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_root_last_snapshot(&reloc_root->root_item) ==
529 root->fs_info->running_transaction->transid - 1)
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; /* For searching extent root */
652 struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */
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;
665 LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */
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 * Parent node found and matches current inline ref, no need to
783 * rebuild this node for this inline ref.
786 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
787 exist->owner == key.offset) ||
788 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
789 exist->bytenr == key.offset))) {
794 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
795 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
796 if (key.objectid == key.offset) {
798 * Only root blocks of reloc trees use backref
799 * pointing to itself.
801 root = find_reloc_root(rc, cur->bytenr);
807 edge = alloc_backref_edge(cache);
812 rb_node = tree_search(&cache->rb_root, key.offset);
814 upper = alloc_backref_node(cache);
816 free_backref_edge(cache, edge);
820 upper->bytenr = key.offset;
821 upper->level = cur->level + 1;
823 * backrefs for the upper level block isn't
824 * cached, add the block to pending list
826 list_add_tail(&edge->list[UPPER], &list);
828 upper = rb_entry(rb_node, struct backref_node,
830 ASSERT(upper->checked);
831 INIT_LIST_HEAD(&edge->list[UPPER]);
833 list_add_tail(&edge->list[LOWER], &cur->upper);
834 edge->node[LOWER] = cur;
835 edge->node[UPPER] = upper;
838 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
840 btrfs_print_v0_err(rc->extent_root->fs_info);
841 btrfs_handle_fs_error(rc->extent_root->fs_info, err,
844 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
849 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
850 * means the root objectid. We need to search the tree to get
853 root = read_fs_root(rc->extent_root->fs_info, key.offset);
859 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
862 if (btrfs_root_level(&root->root_item) == cur->level) {
864 ASSERT(btrfs_root_bytenr(&root->root_item) ==
866 if (should_ignore_root(root))
867 list_add(&cur->list, &useless);
873 level = cur->level + 1;
875 /* Search the tree to find parent blocks referring the block. */
876 path2->search_commit_root = 1;
877 path2->skip_locking = 1;
878 path2->lowest_level = level;
879 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
880 path2->lowest_level = 0;
885 if (ret > 0 && path2->slots[level] > 0)
886 path2->slots[level]--;
888 eb = path2->nodes[level];
889 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
891 btrfs_err(root->fs_info,
892 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
893 cur->bytenr, level - 1,
894 root->root_key.objectid,
895 node_key->objectid, node_key->type,
903 /* Add all nodes and edges in the path */
904 for (; level < BTRFS_MAX_LEVEL; level++) {
905 if (!path2->nodes[level]) {
906 ASSERT(btrfs_root_bytenr(&root->root_item) ==
908 if (should_ignore_root(root))
909 list_add(&lower->list, &useless);
915 edge = alloc_backref_edge(cache);
921 eb = path2->nodes[level];
922 rb_node = tree_search(&cache->rb_root, eb->start);
924 upper = alloc_backref_node(cache);
926 free_backref_edge(cache, edge);
930 upper->bytenr = eb->start;
931 upper->owner = btrfs_header_owner(eb);
932 upper->level = lower->level + 1;
933 if (!test_bit(BTRFS_ROOT_REF_COWS,
938 * if we know the block isn't shared
939 * we can void checking its backrefs.
941 if (btrfs_block_can_be_shared(root, eb))
947 * add the block to pending list if we
948 * need check its backrefs, we only do this once
949 * while walking up a tree as we will catch
950 * anything else later on.
952 if (!upper->checked && need_check) {
954 list_add_tail(&edge->list[UPPER],
959 INIT_LIST_HEAD(&edge->list[UPPER]);
962 upper = rb_entry(rb_node, struct backref_node,
964 ASSERT(upper->checked);
965 INIT_LIST_HEAD(&edge->list[UPPER]);
967 upper->owner = btrfs_header_owner(eb);
969 list_add_tail(&edge->list[LOWER], &lower->upper);
970 edge->node[LOWER] = lower;
971 edge->node[UPPER] = upper;
978 btrfs_release_path(path2);
981 ptr += btrfs_extent_inline_ref_size(key.type);
991 btrfs_release_path(path1);
996 /* the pending list isn't empty, take the first block to process */
997 if (!list_empty(&list)) {
998 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
999 list_del_init(&edge->list[UPPER]);
1000 cur = edge->node[UPPER];
1005 * everything goes well, connect backref nodes and insert backref nodes
1008 ASSERT(node->checked);
1009 cowonly = node->cowonly;
1011 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1014 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1015 list_add_tail(&node->lower, &cache->leaves);
1018 list_for_each_entry(edge, &node->upper, list[LOWER])
1019 list_add_tail(&edge->list[UPPER], &list);
1021 while (!list_empty(&list)) {
1022 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1023 list_del_init(&edge->list[UPPER]);
1024 upper = edge->node[UPPER];
1025 if (upper->detached) {
1026 list_del(&edge->list[LOWER]);
1027 lower = edge->node[LOWER];
1028 free_backref_edge(cache, edge);
1029 if (list_empty(&lower->upper))
1030 list_add(&lower->list, &useless);
1034 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1035 if (upper->lowest) {
1036 list_del_init(&upper->lower);
1040 list_add_tail(&edge->list[UPPER], &upper->lower);
1044 if (!upper->checked) {
1046 * Still want to blow up for developers since this is a
1053 if (cowonly != upper->cowonly) {
1060 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1063 backref_tree_panic(rb_node, -EEXIST,
1067 list_add_tail(&edge->list[UPPER], &upper->lower);
1069 list_for_each_entry(edge, &upper->upper, list[LOWER])
1070 list_add_tail(&edge->list[UPPER], &list);
1073 * process useless backref nodes. backref nodes for tree leaves
1074 * are deleted from the cache. backref nodes for upper level
1075 * tree blocks are left in the cache to avoid unnecessary backref
1078 while (!list_empty(&useless)) {
1079 upper = list_entry(useless.next, struct backref_node, list);
1080 list_del_init(&upper->list);
1081 ASSERT(list_empty(&upper->upper));
1084 if (upper->lowest) {
1085 list_del_init(&upper->lower);
1088 while (!list_empty(&upper->lower)) {
1089 edge = list_entry(upper->lower.next,
1090 struct backref_edge, list[UPPER]);
1091 list_del(&edge->list[UPPER]);
1092 list_del(&edge->list[LOWER]);
1093 lower = edge->node[LOWER];
1094 free_backref_edge(cache, edge);
1096 if (list_empty(&lower->upper))
1097 list_add(&lower->list, &useless);
1099 __mark_block_processed(rc, upper);
1100 if (upper->level > 0) {
1101 list_add(&upper->list, &cache->detached);
1102 upper->detached = 1;
1104 rb_erase(&upper->rb_node, &cache->rb_root);
1105 free_backref_node(cache, upper);
1109 btrfs_free_path(path1);
1110 btrfs_free_path(path2);
1112 while (!list_empty(&useless)) {
1113 lower = list_entry(useless.next,
1114 struct backref_node, list);
1115 list_del_init(&lower->list);
1117 while (!list_empty(&list)) {
1118 edge = list_first_entry(&list, struct backref_edge,
1120 list_del(&edge->list[UPPER]);
1121 list_del(&edge->list[LOWER]);
1122 lower = edge->node[LOWER];
1123 upper = edge->node[UPPER];
1124 free_backref_edge(cache, edge);
1127 * Lower is no longer linked to any upper backref nodes
1128 * and isn't in the cache, we can free it ourselves.
1130 if (list_empty(&lower->upper) &&
1131 RB_EMPTY_NODE(&lower->rb_node))
1132 list_add(&lower->list, &useless);
1134 if (!RB_EMPTY_NODE(&upper->rb_node))
1137 /* Add this guy's upper edges to the list to process */
1138 list_for_each_entry(edge, &upper->upper, list[LOWER])
1139 list_add_tail(&edge->list[UPPER], &list);
1140 if (list_empty(&upper->upper))
1141 list_add(&upper->list, &useless);
1144 while (!list_empty(&useless)) {
1145 lower = list_entry(useless.next,
1146 struct backref_node, list);
1147 list_del_init(&lower->list);
1150 free_backref_node(cache, lower);
1153 free_backref_node(cache, node);
1154 return ERR_PTR(err);
1156 ASSERT(!node || !node->detached);
1161 * helper to add backref node for the newly created snapshot.
