2 * Copyright (C) 2009 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
27 #include "transaction.h"
30 #include "btrfs_inode.h"
31 #include "async-thread.h"
32 #include "free-space-cache.h"
33 #include "inode-map.h"
36 * backref_node, mapping_node and tree_block start with this
39 struct rb_node rb_node;
44 * present a tree block in the backref cache
47 struct rb_node rb_node;
51 /* objectid of tree block owner, can be not uptodate */
53 /* link to pending, changed or detached list */
54 struct list_head list;
55 /* list of upper level blocks reference this block */
56 struct list_head upper;
57 /* list of child blocks in the cache */
58 struct list_head lower;
59 /* NULL if this node is not tree root */
60 struct btrfs_root *root;
61 /* extent buffer got by COW the block */
62 struct extent_buffer *eb;
63 /* level of tree block */
65 /* is the block in non-reference counted tree */
66 unsigned int cowonly:1;
67 /* 1 if no child node in the cache */
68 unsigned int lowest:1;
69 /* is the extent buffer locked */
70 unsigned int locked:1;
71 /* has the block been processed */
72 unsigned int processed:1;
73 /* have backrefs of this block been checked */
74 unsigned int checked:1;
76 * 1 if corresponding block has been cowed but some upper
77 * level block pointers may not point to the new location
79 unsigned int pending:1;
81 * 1 if the backref node isn't connected to any other
84 unsigned int detached:1;
88 * present a block pointer in the backref cache
91 struct list_head list[2];
92 struct backref_node *node[2];
97 #define RELOCATION_RESERVED_NODES 256
99 struct backref_cache {
100 /* red black tree of all backref nodes in the cache */
101 struct rb_root rb_root;
102 /* for passing backref nodes to btrfs_reloc_cow_block */
103 struct backref_node *path[BTRFS_MAX_LEVEL];
105 * list of blocks that have been cowed but some block
106 * pointers in upper level blocks may not reflect the
109 struct list_head pending[BTRFS_MAX_LEVEL];
110 /* list of backref nodes with no child node */
111 struct list_head leaves;
112 /* list of blocks that have been cowed in current transaction */
113 struct list_head changed;
114 /* list of detached backref node. */
115 struct list_head detached;
124 * map address of tree root to tree
126 struct mapping_node {
127 struct rb_node rb_node;
132 struct mapping_tree {
133 struct rb_root rb_root;
138 * present a tree block to process
141 struct rb_node rb_node;
143 struct btrfs_key key;
144 unsigned int level:8;
145 unsigned int key_ready:1;
148 #define MAX_EXTENTS 128
150 struct file_extent_cluster {
153 u64 boundary[MAX_EXTENTS];
157 struct reloc_control {
158 /* block group to relocate */
159 struct btrfs_block_group_cache *block_group;
161 struct btrfs_root *extent_root;
162 /* inode for moving data */
163 struct inode *data_inode;
165 struct btrfs_block_rsv *block_rsv;
167 struct backref_cache backref_cache;
169 struct file_extent_cluster cluster;
170 /* tree blocks have been processed */
171 struct extent_io_tree processed_blocks;
172 /* map start of tree root to corresponding reloc tree */
173 struct mapping_tree reloc_root_tree;
174 /* list of reloc trees */
175 struct list_head reloc_roots;
176 /* size of metadata reservation for merging reloc trees */
177 u64 merging_rsv_size;
178 /* size of relocated tree nodes */
180 /* reserved size for block group relocation*/
186 unsigned int stage:8;
187 unsigned int create_reloc_tree:1;
188 unsigned int merge_reloc_tree:1;
189 unsigned int found_file_extent:1;
192 /* stages of data relocation */
193 #define MOVE_DATA_EXTENTS 0
194 #define UPDATE_DATA_PTRS 1
196 static void remove_backref_node(struct backref_cache *cache,
197 struct backref_node *node);
198 static void __mark_block_processed(struct reloc_control *rc,
199 struct backref_node *node);
201 static void mapping_tree_init(struct mapping_tree *tree)
203 tree->rb_root = RB_ROOT;
204 spin_lock_init(&tree->lock);
207 static void backref_cache_init(struct backref_cache *cache)
210 cache->rb_root = RB_ROOT;
211 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
212 INIT_LIST_HEAD(&cache->pending[i]);
213 INIT_LIST_HEAD(&cache->changed);
214 INIT_LIST_HEAD(&cache->detached);
215 INIT_LIST_HEAD(&cache->leaves);
218 static void backref_cache_cleanup(struct backref_cache *cache)
220 struct backref_node *node;
223 while (!list_empty(&cache->detached)) {
224 node = list_entry(cache->detached.next,
225 struct backref_node, list);
226 remove_backref_node(cache, node);
229 while (!list_empty(&cache->leaves)) {
230 node = list_entry(cache->leaves.next,
231 struct backref_node, lower);
232 remove_backref_node(cache, node);
235 cache->last_trans = 0;
237 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
238 BUG_ON(!list_empty(&cache->pending[i]));
239 BUG_ON(!list_empty(&cache->changed));
240 BUG_ON(!list_empty(&cache->detached));
241 BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
242 BUG_ON(cache->nr_nodes);
243 BUG_ON(cache->nr_edges);
246 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
248 struct backref_node *node;
250 node = kzalloc(sizeof(*node), GFP_NOFS);
252 INIT_LIST_HEAD(&node->list);
253 INIT_LIST_HEAD(&node->upper);
254 INIT_LIST_HEAD(&node->lower);
255 RB_CLEAR_NODE(&node->rb_node);
261 static void free_backref_node(struct backref_cache *cache,
262 struct backref_node *node)
270 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
272 struct backref_edge *edge;
274 edge = kzalloc(sizeof(*edge), GFP_NOFS);
280 static void free_backref_edge(struct backref_cache *cache,
281 struct backref_edge *edge)
289 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
290 struct rb_node *node)
292 struct rb_node **p = &root->rb_node;
293 struct rb_node *parent = NULL;
294 struct tree_entry *entry;
298 entry = rb_entry(parent, struct tree_entry, rb_node);
300 if (bytenr < entry->bytenr)
302 else if (bytenr > entry->bytenr)
308 rb_link_node(node, parent, p);
309 rb_insert_color(node, root);
313 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
315 struct rb_node *n = root->rb_node;
316 struct tree_entry *entry;
319 entry = rb_entry(n, struct tree_entry, rb_node);
321 if (bytenr < entry->bytenr)
323 else if (bytenr > entry->bytenr)
331 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
334 struct btrfs_fs_info *fs_info = NULL;
335 struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
338 fs_info = bnode->root->fs_info;
339 btrfs_panic(fs_info, errno, "Inconsistency in backref cache "
340 "found at offset %llu", bytenr);
344 * walk up backref nodes until reach node presents tree root
346 static struct backref_node *walk_up_backref(struct backref_node *node,
347 struct backref_edge *edges[],
350 struct backref_edge *edge;
353 while (!list_empty(&node->upper)) {
354 edge = list_entry(node->upper.next,
355 struct backref_edge, list[LOWER]);
357 node = edge->node[UPPER];
359 BUG_ON(node->detached);
365 * walk down backref nodes to find start of next reference path
367 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
370 struct backref_edge *edge;
371 struct backref_node *lower;
375 edge = edges[idx - 1];
376 lower = edge->node[LOWER];
377 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
381 edge = list_entry(edge->list[LOWER].next,
382 struct backref_edge, list[LOWER]);
383 edges[idx - 1] = edge;
385 return edge->node[UPPER];
391 static void unlock_node_buffer(struct backref_node *node)
394 btrfs_tree_unlock(node->eb);
399 static void drop_node_buffer(struct backref_node *node)
402 unlock_node_buffer(node);
403 free_extent_buffer(node->eb);
408 static void drop_backref_node(struct backref_cache *tree,
409 struct backref_node *node)
411 BUG_ON(!list_empty(&node->upper));
413 drop_node_buffer(node);
414 list_del(&node->list);
415 list_del(&node->lower);
416 if (!RB_EMPTY_NODE(&node->rb_node))
417 rb_erase(&node->rb_node, &tree->rb_root);
418 free_backref_node(tree, node);
422 * remove a backref node from the backref cache
424 static void remove_backref_node(struct backref_cache *cache,
425 struct backref_node *node)
427 struct backref_node *upper;
428 struct backref_edge *edge;
433 BUG_ON(!node->lowest && !node->detached);
434 while (!list_empty(&node->upper)) {
435 edge = list_entry(node->upper.next, struct backref_edge,
437 upper = edge->node[UPPER];
438 list_del(&edge->list[LOWER]);
439 list_del(&edge->list[UPPER]);
440 free_backref_edge(cache, edge);
442 if (RB_EMPTY_NODE(&upper->rb_node)) {
443 BUG_ON(!list_empty(&node->upper));
444 drop_backref_node(cache, node);
450 * add the node to leaf node list if no other
451 * child block cached.
