_info = root->fs_info;
2282 struct btrfs_root *reloc_root;
2283 struct btrfs_trans_handle *trans;
2284 LIST_HEAD(reloc_roots);
2288 mutex_lock(&fs_info->reloc_mutex);
2289 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2290 rc->merging_rsv_size += rc->nodes_relocated * 2;
2291 mutex_unlock(&fs_info->reloc_mutex);
2295 num_bytes = rc->merging_rsv_size;
2296 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2297 BTRFS_RESERVE_FLUSH_ALL);
2302 trans = btrfs_join_transaction(rc->extent_root);
2303 if (IS_ERR(trans)) {
2305 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2307 return PTR_ERR(trans);
2311 if (num_bytes != rc->merging_rsv_size) {
2312 btrfs_end_transaction(trans);
2313 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2319 rc->merge_reloc_tree = 1;
2321 while (!list_empty(&rc->reloc_roots)) {
2322 reloc_root = list_entry(rc->reloc_roots.next,
2323 struct btrfs_root, root_list);
2324 list_del_init(&reloc_root->root_list);
2326 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2327 BUG_ON(IS_ERR(root));
2328 BUG_ON(root->reloc_root != reloc_root);
2331 * set reference count to 1, so btrfs_recover_relocation
2332 * knows it should resumes merging
2335 btrfs_set_root_refs(&reloc_root->root_item, 1);
2336 btrfs_update_reloc_root(trans, root);
2338 list_add(&reloc_root->root_list, &reloc_roots);
2341 list_splice(&reloc_roots, &rc->reloc_roots);
2344 btrfs_commit_transaction(trans);
2346 btrfs_end_transaction(trans);
2350 static noinline_for_stack
2351 void free_reloc_roots(struct list_head *list)
2353 struct btrfs_root *reloc_root;
2355 while (!list_empty(list)) {
2356 reloc_root = list_entry(list->next, struct btrfs_root,
2358 __del_reloc_root(reloc_root);
2359 free_extent_buffer(reloc_root->node);
2360 free_extent_buffer(reloc_root->commit_root);
2361 reloc_root->node = NULL;
2362 reloc_root->commit_root = NULL;
2366 static noinline_for_stack
2367 void merge_reloc_roots(struct reloc_control *rc)
2369 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2370 struct btrfs_root *root;
2371 struct btrfs_root *reloc_root;
2372 LIST_HEAD(reloc_roots);
2376 root = rc->extent_root;
2379 * this serializes us with btrfs_record_root_in_transaction,
2380 * we have to make sure nobody is in the middle of
2381 * adding their roots to the list while we are
2384 mutex_lock(&fs_info->reloc_mutex);
2385 list_splice_init(&rc->reloc_roots, &reloc_roots);
2386 mutex_unlock(&fs_info->reloc_mutex);
2388 while (!list_empty(&reloc_roots)) {
2390 reloc_root = list_entry(reloc_roots.next,
2391 struct btrfs_root, root_list);
2393 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2394 root = read_fs_root(fs_info,
2395 reloc_root->root_key.offset);
2396 BUG_ON(IS_ERR(root));
2397 BUG_ON(root->reloc_root != reloc_root);
2399 ret = merge_reloc_root(rc, root);
2401 if (list_empty(&reloc_root->root_list))
2402 list_add_tail(&reloc_root->root_list,
2407 list_del_init(&reloc_root->root_list);
2410 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2412 if (list_empty(&reloc_root->root_list))
2413 list_add_tail(&reloc_root->root_list,
2425 btrfs_handle_fs_error(fs_info, ret, NULL);
2426 if (!list_empty(&reloc_roots))
2427 free_reloc_roots(&reloc_roots);
2429 /* new reloc root may be added */
2430 mutex_lock(&fs_info->reloc_mutex);
2431 list_splice_init(&rc->reloc_roots, &reloc_roots);
2432 mutex_unlock(&fs_info->reloc_mutex);
2433 if (!list_empty(&reloc_roots))
2434 free_reloc_roots(&reloc_roots);
2440 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2442 * here, but it's wrong. If we fail to start the transaction in
2443 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
2444 * have actually been removed from the reloc_root_tree rb tree. This is
2445 * fine because we're bailing here, and we hold a reference on the root
2446 * for the list that holds it, so these roots will be cleaned up when we
2447 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
2448 * will be cleaned up on unmount.
2450 * The remaining nodes will be cleaned up by free_reloc_control.
2454 static void free_block_list(struct rb_root *blocks)
2456 struct tree_block *block;
2457 struct rb_node *rb_node;
2458 while ((rb_node = rb_first(blocks))) {
2459 block = rb_entry(rb_node, struct tree_block, rb_node);
2460 rb_erase(rb_node, blocks);
2465 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2466 struct btrfs_root *reloc_root)
2468 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2469 struct btrfs_root *root;
2471 if (reloc_root->last_trans == trans->transid)
2474 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2475 BUG_ON(IS_ERR(root));
2476 BUG_ON(root->reloc_root != reloc_root);
2478 return btrfs_record_root_in_trans(trans, root);
2481 static noinline_for_stack
2482 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2483 struct reloc_control *rc,
2484 struct backref_node *node,
2485 struct backref_edge *edges[])
2487 struct backref_node *next;
2488 struct btrfs_root *root;
2494 next = walk_up_backref(next, edges, &index);
2497 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2499 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2500 record_reloc_root_in_trans(trans, root);
2504 btrfs_record_root_in_trans(trans, root);
2505 root = root->reloc_root;
2507 if (next->new_bytenr != root->node->start) {
2508 BUG_ON(next->new_bytenr);
2509 BUG_ON(!list_empty(&next->list));
2510 next->new_bytenr = root->node->start;
2512 list_add_tail(&next->list,
2513 &rc->backref_cache.changed);
2514 __mark_block_processed(rc, next);
2520 next = walk_down_backref(edges, &index);
2521 if (!next || next->level <= node->level)
2528 /* setup backref node path for btrfs_reloc_cow_block */
2530 rc->backref_cache.path[next->level] = next;
2533 next = edges[index]->node[UPPER];
2539 * select a tree root for relocation. return NULL if the block
2540 * is reference counted. we should use do_relocation() in this
2541 * case. return a tree root pointer if the block isn't reference
2542 * counted. return -ENOENT if the block is root of reloc tree.