1162 * the backref node is created by cloning backref node that
1163 * corresponds to root of source tree
1165 static int clone_backref_node(struct btrfs_trans_handle *trans,
1166 struct reloc_control *rc,
1167 struct btrfs_root *src,
1168 struct btrfs_root *dest)
1170 struct btrfs_root *reloc_root = src->reloc_root;
1171 struct backref_cache *cache = &rc->backref_cache;
1172 struct backref_node *node = NULL;
1173 struct backref_node *new_node;
1174 struct backref_edge *edge;
1175 struct backref_edge *new_edge;
1176 struct rb_node *rb_node;
1178 if (cache->last_trans > 0)
1179 update_backref_cache(trans, cache);
1181 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1183 node = rb_entry(rb_node, struct backref_node, rb_node);
1187 BUG_ON(node->new_bytenr != reloc_root->node->start);
1191 rb_node = tree_search(&cache->rb_root,
1192 reloc_root->commit_root->start);
1194 node = rb_entry(rb_node, struct backref_node,
1196 BUG_ON(node->detached);
1203 new_node = alloc_backref_node(cache);
1207 new_node->bytenr = dest->node->start;
1208 new_node->level = node->level;
1209 new_node->lowest = node->lowest;
1210 new_node->checked = 1;
1211 new_node->root = dest;
1213 if (!node->lowest) {
1214 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1215 new_edge = alloc_backref_edge(cache);
1219 new_edge->node[UPPER] = new_node;
1220 new_edge->node[LOWER] = edge->node[LOWER];
1221 list_add_tail(&new_edge->list[UPPER],
1225 list_add_tail(&new_node->lower, &cache->leaves);
1228 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1229 &new_node->rb_node);
1231 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1233 if (!new_node->lowest) {
1234 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1235 list_add_tail(&new_edge->list[LOWER],
1236 &new_edge->node[LOWER]->upper);
1241 while (!list_empty(&new_node->lower)) {
1242 new_edge = list_entry(new_node->lower.next,
1243 struct backref_edge, list[UPPER]);
1244 list_del(&new_edge->list[UPPER]);
1245 free_backref_edge(cache, new_edge);
1247 free_backref_node(cache, new_node);
1252 * helper to add 'address of tree root -> reloc tree' mapping
1254 static int __must_check __add_reloc_root(struct btrfs_root *root)
1256 struct btrfs_fs_info *fs_info = root->fs_info;
1257 struct rb_node *rb_node;
1258 struct mapping_node *node;
1259 struct reloc_control *rc = fs_info->reloc_ctl;
1261 node = kmalloc(sizeof(*node), GFP_NOFS);
1265 node->bytenr = root->node->start;
1268 spin_lock(&rc->reloc_root_tree.lock);
1269 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1270 node->bytenr, &node->rb_node);
1271 spin_unlock(&rc->reloc_root_tree.lock);
1273 btrfs_panic(fs_info, -EEXIST,
1274 "Duplicate root found for start=%llu while inserting into relocation tree",
1278 list_add_tail(&root->root_list, &rc->reloc_roots);
1283 * helper to delete the 'address of tree root -> reloc tree'
1286 static void __del_reloc_root(struct btrfs_root *root)
1288 struct btrfs_fs_info *fs_info = root->fs_info;
1289 struct rb_node *rb_node;
1290 struct mapping_node *node = NULL;
1291 struct reloc_control *rc = fs_info->reloc_ctl;
1293 if (rc && root->node) {
1294 spin_lock(&rc->reloc_root_tree.lock);
1295 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1298 node = rb_entry(rb_node, struct mapping_node, rb_node);
1299 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1301 spin_unlock(&rc->reloc_root_tree.lock);
1304 BUG_ON((struct btrfs_root *)node->data != root);
1307 spin_lock(&fs_info->trans_lock);
1308 list_del_init(&root->root_list);
1309 spin_unlock(&fs_info->trans_lock);
1314 * helper to update the 'address of tree root -> reloc tree'
1317 static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1319 struct btrfs_fs_info *fs_info = root->fs_info;
1320 struct rb_node *rb_node;
1321 struct mapping_node *node = NULL;
1322 struct reloc_control *rc = fs_info->reloc_ctl;
1324 spin_lock(&rc->reloc_root_tree.lock);
1325 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1328 node = rb_entry(rb_node, struct mapping_node, rb_node);
1329 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1331 spin_unlock(&rc->reloc_root_tree.lock);
1335 BUG_ON((struct btrfs_root *)node->data != root);
1337 spin_lock(&rc->reloc_root_tree.lock);
1338 node->bytenr = new_bytenr;
1339 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1340 node->bytenr, &node->rb_node);
1341 spin_unlock(&rc->reloc_root_tree.lock);
1343 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1347 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1348 struct btrfs_root *root, u64 objectid)
1350 struct btrfs_fs_info *fs_info = root->fs_info;
1351 struct btrfs_root *reloc_root;
1352 struct extent_buffer *eb;
1353 struct btrfs_root_item *root_item;
1354 struct btrfs_key root_key;
1357 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1360 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1361 root_key.type = BTRFS_ROOT_ITEM_KEY;
1362 root_key.offset = objectid;
1364 if (root->root_key.objectid == objectid) {
1365 u64 commit_root_gen;
1367 /* called by btrfs_init_reloc_root */
1368 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1369 BTRFS_TREE_RELOC_OBJECTID);
1372 * Set the last_snapshot field to the generation of the commit
1373 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1374 * correctly (returns true) when the relocation root is created
1375 * either inside the critical section of a transaction commit
1376 * (through transaction.c:qgroup_account_snapshot()) and when
1377 * it's created before the transaction commit is started.
1379 commit_root_gen = btrfs_header_generation(root->commit_root);
1380 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1383 * called by btrfs_reloc_post_snapshot_hook.
1384 * the source tree is a reloc tree, all tree blocks
1385 * modified after it was created have RELOC flag
1386 * set in their headers. so it's OK to not update
1387 * the 'last_snapshot'.
1389 ret = btrfs_copy_root(trans, root, root->node, &eb,
1390 BTRFS_TREE_RELOC_OBJECTID);
1394 memcpy(root_item, &root->root_item, sizeof(*root_item));
1395 btrfs_set_root_bytenr(root_item, eb->start);
1396 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1397 btrfs_set_root_generation(root_item, trans->transid);
1399 if (root->root_key.objectid == objectid) {
1400 btrfs_set_root_refs(root_item, 0);
1401 memset(&root_item->drop_progress, 0,
1402 sizeof(struct btrfs_disk_key));
1403 root_item->drop_level = 0;
1406 btrfs_tree_unlock(eb);
1407 free_extent_buffer(eb);
1409 ret = btrfs_insert_root(trans, fs_info->tree_root,
1410 &root_key, root_item);
1414 reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1415 BUG_ON(IS_ERR(reloc_root));
1416 reloc_root->last_trans = trans->transid;
1421 * create reloc tree for a given fs tree. reloc tree is just a
1422 * snapshot of the fs tree with special root objectid.
1424 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1425 struct btrfs_root *root)
1427 struct btrfs_fs_info *fs_info = root->fs_info;
1428 struct btrfs_root *reloc_root;
1429 struct reloc_control *rc = fs_info->reloc_ctl;
1430 struct btrfs_block_rsv *rsv;
1434 if (root->reloc_root) {
1435 reloc_root = root->reloc_root;
1436 reloc_root->last_trans = trans->transid;
1440 if (!rc || !rc->create_reloc_tree ||
1441 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1444 if (!trans->reloc_reserved) {
1445 rsv = trans->block_rsv;
1446 trans->block_rsv = rc->block_rsv;
1449 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1451 trans->block_rsv = rsv;
1453 ret = __add_reloc_root(reloc_root);
1455 root->reloc_root = reloc_root;
1460 * update root item of reloc tree
1462 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1463 struct btrfs_root *root)
1465 struct btrfs_fs_info *fs_info = root->fs_info;
1466 struct btrfs_root *reloc_root;
1467 struct btrfs_root_item *root_item;
1470 if (!root->reloc_root)
1473 reloc_root = root->reloc_root;
1474 root_item = &reloc_root->root_item;
1476 if (fs_info->reloc_ctl->merge_reloc_tree &&
1477 btrfs_root_refs(root_item) == 0) {
1478 root->reloc_root = NULL;
1479 __del_reloc_root(reloc_root);
1482 if (reloc_root->commit_root != reloc_root->node) {
1483 btrfs_set_root_node(root_item, reloc_root->node);
1484 free_extent_buffer(reloc_root->commit_root);
1485 reloc_root->commit_root = btrfs_root_node(reloc_root);
1488 ret = btrfs_update_root(trans, fs_info->tree_root,
1489 &reloc_root->root_key, root_item);
1497 * helper to find first cached inode with inode number >= objectid
1500 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1502 struct rb_node *node;
1503 struct rb_node *prev;
1504 struct btrfs_inode *entry;
1505 struct inode *inode;
1507 spin_lock(&root->inode_lock);
1509 node = root->inode_tree.rb_node;
1513 entry = rb_entry(node, struct btrfs_inode, rb_node);
1515 if (objectid < btrfs_ino(entry))
1516 node = node->rb_left;
1517 else if (objectid > btrfs_ino(entry))
1518 node = node->rb_right;
1524 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1525 if (objectid <= btrfs_ino(entry)) {
1529 prev = rb_next(prev);
1533 entry = rb_entry(node, struct btrfs_inode, rb_node);
1534 inode = igrab(&entry->vfs_inode);
1536 spin_unlock(&root->inode_lock);
1540 objectid = btrfs_ino(entry) + 1;
1541 if (cond_resched_lock(&root->inode_lock))
1544 node = rb_next(node);
1546 spin_unlock(&root->inode_lock);
1550 static int in_block_group(u64 bytenr,
1551 struct btrfs_block_group_cache *block_group)
1553 if (bytenr >= block_group->key.objectid &&
1554 bytenr < block_group->key.objectid + block_group->key.offset)
1560 * get new location of data
1562 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1563 u64 bytenr, u64 num_bytes)
1565 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1566 struct btrfs_path *path;
1567 struct btrfs_file_extent_item *fi;
1568 struct extent_buffer *leaf;
1571 path = btrfs_alloc_path();
1575 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1576 ret = btrfs_lookup_file_extent(NULL, root, path,
1577 btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1585 leaf = path->nodes[0];
1586 fi = btrfs_item_ptr(leaf, path->slots[0],
1587 struct btrfs_file_extent_item);
1589 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1590 btrfs_file_extent_compression(leaf, fi) ||
1591 btrfs_file_extent_encryption(leaf, fi) ||
1592 btrfs_file_extent_other_encoding(leaf, fi));
1594 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1599 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1602 btrfs_free_path(path);
1607 * update file extent items in the tree leaf to point to
1608 * the new locations.
1610 static noinline_for_stack
1611 int replace_file_extents(struct btrfs_trans_handle *trans,
1612 struct reloc_control *rc,
1613 struct btrfs_root *root,
1614 struct extent_buffer *leaf)
1616 struct btrfs_fs_info *fs_info = root->fs_info;
1617 struct btrfs_key key;
1618 struct btrfs_file_extent_item *fi;
1619 struct inode *inode = NULL;
1631 if (rc->stage != UPDATE_DATA_PTRS)
1634 /* reloc trees always use full backref */
1635 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1636 parent = leaf->start;
1640 nritems = btrfs_header_nritems(leaf);
1641 for (i = 0; i < nritems; i++) {
1643 btrfs_item_key_to_cpu(leaf, &key, i);
1644 if (key.type != BTRFS_EXTENT_DATA_KEY)
1646 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1647 if (btrfs_file_extent_type(leaf, fi) ==
1648 BTRFS_FILE_EXTENT_INLINE)
1650 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1651 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1654 if (!in_block_group(bytenr, rc->block_group))
1658 * if we are modifying block in fs tree, wait for readpage
1659 * to complete and drop the extent cache
1661 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1663 inode = find_next_inode(root, key.objectid);
1665 } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1666 btrfs_add_delayed_iput(inode);
1667 inode = find_next_inode(root, key.objectid);
1669 if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1671 btrfs_file_extent_num_bytes(leaf, fi);
1672 WARN_ON(!IS_ALIGNED(key.offset,
1673 fs_info->sectorsize));
1674 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1676 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1681 btrfs_drop_extent_cache(BTRFS_I(inode),
1682 key.offset, end, 1);
1683 unlock_extent(&BTRFS_I(inode)->io_tree,
1688 ret = get_new_location(rc->data_inode, &new_bytenr,
1692 * Don't have to abort since we've not changed anything
1693 * in the file extent yet.