453 if (list_empty(&upper->lower)) {
454 list_add_tail(&upper->lower, &cache->leaves);
459 drop_backref_node(cache, node);
462 static void update_backref_node(struct backref_cache *cache,
463 struct backref_node *node, u64 bytenr)
465 struct rb_node *rb_node;
466 rb_erase(&node->rb_node, &cache->rb_root);
467 node->bytenr = bytenr;
468 rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
470 backref_tree_panic(rb_node, -EEXIST, bytenr);
474 * update backref cache after a transaction commit
476 static int update_backref_cache(struct btrfs_trans_handle *trans,
477 struct backref_cache *cache)
479 struct backref_node *node;
482 if (cache->last_trans == 0) {
483 cache->last_trans = trans->transid;
487 if (cache->last_trans == trans->transid)
491 * detached nodes are used to avoid unnecessary backref
492 * lookup. transaction commit changes the extent tree.
493 * so the detached nodes are no longer useful.
495 while (!list_empty(&cache->detached)) {
496 node = list_entry(cache->detached.next,
497 struct backref_node, list);
498 remove_backref_node(cache, node);
501 while (!list_empty(&cache->changed)) {
502 node = list_entry(cache->changed.next,
503 struct backref_node, list);
504 list_del_init(&node->list);
505 BUG_ON(node->pending);
506 update_backref_node(cache, node, node->new_bytenr);
510 * some nodes can be left in the pending list if there were
511 * errors during processing the pending nodes.
513 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
514 list_for_each_entry(node, &cache->pending[level], list) {
515 BUG_ON(!node->pending);
516 if (node->bytenr == node->new_bytenr)
518 update_backref_node(cache, node, node->new_bytenr);
522 cache->last_trans = 0;
527 static int should_ignore_root(struct btrfs_root *root)
529 struct btrfs_root *reloc_root;
531 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
534 reloc_root = root->reloc_root;
538 if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
539 root->fs_info->running_transaction->transid - 1)
542 * if there is reloc tree and it was created in previous
543 * transaction backref lookup can find the reloc tree,
544 * so backref node for the fs tree root is useless for
550 * find reloc tree by address of tree root
552 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
555 struct rb_node *rb_node;
556 struct mapping_node *node;
557 struct btrfs_root *root = NULL;
559 spin_lock(&rc->reloc_root_tree.lock);
560 rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
562 node = rb_entry(rb_node, struct mapping_node, rb_node);
563 root = (struct btrfs_root *)node->data;
565 spin_unlock(&rc->reloc_root_tree.lock);
569 static int is_cowonly_root(u64 root_objectid)
571 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
572 root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
573 root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
574 root_objectid == BTRFS_DEV_TREE_OBJECTID ||
575 root_objectid == BTRFS_TREE_LOG_OBJECTID ||
576 root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
577 root_objectid == BTRFS_UUID_TREE_OBJECTID ||
578 root_objectid == BTRFS_QUOTA_TREE_OBJECTID)
583 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
586 struct btrfs_key key;
588 key.objectid = root_objectid;
589 key.type = BTRFS_ROOT_ITEM_KEY;
590 if (is_cowonly_root(root_objectid))
593 key.offset = (u64)-1;
595 return btrfs_get_fs_root(fs_info, &key, false);
598 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
599 static noinline_for_stack
600 struct btrfs_root *find_tree_root(struct reloc_control *rc,
601 struct extent_buffer *leaf,
602 struct btrfs_extent_ref_v0 *ref0)
604 struct btrfs_root *root;
605 u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
606 u64 generation = btrfs_ref_generation_v0(leaf, ref0);
608 BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
610 root = read_fs_root(rc->extent_root->fs_info, root_objectid);
611 BUG_ON(IS_ERR(root));
613 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
614 generation != btrfs_root_generation(&root->root_item))
621 static noinline_for_stack
622 int find_inline_backref(struct extent_buffer *leaf, int slot,
623 unsigned long *ptr, unsigned long *end)
625 struct btrfs_key key;
626 struct btrfs_extent_item *ei;
627 struct btrfs_tree_block_info *bi;
630 btrfs_item_key_to_cpu(leaf, &key, slot);
632 item_size = btrfs_item_size_nr(leaf, slot);
633 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
634 if (item_size < sizeof(*ei)) {
635 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
639 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
640 WARN_ON(!(btrfs_extent_flags(leaf, ei) &
641 BTRFS_EXTENT_FLAG_TREE_BLOCK));
643 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
644 item_size <= sizeof(*ei) + sizeof(*bi)) {
645 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
648 if (key.type == BTRFS_METADATA_ITEM_KEY &&
649 item_size <= sizeof(*ei)) {
650 WARN_ON(item_size < sizeof(*ei));
654 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
655 bi = (struct btrfs_tree_block_info *)(ei + 1);
656 *ptr = (unsigned long)(bi + 1);
658 *ptr = (unsigned long)(ei + 1);
660 *end = (unsigned long)ei + item_size;
665 * build backref tree for a given tree block. root of the backref tree
666 * corresponds the tree block, leaves of the backref tree correspond
667 * roots of b-trees that reference the tree block.
669 * the basic idea of this function is check backrefs of a given block
670 * to find upper level blocks that refernece the block, and then check
671 * bakcrefs of these upper level blocks recursively. the recursion stop
672 * when tree root is reached or backrefs for the block is cached.
674 * NOTE: if we find backrefs for a block are cached, we know backrefs
675 * for all upper level blocks that directly/indirectly reference the
676 * block are also cached.
678 static noinline_for_stack
679 struct backref_node *build_backref_tree(struct reloc_control *rc,
680 struct btrfs_key *node_key,
681 int level, u64 bytenr)
683 struct backref_cache *cache = &rc->backref_cache;
684 struct btrfs_path *path1;
685 struct btrfs_path *path2;
686 struct extent_buffer *eb;
687 struct btrfs_root *root;
688 struct backref_node *cur;
689 struct backref_node *upper;
690 struct backref_node *lower;
691 struct backref_node *node = NULL;
692 struct backref_node *exist = NULL;
693 struct backref_edge *edge;
694 struct rb_node *rb_node;
695 struct btrfs_key key;
703 bool need_check = true;
705 path1 = btrfs_alloc_path();
706 path2 = btrfs_alloc_path();
707 if (!path1 || !path2) {
714 node = alloc_backref_node(cache);
720 node->bytenr = bytenr;
727 key.objectid = cur->bytenr;
728 key.type = BTRFS_METADATA_ITEM_KEY;
729 key.offset = (u64)-1;
731 path1->search_commit_root = 1;
732 path1->skip_locking = 1;
733 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
740 ASSERT(path1->slots[0]);
744 WARN_ON(cur->checked);
745 if (!list_empty(&cur->upper)) {
747 * the backref was added previously when processing
748 * backref of type BTRFS_TREE_BLOCK_REF_KEY
750 ASSERT(list_is_singular(&cur->upper));
751 edge = list_entry(cur->upper.next, struct backref_edge,
753 ASSERT(list_empty(&edge->list[UPPER]));
754 exist = edge->node[UPPER];
756 * add the upper level block to pending list if we need
760 list_add_tail(&edge->list[UPPER], &list);
767 eb = path1->nodes[0];
770 if (path1->slots[0] >= btrfs_header_nritems(eb)) {
771 ret = btrfs_next_leaf(rc->extent_root, path1);
778 eb = path1->nodes[0];
781 btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
782 if (key.objectid != cur->bytenr) {
787 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
788 key.type == BTRFS_METADATA_ITEM_KEY) {
789 ret = find_inline_backref(eb, path1->slots[0],
797 /* update key for inline back ref */
798 struct btrfs_extent_inline_ref *iref;
799 iref = (struct btrfs_extent_inline_ref *)ptr;
800 key.type = btrfs_extent_inline_ref_type(eb, iref);
801 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
802 WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
803 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
807 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
808 exist->owner == key.offset) ||
809 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
810 exist->bytenr == key.offset))) {
815 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
816 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
817 key.type == BTRFS_EXTENT_REF_V0_KEY) {
818 if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
819 struct btrfs_extent_ref_v0 *ref0;
820 ref0 = btrfs_item_ptr(eb, path1->slots[0],
821 struct btrfs_extent_ref_v0);
822 if (key.objectid == key.offset) {
823 root = find_tree_root(rc, eb, ref0);
824 if (root && !should_ignore_root(root))
827 list_add(&cur->list, &useless);
830 if (is_cowonly_root(btrfs_ref_root_v0(eb,
835 ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY);
836 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
838 if (key.objectid == key.offset) {
840 * only root blocks of reloc trees use
841 * backref of this type.