2544 static noinline_for_stack
2545 struct btrfs_root *select_one_root(struct backref_node *node)
2547 struct backref_node *next;
2548 struct btrfs_root *root;
2549 struct btrfs_root *fs_root = NULL;
2550 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2556 next = walk_up_backref(next, edges, &index);
2560 /* no other choice for non-references counted tree */
2561 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2564 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2570 next = walk_down_backref(edges, &index);
2571 if (!next || next->level <= node->level)
2576 return ERR_PTR(-ENOENT);
2580 static noinline_for_stack
2581 u64 calcu_metadata_size(struct reloc_control *rc,
2582 struct backref_node *node, int reserve)
2584 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2585 struct backref_node *next = node;
2586 struct backref_edge *edge;
2587 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2591 BUG_ON(reserve && node->processed);
2596 if (next->processed && (reserve || next != node))
2599 num_bytes += fs_info->nodesize;
2601 if (list_empty(&next->upper))
2604 edge = list_entry(next->upper.next,
2605 struct backref_edge, list[LOWER]);
2606 edges[index++] = edge;
2607 next = edge->node[UPPER];
2609 next = walk_down_backref(edges, &index);
2614 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2615 struct reloc_control *rc,
2616 struct backref_node *node)
2618 struct btrfs_root *root = rc->extent_root;
2619 struct btrfs_fs_info *fs_info = root->fs_info;
2624 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2626 trans->block_rsv = rc->block_rsv;
2627 rc->reserved_bytes += num_bytes;
2630 * We are under a transaction here so we can only do limited flushing.
2631 * If we get an enospc just kick back -EAGAIN so we know to drop the
2632 * transaction and try to refill when we can flush all the things.
2634 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2635 BTRFS_RESERVE_FLUSH_LIMIT);
2637 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2638 while (tmp <= rc->reserved_bytes)
2641 * only one thread can access block_rsv at this point,
2642 * so we don't need hold lock to protect block_rsv.
2643 * we expand more reservation size here to allow enough
2644 * space for relocation and we will return eailer in
2647 rc->block_rsv->size = tmp + fs_info->nodesize *
2648 RELOCATION_RESERVED_NODES;
2656 * relocate a block tree, and then update pointers in upper level
2657 * blocks that reference the block to point to the new location.
2659 * if called by link_to_upper, the block has already been relocated.
2660 * in that case this function just updates pointers.
2662 static int do_relocation(struct btrfs_trans_handle *trans,
2663 struct reloc_control *rc,
2664 struct backref_node *node,
2665 struct btrfs_key *key,
2666 struct btrfs_path *path, int lowest)
2668 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2669 struct backref_node *upper;
2670 struct backref_edge *edge;
2671 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2672 struct btrfs_root *root;
2673 struct extent_buffer *eb;
2681 BUG_ON(lowest && node->eb);
2683 path->lowest_level = node->level + 1;
2684 rc->backref_cache.path[node->level] = node;
2685 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2686 struct btrfs_key first_key;
2690 upper = edge->node[UPPER];
2691 root = select_reloc_root(trans, rc, upper, edges);
2694 if (upper->eb && !upper->locked) {
2696 ret = btrfs_bin_search(upper->eb, key,
2697 upper->level, &slot);
2699 bytenr = btrfs_node_blockptr(upper->eb, slot);
2700 if (node->eb->start == bytenr)
2703 drop_node_buffer(upper);
2707 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2714 btrfs_release_path(path);
2719 upper->eb = path->nodes[upper->level];
2720 path->nodes[upper->level] = NULL;
2722 BUG_ON(upper->eb != path->nodes[upper->level]);
2726 path->locks[upper->level] = 0;
2728 slot = path->slots[upper->level];
2729 btrfs_release_path(path);
2731 ret = btrfs_bin_search(upper->eb, key, upper->level,
2736 bytenr = btrfs_node_blockptr(upper->eb, slot);
2738 if (bytenr != node->bytenr) {
2739 btrfs_err(root->fs_info,
2740 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2741 bytenr, node->bytenr, slot,
2747 if (node->eb->start == bytenr)
2751 blocksize = root->fs_info->nodesize;
2752 generation = btrfs_node_ptr_generation(upper->eb, slot);
2753 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2754 eb = read_tree_block(fs_info, bytenr, generation,
2755 upper->level - 1, &first_key);
2759 } else if (!extent_buffer_uptodate(eb)) {
2760 free_extent_buffer(eb);
2764 btrfs_tree_lock(eb);
2765 btrfs_set_lock_blocking(eb);
2768 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2770 btrfs_tree_unlock(eb);
2771 free_extent_buffer(eb);
2776 BUG_ON(node->eb != eb);
2778 btrfs_set_node_blockptr(upper->eb, slot,
2780 btrfs_set_node_ptr_generation(upper->eb, slot,
2782 btrfs_mark_buffer_dirty(upper->eb);
2784 ret = btrfs_inc_extent_ref(trans, root,
2785 node->eb->start, blocksize,
2787 btrfs_header_owner(upper->eb),
2791 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2795 if (!upper->pending)
2796 drop_node_buffer(upper);
2798 unlock_node_buffer(upper);
2803 if (!err && node->pending) {
2804 drop_node_buffer(node);
2805 list_move_tail(&node->list, &rc->backref_cache.changed);
2809 path->lowest_level = 0;
2810 BUG_ON(err == -ENOSPC);
2814 static int link_to_upper(struct btrfs_trans_handle *trans,
2815 struct reloc_control *rc,
2816 struct backref_node *node,
2817 struct btrfs_path *path)
2819 struct btrfs_key key;
2821 btrfs_node_key_to_cpu(node->eb, &key, 0);
2822 return do_relocation(trans, rc, node, &key, path, 0);
2825 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2826 struct reloc_control *rc,
2827 struct btrfs_path *path, int err)
2830 struct backref_cache *cache = &rc->backref_cache;
2831 struct backref_node *node;
2835 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2836 while (!list_empty(&cache->pending[level])) {
2837 node = list_entry(cache->pending[level].next,
2838 struct backref_node, list);
2839 list_move_tail(&node->list, &list);
2840 BUG_ON(!node->pending);
2843 ret = link_to_upper(trans, rc, node, path);
2848 list_splice_init(&list, &cache->pending[level]);
2853 static void mark_block_processed(struct reloc_control *rc,
2854 u64 bytenr, u32 blocksize)
2856 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2860 static void __mark_block_processed(struct reloc_control *rc,
2861 struct backref_node *node)
2864 if (node->level == 0 ||
2865 in_block_group(node->bytenr, rc->block_group)) {
2866 blocksize = rc->extent_root->fs_info->nodesize;
2867 mark_block_processed(rc, node->bytenr, blocksize);
2869 node->processed = 1;
2873 * mark a block and all blocks directly/indirectly reference the block