1698 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1701 key.offset -= btrfs_file_extent_offset(leaf, fi);
1702 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1704 btrfs_header_owner(leaf),
1705 key.objectid, key.offset);
1707 btrfs_abort_transaction(trans, ret);
1711 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1712 parent, btrfs_header_owner(leaf),
1713 key.objectid, key.offset);
1715 btrfs_abort_transaction(trans, ret);
1720 btrfs_mark_buffer_dirty(leaf);
1722 btrfs_add_delayed_iput(inode);
1726 static noinline_for_stack
1727 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1728 struct btrfs_path *path, int level)
1730 struct btrfs_disk_key key1;
1731 struct btrfs_disk_key key2;
1732 btrfs_node_key(eb, &key1, slot);
1733 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1734 return memcmp(&key1, &key2, sizeof(key1));
1738 * try to replace tree blocks in fs tree with the new blocks
1739 * in reloc tree. tree blocks haven't been modified since the
1740 * reloc tree was create can be replaced.
1742 * if a block was replaced, level of the block + 1 is returned.
1743 * if no block got replaced, 0 is returned. if there are other
1744 * errors, a negative error number is returned.
1746 static noinline_for_stack
1747 int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1748 struct btrfs_root *dest, struct btrfs_root *src,
1749 struct btrfs_path *path, struct btrfs_key *next_key,
1750 int lowest_level, int max_level)
1752 struct btrfs_fs_info *fs_info = dest->fs_info;
1753 struct extent_buffer *eb;
1754 struct extent_buffer *parent;
1755 struct btrfs_key key;
1767 BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1768 BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1770 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1772 slot = path->slots[lowest_level];
1773 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1775 eb = btrfs_lock_root_node(dest);
1776 btrfs_set_lock_blocking(eb);
1777 level = btrfs_header_level(eb);
1779 if (level < lowest_level) {
1780 btrfs_tree_unlock(eb);
1781 free_extent_buffer(eb);
1786 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1789 btrfs_set_lock_blocking(eb);
1792 next_key->objectid = (u64)-1;
1793 next_key->type = (u8)-1;
1794 next_key->offset = (u64)-1;
1799 struct btrfs_key first_key;
1801 level = btrfs_header_level(parent);
1802 BUG_ON(level < lowest_level);
1804 ret = btrfs_bin_search(parent, &key, level, &slot);
1805 if (ret && slot > 0)
1808 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1809 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1811 old_bytenr = btrfs_node_blockptr(parent, slot);
1812 blocksize = fs_info->nodesize;
1813 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1814 btrfs_node_key_to_cpu(parent, &first_key, slot);
1816 if (level <= max_level) {
1817 eb = path->nodes[level];
1818 new_bytenr = btrfs_node_blockptr(eb,
1819 path->slots[level]);
1820 new_ptr_gen = btrfs_node_ptr_generation(eb,
1821 path->slots[level]);
1827 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1832 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1833 memcmp_node_keys(parent, slot, path, level)) {
1834 if (level <= lowest_level) {
1839 eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1840 level - 1, &first_key);
1844 } else if (!extent_buffer_uptodate(eb)) {
1846 free_extent_buffer(eb);
1849 btrfs_tree_lock(eb);
1851 ret = btrfs_cow_block(trans, dest, eb, parent,
1855 btrfs_set_lock_blocking(eb);
1857 btrfs_tree_unlock(parent);
1858 free_extent_buffer(parent);
1865 btrfs_tree_unlock(parent);
1866 free_extent_buffer(parent);
1871 btrfs_node_key_to_cpu(path->nodes[level], &key,
1872 path->slots[level]);
1873 btrfs_release_path(path);
1875 path->lowest_level = level;
1876 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1877 path->lowest_level = 0;
1881 * Info qgroup to trace both subtrees.
1883 * We must trace both trees.
1884 * 1) Tree reloc subtree
1885 * If not traced, we will leak data numbers
1887 * If not traced, we will double count old data
1888 * and tree block numbers, if current trans doesn't free
1889 * data reloc tree inode.
1891 ret = btrfs_qgroup_trace_subtree_swap(trans, rc->block_group,
1892 parent, slot, path->nodes[level],
1893 path->slots[level], last_snapshot);
1898 * swap blocks in fs tree and reloc tree.
1900 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1901 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1902 btrfs_mark_buffer_dirty(parent);
1904 btrfs_set_node_blockptr(path->nodes[level],
1905 path->slots[level], old_bytenr);
1906 btrfs_set_node_ptr_generation(path->nodes[level],
1907 path->slots[level], old_ptr_gen);
1908 btrfs_mark_buffer_dirty(path->nodes[level]);
1910 ret = btrfs_inc_extent_ref(trans, src, old_bytenr,
1911 blocksize, path->nodes[level]->start,
1912 src->root_key.objectid, level - 1, 0);
1914 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr,
1915 blocksize, 0, dest->root_key.objectid,
1919 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1920 path->nodes[level]->start,
1921 src->root_key.objectid, level - 1, 0);
1924 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1925 0, dest->root_key.objectid, level - 1,
1929 btrfs_unlock_up_safe(path, 0);
1934 btrfs_tree_unlock(parent);
1935 free_extent_buffer(parent);
1940 * helper to find next relocated block in reloc tree
1942 static noinline_for_stack
1943 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1946 struct extent_buffer *eb;
1951 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1953 for (i = 0; i < *level; i++) {
1954 free_extent_buffer(path->nodes[i]);
1955 path->nodes[i] = NULL;
1958 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1959 eb = path->nodes[i];
1960 nritems = btrfs_header_nritems(eb);
1961 while (path->slots[i] + 1 < nritems) {
1963 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1970 free_extent_buffer(path->nodes[i]);
1971 path->nodes[i] = NULL;
1977 * walk down reloc tree to find relocated block of lowest level
1979 static noinline_for_stack
1980 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1983 struct btrfs_fs_info *fs_info = root->fs_info;
1984 struct extent_buffer *eb = NULL;
1991 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1993 for (i = *level; i > 0; i--) {
1994 struct btrfs_key first_key;
1996 eb = path->nodes[i];
1997 nritems = btrfs_header_nritems(eb);
1998 while (path->slots[i] < nritems) {
1999 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2000 if (ptr_gen > last_snapshot)
2004 if (path->slots[i] >= nritems) {
2015 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2016 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2017 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2021 } else if (!extent_buffer_uptodate(eb)) {
2022 free_extent_buffer(eb);
2025 BUG_ON(btrfs_header_level(eb) != i - 1);
2026 path->nodes[i - 1] = eb;
2027 path->slots[i - 1] = 0;
2033 * invalidate extent cache for file extents whose key in range of
2034 * [min_key, max_key)
2036 static int invalidate_extent_cache(struct btrfs_root *root,
2037 struct btrfs_key *min_key,
2038 struct btrfs_key *max_key)
2040 struct btrfs_fs_info *fs_info = root->fs_info;
2041 struct inode *inode = NULL;
2046 objectid = min_key->objectid;
2051 if (objectid > max_key->objectid)
2054 inode = find_next_inode(root, objectid);
2057 ino = btrfs_ino(BTRFS_I(inode));
2059 if (ino > max_key->objectid) {
2065 if (!S_ISREG(inode->i_mode))
2068 if (unlikely(min_key->objectid == ino)) {
2069 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2071 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2074 start = min_key->offset;
2075 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2081 if (unlikely(max_key->objectid == ino)) {
2082 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2084 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2087 if (max_key->offset == 0)
2089 end = max_key->offset;
2090 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2097 /* the lock_extent waits for readpage to complete */
2098 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2099 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2100 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2105 static int find_next_key(struct btrfs_path *path, int level,
2106 struct btrfs_key *key)
2109 while (level < BTRFS_MAX_LEVEL) {
2110 if (!path->nodes[level])
2112 if (path->slots[level] + 1 <
2113 btrfs_header_nritems(path->nodes[level])) {
2114 btrfs_node_key_to_cpu(path->nodes[level], key,
2115 path->slots[level] + 1);
2124 * merge the relocated tree blocks in reloc tree with corresponding
2127 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2128 struct btrfs_root *root)
2130 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2131 LIST_HEAD(inode_list);
2132 struct btrfs_key key;
2133 struct btrfs_key next_key;
2134 struct btrfs_trans_handle *trans = NULL;
2135 struct btrfs_root *reloc_root;
2136 struct btrfs_root_item *root_item;
2137 struct btrfs_path *path;
2138 struct extent_buffer *leaf;
2146 path = btrfs_alloc_path();
2149 path->reada = READA_FORWARD;
2151 reloc_root = root->reloc_root;
2152 root_item = &reloc_root->root_item;
2154 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2155 level = btrfs_root_level(root_item);
2156 extent_buffer_get(reloc_root->node);
2157 path->nodes[level] = reloc_root->node;
2158 path->slots[level] = 0;
2160 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2162 level = root_item->drop_level;
2164 path->lowest_level = level;
2165 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2166 path->lowest_level = 0;
2168 btrfs_free_path(path);
2172 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2173 path->slots[level]);
2174 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2176 btrfs_unlock_up_safe(path, 0);
2179 min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2180 memset(&next_key, 0, sizeof(next_key));
2183 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2184 BTRFS_RESERVE_FLUSH_ALL);
2189 trans = btrfs_start_transaction(root, 0);
2190 if (IS_ERR(trans)) {
2191 err = PTR_ERR(trans);
2195 trans->block_rsv = rc->block_rsv;
2200 ret = walk_down_reloc_tree(reloc_root, path, &level);
2208 if (!find_next_key(path, level, &key) &&
2209 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2212 ret = replace_path(trans, rc, root, reloc_root, path,
2213 &next_key, level, max_level);
2222 btrfs_node_key_to_cpu(path->nodes[level], &key,
2223 path->slots[level]);
2227 ret = walk_up_reloc_tree(reloc_root, path, &level);
2233 * save the merging progress in the drop_progress.