843 root = find_reloc_root(rc, cur->bytenr);
849 edge = alloc_backref_edge(cache);
854 rb_node = tree_search(&cache->rb_root, key.offset);
856 upper = alloc_backref_node(cache);
858 free_backref_edge(cache, edge);
862 upper->bytenr = key.offset;
863 upper->level = cur->level + 1;
865 * backrefs for the upper level block isn't
866 * cached, add the block to pending list
868 list_add_tail(&edge->list[UPPER], &list);
870 upper = rb_entry(rb_node, struct backref_node,
872 ASSERT(upper->checked);
873 INIT_LIST_HEAD(&edge->list[UPPER]);
875 list_add_tail(&edge->list[LOWER], &cur->upper);
876 edge->node[LOWER] = cur;
877 edge->node[UPPER] = upper;
880 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
884 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
885 root = read_fs_root(rc->extent_root->fs_info, key.offset);
891 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
894 if (btrfs_root_level(&root->root_item) == cur->level) {
896 ASSERT(btrfs_root_bytenr(&root->root_item) ==
898 if (should_ignore_root(root))
899 list_add(&cur->list, &useless);
905 level = cur->level + 1;
908 * searching the tree to find upper level blocks
909 * reference the block.
911 path2->search_commit_root = 1;
912 path2->skip_locking = 1;
913 path2->lowest_level = level;
914 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
915 path2->lowest_level = 0;
920 if (ret > 0 && path2->slots[level] > 0)
921 path2->slots[level]--;
923 eb = path2->nodes[level];
924 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
926 btrfs_err(root->fs_info,
927 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
928 cur->bytenr, level - 1, root->objectid,
929 node_key->objectid, node_key->type,
936 for (; level < BTRFS_MAX_LEVEL; level++) {
937 if (!path2->nodes[level]) {
938 ASSERT(btrfs_root_bytenr(&root->root_item) ==
940 if (should_ignore_root(root))
941 list_add(&lower->list, &useless);
947 edge = alloc_backref_edge(cache);
953 eb = path2->nodes[level];
954 rb_node = tree_search(&cache->rb_root, eb->start);
956 upper = alloc_backref_node(cache);
958 free_backref_edge(cache, edge);
962 upper->bytenr = eb->start;
963 upper->owner = btrfs_header_owner(eb);
964 upper->level = lower->level + 1;
965 if (!test_bit(BTRFS_ROOT_REF_COWS,
970 * if we know the block isn't shared
971 * we can void checking its backrefs.
973 if (btrfs_block_can_be_shared(root, eb))
979 * add the block to pending list if we
980 * need check its backrefs, we only do this once
981 * while walking up a tree as we will catch
982 * anything else later on.
984 if (!upper->checked && need_check) {
986 list_add_tail(&edge->list[UPPER],
991 INIT_LIST_HEAD(&edge->list[UPPER]);
994 upper = rb_entry(rb_node, struct backref_node,
996 ASSERT(upper->checked);
997 INIT_LIST_HEAD(&edge->list[UPPER]);
999 upper->owner = btrfs_header_owner(eb);
1001 list_add_tail(&edge->list[LOWER], &lower->upper);
1002 edge->node[LOWER] = lower;
1003 edge->node[UPPER] = upper;
1010 btrfs_release_path(path2);
1013 ptr += btrfs_extent_inline_ref_size(key.type);
1023 btrfs_release_path(path1);
1028 /* the pending list isn't empty, take the first block to process */
1029 if (!list_empty(&list)) {
1030 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1031 list_del_init(&edge->list[UPPER]);
1032 cur = edge->node[UPPER];
1037 * everything goes well, connect backref nodes and insert backref nodes
1040 ASSERT(node->checked);
1041 cowonly = node->cowonly;
1043 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1046 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1047 list_add_tail(&node->lower, &cache->leaves);
1050 list_for_each_entry(edge, &node->upper, list[LOWER])
1051 list_add_tail(&edge->list[UPPER], &list);
1053 while (!list_empty(&list)) {
1054 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1055 list_del_init(&edge->list[UPPER]);
1056 upper = edge->node[UPPER];
1057 if (upper->detached) {
1058 list_del(&edge->list[LOWER]);
1059 lower = edge->node[LOWER];
1060 free_backref_edge(cache, edge);
1061 if (list_empty(&lower->upper))
1062 list_add(&lower->list, &useless);
1066 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1067 if (upper->lowest) {
1068 list_del_init(&upper->lower);
1072 list_add_tail(&edge->list[UPPER], &upper->lower);
1076 if (!upper->checked) {
1078 * Still want to blow up for developers since this is a
1085 if (cowonly != upper->cowonly) {
1092 rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1095 backref_tree_panic(rb_node, -EEXIST,
1099 list_add_tail(&edge->list[UPPER], &upper->lower);
1101 list_for_each_entry(edge, &upper->upper, list[LOWER])
1102 list_add_tail(&edge->list[UPPER], &list);
1105 * process useless backref nodes. backref nodes for tree leaves
1106 * are deleted from the cache. backref nodes for upper level
1107 * tree blocks are left in the cache to avoid unnecessary backref
1110 while (!list_empty(&useless)) {
1111 upper = list_entry(useless.next, struct backref_node, list);
1112 list_del_init(&upper->list);
1113 ASSERT(list_empty(&upper->upper));
1116 if (upper->lowest) {
1117 list_del_init(&upper->lower);
1120 while (!list_empty(&upper->lower)) {
1121 edge = list_entry(upper->lower.next,
1122 struct backref_edge, list[UPPER]);
1123 list_del(&edge->list[UPPER]);
1124 list_del(&edge->list[LOWER]);
1125 lower = edge->node[LOWER];
1126 free_backref_edge(cache, edge);
1128 if (list_empty(&lower->upper))
1129 list_add(&lower->list, &useless);
1131 __mark_block_processed(rc, upper);
1132 if (upper->level > 0) {
1133 list_add(&upper->list, &cache->detached);
1134 upper->detached = 1;
1136 rb_erase(&upper->rb_node, &cache->rb_root);
1137 free_backref_node(cache, upper);
1141 btrfs_free_path(path1);
1142 btrfs_free_path(path2);
1144 while (!list_empty(&useless)) {
1145 lower = list_entry(useless.next,
1146 struct backref_node, list);
1147 list_del_init(&lower->list);
1149 while (!list_empty(&list)) {
1150 edge = list_first_entry(&list, struct backref_edge,
1152 list_del(&edge->list[UPPER]);
1153 list_del(&edge->list[LOWER]);
1154 lower = edge->node[LOWER];
1155 upper = edge->node[UPPER];
1156 free_backref_edge(cache, edge);
1159 * Lower is no longer linked to any upper backref nodes
1160 * and isn't in the cache, we can free it ourselves.
1162 if (list_empty(&lower->upper) &&
1163 RB_EMPTY_NODE(&lower->rb_node))
1164 list_add(&lower->list, &useless);
1166 if (!RB_EMPTY_NODE(&upper->rb_node))
1169 /* Add this guy's upper edges to the list to proces */
1170 list_for_each_entry(edge, &upper->upper, list[LOWER])
1171 list_add_tail(&edge->list[UPPER], &list);
1172 if (list_empty(&upper->upper))
1173 list_add(&upper->list, &useless);
1176 while (!list_empty(&useless)) {
1177 lower = list_entry(useless.next,
1178 struct backref_node, list);
1179 list_del_init(&lower->list);
1180 free_backref_node(cache, lower);
1182 return ERR_PTR(err);
1184 ASSERT(!node || !node->detached);
1189 * helper to add backref node for the newly created snapshot.