2876 static void update_processed_blocks(struct reloc_control *rc,
2877 struct backref_node *node)
2879 struct backref_node *next = node;
2880 struct backref_edge *edge;
2881 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2887 if (next->processed)
2890 __mark_block_processed(rc, next);
2892 if (list_empty(&next->upper))
2895 edge = list_entry(next->upper.next,
2896 struct backref_edge, list[LOWER]);
2897 edges[index++] = edge;
2898 next = edge->node[UPPER];
2900 next = walk_down_backref(edges, &index);
2904 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2906 u32 blocksize = rc->extent_root->fs_info->nodesize;
2908 if (test_range_bit(&rc->processed_blocks, bytenr,
2909 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2914 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2915 struct tree_block *block)
2917 struct extent_buffer *eb;
2919 BUG_ON(block->key_ready);
2920 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2921 block->level, NULL);
2924 } else if (!extent_buffer_uptodate(eb)) {
2925 free_extent_buffer(eb);
2928 WARN_ON(btrfs_header_level(eb) != block->level);
2929 if (block->level == 0)
2930 btrfs_item_key_to_cpu(eb, &block->key, 0);
2932 btrfs_node_key_to_cpu(eb, &block->key, 0);
2933 free_extent_buffer(eb);
2934 block->key_ready = 1;
2939 * helper function to relocate a tree block
2941 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2942 struct reloc_control *rc,
2943 struct backref_node *node,
2944 struct btrfs_key *key,
2945 struct btrfs_path *path)
2947 struct btrfs_root *root;
2953 BUG_ON(node->processed);
2954 root = select_one_root(node);
2955 if (root == ERR_PTR(-ENOENT)) {
2956 update_processed_blocks(rc, node);
2960 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2961 ret = reserve_metadata_space(trans, rc, node);
2967 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2968 BUG_ON(node->new_bytenr);
2969 BUG_ON(!list_empty(&node->list));
2970 btrfs_record_root_in_trans(trans, root);
2971 root = root->reloc_root;
2972 node->new_bytenr = root->node->start;
2974 list_add_tail(&node->list, &rc->backref_cache.changed);
2976 path->lowest_level = node->level;
2977 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2978 btrfs_release_path(path);
2983 update_processed_blocks(rc, node);
2985 ret = do_relocation(trans, rc, node, key, path, 1);
2988 if (ret || node->level == 0 || node->cowonly)
2989 remove_backref_node(&rc->backref_cache, node);
2994 * relocate a list of blocks
2996 static noinline_for_stack
2997 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2998 struct reloc_control *rc, struct rb_root *blocks)
3000 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3001 struct backref_node *node;
3002 struct btrfs_path *path;
3003 struct tree_block *block;
3004 struct rb_node *rb_node;
3008 path = btrfs_alloc_path();
3011 goto out_free_blocks;
3014 rb_node = rb_first(blocks);
3016 block = rb_entry(rb_node, struct tree_block, rb_node);
3017 if (!block->key_ready)
3018 readahead_tree_block(fs_info, block->bytenr);
3019 rb_node = rb_next(rb_node);
3022 rb_node = rb_first(blocks);
3024 block = rb_entry(rb_node, struct tree_block, rb_node);
3025 if (!block->key_ready) {
3026 err = get_tree_block_key(fs_info, block);
3030 rb_node = rb_next(rb_node);
3033 rb_node = rb_first(blocks);
3035 block = rb_entry(rb_node, struct tree_block, rb_node);
3037 node = build_backref_tree(rc, &block->key,
3038 block->level, block->bytenr);
3040 err = PTR_ERR(node);
3044 ret = relocate_tree_block(trans, rc, node, &block->key,
3047 if (ret != -EAGAIN || rb_node == rb_first(blocks))
3051 rb_node = rb_next(rb_node);
3054 err = finish_pending_nodes(trans, rc, path, err);
3057 btrfs_free_path(path);
3059 free_block_list(blocks);
3063 static noinline_for_stack
3064 int prealloc_file_extent_cluster(struct inode *inode,
3065 struct file_extent_cluster *cluster)
3070 u64 offset = BTRFS_I(inode)->index_cnt;
3074 u64 prealloc_start = cluster->start - offset;
3075 u64 prealloc_end = cluster->end - offset;
3077 struct extent_changeset *data_reserved = NULL;
3079 BUG_ON(cluster->start != cluster->boundary[0]);
3082 ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3083 prealloc_end + 1 - prealloc_start);
3087 cur_offset = prealloc_start;
3088 while (nr < cluster->nr) {
3089 start = cluster->boundary[nr] - offset;
3090 if (nr + 1 < cluster->nr)
3091 end = cluster->boundary[nr + 1] - 1 - offset;
3093 end = cluster->end - offset;
3095 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3096 num_bytes = end + 1 - start;
3097 if (cur_offset < start)
3098 btrfs_free_reserved_data_space(inode, data_reserved,
3099 cur_offset, start - cur_offset);
3100 ret = btrfs_prealloc_file_range(inode, 0, start,
3101 num_bytes, num_bytes,
3102 end + 1, &alloc_hint);
3103 cur_offset = end + 1;
3104 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3109 if (cur_offset < prealloc_end)
3110 btrfs_free_reserved_data_space(inode, data_reserved,
3111 cur_offset, prealloc_end + 1 - cur_offset);
3113 inode_unlock(inode);
3114 extent_changeset_free(data_reserved);
3118 static noinline_for_stack
3119 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3122 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3123 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3124 struct extent_map *em;
3127 em = alloc_extent_map();
3132 em->len = end + 1 - start;
3133 em->block_len = em->len;
3134 em->block_start = block_start;
3135 em->bdev = fs_info->fs_devices->latest_bdev;
3136 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3138 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3140 write_lock(&em_tree->lock);
3141 ret = add_extent_mapping(em_tree, em, 0);
3142 write_unlock(&em_tree->lock);
3143 if (ret != -EEXIST) {
3144 free_extent_map(em);
3147 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3149 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3153 static int relocate_file_extent_cluster(struct inode *inode,
3154 struct file_extent_cluster *cluster)
3156 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3159 u64 offset = BTRFS_I(inode)->index_cnt;
3160 unsigned long index;
3161 unsigned long last_index;
3163 struct file_ra_state *ra;
3164 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3171 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3175 ret = prealloc_file_extent_cluster(inode, cluster);
3179 file_ra_state_init(ra, inode->i_mapping);
3181 ret = setup_extent_mapping(inode, cluster->start - offset,
3182 cluster->end - offset, cluster->start);
3186 index = (cluster->start - offset) >> PAGE_SHIFT;
3187 last_index = (cluster->end - offset) >> PAGE_SHIFT;
3188 while (index <= last_index) {
3189 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3194 page = find_lock_page(inode->i_mapping, index);
3196 page_cache_sync_readahead(inode->i_mapping,