2234 * this is OK since root refs == 1 in this case.
2236 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2237 path->slots[level]);
2238 root_item->drop_level = level;
2240 btrfs_end_transaction_throttle(trans);
2243 btrfs_btree_balance_dirty(fs_info);
2245 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2246 invalidate_extent_cache(root, &key, &next_key);
2250 * handle the case only one block in the fs tree need to be
2251 * relocated and the block is tree root.
2253 leaf = btrfs_lock_root_node(root);
2254 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2255 btrfs_tree_unlock(leaf);
2256 free_extent_buffer(leaf);
2260 btrfs_free_path(path);
2263 memset(&root_item->drop_progress, 0,
2264 sizeof(root_item->drop_progress));
2265 root_item->drop_level = 0;
2266 btrfs_set_root_refs(root_item, 0);
2267 btrfs_update_reloc_root(trans, root);
2271 btrfs_end_transaction_throttle(trans);
2273 btrfs_btree_balance_dirty(fs_info);
2275 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2276 invalidate_extent_cache(root, &key, &next_key);
2281 static noinline_for_stack
2282 int prepare_to_merge(struct reloc_control *rc, int err)
2284 struct btrfs_root *root = rc->extent_root;
2285 struct btrfs_fs_info *fs_info = root->fs_info;
2286 struct btrfs_root *reloc_root;
2287 struct btrfs_trans_handle *trans;
2288 LIST_HEAD(reloc_roots);
2292 mutex_lock(&fs_info->reloc_mutex);
2293 rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2294 rc->merging_rsv_size += rc->nodes_relocated * 2;
2295 mutex_unlock(&fs_info->reloc_mutex);
2299 num_bytes = rc->merging_rsv_size;
2300 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2301 BTRFS_RESERVE_FLUSH_ALL);
2306 trans = btrfs_join_transaction(rc->extent_root);
2307 if (IS_ERR(trans)) {
2309 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2311 return PTR_ERR(trans);
2315 if (num_bytes != rc->merging_rsv_size) {
2316 btrfs_end_transaction(trans);
2317 btrfs_block_rsv_release(fs_info, rc->block_rsv,
2323 rc->merge_reloc_tree = 1;
2325 while (!list_empty(&rc->reloc_roots)) {
2326 reloc_root = list_entry(rc->reloc_roots.next,
2327 struct btrfs_root, root_list);
2328 list_del_init(&reloc_root->root_list);
2330 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2331 BUG_ON(IS_ERR(root));
2332 BUG_ON(root->reloc_root != reloc_root);
2335 * set reference count to 1, so btrfs_recover_relocation
2336 * knows it should resumes merging
2339 btrfs_set_root_refs(&reloc_root->root_item, 1);
2340 btrfs_update_reloc_root(trans, root);
2342 list_add(&reloc_root->root_list, &reloc_roots);
2345 list_splice(&reloc_roots, &rc->reloc_roots);
2348 btrfs_commit_transaction(trans);
2350 btrfs_end_transaction(trans);
2354 static noinline_for_stack
2355 void free_reloc_roots(struct list_head *list)
2357 struct btrfs_root *reloc_root;
2359 while (!list_empty(list)) {
2360 reloc_root = list_entry(list->next, struct btrfs_root,
2362 __del_reloc_root(reloc_root);
2363 free_extent_buffer(reloc_root->node);
2364 free_extent_buffer(reloc_root->commit_root);
2365 reloc_root->node = NULL;
2366 reloc_root->commit_root = NULL;
2370 static noinline_for_stack
2371 void merge_reloc_roots(struct reloc_control *rc)
2373 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2374 struct btrfs_root *root;
2375 struct btrfs_root *reloc_root;
2376 LIST_HEAD(reloc_roots);
2380 root = rc->extent_root;
2383 * this serializes us with btrfs_record_root_in_transaction,
2384 * we have to make sure nobody is in the middle of
2385 * adding their roots to the list while we are
2388 mutex_lock(&fs_info->reloc_mutex);
2389 list_splice_init(&rc->reloc_roots, &reloc_roots);
2390 mutex_unlock(&fs_info->reloc_mutex);
2392 while (!list_empty(&reloc_roots)) {
2394 reloc_root = list_entry(reloc_roots.next,
2395 struct btrfs_root, root_list);
2397 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2398 root = read_fs_root(fs_info,
2399 reloc_root->root_key.offset);
2400 BUG_ON(IS_ERR(root));
2401 BUG_ON(root->reloc_root != reloc_root);
2403 ret = merge_reloc_root(rc, root);
2405 if (list_empty(&reloc_root->root_list))
2406 list_add_tail(&reloc_root->root_list,
2411 list_del_init(&reloc_root->root_list);
2414 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2416 if (list_empty(&reloc_root->root_list))
2417 list_add_tail(&reloc_root->root_list,
2429 btrfs_handle_fs_error(fs_info, ret, NULL);
2430 if (!list_empty(&reloc_roots))
2431 free_reloc_roots(&reloc_roots);
2433 /* new reloc root may be added */
2434 mutex_lock(&fs_info->reloc_mutex);
2435 list_splice_init(&rc->reloc_roots, &reloc_roots);
2436 mutex_unlock(&fs_info->reloc_mutex);
2437 if (!list_empty(&reloc_roots))
2438 free_reloc_roots(&reloc_roots);
2441 BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2444 static void free_block_list(struct rb_root *blocks)
2446 struct tree_block *block;
2447 struct rb_node *rb_node;
2448 while ((rb_node = rb_first(blocks))) {
2449 block = rb_entry(rb_node, struct tree_block, rb_node);
2450 rb_erase(rb_node, blocks);
2455 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2456 struct btrfs_root *reloc_root)
2458 struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2459 struct btrfs_root *root;
2461 if (reloc_root->last_trans == trans->transid)
2464 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2465 BUG_ON(IS_ERR(root));
2466 BUG_ON(root->reloc_root != reloc_root);
2468 return btrfs_record_root_in_trans(trans, root);
2471 static noinline_for_stack
2472 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2473 struct reloc_control *rc,
2474 struct backref_node *node,
2475 struct backref_edge *edges[])
2477 struct backref_node *next;
2478 struct btrfs_root *root;
2484 next = walk_up_backref(next, edges, &index);
2487 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2489 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2490 record_reloc_root_in_trans(trans, root);
2494 btrfs_record_root_in_trans(trans, root);
2495 root = root->reloc_root;
2497 if (next->new_bytenr != root->node->start) {
2498 BUG_ON(next->new_bytenr);
2499 BUG_ON(!list_empty(&next->list));
2500 next->new_bytenr = root->node->start;
2502 list_add_tail(&next->list,
2503 &rc->backref_cache.changed);
2504 __mark_block_processed(rc, next);
2510 next = walk_down_backref(edges, &index);
2511 if (!next || next->level <= node->level)
2518 /* setup backref node path for btrfs_reloc_cow_block */
2520 rc->backref_cache.path[next->level] = next;
2523 next = edges[index]->node[UPPER];
2529 * select a tree root for relocation. return NULL if the block
2530 * is reference counted. we should use do_relocation() in this
2531 * case. return a tree root pointer if the block isn't reference
2532 * counted. return -ENOENT if the block is root of reloc tree.
2534 static noinline_for_stack
2535 struct btrfs_root *select_one_root(struct backref_node *node)
2537 struct backref_node *next;
2538 struct btrfs_root *root;
2539 struct btrfs_root *fs_root = NULL;
2540 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2546 next = walk_up_backref(next, edges, &index);
2550 /* no other choice for non-references counted tree */
2551 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2554 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2560 next = walk_down_backref(edges, &index);
2561 if (!next || next->level <= node->level)
2566 return ERR_PTR(-ENOENT);
2570 static noinline_for_stack
2571 u64 calcu_metadata_size(struct reloc_control *rc,
2572 struct backref_node *node, int reserve)
2574 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2575 struct backref_node *next = node;
2576 struct backref_edge *edge;
2577 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2581 BUG_ON(reserve && node->processed);
2586 if (next->processed && (reserve || next != node))
2589 num_bytes += fs_info->nodesize;
2591 if (list_empty(&next->upper))
2594 edge = list_entry(next->upper.next,
2595 struct backref_edge, list[LOWER]);
2596 edges[index++] = edge;
2597 next = edge->node[UPPER];
2599 next = walk_down_backref(edges, &index);
2604 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2605 struct reloc_control *rc,
2606 struct backref_node *node)
2608 struct btrfs_root *root = rc->extent_root;
2609 struct btrfs_fs_info *fs_info = root->fs_info;
2614 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2616 trans->block_rsv = rc->block_rsv;
2617 rc->reserved_bytes += num_bytes;
2620 * We are under a transaction here so we can only do limited flushing.
2621 * If we get an enospc just kick back -EAGAIN so we know to drop the
2622 * transaction and try to refill when we can flush all the things.
2624 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2625 BTRFS_RESERVE_FLUSH_LIMIT);
2627 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2628 while (tmp <= rc->reserved_bytes)
2631 * only one thread can access block_rsv at this point,
2632 * so we don't need hold lock to protect block_rsv.
2633 * we expand more reservation size here to allow enough
2634 * space for relocation and we will return eailer in
2637 rc->block_rsv->size = tmp + fs_info->nodesize *
2638 RELOCATION_RESERVED_NODES;
2646 * relocate a block tree, and then update pointers in upper level
2647 * blocks that reference the block to point to the new location.
2649 * if called by link_to_upper, the block has already been relocated.
2650 * in that case this function just updates pointers.