1190 * the backref node is created by cloning backref node that
1191 * corresponds to root of source tree
1193 static int clone_backref_node(struct btrfs_trans_handle *trans,
1194 struct reloc_control *rc,
1195 struct btrfs_root *src,
1196 struct btrfs_root *dest)
1198 struct btrfs_root *reloc_root = src->reloc_root;
1199 struct backref_cache *cache = &rc->backref_cache;
1200 struct backref_node *node = NULL;
1201 struct backref_node *new_node;
1202 struct backref_edge *edge;
1203 struct backref_edge *new_edge;
1204 struct rb_node *rb_node;
1206 if (cache->last_trans > 0)
1207 update_backref_cache(trans, cache);
1209 rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1211 node = rb_entry(rb_node, struct backref_node, rb_node);
1215 BUG_ON(node->new_bytenr != reloc_root->node->start);
1219 rb_node = tree_search(&cache->rb_root,
1220 reloc_root->commit_root->start);
1222 node = rb_entry(rb_node, struct backref_node,
1224 BUG_ON(node->detached);
1231 new_node = alloc_backref_node(cache);
1235 new_node->bytenr = dest->node->start;
1236 new_node->level = node->level;
1237 new_node->lowest = node->lowest;
1238 new_node->checked = 1;
1239 new_node->root = dest;
1241 if (!node->lowest) {
1242 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1243 new_edge = alloc_backref_edge(cache);
1247 new_edge->node[UPPER] = new_node;
1248 new_edge->node[LOWER] = edge->node[LOWER];
1249 list_add_tail(&new_edge->list[UPPER],
1253 list_add_tail(&new_node->lower, &cache->leaves);
1256 rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1257 &new_node->rb_node);
1259 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1261 if (!new_node->lowest) {
1262 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1263 list_add_tail(&new_edge->list[LOWER],
1264 &new_edge->node[LOWER]->upper);
1269 while (!list_empty(&new_node->lower)) {
1270 new_edge = list_entry(new_node->lower.next,
1271 struct backref_edge, list[UPPER]);
1272 list_del(&new_edge->list[UPPER]);
1273 free_backref_edge(cache, new_edge);
1275 free_backref_node(cache, new_node);
1280 * helper to add 'address of tree root -> reloc tree' mapping
1282 static int __must_check __add_reloc_root(struct btrfs_root *root)
1284 struct rb_node *rb_node;
1285 struct mapping_node *node;
1286 struct reloc_control *rc = root->fs_info->reloc_ctl;
1288 node = kmalloc(sizeof(*node), GFP_NOFS);
1292 node->bytenr = root->commit_root->start;
1295 spin_lock(&rc->reloc_root_tree.lock);
1296 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1297 node->bytenr, &node->rb_node);
1298 spin_unlock(&rc->reloc_root_tree.lock);
1300 btrfs_panic(root->fs_info, -EEXIST, "Duplicate root found "
1301 "for start=%llu while inserting into relocation "
1302 "tree", node->bytenr);
1307 list_add_tail(&root->root_list, &rc->reloc_roots);
1312 * helper to delete the 'address of tree root -> reloc tree'
1315 static void __del_reloc_root(struct btrfs_root *root)
1317 struct rb_node *rb_node;
1318 struct mapping_node *node = NULL;
1319 struct reloc_control *rc = root->fs_info->reloc_ctl;
1321 if (rc && root->node) {
1322 spin_lock(&rc->reloc_root_tree.lock);
1323 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1324 root->commit_root->start);
1326 node = rb_entry(rb_node, struct mapping_node, rb_node);
1327 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1328 RB_CLEAR_NODE(&node->rb_node);
1330 spin_unlock(&rc->reloc_root_tree.lock);
1331 ASSERT(!node || (struct btrfs_root *)node->data == root);
1334 spin_lock(&root->fs_info->trans_lock);
1335 list_del_init(&root->root_list);
1336 spin_unlock(&root->fs_info->trans_lock);
1341 * helper to update the 'address of tree root -> reloc tree'
1344 static int __update_reloc_root(struct btrfs_root *root)
1346 struct rb_node *rb_node;
1347 struct mapping_node *node = NULL;
1348 struct reloc_control *rc = root->fs_info->reloc_ctl;
1350 spin_lock(&rc->reloc_root_tree.lock);
1351 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1352 root->commit_root->start);
1354 node = rb_entry(rb_node, struct mapping_node, rb_node);
1355 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1357 spin_unlock(&rc->reloc_root_tree.lock);
1361 BUG_ON((struct btrfs_root *)node->data != root);
1363 spin_lock(&rc->reloc_root_tree.lock);
1364 node->bytenr = root->node->start;
1365 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1366 node->bytenr, &node->rb_node);
1367 spin_unlock(&rc->reloc_root_tree.lock);
1369 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1373 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1374 struct btrfs_root *root, u64 objectid)
1376 struct btrfs_root *reloc_root;
1377 struct extent_buffer *eb;
1378 struct btrfs_root_item *root_item;
1379 struct btrfs_key root_key;
1383 root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1386 root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1387 root_key.type = BTRFS_ROOT_ITEM_KEY;
1388 root_key.offset = objectid;
1390 if (root->root_key.objectid == objectid) {
1391 /* called by btrfs_init_reloc_root */
1392 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1393 BTRFS_TREE_RELOC_OBJECTID);
1396 last_snap = btrfs_root_last_snapshot(&root->root_item);
1397 btrfs_set_root_last_snapshot(&root->root_item,
1398 trans->transid - 1);
1401 * called by btrfs_reloc_post_snapshot_hook.
1402 * the source tree is a reloc tree, all tree blocks
1403 * modified after it was created have RELOC flag
1404 * set in their headers. so it's OK to not update
1405 * the 'last_snapshot'.
1407 ret = btrfs_copy_root(trans, root, root->node, &eb,
1408 BTRFS_TREE_RELOC_OBJECTID);
1412 memcpy(root_item, &root->root_item, sizeof(*root_item));
1413 btrfs_set_root_bytenr(root_item, eb->start);
1414 btrfs_set_root_level(root_item, btrfs_header_level(eb));
1415 btrfs_set_root_generation(root_item, trans->transid);
1417 if (root->root_key.objectid == objectid) {
1418 btrfs_set_root_refs(root_item, 0);
1419 memset(&root_item->drop_progress, 0,
1420 sizeof(struct btrfs_disk_key));
1421 root_item->drop_level = 0;
1423 * abuse rtransid, it is safe because it is impossible to
1424 * receive data into a relocation tree.
1426 btrfs_set_root_rtransid(root_item, last_snap);
1427 btrfs_set_root_otransid(root_item, trans->transid);
1430 btrfs_tree_unlock(eb);
1431 free_extent_buffer(eb);
1433 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1434 &root_key, root_item);
1438 reloc_root = btrfs_read_fs_root(root->fs_info->tree_root, &root_key);
1439 BUG_ON(IS_ERR(reloc_root));
1440 reloc_root->last_trans = trans->transid;
1445 * create reloc tree for a given fs tree. reloc tree is just a
1446 * snapshot of the fs tree with special root objectid.