3198 last_index + 1 - index);
3199 page = find_or_create_page(inode->i_mapping, index,
3202 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3204 btrfs_delalloc_release_extents(BTRFS_I(inode),
3211 if (PageReadahead(page)) {
3212 page_cache_async_readahead(inode->i_mapping,
3213 ra, NULL, page, index,
3214 last_index + 1 - index);
3217 if (!PageUptodate(page)) {
3218 btrfs_readpage(NULL, page);
3220 if (!PageUptodate(page)) {
3223 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3225 btrfs_delalloc_release_extents(BTRFS_I(inode),
3232 page_start = page_offset(page);
3233 page_end = page_start + PAGE_SIZE - 1;
3235 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3237 set_page_extent_mapped(page);
3239 if (nr < cluster->nr &&
3240 page_start + offset == cluster->boundary[nr]) {
3241 set_extent_bits(&BTRFS_I(inode)->io_tree,
3242 page_start, page_end,
3247 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3252 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3254 btrfs_delalloc_release_extents(BTRFS_I(inode),
3257 clear_extent_bits(&BTRFS_I(inode)->io_tree,
3258 page_start, page_end,
3259 EXTENT_LOCKED | EXTENT_BOUNDARY);
3263 set_page_dirty(page);
3265 unlock_extent(&BTRFS_I(inode)->io_tree,
3266 page_start, page_end);
3271 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
3272 balance_dirty_pages_ratelimited(inode->i_mapping);
3273 btrfs_throttle(fs_info);
3275 WARN_ON(nr != cluster->nr);
3281 static noinline_for_stack
3282 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3283 struct file_extent_cluster *cluster)
3287 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3288 ret = relocate_file_extent_cluster(inode, cluster);
3295 cluster->start = extent_key->objectid;
3297 BUG_ON(cluster->nr >= MAX_EXTENTS);
3298 cluster->end = extent_key->objectid + extent_key->offset - 1;
3299 cluster->boundary[cluster->nr] = extent_key->objectid;
3302 if (cluster->nr >= MAX_EXTENTS) {
3303 ret = relocate_file_extent_cluster(inode, cluster);
3312 * helper to add a tree block to the list.
3313 * the major work is getting the generation and level of the block
3315 static int add_tree_block(struct reloc_control *rc,
3316 struct btrfs_key *extent_key,
3317 struct btrfs_path *path,
3318 struct rb_root *blocks)
3320 struct extent_buffer *eb;
3321 struct btrfs_extent_item *ei;
3322 struct btrfs_tree_block_info *bi;
3323 struct tree_block *block;
3324 struct rb_node *rb_node;
3329 eb = path->nodes[0];
3330 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3332 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3333 item_size >= sizeof(*ei) + sizeof(*bi)) {
3334 ei = btrfs_item_ptr(eb, path->slots[0],
3335 struct btrfs_extent_item);
3336 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3337 bi = (struct btrfs_tree_block_info *)(ei + 1);
3338 level = btrfs_tree_block_level(eb, bi);
3340 level = (int)extent_key->offset;
3342 generation = btrfs_extent_generation(eb, ei);
3343 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3344 btrfs_print_v0_err(eb->fs_info);
3345 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3351 btrfs_release_path(path);
3353 BUG_ON(level == -1);
3355 block = kmalloc(sizeof(*block), GFP_NOFS);
3359 block->bytenr = extent_key->objectid;
3360 block->key.objectid = rc->extent_root->fs_info->nodesize;
3361 block->key.offset = generation;
3362 block->level = level;
3363 block->key_ready = 0;
3365 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3367 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3373 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3375 static int __add_tree_block(struct reloc_control *rc,
3376 u64 bytenr, u32 blocksize,
3377 struct rb_root *blocks)
3379 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3380 struct btrfs_path *path;
3381 struct btrfs_key key;
3383 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3385 if (tree_block_processed(bytenr, rc))
3388 if (tree_search(blocks, bytenr))
3391 path = btrfs_alloc_path();
3395 key.objectid = bytenr;
3397 key.type = BTRFS_METADATA_ITEM_KEY;
3398 key.offset = (u64)-1;
3400 key.type = BTRFS_EXTENT_ITEM_KEY;
3401 key.offset = blocksize;
3404 path->search_commit_root = 1;
3405 path->skip_locking = 1;
3406 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3410 if (ret > 0 && skinny) {
3411 if (path->slots[0]) {
3413 btrfs_item_key_to_cpu(path->nodes[0], &key,
3415 if (key.objectid == bytenr &&
3416 (key.type == BTRFS_METADATA_ITEM_KEY ||
3417 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3418 key.offset == blocksize)))
3424 btrfs_release_path(path);
3430 btrfs_print_leaf(path->nodes[0]);
3432 "tree block extent item (%llu) is not found in extent tree",
3439 ret = add_tree_block(rc, &key, path, blocks);
3441 btrfs_free_path(path);
3446 * helper to check if the block use full backrefs for pointers in it
3448 static int block_use_full_backref(struct reloc_control *rc,
3449 struct extent_buffer *eb)
3454 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3455 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3458 ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3459 eb->start, btrfs_header_level(eb), 1,
3463 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3470 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3471 struct btrfs_block_group_cache *block_group,
3472 struct inode *inode,
3475 struct btrfs_key key;
3476 struct btrfs_root *root = fs_info->tree_root;
3477 struct btrfs_trans_handle *trans;
3484 key.type = BTRFS_INODE_ITEM_KEY;
3487 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3492 ret = btrfs_check_trunc_cache_free_space(fs_info,
3493 &fs_info->global_block_rsv);
3497 trans = btrfs_join_transaction(root);
3498 if (IS_ERR(trans)) {
3499 ret = PTR_ERR(trans);
3503 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3505 btrfs_end_transaction(trans);
3506 btrfs_btree_balance_dirty(fs_info);
3513 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3514 * this function scans fs tree to find blocks reference the data extent
3516 static int find_data_references(struct reloc_control *rc,
3517 struct btrfs_key *extent_key,
3518 struct extent_buffer *leaf,
3519 struct btrfs_extent_data_ref *ref,
3520 struct rb_root *blocks)
3522 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3523 struct btrfs_path *path;
3524 struct tree_block *block;
3525 struct btrfs_root *root;
3526 struct btrfs_file_extent_item *fi;
3527 struct rb_node *rb_node;
3528 struct btrfs_key key;
3539 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3540 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3541 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3542 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3545 * This is an extent belonging to the free space cache, lets just delete
3546 * it and redo the search.