2652 static int do_relocation(struct btrfs_trans_handle *trans,
2653 struct reloc_control *rc,
2654 struct backref_node *node,
2655 struct btrfs_key *key,
2656 struct btrfs_path *path, int lowest)
2658 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2659 struct backref_node *upper;
2660 struct backref_edge *edge;
2661 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2662 struct btrfs_root *root;
2663 struct extent_buffer *eb;
2671 BUG_ON(lowest && node->eb);
2673 path->lowest_level = node->level + 1;
2674 rc->backref_cache.path[node->level] = node;
2675 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2676 struct btrfs_key first_key;
2680 upper = edge->node[UPPER];
2681 root = select_reloc_root(trans, rc, upper, edges);
2684 if (upper->eb && !upper->locked) {
2686 ret = btrfs_bin_search(upper->eb, key,
2687 upper->level, &slot);
2689 bytenr = btrfs_node_blockptr(upper->eb, slot);
2690 if (node->eb->start == bytenr)
2693 drop_node_buffer(upper);
2697 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2704 btrfs_release_path(path);
2709 upper->eb = path->nodes[upper->level];
2710 path->nodes[upper->level] = NULL;
2712 BUG_ON(upper->eb != path->nodes[upper->level]);
2716 path->locks[upper->level] = 0;
2718 slot = path->slots[upper->level];
2719 btrfs_release_path(path);
2721 ret = btrfs_bin_search(upper->eb, key, upper->level,
2726 bytenr = btrfs_node_blockptr(upper->eb, slot);
2728 if (bytenr != node->bytenr) {
2729 btrfs_err(root->fs_info,
2730 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2731 bytenr, node->bytenr, slot,
2737 if (node->eb->start == bytenr)
2741 blocksize = root->fs_info->nodesize;
2742 generation = btrfs_node_ptr_generation(upper->eb, slot);
2743 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2744 eb = read_tree_block(fs_info, bytenr, generation,
2745 upper->level - 1, &first_key);
2749 } else if (!extent_buffer_uptodate(eb)) {
2750 free_extent_buffer(eb);
2754 btrfs_tree_lock(eb);
2755 btrfs_set_lock_blocking(eb);
2758 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2760 btrfs_tree_unlock(eb);
2761 free_extent_buffer(eb);
2766 BUG_ON(node->eb != eb);
2768 btrfs_set_node_blockptr(upper->eb, slot,
2770 btrfs_set_node_ptr_generation(upper->eb, slot,
2772 btrfs_mark_buffer_dirty(upper->eb);
2774 ret = btrfs_inc_extent_ref(trans, root,
2775 node->eb->start, blocksize,
2777 btrfs_header_owner(upper->eb),
2781 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2785 if (!upper->pending)
2786 drop_node_buffer(upper);
2788 unlock_node_buffer(upper);
2793 if (!err && node->pending) {
2794 drop_node_buffer(node);
2795 list_move_tail(&node->list, &rc->backref_cache.changed);
2799 path->lowest_level = 0;
2800 BUG_ON(err == -ENOSPC);
2804 static int link_to_upper(struct btrfs_trans_handle *trans,
2805 struct reloc_control *rc,
2806 struct backref_node *node,
2807 struct btrfs_path *path)
2809 struct btrfs_key key;
2811 btrfs_node_key_to_cpu(node->eb, &key, 0);
2812 return do_relocation(trans, rc, node, &key, path, 0);
2815 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2816 struct reloc_control *rc,
2817 struct btrfs_path *path, int err)
2820 struct backref_cache *cache = &rc->backref_cache;
2821 struct backref_node *node;
2825 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2826 while (!list_empty(&cache->pending[level])) {
2827 node = list_entry(cache->pending[level].next,
2828 struct backref_node, list);
2829 list_move_tail(&node->list, &list);
2830 BUG_ON(!node->pending);
2833 ret = link_to_upper(trans, rc, node, path);
2838 list_splice_init(&list, &cache->pending[level]);
2843 static void mark_block_processed(struct reloc_control *rc,
2844 u64 bytenr, u32 blocksize)
2846 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2850 static void __mark_block_processed(struct reloc_control *rc,
2851 struct backref_node *node)
2854 if (node->level == 0 ||
2855 in_block_group(node->bytenr, rc->block_group)) {
2856 blocksize = rc->extent_root->fs_info->nodesize;
2857 mark_block_processed(rc, node->bytenr, blocksize);
2859 node->processed = 1;
2863 * mark a block and all blocks directly/indirectly reference the block
2866 static void update_processed_blocks(struct reloc_control *rc,
2867 struct backref_node *node)
2869 struct backref_node *next = node;
2870 struct backref_edge *edge;
2871 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2877 if (next->processed)
2880 __mark_block_processed(rc, next);
2882 if (list_empty(&next->upper))
2885 edge = list_entry(next->upper.next,
2886 struct backref_edge, list[LOWER]);
2887 edges[index++] = edge;
2888 next = edge->node[UPPER];
2890 next = walk_down_backref(edges, &index);
2894 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2896 u32 blocksize = rc->extent_root->fs_info->nodesize;
2898 if (test_range_bit(&rc->processed_blocks, bytenr,
2899 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2904 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2905 struct tree_block *block)
2907 struct extent_buffer *eb;
2909 BUG_ON(block->key_ready);
2910 eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2911 block->level, NULL);
2914 } else if (!extent_buffer_uptodate(eb)) {
2915 free_extent_buffer(eb);
2918 WARN_ON(btrfs_header_level(eb) != block->level);
2919 if (block->level == 0)
2920 btrfs_item_key_to_cpu(eb, &block->key, 0);
2922 btrfs_node_key_to_cpu(eb, &block->key, 0);
2923 free_extent_buffer(eb);
2924 block->key_ready = 1;
2929 * helper function to relocate a tree block
2931 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2932 struct reloc_control *rc,
2933 struct backref_node *node,
2934 struct btrfs_key *key,
2935 struct btrfs_path *path)
2937 struct btrfs_root *root;
2943 BUG_ON(node->processed);
2944 root = select_one_root(node);
2945 if (root == ERR_PTR(-ENOENT)) {
2946 update_processed_blocks(rc, node);
2950 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2951 ret = reserve_metadata_space(trans, rc, node);
2957 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2958 BUG_ON(node->new_bytenr);
2959 BUG_ON(!list_empty(&node->list));
2960 btrfs_record_root_in_trans(trans, root);
2961 root = root->reloc_root;
2962 node->new_bytenr = root->node->start;
2964 list_add_tail(&node->list, &rc->backref_cache.changed);
2966 path->lowest_level = node->level;
2967 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2968 btrfs_release_path(path);
2973 update_processed_blocks(rc, node);
2975 ret = do_relocation(trans, rc, node, key, path, 1);
2978 if (ret || node->level == 0 || node->cowonly)
2979 remove_backref_node(&rc->backref_cache, node);
2984 * relocate a list of blocks
2986 static noinline_for_stack
2987 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2988 struct reloc_control *rc, struct rb_root *blocks)
2990 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2991 struct backref_node *node;
2992 struct btrfs_path *path;
2993 struct tree_block *block;
2994 struct rb_node *rb_node;
2998 path = btrfs_alloc_path();
3001 goto out_free_blocks;
3004 rb_node = rb_first(blocks);
3006 block = rb_entry(rb_node, struct tree_block, rb_node);
3007 if (!block->key_ready)
3008 readahead_tree_block(fs_info, block->bytenr);
3009 rb_node = rb_next(rb_node);
3012 rb_node = rb_first(blocks);
3014 block = rb_entry(rb_node, struct tree_block, rb_node);
3015 if (!block->key_ready) {
3016 err = get_tree_block_key(fs_info, block);
3020 rb_node = rb_next(rb_node);
3023 rb_node = rb_first(blocks);
3025 block = rb_entry(rb_node, struct tree_block, rb_node);
3027 node = build_backref_tree(rc, &block->key,
3028 block->level, block->bytenr);
3030 err = PTR_ERR(node);
3034 ret = relocate_tree_block(trans, rc, node, &block->key,
3037 if (ret != -EAGAIN || rb_node == rb_first(blocks))
3041 rb_node = rb_next(rb_node);
3044 err = finish_pending_nodes(trans, rc, path, err);
3047 btrfs_free_path(path);
3049 free_block_list(blocks);
3053 static noinline_for_stack
3054 int prealloc_file_extent_cluster(struct inode *inode,
3055 struct file_extent_cluster *cluster)
3060 u64 offset = BTRFS_I(inode)->index_cnt;
3064 u64 prealloc_start = cluster->start - offset;
3065 u64 prealloc_end = cluster->end - offset;
3067 struct extent_changeset *data_reserved = NULL;
3069 BUG_ON(cluster->start != cluster->boundary[0]);
3072 ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3073 prealloc_end + 1 - prealloc_start);
3077 cur_offset = prealloc_start;
3078 while (nr < cluster->nr) {
3079 start = cluster->boundary[nr] - offset;
3080 if (nr + 1 < cluster->nr)
3081 end = cluster->boundary[nr + 1] - 1 - offset;
3083 end = cluster->end - offset;
3085 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3086 num_bytes = end + 1 - start;
3087 if (cur_offset < start)
3088 btrfs_free_reserved_data_space(inode, data_reserved,
3089 cur_offset, start - cur_offset);
3090 ret = btrfs_prealloc_file_range(inode, 0, start,
3091 num_bytes, num_bytes,
3092 end + 1, &alloc_hint);
3093 cur_offset = end + 1;
3094 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3099 if (cur_offset < prealloc_end)
3100 btrfs_free_reserved_data_space(inode, data_reserved,
3101 cur_offset, prealloc_end + 1 - cur_offset);
3103 inode_unlock(inode);
3104 extent_changeset_free(data_reserved);
3108 static noinline_for_stack
3109 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3112 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3113 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3114 struct extent_map *em;
3117 em = alloc_extent_map();
3122 em->len = end + 1 - start;
3123 em->block_len = em->len;
3124 em->block_start = block_start;
3125 em->bdev = fs_info->fs_devices->latest_bdev;
3126 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3128 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3130 write_lock(&em_tree->lock);
3131 ret = add_extent_mapping(em_tree, em, 0);
3132 write_unlock(&em_tree->lock);
3133 if (ret != -EEXIST) {
3134 free_extent_map(em);
3137 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3139 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3143 static int relocate_file_extent_cluster(struct inode *inode,
3144 struct file_extent_cluster *cluster)
3146 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3149 u64 offset = BTRFS_I(inode)->index_cnt;
3150 unsigned long index;
3151 unsigned long last_index;
3153 struct file_ra_state *ra;
3154 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3161 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3165 ret = prealloc_file_extent_cluster(inode, cluster);
3169 file_ra_state_init(ra, inode->i_mapping);
3171 ret = setup_extent_mapping(inode, cluster->start - offset,
3172 cluster->end - offset, cluster->start);
3176 index = (cluster->start - offset) >> PAGE_SHIFT;
3177 last_index = (cluster->end - offset) >> PAGE_SHIFT;
3178 while (index <= last_index) {
3179 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3184 page = find_lock_page(inode->i_mapping, index);
3186 page_cache_sync_readahead(inode->i_mapping,
3188 last_index + 1 - index);
3189 page = find_or_create_page(inode->i_mapping, index,
3192 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3199 if (PageReadahead(page)) {
3200 page_cache_async_readahead(inode->i_mapping,
3201 ra, NULL, page, index,
3202 last_index + 1 - index);
3205 if (!PageUptodate(page)) {
3206 btrfs_readpage(NULL, page);
3208 if (!PageUptodate(page)) {
3211 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3213 btrfs_delalloc_release_extents(BTRFS_I(inode),
3220 page_start = page_offset(page);
3221 page_end = page_start + PAGE_SIZE - 1;
3223 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3225 set_page_extent_mapped(page);
3227 if (nr < cluster->nr &&
3228 page_start + offset == cluster->boundary[nr]) {
3229 set_extent_bits(&BTRFS_I(inode)->io_tree,
3230 page_start, page_end,
3235 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3240 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3242 btrfs_delalloc_release_extents(BTRFS_I(inode),
3245 clear_extent_bits(&BTRFS_I(inode)->io_tree,
3246 page_start, page_end,
3247 EXTENT_LOCKED | EXTENT_BOUNDARY);
3251 set_page_dirty(page);
3253 unlock_extent(&BTRFS_I(inode)->io_tree,
3254 page_start, page_end);
3259 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE,
3261 balance_dirty_pages_ratelimited(inode->i_mapping);
3262 btrfs_throttle(fs_info);
3264 WARN_ON(nr != cluster->nr);
3270 static noinline_for_stack
3271 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3272 struct file_extent_cluster *cluster)
3276 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3277 ret = relocate_file_extent_cluster(inode, cluster);
3284 cluster->start = extent_key->objectid;
3286 BUG_ON(cluster->nr >= MAX_EXTENTS);
3287 cluster->end = extent_key->objectid + extent_key->offset - 1;
3288 cluster->boundary[cluster->nr] = extent_key->objectid;
3291 if (cluster->nr >= MAX_EXTENTS) {
3292 ret = relocate_file_extent_cluster(inode, cluster);
3301 * helper to add a tree block to the list.