1448 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1449 struct btrfs_root *root)
1451 struct btrfs_root *reloc_root;
1452 struct reloc_control *rc = root->fs_info->reloc_ctl;
1453 struct btrfs_block_rsv *rsv;
1457 if (root->reloc_root) {
1458 reloc_root = root->reloc_root;
1459 reloc_root->last_trans = trans->transid;
1463 if (!rc || !rc->create_reloc_tree ||
1464 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1467 if (!trans->reloc_reserved) {
1468 rsv = trans->block_rsv;
1469 trans->block_rsv = rc->block_rsv;
1472 reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1474 trans->block_rsv = rsv;
1476 ret = __add_reloc_root(reloc_root);
1478 root->reloc_root = reloc_root;
1483 * update root item of reloc tree
1485 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1486 struct btrfs_root *root)
1488 struct btrfs_root *reloc_root;
1489 struct btrfs_root_item *root_item;
1492 if (!root->reloc_root)
1495 reloc_root = root->reloc_root;
1496 root_item = &reloc_root->root_item;
1498 if (root->fs_info->reloc_ctl->merge_reloc_tree &&
1499 btrfs_root_refs(root_item) == 0) {
1500 root->reloc_root = NULL;
1501 __del_reloc_root(reloc_root);
1504 if (reloc_root->commit_root != reloc_root->node) {
1505 __update_reloc_root(reloc_root);
1506 btrfs_set_root_node(root_item, reloc_root->node);
1507 free_extent_buffer(reloc_root->commit_root);
1508 reloc_root->commit_root = btrfs_root_node(reloc_root);
1511 ret = btrfs_update_root(trans, root->fs_info->tree_root,
1512 &reloc_root->root_key, root_item);
1520 * helper to find first cached inode with inode number >= objectid
1523 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1525 struct rb_node *node;
1526 struct rb_node *prev;
1527 struct btrfs_inode *entry;
1528 struct inode *inode;
1530 spin_lock(&root->inode_lock);
1532 node = root->inode_tree.rb_node;
1536 entry = rb_entry(node, struct btrfs_inode, rb_node);
1538 if (objectid < btrfs_ino(&entry->vfs_inode))
1539 node = node->rb_left;
1540 else if (objectid > btrfs_ino(&entry->vfs_inode))
1541 node = node->rb_right;
1547 entry = rb_entry(prev, struct btrfs_inode, rb_node);
1548 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
1552 prev = rb_next(prev);
1556 entry = rb_entry(node, struct btrfs_inode, rb_node);
1557 inode = igrab(&entry->vfs_inode);
1559 spin_unlock(&root->inode_lock);
1563 objectid = btrfs_ino(&entry->vfs_inode) + 1;
1564 if (cond_resched_lock(&root->inode_lock))
1567 node = rb_next(node);
1569 spin_unlock(&root->inode_lock);
1573 static int in_block_group(u64 bytenr,
1574 struct btrfs_block_group_cache *block_group)
1576 if (bytenr >= block_group->key.objectid &&
1577 bytenr < block_group->key.objectid + block_group->key.offset)
1583 * get new location of data
1585 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1586 u64 bytenr, u64 num_bytes)
1588 struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1589 struct btrfs_path *path;
1590 struct btrfs_file_extent_item *fi;
1591 struct extent_buffer *leaf;
1594 path = btrfs_alloc_path();
1598 bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1599 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode),
1608 leaf = path->nodes[0];
1609 fi = btrfs_item_ptr(leaf, path->slots[0],
1610 struct btrfs_file_extent_item);
1612 BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1613 btrfs_file_extent_compression(leaf, fi) ||
1614 btrfs_file_extent_encryption(leaf, fi) ||
1615 btrfs_file_extent_other_encoding(leaf, fi));
1617 if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1622 *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1625 btrfs_free_path(path);
1630 * update file extent items in the tree leaf to point to
1631 * the new locations.
1633 static noinline_for_stack
1634 int replace_file_extents(struct btrfs_trans_handle *trans,
1635 struct reloc_control *rc,
1636 struct btrfs_root *root,
1637 struct extent_buffer *leaf)
1639 struct btrfs_key key;
1640 struct btrfs_file_extent_item *fi;
1641 struct inode *inode = NULL;
1653 if (rc->stage != UPDATE_DATA_PTRS)
1656 /* reloc trees always use full backref */
1657 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1658 parent = leaf->start;
1662 nritems = btrfs_header_nritems(leaf);
1663 for (i = 0; i < nritems; i++) {
1665 btrfs_item_key_to_cpu(leaf, &key, i);
1666 if (key.type != BTRFS_EXTENT_DATA_KEY)
1668 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1669 if (btrfs_file_extent_type(leaf, fi) ==
1670 BTRFS_FILE_EXTENT_INLINE)
1672 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1673 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1676 if (!in_block_group(bytenr, rc->block_group))
1680 * if we are modifying block in fs tree, wait for readpage
1681 * to complete and drop the extent cache
1683 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1685 inode = find_next_inode(root, key.objectid);
1687 } else if (inode && btrfs_ino(inode) < key.objectid) {
1688 btrfs_add_delayed_iput(inode);
1689 inode = find_next_inode(root, key.objectid);
1691 if (inode && btrfs_ino(inode) == key.objectid) {
1693 btrfs_file_extent_num_bytes(leaf, fi);
1694 WARN_ON(!IS_ALIGNED(key.offset,
1696 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1698 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1703 btrfs_drop_extent_cache(inode, key.offset, end,
1705 unlock_extent(&BTRFS_I(inode)->io_tree,
1710 ret = get_new_location(rc->data_inode, &new_bytenr,
1714 * Don't have to abort since we've not changed anything
1715 * in the file extent yet.
1720 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1723 key.offset -= btrfs_file_extent_offset(leaf, fi);
1724 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1726 btrfs_header_owner(leaf),
1727 key.objectid, key.offset);
1729 btrfs_abort_transaction(trans, root, ret);
1733 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1734 parent, btrfs_header_owner(leaf),
1735 key.objectid, key.offset);
1737 btrfs_abort_transaction(trans, root, ret);
1742 btrfs_mark_buffer_dirty(leaf);
1744 btrfs_add_delayed_iput(inode);
1748 static noinline_for_stack
1749 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1750 struct btrfs_path *path, int level)
1752 struct btrfs_disk_key key1;
1753 struct btrfs_disk_key key2;
1754 btrfs_node_key(eb, &key1, slot);
1755 btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1756 return memcmp(&key1, &key2, sizeof(key1));
1760 * try to replace tree blocks in fs tree with the new blocks
1761 * in reloc tree. tree blocks haven't been modified since the
1762 * reloc tree was create can be replaced.
1764 * if a block was replaced, level of the block + 1 is returned.
1765 * if no block got replaced, 0 is returned. if there are other
1766 * errors, a negative error number is returned.
1768 static noinline_for_stack
1769 int replace_path(struct btrfs_trans_handle *trans,
1770 struct btrfs_root *dest, struct btrfs_root *src,
1771 struct btrfs_path *path, struct btrfs_key *next_key,
1772 int lowest_level, int max_level)
1774 struct extent_buffer *eb;
1775 struct extent_buffer *parent;
1776 struct btrfs_key key;
1788 ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1789 ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1791 last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1793 slot = path->slots[lowest_level];
1794 btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1796 eb = btrfs_lock_root_node(dest);
1797 btrfs_set_lock_blocking(eb);
1798 level = btrfs_header_level(eb);
1800 if (level < lowest_level) {
1801 btrfs_tree_unlock(eb);
1802 free_extent_buffer(eb);
1807 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1810 btrfs_set_lock_blocking(eb);
1813 next_key->objectid = (u64)-1;
1814 next_key->type = (u8)-1;
1815 next_key->offset = (u64)-1;
1820 level = btrfs_header_level(parent);
1821 ASSERT(level >= lowest_level);
1823 ret = btrfs_bin_search(parent, &key, level, &slot);
1824 if (ret && slot > 0)
1827 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1828 btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1830 old_bytenr = btrfs_node_blockptr(parent, slot);
1831 blocksize = dest->nodesize;
1832 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1834 if (level <= max_level) {
1835 eb = path->nodes[level];
1836 new_bytenr = btrfs_node_blockptr(eb,
1837 path->slots[level]);
1838 new_ptr_gen = btrfs_node_ptr_generation(eb,
1839 path->slots[level]);
1845 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1850 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1851 memcmp_node_keys(parent, slot, path, level)) {
1852 if (level <= lowest_level) {
1857 eb = read_tree_block(dest, old_bytenr, old_ptr_gen);
1860 } else if (!extent_buffer_uptodate(eb)) {
1862 free_extent_buffer(eb);
1865 btrfs_tree_lock(eb);
1867 ret = btrfs_cow_block(trans, dest, eb, parent,
1871 btrfs_set_lock_blocking(eb);
1873 btrfs_tree_unlock(parent);
1874 free_extent_buffer(parent);
1881 btrfs_tree_unlock(parent);
1882 free_extent_buffer(parent);
1887 btrfs_node_key_to_cpu(path->nodes[level], &key,
1888 path->slots[level]);
1889 btrfs_release_path(path);
1891 path->lowest_level = level;
1892 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1893 path->lowest_level = 0;
1897 * swap blocks in fs tree and reloc tree.