3548 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3549 ret = delete_block_group_cache(fs_info, rc->block_group,
3550 NULL, ref_objectid);
3556 path = btrfs_alloc_path();
3559 path->reada = READA_FORWARD;
3561 root = read_fs_root(fs_info, ref_root);
3563 err = PTR_ERR(root);
3567 key.objectid = ref_objectid;
3568 key.type = BTRFS_EXTENT_DATA_KEY;
3569 if (ref_offset > ((u64)-1 << 32))
3572 key.offset = ref_offset;
3574 path->search_commit_root = 1;
3575 path->skip_locking = 1;
3576 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3582 leaf = path->nodes[0];
3583 nritems = btrfs_header_nritems(leaf);
3585 * the references in tree blocks that use full backrefs
3586 * are not counted in
3588 if (block_use_full_backref(rc, leaf))
3592 rb_node = tree_search(blocks, leaf->start);
3597 path->slots[0] = nritems;
3600 while (ref_count > 0) {
3601 while (path->slots[0] >= nritems) {
3602 ret = btrfs_next_leaf(root, path);
3607 if (WARN_ON(ret > 0))
3610 leaf = path->nodes[0];
3611 nritems = btrfs_header_nritems(leaf);
3614 if (block_use_full_backref(rc, leaf))
3618 rb_node = tree_search(blocks, leaf->start);
3623 path->slots[0] = nritems;
3627 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3628 if (WARN_ON(key.objectid != ref_objectid ||
3629 key.type != BTRFS_EXTENT_DATA_KEY))
3632 fi = btrfs_item_ptr(leaf, path->slots[0],
3633 struct btrfs_file_extent_item);
3635 if (btrfs_file_extent_type(leaf, fi) ==
3636 BTRFS_FILE_EXTENT_INLINE)
3639 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3640 extent_key->objectid)
3643 key.offset -= btrfs_file_extent_offset(leaf, fi);
3644 if (key.offset != ref_offset)
3652 if (!tree_block_processed(leaf->start, rc)) {
3653 block = kmalloc(sizeof(*block), GFP_NOFS);
3658 block->bytenr = leaf->start;
3659 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3661 block->key_ready = 1;
3662 rb_node = tree_insert(blocks, block->bytenr,
3665 backref_tree_panic(rb_node, -EEXIST,
3671 path->slots[0] = nritems;
3677 btrfs_free_path(path);
3682 * helper to find all tree blocks that reference a given data extent
3684 static noinline_for_stack
3685 int add_data_references(struct reloc_control *rc,
3686 struct btrfs_key *extent_key,
3687 struct btrfs_path *path,
3688 struct rb_root *blocks)
3690 struct btrfs_key key;
3691 struct extent_buffer *eb;
3692 struct btrfs_extent_data_ref *dref;
3693 struct btrfs_extent_inline_ref *iref;
3696 u32 blocksize = rc->extent_root->fs_info->nodesize;
3700 eb = path->nodes[0];
3701 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3702 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3703 ptr += sizeof(struct btrfs_extent_item);
3706 iref = (struct btrfs_extent_inline_ref *)ptr;
3707 key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3708 BTRFS_REF_TYPE_DATA);
3709 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3710 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3711 ret = __add_tree_block(rc, key.offset, blocksize,
3713 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3714 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3715 ret = find_data_references(rc, extent_key,
3719 btrfs_err(rc->extent_root->fs_info,
3720 "extent %llu slot %d has an invalid inline ref type",
3721 eb->start, path->slots[0]);
3727 ptr += btrfs_extent_inline_ref_size(key.type);
3733 eb = path->nodes[0];
3734 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3735 ret = btrfs_next_leaf(rc->extent_root, path);
3742 eb = path->nodes[0];
3745 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3746 if (key.objectid != extent_key->objectid)
3749 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3750 ret = __add_tree_block(rc, key.offset, blocksize,
3752 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3753 dref = btrfs_item_ptr(eb, path->slots[0],
3754 struct btrfs_extent_data_ref);
3755 ret = find_data_references(rc, extent_key,
3757 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3758 btrfs_print_v0_err(eb->fs_info);
3759 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3771 btrfs_release_path(path);
3773 free_block_list(blocks);
3778 * helper to find next unprocessed extent
3780 static noinline_for_stack
3781 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3782 struct btrfs_key *extent_key)
3784 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3785 struct btrfs_key key;
3786 struct extent_buffer *leaf;
3787 u64 start, end, last;
3790 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3793 if (rc->search_start >= last) {
3798 key.objectid = rc->search_start;
3799 key.type = BTRFS_EXTENT_ITEM_KEY;
3802 path->search_commit_root = 1;
3803 path->skip_locking = 1;
3804 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3809 leaf = path->nodes[0];
3810 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3811 ret = btrfs_next_leaf(rc->extent_root, path);
3814 leaf = path->nodes[0];
3817 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3818 if (key.objectid >= last) {
3823 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3824 key.type != BTRFS_METADATA_ITEM_KEY) {
3829 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3830 key.objectid + key.offset <= rc->search_start) {
3835 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3836 key.objectid + fs_info->nodesize <=
3842 ret = find_first_extent_bit(&rc->processed_blocks,
3843 key.objectid, &start, &end,
3844 EXTENT_DIRTY, NULL);
3846 if (ret == 0 && start <= key.objectid) {
3847 btrfs_release_path(path);
3848 rc->search_start = end + 1;
3850 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3851 rc->search_start = key.objectid + key.offset;
3853 rc->search_start = key.objectid +
3855 memcpy(extent_key, &key, sizeof(key));
3859 btrfs_release_path(path);
3863 static void set_reloc_control(struct reloc_control *rc)