3302 * the major work is getting the generation and level of the block
3304 static int add_tree_block(struct reloc_control *rc,
3305 struct btrfs_key *extent_key,
3306 struct btrfs_path *path,
3307 struct rb_root *blocks)
3309 struct extent_buffer *eb;
3310 struct btrfs_extent_item *ei;
3311 struct btrfs_tree_block_info *bi;
3312 struct tree_block *block;
3313 struct rb_node *rb_node;
3318 eb = path->nodes[0];
3319 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3321 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3322 item_size >= sizeof(*ei) + sizeof(*bi)) {
3323 ei = btrfs_item_ptr(eb, path->slots[0],
3324 struct btrfs_extent_item);
3325 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3326 bi = (struct btrfs_tree_block_info *)(ei + 1);
3327 level = btrfs_tree_block_level(eb, bi);
3329 level = (int)extent_key->offset;
3331 generation = btrfs_extent_generation(eb, ei);
3332 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3333 btrfs_print_v0_err(eb->fs_info);
3334 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3340 btrfs_release_path(path);
3342 BUG_ON(level == -1);
3344 block = kmalloc(sizeof(*block), GFP_NOFS);
3348 block->bytenr = extent_key->objectid;
3349 block->key.objectid = rc->extent_root->fs_info->nodesize;
3350 block->key.offset = generation;
3351 block->level = level;
3352 block->key_ready = 0;
3354 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3356 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3362 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3364 static int __add_tree_block(struct reloc_control *rc,
3365 u64 bytenr, u32 blocksize,
3366 struct rb_root *blocks)
3368 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3369 struct btrfs_path *path;
3370 struct btrfs_key key;
3372 bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3374 if (tree_block_processed(bytenr, rc))
3377 if (tree_search(blocks, bytenr))
3380 path = btrfs_alloc_path();
3384 key.objectid = bytenr;
3386 key.type = BTRFS_METADATA_ITEM_KEY;
3387 key.offset = (u64)-1;
3389 key.type = BTRFS_EXTENT_ITEM_KEY;
3390 key.offset = blocksize;
3393 path->search_commit_root = 1;
3394 path->skip_locking = 1;
3395 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3399 if (ret > 0 && skinny) {
3400 if (path->slots[0]) {
3402 btrfs_item_key_to_cpu(path->nodes[0], &key,
3404 if (key.objectid == bytenr &&
3405 (key.type == BTRFS_METADATA_ITEM_KEY ||
3406 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3407 key.offset == blocksize)))
3413 btrfs_release_path(path);
3419 btrfs_print_leaf(path->nodes[0]);
3421 "tree block extent item (%llu) is not found in extent tree",
3428 ret = add_tree_block(rc, &key, path, blocks);
3430 btrfs_free_path(path);
3435 * helper to check if the block use full backrefs for pointers in it
3437 static int block_use_full_backref(struct reloc_control *rc,
3438 struct extent_buffer *eb)
3443 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3444 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3447 ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3448 eb->start, btrfs_header_level(eb), 1,
3452 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3459 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3460 struct btrfs_block_group_cache *block_group,
3461 struct inode *inode,
3464 struct btrfs_key key;
3465 struct btrfs_root *root = fs_info->tree_root;
3466 struct btrfs_trans_handle *trans;
3473 key.type = BTRFS_INODE_ITEM_KEY;
3476 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3481 ret = btrfs_check_trunc_cache_free_space(fs_info,
3482 &fs_info->global_block_rsv);
3486 trans = btrfs_join_transaction(root);
3487 if (IS_ERR(trans)) {
3488 ret = PTR_ERR(trans);
3492 ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3494 btrfs_end_transaction(trans);
3495 btrfs_btree_balance_dirty(fs_info);
3502 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3503 * this function scans fs tree to find blocks reference the data extent
3505 static int find_data_references(struct reloc_control *rc,
3506 struct btrfs_key *extent_key,
3507 struct extent_buffer *leaf,
3508 struct btrfs_extent_data_ref *ref,
3509 struct rb_root *blocks)
3511 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3512 struct btrfs_path *path;
3513 struct tree_block *block;
3514 struct btrfs_root *root;
3515 struct btrfs_file_extent_item *fi;
3516 struct rb_node *rb_node;
3517 struct btrfs_key key;
3528 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3529 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3530 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3531 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3534 * This is an extent belonging to the free space cache, lets just delete
3535 * it and redo the search.
3537 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3538 ret = delete_block_group_cache(fs_info, rc->block_group,
3539 NULL, ref_objectid);
3545 path = btrfs_alloc_path();
3548 path->reada = READA_FORWARD;
3550 root = read_fs_root(fs_info, ref_root);
3552 err = PTR_ERR(root);
3556 key.objectid = ref_objectid;
3557 key.type = BTRFS_EXTENT_DATA_KEY;
3558 if (ref_offset > ((u64)-1 << 32))
3561 key.offset = ref_offset;
3563 path->search_commit_root = 1;
3564 path->skip_locking = 1;
3565 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3571 leaf = path->nodes[0];
3572 nritems = btrfs_header_nritems(leaf);
3574 * the references in tree blocks that use full backrefs
3575 * are not counted in
3577 if (block_use_full_backref(rc, leaf))
3581 rb_node = tree_search(blocks, leaf->start);
3586 path->slots[0] = nritems;
3589 while (ref_count > 0) {
3590 while (path->slots[0] >= nritems) {
3591 ret = btrfs_next_leaf(root, path);
3596 if (WARN_ON(ret > 0))
3599 leaf = path->nodes[0];
3600 nritems = btrfs_header_nritems(leaf);
3603 if (block_use_full_backref(rc, leaf))
3607 rb_node = tree_search(blocks, leaf->start);
3612 path->slots[0] = nritems;
3616 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3617 if (WARN_ON(key.objectid != ref_objectid ||
3618 key.type != BTRFS_EXTENT_DATA_KEY))
3621 fi = btrfs_item_ptr(leaf, path->slots[0],
3622 struct btrfs_file_extent_item);
3624 if (btrfs_file_extent_type(leaf, fi) ==
3625 BTRFS_FILE_EXTENT_INLINE)
3628 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3629 extent_key->objectid)
3632 key.offset -= btrfs_file_extent_offset(leaf, fi);
3633 if (key.offset != ref_offset)
3641 if (!tree_block_processed(leaf->start, rc)) {
3642 block = kmalloc(sizeof(*block), GFP_NOFS);
3647 block->bytenr = leaf->start;
3648 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3650 block->key_ready = 1;
3651 rb_node = tree_insert(blocks, block->bytenr,
3654 backref_tree_panic(rb_node, -EEXIST,
3660 path->slots[0] = nritems;
3666 btrfs_free_path(path);
3671 * helper to find all tree blocks that reference a given data extent
3673 static noinline_for_stack
3674 int add_data_references(struct reloc_control *rc,
3675 struct btrfs_key *extent_key,
3676 struct btrfs_path *path,
3677 struct rb_root *blocks)
3679 struct btrfs_key key;
3680 struct extent_buffer *eb;
3681 struct btrfs_extent_data_ref *dref;
3682 struct btrfs_extent_inline_ref *iref;
3685 u32 blocksize = rc->extent_root->fs_info->nodesize;
3689 eb = path->nodes[0];
3690 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3691 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3692 ptr += sizeof(struct btrfs_extent_item);
3695 iref = (struct btrfs_extent_inline_ref *)ptr;
3696 key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3697 BTRFS_REF_TYPE_DATA);
3698 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3699 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3700 ret = __add_tree_block(rc, key.offset, blocksize,
3702 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3703 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3704 ret = find_data_references(rc, extent_key,
3708 btrfs_err(rc->extent_root->fs_info,
3709 "extent %llu slot %d has an invalid inline ref type",
3710 eb->start, path->slots[0]);
3716 ptr += btrfs_extent_inline_ref_size(key.type);
3722 eb = path->nodes[0];
3723 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3724 ret = btrfs_next_leaf(rc->extent_root, path);
3731 eb = path->nodes[0];
3734 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3735 if (key.objectid != extent_key->objectid)
3738 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3739 ret = __add_tree_block(rc, key.offset, blocksize,
3741 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3742 dref = btrfs_item_ptr(eb, path->slots[0],
3743 struct btrfs_extent_data_ref);
3744 ret = find_data_references(rc, extent_key,
3746 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3747 btrfs_print_v0_err(eb->fs_info);
3748 btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3760 btrfs_release_path(path);
3762 free_block_list(blocks);
3767 * helper to find next unprocessed extent
3769 static noinline_for_stack
3770 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3771 struct btrfs_key *extent_key)
3773 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3774 struct btrfs_key key;
3775 struct extent_buffer *leaf;
3776 u64 start, end, last;
3779 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3782 if (rc->search_start >= last) {
3787 key.objectid = rc->search_start;
3788 key.type = BTRFS_EXTENT_ITEM_KEY;