1899 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1900 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1901 btrfs_mark_buffer_dirty(parent);
1903 btrfs_set_node_blockptr(path->nodes[level],
1904 path->slots[level], old_bytenr);
1905 btrfs_set_node_ptr_generation(path->nodes[level],
1906 path->slots[level], old_ptr_gen);
1907 btrfs_mark_buffer_dirty(path->nodes[level]);
1909 ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
1910 path->nodes[level]->start,
1911 src->root_key.objectid, level - 1, 0);
1913 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
1914 0, dest->root_key.objectid, level - 1,
1918 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1919 path->nodes[level]->start,
1920 src->root_key.objectid, level - 1, 0);
1923 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1924 0, dest->root_key.objectid, level - 1,
1928 btrfs_unlock_up_safe(path, 0);
1933 btrfs_tree_unlock(parent);
1934 free_extent_buffer(parent);
1939 * helper to find next relocated block in reloc tree
1941 static noinline_for_stack
1942 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1945 struct extent_buffer *eb;
1950 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1952 for (i = 0; i < *level; i++) {
1953 free_extent_buffer(path->nodes[i]);
1954 path->nodes[i] = NULL;
1957 for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1958 eb = path->nodes[i];
1959 nritems = btrfs_header_nritems(eb);
1960 while (path->slots[i] + 1 < nritems) {
1962 if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1969 free_extent_buffer(path->nodes[i]);
1970 path->nodes[i] = NULL;
1976 * walk down reloc tree to find relocated block of lowest level
1978 static noinline_for_stack
1979 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1982 struct extent_buffer *eb = NULL;
1989 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1991 for (i = *level; i > 0; i--) {
1992 eb = path->nodes[i];
1993 nritems = btrfs_header_nritems(eb);
1994 while (path->slots[i] < nritems) {
1995 ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1996 if (ptr_gen > last_snapshot)
2000 if (path->slots[i] >= nritems) {
2011 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2012 eb = read_tree_block(root, bytenr, ptr_gen);
2015 } else if (!extent_buffer_uptodate(eb)) {
2016 free_extent_buffer(eb);
2019 BUG_ON(btrfs_header_level(eb) != i - 1);
2020 path->nodes[i - 1] = eb;
2021 path->slots[i - 1] = 0;
2027 * invalidate extent cache for file extents whose key in range of
2028 * [min_key, max_key)
2030 static int invalidate_extent_cache(struct btrfs_root *root,
2031 struct btrfs_key *min_key,
2032 struct btrfs_key *max_key)
2034 struct inode *inode = NULL;
2039 objectid = min_key->objectid;
2044 if (objectid > max_key->objectid)
2047 inode = find_next_inode(root, objectid);
2050 ino = btrfs_ino(inode);
2052 if (ino > max_key->objectid) {
2058 if (!S_ISREG(inode->i_mode))
2061 if (unlikely(min_key->objectid == ino)) {
2062 if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2064 if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2067 start = min_key->offset;
2068 WARN_ON(!IS_ALIGNED(start, root->sectorsize));
2074 if (unlikely(max_key->objectid == ino)) {
2075 if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2077 if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2080 if (max_key->offset == 0)
2082 end = max_key->offset;
2083 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
2090 /* the lock_extent waits for readpage to complete */
2091 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2092 btrfs_drop_extent_cache(inode, start, end, 1);
2093 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2098 static int find_next_key(struct btrfs_path *path, int level,
2099 struct btrfs_key *key)
2102 while (level < BTRFS_MAX_LEVEL) {
2103 if (!path->nodes[level])
2105 if (path->slots[level] + 1 <
2106 btrfs_header_nritems(path->nodes[level])) {
2107 btrfs_node_key_to_cpu(path->nodes[level], key,
2108 path->slots[level] + 1);
2117 * merge the relocated tree blocks in reloc tree with corresponding
2120 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2121 struct btrfs_root *root)
2123 LIST_HEAD(inode_list);
2124 struct btrfs_key key;
2125 struct btrfs_key next_key;
2126 struct btrfs_trans_handle *trans = NULL;
2127 struct btrfs_root *reloc_root;
2128 struct btrfs_root_item *root_item;
2129 struct btrfs_path *path;
2130 struct extent_buffer *leaf;
2138 path = btrfs_alloc_path();
2143 reloc_root = root->reloc_root;
2144 root_item = &reloc_root->root_item;
2146 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2147 level = btrfs_root_level(root_item);
2148 extent_buffer_get(reloc_root->node);
2149 path->nodes[level] = reloc_root->node;
2150 path->slots[level] = 0;
2152 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2154 level = root_item->drop_level;
2156 path->lowest_level = level;
2157 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2158 path->lowest_level = 0;
2160 btrfs_free_path(path);
2164 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2165 path->slots[level]);
2166 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2168 btrfs_unlock_up_safe(path, 0);
2171 min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2172 memset(&next_key, 0, sizeof(next_key));
2175 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2176 BTRFS_RESERVE_FLUSH_ALL);
2181 trans = btrfs_start_transaction(root, 0);
2182 if (IS_ERR(trans)) {
2183 err = PTR_ERR(trans);
2187 trans->block_rsv = rc->block_rsv;
2192 ret = walk_down_reloc_tree(reloc_root, path, &level);
2200 if (!find_next_key(path, level, &key) &&
2201 btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2204 ret = replace_path(trans, root, reloc_root, path,
2205 &next_key, level, max_level);
2214 btrfs_node_key_to_cpu(path->nodes[level], &key,
2215 path->slots[level]);
2219 ret = walk_up_reloc_tree(reloc_root, path, &level);
2225 * save the merging progress in the drop_progress.
2226 * this is OK since root refs == 1 in this case.
2228 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2229 path->slots[level]);
2230 root_item->drop_level = level;
2232 btrfs_end_transaction_throttle(trans, root);
2235 btrfs_btree_balance_dirty(root);
2237 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2238 invalidate_extent_cache(root, &key, &next_key);
2242 * handle the case only one block in the fs tree need to be
2243 * relocated and the block is tree root.
2245 leaf = btrfs_lock_root_node(root);
2246 ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2247 btrfs_tree_unlock(leaf);
2248 free_extent_buffer(leaf);
2252 btrfs_free_path(path);
2255 memset(&root_item->drop_progress, 0,
2256 sizeof(root_item->drop_progress));
2257 root_item->drop_level = 0;
2258 btrfs_set_root_refs(root_item, 0);
2259 btrfs_update_reloc_root(trans, root);
2263 btrfs_end_transaction_throttle(trans, root);
2265 btrfs_btree_balance_dirty(root);
2267 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2268 invalidate_extent_cache(root, &key, &next_key);
2273 static noinline_for_stack
2274 int prepare_to_merge(struct reloc_control *rc, int err)
2276 struct btrfs_root *root = rc->extent_root;
2277 struct btrfs_root *reloc_root;
2278 struct btrfs_trans_handle *trans;
2279 LIST_HEAD(reloc_roots);
2283 mutex_lock(&root->fs_info->reloc_mutex);
2284 rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2285 rc->merging_rsv_size += rc->nodes_relocated * 2;
2286 mutex_unlock(&root->fs_info->reloc_mutex);
2290 num_bytes = rc->merging_rsv_size;
2291 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2292 BTRFS_RESERVE_FLUSH_ALL);
2297 trans = btrfs_join_transaction(rc->extent_root);
2298 if (IS_ERR(trans)) {
2300 btrfs_block_rsv_release(rc->extent_root,
2301 rc->block_rsv, num_bytes);
2302 return PTR_ERR(trans);
2306 if (num_bytes != rc->merging_rsv_size) {
2307 btrfs_end_transaction(trans, rc->extent_root);
2308 btrfs_block_rsv_release(rc->extent_root,
2309 rc->block_rsv, num_bytes);
2314 rc->merge_reloc_tree = 1;
2316 while (!list_empty(&rc->reloc_roots)) {
2317 reloc_root = list_entry(rc->reloc_roots.next,
2318 struct btrfs_root, root_list);
2319 list_del_init(&reloc_root->root_list);
2321 root = read_fs_root(reloc_root->fs_info,
2322 reloc_root->root_key.offset);
2323 BUG_ON(IS_ERR(root));
2324 BUG_ON(root->reloc_root != reloc_root);
2327 * set reference count to 1, so btrfs_recover_relocation
2328 * knows it should resumes merging
2331 btrfs_set_root_refs(&reloc_root->root_item, 1);
2332 btrfs_update_reloc_root(trans, root);
2334 list_add(&reloc_root->root_list, &reloc_roots);
2337 list_splice(&reloc_roots, &rc->reloc_roots);
2340 btrfs_commit_transaction(trans, rc->extent_root);
2342 btrfs_end_transaction(trans, rc->extent_root);
2346 static noinline_for_stack
2347 void free_reloc_roots(struct list_head *list)
2349 struct btrfs_root *reloc_root;
2351 while (!list_empty(list)) {
2352 reloc_root = list_entry(list->next, struct btrfs_root,
2354 __del_reloc_root(reloc_root);
2355 free_extent_buffer(reloc_root->node);
2356 free_extent_buffer(reloc_root->commit_root);
2357 reloc_root->node = NULL;
2358 reloc_root->commit_root = NULL;
2362 static noinline_for_stack
2363 void merge_reloc_roots(struct reloc_control *rc)
2365 struct btrfs_root *root;
2366 struct btrfs_root *reloc_root;
2370 LIST_HEAD(reloc_roots);
2374 root = rc->extent_root;
2377 * this serializes us with btrfs_record_root_in_transaction,
2378 * we have to make sure nobody is in the middle of
2379 * adding their roots to the list while we are
2382 mutex_lock(&root->fs_info->reloc_mutex);
2383 list_splice_init(&rc->reloc_roots, &reloc_roots);
2384 mutex_unlock(&root->fs_info->reloc_mutex);
2386 while (!list_empty(&reloc_roots)) {
2388 reloc_root = list_entry(reloc_roots.next,
2389 struct btrfs_root, root_list);
2391 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2392 root = read_fs_root(reloc_root->fs_info,
2393 reloc_root->root_key.offset);
2394 BUG_ON(IS_ERR(root));
2395 BUG_ON(root->reloc_root != reloc_root);
2397 ret = merge_reloc_root(rc, root);
2399 if (list_empty(&reloc_root->root_list))
2400 list_add_tail(&reloc_root->root_list,
2405 list_del_init(&reloc_root->root_list);
2409 * we keep the old last snapshod transid in rtranid when we
2410 * created the relocation tree.