3865 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3867 mutex_lock(&fs_info->reloc_mutex);
3868 fs_info->reloc_ctl = rc;
3869 mutex_unlock(&fs_info->reloc_mutex);
3872 static void unset_reloc_control(struct reloc_control *rc)
3874 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3876 mutex_lock(&fs_info->reloc_mutex);
3877 fs_info->reloc_ctl = NULL;
3878 mutex_unlock(&fs_info->reloc_mutex);
3881 static int check_extent_flags(u64 flags)
3883 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3884 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3886 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3887 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3889 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3890 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3895 static noinline_for_stack
3896 int prepare_to_relocate(struct reloc_control *rc)
3898 struct btrfs_trans_handle *trans;
3901 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3902 BTRFS_BLOCK_RSV_TEMP);
3906 memset(&rc->cluster, 0, sizeof(rc->cluster));
3907 rc->search_start = rc->block_group->key.objectid;
3908 rc->extents_found = 0;
3909 rc->nodes_relocated = 0;
3910 rc->merging_rsv_size = 0;
3911 rc->reserved_bytes = 0;
3912 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3913 RELOCATION_RESERVED_NODES;
3914 ret = btrfs_block_rsv_refill(rc->extent_root,
3915 rc->block_rsv, rc->block_rsv->size,
3916 BTRFS_RESERVE_FLUSH_ALL);
3920 rc->create_reloc_tree = 1;
3921 set_reloc_control(rc);
3923 trans = btrfs_join_transaction(rc->extent_root);
3924 if (IS_ERR(trans)) {
3925 unset_reloc_control(rc);
3927 * extent tree is not a ref_cow tree and has no reloc_root to
3928 * cleanup. And callers are responsible to free the above
3931 return PTR_ERR(trans);
3933 btrfs_commit_transaction(trans);
3937 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3939 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3940 struct rb_root blocks = RB_ROOT;
3941 struct btrfs_key key;
3942 struct btrfs_trans_handle *trans = NULL;
3943 struct btrfs_path *path;
3944 struct btrfs_extent_item *ei;
3951 path = btrfs_alloc_path();
3954 path->reada = READA_FORWARD;
3956 ret = prepare_to_relocate(rc);
3963 rc->reserved_bytes = 0;
3964 ret = btrfs_block_rsv_refill(rc->extent_root,
3965 rc->block_rsv, rc->block_rsv->size,
3966 BTRFS_RESERVE_FLUSH_ALL);
3972 trans = btrfs_start_transaction(rc->extent_root, 0);
3973 if (IS_ERR(trans)) {
3974 err = PTR_ERR(trans);
3979 if (update_backref_cache(trans, &rc->backref_cache)) {
3980 btrfs_end_transaction(trans);
3985 ret = find_next_extent(rc, path, &key);
3991 rc->extents_found++;
3993 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3994 struct btrfs_extent_item);
3995 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3996 if (item_size >= sizeof(*ei)) {
3997 flags = btrfs_extent_flags(path->nodes[0], ei);
3998 ret = check_extent_flags(flags);
4000 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
4002 btrfs_print_v0_err(trans->fs_info);
4003 btrfs_abort_transaction(trans, err);
4009 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4010 ret = add_tree_block(rc, &key, path, &blocks);
4011 } else if (rc->stage == UPDATE_DATA_PTRS &&
4012 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4013 ret = add_data_references(rc, &key, path, &blocks);
4015 btrfs_release_path(path);
4023 if (!RB_EMPTY_ROOT(&blocks)) {
4024 ret = relocate_tree_blocks(trans, rc, &blocks);
4027 * if we fail to relocate tree blocks, force to update
4028 * backref cache when committing transaction.
4030 rc->backref_cache.last_trans = trans->transid - 1;
4032 if (ret != -EAGAIN) {
4036 rc->extents_found--;
4037 rc->search_start = key.objectid;
4041 btrfs_end_transaction_throttle(trans);
4042 btrfs_btree_balance_dirty(fs_info);
4045 if (rc->stage == MOVE_DATA_EXTENTS &&
4046 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4047 rc->found_file_extent = 1;
4048 ret = relocate_data_extent(rc->data_inode,
4049 &key, &rc->cluster);
4056 if (trans && progress && err == -ENOSPC) {
4057 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
4065 btrfs_release_path(path);
4066 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4069 btrfs_end_transaction_throttle(trans);
4070 btrfs_btree_balance_dirty(fs_info);
4074 ret = relocate_file_extent_cluster(rc->data_inode,
4080 rc->create_reloc_tree = 0;
4081 set_reloc_control(rc);
4083 backref_cache_cleanup(&rc->backref_cache);
4084 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4086 err = prepare_to_merge(rc, err);
4088 merge_reloc_roots(rc);
4090 rc->merge_reloc_tree = 0;
4091 unset_reloc_control(rc);
4092 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4094 /* get rid of pinned extents */
4095 trans = btrfs_join_transaction(rc->extent_root);
4096 if (IS_ERR(trans)) {
4097 err = PTR_ERR(trans);
4100 btrfs_commit_transaction(trans);
4102 btrfs_free_block_rsv(fs_info, rc->block_rsv);
4103 btrfs_free_path(path);
4107 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4108 struct btrfs_root *root, u64 objectid)
4110 struct btrfs_path *path;
4111 struct btrfs_inode_item *item;
4112 struct extent_buffer *leaf;
4115 path = btrfs_alloc_path();
4119 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4123 leaf = path->nodes[0];
4124 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4125 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4126 btrfs_set_inode_generation(leaf, item, 1);
4127 btrfs_set_inode_size(leaf, item, 0);
4128 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4129 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4130 BTRFS_INODE_PREALLOC);
4131 btrfs_mark_buffer_dirty(leaf);
4133 btrfs_free_path(path);
4138 * helper to create inode for data relocation.