3791 path->search_commit_root = 1;
3792 path->skip_locking = 1;
3793 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3798 leaf = path->nodes[0];
3799 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3800 ret = btrfs_next_leaf(rc->extent_root, path);
3803 leaf = path->nodes[0];
3806 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3807 if (key.objectid >= last) {
3812 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3813 key.type != BTRFS_METADATA_ITEM_KEY) {
3818 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3819 key.objectid + key.offset <= rc->search_start) {
3824 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3825 key.objectid + fs_info->nodesize <=
3831 ret = find_first_extent_bit(&rc->processed_blocks,
3832 key.objectid, &start, &end,
3833 EXTENT_DIRTY, NULL);
3835 if (ret == 0 && start <= key.objectid) {
3836 btrfs_release_path(path);
3837 rc->search_start = end + 1;
3839 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3840 rc->search_start = key.objectid + key.offset;
3842 rc->search_start = key.objectid +
3844 memcpy(extent_key, &key, sizeof(key));
3848 btrfs_release_path(path);
3852 static void set_reloc_control(struct reloc_control *rc)
3854 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3856 mutex_lock(&fs_info->reloc_mutex);
3857 fs_info->reloc_ctl = rc;
3858 mutex_unlock(&fs_info->reloc_mutex);
3861 static void unset_reloc_control(struct reloc_control *rc)
3863 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3865 mutex_lock(&fs_info->reloc_mutex);
3866 fs_info->reloc_ctl = NULL;
3867 mutex_unlock(&fs_info->reloc_mutex);
3870 static int check_extent_flags(u64 flags)
3872 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3873 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3875 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3876 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3878 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3879 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3884 static noinline_for_stack
3885 int prepare_to_relocate(struct reloc_control *rc)
3887 struct btrfs_trans_handle *trans;
3890 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3891 BTRFS_BLOCK_RSV_TEMP);
3895 memset(&rc->cluster, 0, sizeof(rc->cluster));
3896 rc->search_start = rc->block_group->key.objectid;
3897 rc->extents_found = 0;
3898 rc->nodes_relocated = 0;
3899 rc->merging_rsv_size = 0;
3900 rc->reserved_bytes = 0;
3901 rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3902 RELOCATION_RESERVED_NODES;
3903 ret = btrfs_block_rsv_refill(rc->extent_root,
3904 rc->block_rsv, rc->block_rsv->size,
3905 BTRFS_RESERVE_FLUSH_ALL);
3909 rc->create_reloc_tree = 1;
3910 set_reloc_control(rc);
3912 trans = btrfs_join_transaction(rc->extent_root);
3913 if (IS_ERR(trans)) {
3914 unset_reloc_control(rc);
3916 * extent tree is not a ref_cow tree and has no reloc_root to
3917 * cleanup. And callers are responsible to free the above
3920 return PTR_ERR(trans);
3922 btrfs_commit_transaction(trans);
3926 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3928 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3929 struct rb_root blocks = RB_ROOT;
3930 struct btrfs_key key;
3931 struct btrfs_trans_handle *trans = NULL;
3932 struct btrfs_path *path;
3933 struct btrfs_extent_item *ei;
3940 path = btrfs_alloc_path();
3943 path->reada = READA_FORWARD;
3945 ret = prepare_to_relocate(rc);
3952 rc->reserved_bytes = 0;
3953 ret = btrfs_block_rsv_refill(rc->extent_root,
3954 rc->block_rsv, rc->block_rsv->size,
3955 BTRFS_RESERVE_FLUSH_ALL);
3961 trans = btrfs_start_transaction(rc->extent_root, 0);
3962 if (IS_ERR(trans)) {
3963 err = PTR_ERR(trans);
3968 if (update_backref_cache(trans, &rc->backref_cache)) {
3969 btrfs_end_transaction(trans);
3973 ret = find_next_extent(rc, path, &key);
3979 rc->extents_found++;
3981 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3982 struct btrfs_extent_item);
3983 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3984 if (item_size >= sizeof(*ei)) {
3985 flags = btrfs_extent_flags(path->nodes[0], ei);
3986 ret = check_extent_flags(flags);
3988 } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3990 btrfs_print_v0_err(trans->fs_info);
3991 btrfs_abort_transaction(trans, err);
3997 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3998 ret = add_tree_block(rc, &key, path, &blocks);
3999 } else if (rc->stage == UPDATE_DATA_PTRS &&
4000 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4001 ret = add_data_references(rc, &key, path, &blocks);
4003 btrfs_release_path(path);
4011 if (!RB_EMPTY_ROOT(&blocks)) {
4012 ret = relocate_tree_blocks(trans, rc, &blocks);
4015 * if we fail to relocate tree blocks, force to update
4016 * backref cache when committing transaction.
4018 rc->backref_cache.last_trans = trans->transid - 1;
4020 if (ret != -EAGAIN) {
4024 rc->extents_found--;
4025 rc->search_start = key.objectid;
4029 btrfs_end_transaction_throttle(trans);
4030 btrfs_btree_balance_dirty(fs_info);
4033 if (rc->stage == MOVE_DATA_EXTENTS &&
4034 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4035 rc->found_file_extent = 1;
4036 ret = relocate_data_extent(rc->data_inode,
4037 &key, &rc->cluster);
4044 if (trans && progress && err == -ENOSPC) {
4045 ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
4053 btrfs_release_path(path);
4054 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4057 btrfs_end_transaction_throttle(trans);
4058 btrfs_btree_balance_dirty(fs_info);
4062 ret = relocate_file_extent_cluster(rc->data_inode,
4068 rc->create_reloc_tree = 0;
4069 set_reloc_control(rc);
4071 backref_cache_cleanup(&rc->backref_cache);
4072 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4074 err = prepare_to_merge(rc, err);
4076 merge_reloc_roots(rc);
4078 rc->merge_reloc_tree = 0;
4079 unset_reloc_control(rc);
4080 btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4082 /* get rid of pinned extents */
4083 trans = btrfs_join_transaction(rc->extent_root);
4084 if (IS_ERR(trans)) {
4085 err = PTR_ERR(trans);
4088 btrfs_commit_transaction(trans);
4090 btrfs_free_block_rsv(fs_info, rc->block_rsv);
4091 btrfs_free_path(path);
4095 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4096 struct btrfs_root *root, u64 objectid)
4098 struct btrfs_path *path;
4099 struct btrfs_inode_item *item;
4100 struct extent_buffer *leaf;
4103 path = btrfs_alloc_path();
4107 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4111 leaf = path->nodes[0];
4112 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4113 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4114 btrfs_set_inode_generation(leaf, item, 1);
4115 btrfs_set_inode_size(leaf, item, 0);
4116 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4117 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4118 BTRFS_INODE_PREALLOC);
4119 btrfs_mark_buffer_dirty(leaf);
4121 btrfs_free_path(path);
4126 * helper to create inode for data relocation.
4127 * the inode is in data relocation tree and its link count is 0
4129 static noinline_for_stack
4130 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4131 struct btrfs_block_group_cache *group)
4133 struct inode *inode = NULL;
4134 struct btrfs_trans_handle *trans;
4135 struct btrfs_root *root;
4136 struct btrfs_key key;
4140 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4142 return ERR_CAST(root);
4144 trans = btrfs_start_transaction(root, 6);
4146 return ERR_CAST(trans);
4148 err = btrfs_find_free_objectid(root, &objectid);
4152 err = __insert_orphan_inode(trans, root, objectid);
4155 key.objectid = objectid;
4156 key.type = BTRFS_INODE_ITEM_KEY;
4158 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4159 BUG_ON(IS_ERR(inode));
4160 BTRFS_I(inode)->index_cnt = group->key.objectid;
4162 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4164 btrfs_end_transaction(trans);
4165 btrfs_btree_balance_dirty(fs_info);
4169 inode = ERR_PTR(err);
4174 static struct reloc_control *alloc_reloc_control(void)
4176 struct reloc_control *rc;
4178 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4182 INIT_LIST_HEAD(&rc->reloc_roots);
4183 backref_cache_init(&rc->backref_cache);
4184 mapping_tree_init(&rc->reloc_root_tree);
4185 extent_io_tree_init(&rc->processed_blocks, NULL);
4190 * Print the block group being relocated
4192 static void describe_relocation(struct btrfs_fs_info *fs_info,
4193 struct btrfs_block_group_cache *block_group)
4195 char buf[128]; /* prefixed by a '|' that'll be dropped */
4196 u64 flags = block_group->flags;
4198 /* Shouldn't happen */
4200 strcpy(buf, "|NONE");
4204 #define DESCRIBE_FLAG(f, d) \
4205 if (flags & BTRFS_BLOCK_GROUP_##f) { \
4206 bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \
4207 flags &= ~BTRFS_BLOCK_GROUP_##f; \
4209 DESCRIBE_FLAG(DATA, "data");
4210 DESCRIBE_FLAG(SYSTEM, "system");
4211 DESCRIBE_FLAG(METADATA, "metadata");
4212 DESCRIBE_FLAG(RAID0, "raid0");
4213 DESCRIBE_FLAG(RAID1, "raid1");
4214 DESCRIBE_FLAG(DUP, "dup");
4215 DESCRIBE_FLAG(RAID10, "raid10");
4216 DESCRIBE_FLAG(RAID5, "raid5");
4217 DESCRIBE_FLAG(RAID6, "raid6");
4219 snprintf(bp, buf - bp + sizeof(buf), "|0x%llx", flags);
4220 #undef DESCRIBE_FLAG
4224 "relocating block group %llu flags %s",
4225 block_group->key.objectid, buf + 1);
4229 * function to relocate all extents in a block group.