2412 last_snap = btrfs_root_rtransid(&reloc_root->root_item);
2413 otransid = btrfs_root_otransid(&reloc_root->root_item);
2414 objectid = reloc_root->root_key.offset;
2416 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2418 if (list_empty(&reloc_root->root_list))
2419 list_add_tail(&reloc_root->root_list,
2431 btrfs_std_error(root->fs_info, ret, NULL);
2432 if (!list_empty(&reloc_roots))
2433 free_reloc_roots(&reloc_roots);
2435 /* new reloc root may be added */
2436 mutex_lock(&root->fs_info->reloc_mutex);
2437 list_splice_init(&rc->reloc_roots, &reloc_roots);
2438 mutex_unlock(&root->fs_info->reloc_mutex);
2439 if (!list_empty(&reloc_roots))
2440 free_reloc_roots(&reloc_roots);
2446 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2448 * here, but it's wrong. If we fail to start the transaction in
2449 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
2450 * have actually been removed from the reloc_root_tree rb tree. This is
2451 * fine because we're bailing here, and we hold a reference on the root
2452 * for the list that holds it, so these roots will be cleaned up when we
2453 * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root
2454 * will be cleaned up on unmount.
2456 * The remaining nodes will be cleaned up by free_reloc_control.
2460 static void free_block_list(struct rb_root *blocks)
2462 struct tree_block *block;
2463 struct rb_node *rb_node;
2464 while ((rb_node = rb_first(blocks))) {
2465 block = rb_entry(rb_node, struct tree_block, rb_node);
2466 rb_erase(rb_node, blocks);
2471 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2472 struct btrfs_root *reloc_root)
2474 struct btrfs_root *root;
2476 if (reloc_root->last_trans == trans->transid)
2479 root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset);
2480 BUG_ON(IS_ERR(root));
2481 BUG_ON(root->reloc_root != reloc_root);
2483 return btrfs_record_root_in_trans(trans, root);
2486 static noinline_for_stack
2487 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2488 struct reloc_control *rc,
2489 struct backref_node *node,
2490 struct backref_edge *edges[])
2492 struct backref_node *next;
2493 struct btrfs_root *root;
2499 next = walk_up_backref(next, edges, &index);
2502 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2504 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2505 record_reloc_root_in_trans(trans, root);
2509 btrfs_record_root_in_trans(trans, root);
2510 root = root->reloc_root;
2512 if (next->new_bytenr != root->node->start) {
2513 BUG_ON(next->new_bytenr);
2514 BUG_ON(!list_empty(&next->list));
2515 next->new_bytenr = root->node->start;
2517 list_add_tail(&next->list,
2518 &rc->backref_cache.changed);
2519 __mark_block_processed(rc, next);
2525 next = walk_down_backref(edges, &index);
2526 if (!next || next->level <= node->level)
2533 /* setup backref node path for btrfs_reloc_cow_block */
2535 rc->backref_cache.path[next->level] = next;
2538 next = edges[index]->node[UPPER];
2544 * select a tree root for relocation. return NULL if the block
2545 * is reference counted. we should use do_relocation() in this
2546 * case. return a tree root pointer if the block isn't reference
2547 * counted. return -ENOENT if the block is root of reloc tree.
2549 static noinline_for_stack
2550 struct btrfs_root *select_one_root(struct backref_node *node)
2552 struct backref_node *next;
2553 struct btrfs_root *root;
2554 struct btrfs_root *fs_root = NULL;
2555 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2561 next = walk_up_backref(next, edges, &index);
2565 /* no other choice for non-references counted tree */
2566 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2569 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2575 next = walk_down_backref(edges, &index);
2576 if (!next || next->level <= node->level)
2581 return ERR_PTR(-ENOENT);
2585 static noinline_for_stack
2586 u64 calcu_metadata_size(struct reloc_control *rc,
2587 struct backref_node *node, int reserve)
2589 struct backref_node *next = node;
2590 struct backref_edge *edge;
2591 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2595 BUG_ON(reserve && node->processed);
2600 if (next->processed && (reserve || next != node))
2603 num_bytes += rc->extent_root->nodesize;
2605 if (list_empty(&next->upper))
2608 edge = list_entry(next->upper.next,
2609 struct backref_edge, list[LOWER]);
2610 edges[index++] = edge;
2611 next = edge->node[UPPER];
2613 next = walk_down_backref(edges, &index);
2618 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2619 struct reloc_control *rc,
2620 struct backref_node *node)
2622 struct btrfs_root *root = rc->extent_root;
2627 num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2629 trans->block_rsv = rc->block_rsv;
2630 rc->reserved_bytes += num_bytes;
2631 ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2632 BTRFS_RESERVE_FLUSH_ALL);
2634 if (ret == -EAGAIN) {
2635 tmp = rc->extent_root->nodesize *
2636 RELOCATION_RESERVED_NODES;
2637 while (tmp <= rc->reserved_bytes)
2640 * only one thread can access block_rsv at this point,
2641 * so we don't need hold lock to protect block_rsv.
2642 * we expand more reservation size here to allow enough
2643 * space for relocation and we will return eailer in
2646 rc->block_rsv->size = tmp + rc->extent_root->nodesize *
2647 RELOCATION_RESERVED_NODES;
2656 * relocate a block tree, and then update pointers in upper level
2657 * blocks that reference the block to point to the new location.
2659 * if called by link_to_upper, the block has already been relocated.
2660 * in that case this function just updates pointers.