4139 * the inode is in data relocation tree and its link count is 0
4141 static noinline_for_stack
4142 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4143 struct btrfs_block_group_cache *group)
4145 struct inode *inode = NULL;
4146 struct btrfs_trans_handle *trans;
4147 struct btrfs_root *root;
4148 struct btrfs_key key;
4152 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4154 return ERR_CAST(root);
4156 trans = btrfs_start_transaction(root, 6);
4158 return ERR_CAST(trans);
4160 err = btrfs_find_free_objectid(root, &objectid);
4164 err = __insert_orphan_inode(trans, root, objectid);
4167 key.objectid = objectid;
4168 key.type = BTRFS_INODE_ITEM_KEY;
4170 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4171 BUG_ON(IS_ERR(inode));
4172 BTRFS_I(inode)->index_cnt = group->key.objectid;
4174 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4176 btrfs_end_transaction(trans);
4177 btrfs_btree_balance_dirty(fs_info);
4181 inode = ERR_PTR(err);
4186 static struct reloc_control *alloc_reloc_control(void)
4188 struct reloc_control *rc;
4190 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4194 INIT_LIST_HEAD(&rc->reloc_roots);
4195 backref_cache_init(&rc->backref_cache);
4196 mapping_tree_init(&rc->reloc_root_tree);
4197 extent_io_tree_init(&rc->processed_blocks, NULL);
4202 * Print the block group being relocated
4204 static void describe_relocation(struct btrfs_fs_info *fs_info,
4205 struct btrfs_block_group_cache *block_group)
4207 char buf[128]; /* prefixed by a '|' that'll be dropped */
4208 u64 flags = block_group->flags;
4210 /* Shouldn't happen */
4212 strcpy(buf, "|NONE");
4216 #define DESCRIBE_FLAG(f, d) \
4217 if (flags & BTRFS_BLOCK_GROUP_##f) { \
4218 bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \
4219 flags &= ~BTRFS_BLOCK_GROUP_##f; \
4221 DESCRIBE_FLAG(DATA, "data");
4222 DESCRIBE_FLAG(SYSTEM, "system");
4223 DESCRIBE_FLAG(METADATA, "metadata");
4224 DESCRIBE_FLAG(RAID0, "raid0");
4225 DESCRIBE_FLAG(RAID1, "raid1");
4226 DESCRIBE_FLAG(DUP, "dup");
4227 DESCRIBE_FLAG(RAID10, "raid10");
4228 DESCRIBE_FLAG(RAID5, "raid5");
4229 DESCRIBE_FLAG(RAID6, "raid6");
4231 snprintf(bp, buf - bp + sizeof(buf), "|0x%llx", flags);
4232 #undef DESCRIBE_FLAG
4236 "relocating block group %llu flags %s",
4237 block_group->key.objectid, buf + 1);
4241 * function to relocate all extents in a block group.
4243 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4245 struct btrfs_root *extent_root = fs_info->extent_root;
4246 struct reloc_control *rc;
4247 struct inode *inode;
4248 struct btrfs_path *path;
4253 rc = alloc_reloc_control();
4257 rc->extent_root = extent_root;
4259 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4260 BUG_ON(!rc->block_group);
4262 ret = btrfs_inc_block_group_ro(rc->block_group);
4269 path = btrfs_alloc_path();
4275 inode = lookup_free_space_inode(fs_info, rc->block_group, path);
4276 btrfs_free_path(path);
4279 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4281 ret = PTR_ERR(inode);
4283 if (ret && ret != -ENOENT) {
4288 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4289 if (IS_ERR(rc->data_inode)) {
4290 err = PTR_ERR(rc->data_inode);
4291 rc->data_inode = NULL;
4295 describe_relocation(fs_info, rc->block_group);
4297 btrfs_wait_block_group_reservations(rc->block_group);
4298 btrfs_wait_nocow_writers(rc->block_group);
4299 btrfs_wait_ordered_roots(fs_info, U64_MAX,
4300 rc->block_group->key.objectid,
4301 rc->block_group->key.offset);
4304 mutex_lock(&fs_info->cleaner_mutex);
4305 ret = relocate_block_group(rc);
4306 mutex_unlock(&fs_info->cleaner_mutex);
4311 * We may have gotten ENOSPC after we already dirtied some
4312 * extents. If writeout happens while we're relocating a
4313 * different block group we could end up hitting the
4314 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4315 * btrfs_reloc_cow_block. Make sure we write everything out
4316 * properly so we don't trip over this problem, and then break
4317 * out of the loop if we hit an error.
4319 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4320 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4324 invalidate_mapping_pages(rc->data_inode->i_mapping,
4326 rc->stage = UPDATE_DATA_PTRS;
4332 if (rc->extents_found == 0)
4335 btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4339 WARN_ON(rc->block_group->pinned > 0);
4340 WARN_ON(rc->block_group->reserved > 0);
4341 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4344 btrfs_dec_block_group_ro(rc->block_group);
4345 iput(rc->data_inode);
4346 btrfs_put_block_group(rc->block_group);
4351 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4353 struct btrfs_fs_info *fs_info = root->fs_info;
4354 struct btrfs_trans_handle *trans;
4357 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4359 return PTR_ERR(trans);
4361 memset(&root->root_item.drop_progress, 0,
4362 sizeof(root->root_item.drop_progress));
4363 root->root_item.drop_level = 0;
4364 btrfs_set_root_refs(&root->root_item, 0);
4365 ret = btrfs_update_root(trans, fs_info->tree_root,
4366 &root->root_key, &root->root_item);
4368 err = btrfs_end_transaction(trans);
4375 * recover relocation interrupted by system crash.
4377 * this function resumes merging reloc trees with corresponding fs trees.