4231 int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4233 struct btrfs_root *extent_root = fs_info->extent_root;
4234 struct reloc_control *rc;
4235 struct inode *inode;
4236 struct btrfs_path *path;
4241 rc = alloc_reloc_control();
4245 rc->extent_root = extent_root;
4247 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4248 BUG_ON(!rc->block_group);
4250 ret = btrfs_inc_block_group_ro(rc->block_group);
4257 path = btrfs_alloc_path();
4263 inode = lookup_free_space_inode(fs_info, rc->block_group, path);
4264 btrfs_free_path(path);
4267 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4269 ret = PTR_ERR(inode);
4271 if (ret && ret != -ENOENT) {
4276 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4277 if (IS_ERR(rc->data_inode)) {
4278 err = PTR_ERR(rc->data_inode);
4279 rc->data_inode = NULL;
4283 describe_relocation(fs_info, rc->block_group);
4285 btrfs_wait_block_group_reservations(rc->block_group);
4286 btrfs_wait_nocow_writers(rc->block_group);
4287 btrfs_wait_ordered_roots(fs_info, U64_MAX,
4288 rc->block_group->key.objectid,
4289 rc->block_group->key.offset);
4292 mutex_lock(&fs_info->cleaner_mutex);
4293 ret = relocate_block_group(rc);
4294 mutex_unlock(&fs_info->cleaner_mutex);
4300 if (rc->extents_found == 0)
4303 btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4305 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4306 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4312 invalidate_mapping_pages(rc->data_inode->i_mapping,
4314 rc->stage = UPDATE_DATA_PTRS;
4318 WARN_ON(rc->block_group->pinned > 0);
4319 WARN_ON(rc->block_group->reserved > 0);
4320 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4323 btrfs_dec_block_group_ro(rc->block_group);
4324 iput(rc->data_inode);
4325 btrfs_put_block_group(rc->block_group);
4330 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4332 struct btrfs_fs_info *fs_info = root->fs_info;
4333 struct btrfs_trans_handle *trans;
4336 trans = btrfs_start_transaction(fs_info->tree_root, 0);
4338 return PTR_ERR(trans);
4340 memset(&root->root_item.drop_progress, 0,
4341 sizeof(root->root_item.drop_progress));
4342 root->root_item.drop_level = 0;
4343 btrfs_set_root_refs(&root->root_item, 0);
4344 ret = btrfs_update_root(trans, fs_info->tree_root,
4345 &root->root_key, &root->root_item);
4347 err = btrfs_end_transaction(trans);
4354 * recover relocation interrupted by system crash.
4356 * this function resumes merging reloc trees with corresponding fs trees.
4357 * this is important for keeping the sharing of tree blocks
4359 int btrfs_recover_relocation(struct btrfs_root *root)
4361 struct btrfs_fs_info *fs_info = root->fs_info;
4362 LIST_HEAD(reloc_roots);
4363 struct btrfs_key key;
4364 struct btrfs_root *fs_root;
4365 struct btrfs_root *reloc_root;
4366 struct btrfs_path *path;
4367 struct extent_buffer *leaf;
4368 struct reloc_control *rc = NULL;
4369 struct btrfs_trans_handle *trans;
4373 path = btrfs_alloc_path();
4376 path->reada = READA_BACK;
4378 key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4379 key.type = BTRFS_ROOT_ITEM_KEY;
4380 key.offset = (u64)-1;
4383 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4390 if (path->slots[0] == 0)
4394 leaf = path->nodes[0];
4395 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4396 btrfs_release_path(path);
4398 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4399 key.type != BTRFS_ROOT_ITEM_KEY)
4402 reloc_root = btrfs_read_fs_root(root, &key);
4403 if (IS_ERR(reloc_root)) {
4404 err = PTR_ERR(reloc_root);
4408 list_add(&reloc_root->root_list, &reloc_roots);
4410 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4411 fs_root = read_fs_root(fs_info,
4412 reloc_root->root_key.offset);
4413 if (IS_ERR(fs_root)) {
4414 ret = PTR_ERR(fs_root);
4415 if (ret != -ENOENT) {
4419 ret = mark_garbage_root(reloc_root);
4427 if (key.offset == 0)
4432 btrfs_release_path(path);
4434 if (list_empty(&reloc_roots))
4437 rc = alloc_reloc_control();
4443 rc->extent_root = fs_info->extent_root;
4445 set_reloc_control(rc);
4447 trans = btrfs_join_transaction(rc->extent_root);
4448 if (IS_ERR(trans)) {
4449 unset_reloc_control(rc);
4450 err = PTR_ERR(trans);
4454 rc->merge_reloc_tree = 1;
4456 while (!list_empty(&reloc_roots)) {
4457 reloc_root = list_entry(reloc_roots.next,
4458 struct btrfs_root, root_list);
4459 list_del(&reloc_root->root_list);
4461 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4462 list_add_tail(&reloc_root->root_list,
4467 fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
4468 if (IS_ERR(fs_root)) {
4469 err = PTR_ERR(fs_root);
4473 err = __add_reloc_root(reloc_root);
4474 BUG_ON(err < 0); /* -ENOMEM or logic error */
4475 fs_root->reloc_root = reloc_root;
4478 err = btrfs_commit_transaction(trans);
4482 merge_reloc_roots(rc);
4484 unset_reloc_control(rc);
4486 trans = btrfs_join_transaction(rc->extent_root);
4487 if (IS_ERR(trans)) {
4488 err = PTR_ERR(trans);
4491 err = btrfs_commit_transaction(trans);
4495 if (!list_empty(&reloc_roots))
4496 free_reloc_roots(&reloc_roots);
4498 btrfs_free_path(path);
4501 /* cleanup orphan inode in data relocation tree */
4502 fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4503 if (IS_ERR(fs_root))
4504 err = PTR_ERR(fs_root);
4506 err = btrfs_orphan_cleanup(fs_root);
4512 * helper to add ordered checksum for data relocation.
4514 * cloning checksum properly handles the nodatasum extents.
4515 * it also saves CPU time to re-calculate the checksum.
4517 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4519 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4520 struct btrfs_ordered_sum *sums;
4521 struct btrfs_ordered_extent *ordered;
4527 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4528 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4530 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4531 ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4532 disk_bytenr + len - 1, &list, 0);
4536 while (!list_empty(&list)) {
4537 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4538 list_del_init(&sums->list);
4541 * We need to offset the new_bytenr based on where the csum is.
4542 * We need to do this because we will read in entire prealloc
4543 * extents but we may have written to say the middle of the
4544 * prealloc extent, so we need to make sure the csum goes with
4545 * the right disk offset.
4547 * We can do this because the data reloc inode refers strictly
4548 * to the on disk bytes, so we don't have to worry about
4549 * disk_len vs real len like with real inodes since it's all
4552 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4553 sums->bytenr = new_bytenr;
4555 btrfs_add_ordered_sum(inode, ordered, sums);
4558 btrfs_put_ordered_extent(ordered);
4562 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4563 struct btrfs_root *root, struct extent_buffer *buf,
4564 struct extent_buffer *cow)
4566 struct btrfs_fs_info *fs_info = root->fs_info;
4567 struct reloc_control *rc;
4568 struct backref_node *node;
4573 rc = fs_info->reloc_ctl;
4577 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4578 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4580 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4581 if (buf == root->node)
4582 __update_reloc_root(root, cow->start);
4585 level = btrfs_header_level(buf);
4586 if (btrfs_header_generation(buf) <=
4587 btrfs_root_last_snapshot(&root->root_item))
4590 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4591 rc->create_reloc_tree) {
4592 WARN_ON(!first_cow && level == 0);
4594 node = rc->backref_cache.path[level];
4595 BUG_ON(node->bytenr != buf->start &&
4596 node->new_bytenr != buf->start);
4598 drop_node_buffer(node);
4599 extent_buffer_get(cow);
4601 node->new_bytenr = cow->start;
4603 if (!node->pending) {
4604 list_move_tail(&node->list,
4605 &rc->backref_cache.pending[level]);
4610 __mark_block_processed(rc, node);
4612 if (first_cow && level > 0)
4613 rc->nodes_relocated += buf->len;
4616 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4617 ret = replace_file_extents(trans, rc, root, cow);
4622 * called before creating snapshot. it calculates metadata reservation
4623 * required for relocating tree blocks in the snapshot
4625 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4626 u64 *bytes_to_reserve)
4628 struct btrfs_root *root;
4629 struct reloc_control *rc;
4631 root = pending->root;
4632 if (!root->reloc_root)
4635 rc = root->fs_info->reloc_ctl;
4636 if (!rc->merge_reloc_tree)
4639 root = root->reloc_root;
4640 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4642 * relocation is in the stage of merging trees. the space
4643 * used by merging a reloc tree is twice the size of
4644 * relocated tree nodes in the worst case. half for cowing
4645 * the reloc tree, half for cowing the fs tree. the space
4646 * used by cowing the reloc tree will be freed after the
4647 * tree is dropped. if we create snapshot, cowing the fs
4648 * tree may use more space than it frees. so we need
4649 * reserve extra space.
4651 *bytes_to_reserve += rc->nodes_relocated;
4655 * called after snapshot is created. migrate block reservation
4656 * and create reloc root for the newly created snapshot
4658 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4659 struct btrfs_pending_snapshot *pending)
4661 struct btrfs_root *root = pending->root;
4662 struct btrfs_root *reloc_root;
4663 struct btrfs_root *new_root;
4664 struct reloc_control *rc;
4667 if (!root->reloc_root)
4670 rc = root->fs_info->reloc_ctl;
4671 rc->merging_rsv_size += rc->nodes_relocated;
4673 if (rc->merge_reloc_tree) {
4674 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4676 rc->nodes_relocated, true);
4681 new_root = pending->snap;
4682 reloc_root = create_reloc_root(trans, root->reloc_root,
4683 new_root->root_key.objectid);
4684 if (IS_ERR(reloc_root))
4685 return PTR_ERR(reloc_root);
4687 ret = __add_reloc_root(reloc_root);
4689 new_root->reloc_root = reloc_root;
4691 if (rc->create_reloc_tree)
4692 ret = clone_backref_node(trans, rc, root, reloc_root);