4378 * this is important for keeping the sharing of tree blocks
4380 int btrfs_recover_relocation(struct btrfs_root *root)
4382 struct btrfs_fs_info *fs_info = root->fs_info;
4383 LIST_HEAD(reloc_roots);
4384 struct btrfs_key key;
4385 struct btrfs_root *fs_root;
4386 struct btrfs_root *reloc_root;
4387 struct btrfs_path *path;
4388 struct extent_buffer *leaf;
4389 struct reloc_control *rc = NULL;
4390 struct btrfs_trans_handle *trans;
4394 path = btrfs_alloc_path();
4397 path->reada = READA_BACK;
4399 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4400 key.type = BTRFS_ROOT_ITEM_KEY;
4401 key.offset = (u64)-1;
4404 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4411 if (path->slots[0] == 0)
4415 leaf = path->nodes[0];
4416 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4417 btrfs_release_path(path);
4419 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4420 key.type != BTRFS_ROOT_ITEM_KEY)
4423 reloc_root = btrfs_read_fs_root(root, &key);
4424 if (IS_ERR(reloc_root)) {
4425 err = PTR_ERR(reloc_root);
4429 list_add(&reloc_root->root_list, &reloc_roots);
4431 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4432 fs_root = read_fs_root(fs_info,
4433 reloc_root->root_key.offset);
4434 if (IS_ERR(fs_root)) {
4435 ret = PTR_ERR(fs_root);
4436 if (ret != -ENOENT) {
4440 ret = mark_garbage_root(reloc_root);
4448 if (key.offset == 0)
4453 btrfs_release_path(path);
4455 if (list_empty(&reloc_roots))
4458 rc = alloc_reloc_control();
4464 rc->extent_root = fs_info->extent_root;
4466 set_reloc_control(rc);
4468 trans = btrfs_join_transaction(rc->extent_root);
4469 if (IS_ERR(trans)) {
4470 unset_reloc_control(rc);
4471 err = PTR_ERR(trans);
4475 rc->merge_reloc_tree = 1;
4477 while (!list_empty(&reloc_roots)) {
4478 reloc_root = list_entry(reloc_roots.next,
4479 struct btrfs_root, root_list);
4480 list_del(&reloc_root->root_list);
4482 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4483 list_add_tail(&reloc_root->root_list,
4488 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
4489 if (IS_ERR(fs_root)) {
4490 err = PTR_ERR(fs_root);
4491 list_add_tail(&reloc_root->root_list, &reloc_roots);
4495 err = __add_reloc_root(reloc_root);
4496 BUG_ON(err < 0); /* -ENOMEM or logic error */
4497 fs_root->reloc_root = reloc_root;
4500 err = btrfs_commit_transaction(trans);
4504 merge_reloc_roots(rc);
4506 unset_reloc_control(rc);
4508 trans = btrfs_join_transaction(rc->extent_root);
4509 if (IS_ERR(trans)) {
4510 err = PTR_ERR(trans);
4513 err = btrfs_commit_transaction(trans);
4517 if (!list_empty(&reloc_roots))
4518 free_reloc_roots(&reloc_roots);
4520 btrfs_free_path(path);
4523 /* cleanup orphan inode in data relocation tree */
4524 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4525 if (IS_ERR(fs_root))
4526 err = PTR_ERR(fs_root);
4528 err = btrfs_orphan_cleanup(fs_root);
4534 * helper to add ordered checksum for data relocation.
4536 * cloning checksum properly handles the nodatasum extents.
4537 * it also saves CPU time to re-calculate the checksum.
4539 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4541 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4542 struct btrfs_ordered_sum *sums;
4543 struct btrfs_ordered_extent *ordered;
4549 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4550 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4552 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4553 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4554 disk_bytenr + len - 1, &list, 0);
4558 while (!list_empty(&list)) {
4559 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4560 list_del_init(&sums->list);
4563 * We need to offset the new_bytenr based on where the csum is.
4564 * We need to do this because we will read in entire prealloc
4565 * extents but we may have written to say the middle of the
4566 * prealloc extent, so we need to make sure the csum goes with
4567 * the right disk offset.
4569 * We can do this because the data reloc inode refers strictly
4570 * to the on disk bytes, so we don't have to worry about
4571 * disk_len vs real len like with real inodes since it's all
4574 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4575 sums->bytenr = new_bytenr;
4577 btrfs_add_ordered_sum(inode, ordered, sums);
4580 btrfs_put_ordered_extent(ordered);
4584 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4585 struct btrfs_root *root, struct extent_buffer *buf,
4586 struct extent_buffer *cow)
4588 struct btrfs_fs_info *fs_info = root->fs_info;
4589 struct reloc_control *rc;
4590 struct backref_node *node;
4595 rc = fs_info->reloc_ctl;
4599 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4600 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4602 level = btrfs_header_level(buf);
4603 if (btrfs_header_generation(buf) <=
4604 btrfs_root_last_snapshot(&root->root_item))
4607 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4608 rc->create_reloc_tree) {
4609 WARN_ON(!first_cow && level == 0);
4611 node = rc->backref_cache.path[level];
4612 BUG_ON(node->bytenr != buf->start &&
4613 node->new_bytenr != buf->start);
4615 drop_node_buffer(node);
4616 extent_buffer_get(cow);
4618 node->new_bytenr = cow->start;
4620 if (!node->pending) {
4621 list_move_tail(&node->list,
4622 &rc->backref_cache.pending[level]);
4627 __mark_block_processed(rc, node);
4629 if (first_cow && level > 0)
4630 rc->nodes_relocated += buf->len;
4633 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4634 ret = replace_file_extents(trans, rc, root, cow);
4639 * called before creating snapshot. it calculates metadata reservation
4640 * required for relocating tree blocks in the snapshot
4642 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4643 u64 *bytes_to_reserve)
4645 struct btrfs_root *root;
4646 struct reloc_control *rc;
4648 root = pending->root;
4649 if (!root->reloc_root)
4652 rc = root->fs_info->reloc_ctl;
4653 if (!rc->merge_reloc_tree)
4656 root = root->reloc_root;
4657 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4659 * relocation is in the stage of merging trees. the space
4660 * used by merging a reloc tree is twice the size of
4661 * relocated tree nodes in the worst case. half for cowing
4662 * the reloc tree, half for cowing the fs tree. the space
4663 * used by cowing the reloc tree will be freed after the
4664 * tree is dropped. if we create snapshot, cowing the fs
4665 * tree may use more space than it frees. so we need
4666 * reserve extra space.
4668 *bytes_to_reserve += rc->nodes_relocated;
4672 * called after snapshot is created. migrate block reservation
4673 * and create reloc root for the newly created snapshot
4675 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4676 struct btrfs_pending_snapshot *pending)
4678 struct btrfs_root *root = pending->root;
4679 struct btrfs_root *reloc_root;
4680 struct btrfs_root *new_root;
4681 struct reloc_control *rc;
4684 if (!root->reloc_root)
4687 rc = root->fs_info->reloc_ctl;
4688 rc->merging_rsv_size += rc->nodes_relocated;
4690 if (rc->merge_reloc_tree) {
4691 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4693 rc->nodes_relocated, 1);
4698 new_root = pending->snap;
4699 reloc_root = create_reloc_root(trans, root->reloc_root,
4700 new_root->root_key.objectid);
4701 if (IS_ERR(reloc_root))
4702 return PTR_ERR(reloc_root);
4704 ret = __add_reloc_root(reloc_root);
4706 new_root->reloc_root = reloc_root;
4708 if (rc->create_reloc_tree)
4709 ret = clone_backref_node(trans, rc, root, reloc_root);