2662 static int do_relocation(struct btrfs_trans_handle *trans,
2663 struct reloc_control *rc,
2664 struct backref_node *node,
2665 struct btrfs_key *key,
2666 struct btrfs_path *path, int lowest)
2668 struct backref_node *upper;
2669 struct backref_edge *edge;
2670 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2671 struct btrfs_root *root;
2672 struct extent_buffer *eb;
2680 BUG_ON(lowest && node->eb);
2682 path->lowest_level = node->level + 1;
2683 rc->backref_cache.path[node->level] = node;
2684 list_for_each_entry(edge, &node->upper, list[LOWER]) {
2687 upper = edge->node[UPPER];
2688 root = select_reloc_root(trans, rc, upper, edges);
2691 if (upper->eb && !upper->locked) {
2693 ret = btrfs_bin_search(upper->eb, key,
2694 upper->level, &slot);
2696 bytenr = btrfs_node_blockptr(upper->eb, slot);
2697 if (node->eb->start == bytenr)
2700 drop_node_buffer(upper);
2704 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2711 btrfs_release_path(path);
2716 upper->eb = path->nodes[upper->level];
2717 path->nodes[upper->level] = NULL;
2719 BUG_ON(upper->eb != path->nodes[upper->level]);
2723 path->locks[upper->level] = 0;
2725 slot = path->slots[upper->level];
2726 btrfs_release_path(path);
2728 ret = btrfs_bin_search(upper->eb, key, upper->level,
2733 bytenr = btrfs_node_blockptr(upper->eb, slot);
2735 BUG_ON(bytenr != node->bytenr);
2737 if (node->eb->start == bytenr)
2741 blocksize = root->nodesize;
2742 generation = btrfs_node_ptr_generation(upper->eb, slot);
2743 eb = read_tree_block(root, bytenr, generation);
2747 } else if (!extent_buffer_uptodate(eb)) {
2748 free_extent_buffer(eb);
2752 btrfs_tree_lock(eb);
2753 btrfs_set_lock_blocking(eb);
2756 ret = btrfs_cow_block(trans, root, eb, upper->eb,
2758 btrfs_tree_unlock(eb);
2759 free_extent_buffer(eb);
2764 BUG_ON(node->eb != eb);
2766 btrfs_set_node_blockptr(upper->eb, slot,
2768 btrfs_set_node_ptr_generation(upper->eb, slot,
2770 btrfs_mark_buffer_dirty(upper->eb);
2772 ret = btrfs_inc_extent_ref(trans, root,
2773 node->eb->start, blocksize,
2775 btrfs_header_owner(upper->eb),
2779 ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2783 if (!upper->pending)
2784 drop_node_buffer(upper);
2786 unlock_node_buffer(upper);
2791 if (!err && node->pending) {
2792 drop_node_buffer(node);
2793 list_move_tail(&node->list, &rc->backref_cache.changed);
2797 path->lowest_level = 0;
2798 BUG_ON(err == -ENOSPC);
2802 static int link_to_upper(struct btrfs_trans_handle *trans,
2803 struct reloc_control *rc,
2804 struct backref_node *node,
2805 struct btrfs_path *path)
2807 struct btrfs_key key;
2809 btrfs_node_key_to_cpu(node->eb, &key, 0);
2810 return do_relocation(trans, rc, node, &key, path, 0);
2813 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2814 struct reloc_control *rc,
2815 struct btrfs_path *path, int err)
2818 struct backref_cache *cache = &rc->backref_cache;
2819 struct backref_node *node;
2823 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2824 while (!list_empty(&cache->pending[level])) {
2825 node = list_entry(cache->pending[level].next,
2826 struct backref_node, list);
2827 list_move_tail(&node->list, &list);
2828 BUG_ON(!node->pending);
2831 ret = link_to_upper(trans, rc, node, path);
2836 list_splice_init(&list, &cache->pending[level]);
2841 static void mark_block_processed(struct reloc_control *rc,
2842 u64 bytenr, u32 blocksize)
2844 set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2845 EXTENT_DIRTY, GFP_NOFS);
2848 static void __mark_block_processed(struct reloc_control *rc,
2849 struct backref_node *node)
2852 if (node->level == 0 ||
2853 in_block_group(node->bytenr, rc->block_group)) {
2854 blocksize = rc->extent_root->nodesize;
2855 mark_block_processed(rc, node->bytenr, blocksize);
2857 node->processed = 1;
2861 * mark a block and all blocks directly/indirectly reference the block
2864 static void update_processed_blocks(struct reloc_control *rc,
2865 struct backref_node *node)
2867 struct backref_node *next = node;
2868 struct backref_edge *edge;
2869 struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2875 if (next->processed)
2878 __mark_block_processed(rc, next);
2880 if (list_empty(&next->upper))
2883 edge = list_entry(next->upper.next,
2884 struct backref_edge, list[LOWER]);
2885 edges[index++] = edge;
2886 next = edge->node[UPPER];
2888 next = walk_down_backref(edges, &index);
2892 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2894 u32 blocksize = rc->extent_root->nodesize;
2896 if (test_range_bit(&rc->processed_blocks, bytenr,
2897 bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2902 static int get_tree_block_key(struct reloc_control *rc,
2903 struct tree_block *block)
2905 struct extent_buffer *eb;
2907 BUG_ON(block->key_ready);
2908 eb = read_tree_block(rc->extent_root, block->bytenr,
2912 } else if (!extent_buffer_uptodate(eb)) {
2913 free_extent_buffer(eb);
2916 WARN_ON(btrfs_header_level(eb) != block->level);
2917 if (block->level == 0)
2918 btrfs_item_key_to_cpu(eb, &block->key, 0);
2920 btrfs_node_key_to_cpu(eb, &block->key, 0);
2921 free_extent_buffer(eb);
2922 block->key_ready = 1;
2927 * helper function to relocate a tree block
2929 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2930 struct reloc_control *rc,
2931 struct backref_node *node,
2932 struct btrfs_key *key,
2933 struct btrfs_path *path)
2935 struct btrfs_root *root;
2941 BUG_ON(node->processed);
2942 root = select_one_root(node);
2943 if (root == ERR_PTR(-ENOENT)) {
2944 update_processed_blocks(rc, node);
2948 if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2949 ret = reserve_metadata_space(trans, rc, node);
2955 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2956 BUG_ON(node->new_bytenr);
2957 BUG_ON(!list_empty(&node->list));
2958 btrfs_record_root_in_trans(trans, root);
2959 root = root->reloc_root;
2960 node->new_bytenr = root->node->start;
2962 list_add_tail(&node->list, &rc->backref_cache.changed);
2964 path->lowest_level = node->level;
2965 ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2966 btrfs_release_path(path);
2971 update_processed_blocks(rc, node);
2973 ret = do_relocation(trans, rc, node, key, path, 1);
2976 if (ret || node->level == 0 || node->cowonly)
2977 remove_backref_node(&rc->backref_cache, node);
2982 * relocate a list of blocks
2984 static noinline_for_stack
2985 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2986 struct reloc_control *rc, struct rb_root *blocks)
2988 struct backref_node *node;
2989 struct btrfs_path *path;
2990 struct tree_block *block;
2991 struct rb_node *rb_node;
2995 path = btrfs_alloc_path();
2998 goto out_free_blocks;
3001 rb_node = rb_first(blocks);
3003 block = rb_entry(rb_node, struct tree_block, rb_node);
3004 if (!block->key_ready)
3005 readahead_tree_block(rc->extent_root, block->bytenr);
3006 rb_node = rb_next(rb_node);
3009 rb_node = rb_first(blocks);
3011 block = rb_entry(rb_node, struct tree_block, rb_node);
3012 if (!block->key_ready) {
3013 err = get_tree_block_key(rc, block);
3017 rb_node = rb_next(rb_node);
3020 rb_node = rb_first(blocks);
3022 block = rb_entry(rb_node, struct tree_block, rb_node);
3024 node = build_backref_tree(rc, &block->key,
3025 block->level, block->bytenr);
3027 err = PTR_ERR(node);
3031 ret = relocate_tree_block(trans, rc, node, &block->key,
3034 if (ret != -EAGAIN || rb_node == rb_first(blocks))
3038 rb_node = rb_next(rb_node);
3041 err = finish_pending_nodes(trans, rc, path, err);
3044 btrfs_free_path(path);
3046 free_block_list(blocks);
3050 static noinline_for_stack
3051 int prealloc_file_extent_cluster(struct inode *inode,
3052 struct file_extent_cluster *cluster)
3057 u64 offset = BTRFS_I(inode)->index_cnt;
3062 BUG_ON(cluster->start != cluster->boundary[0]);
3063 mutex_lock(&inode->i_mutex);
3065 ret = btrfs_check_data_free_space(inode, cluster->start,
3066 cluster->end + 1 - cluster->start);
3070 while (nr < cluster->nr) {
3071 start = cluster->boundary[nr] - offset;
3072 if (nr + 1 < cluster->nr)
3073 end = cluster->boundary[nr + 1] - 1 - offset;
3075 end = cluster->end - offset;
3077 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3078 num_bytes = end + 1 - start;
3079 ret = btrfs_prealloc_file_range(inode, 0, start,
3080 num_bytes, num_bytes,
3081 end + 1, &alloc_hint);
3082 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3087 btrfs_free_reserved_data_space(inode, cluster->start,
3088 cluster->end + 1 - cluster->start);
3090 mutex_unlock(&inode->i_mutex);
3094 static noinline_for_stack
3095 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3098 struct btrfs_root *root = BTRFS_I(inode)->root;
3099 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3100 struct extent_map *em;
3103 em = alloc_extent_map();
3108 em->len = end + 1 - start;
3109 em->block_len = em->len;
3110 em->block_start = block_start;
3111 em->bdev = root->fs_info->fs_devices->latest_bdev;
3112 set_bit(EXTENT_FLAG_PINNED, &em->flags);
3114 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3116 write_lock(&em_tree->lock);
3117 ret = add_extent_mapping(em_tree, em, 0);
3118 write_unlock(&em_tree->lock);
3119 if (ret != -EEXIST) {
3120 free_extent_map(em);
3123 btrfs_drop_extent_cache(inode, start, end, 0);
3125 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3129 static int relocate_file_extent_cluster(struct inode *inode,
3130 struct file_extent_cluster *cluster)
3134 u64 offset = BTRFS_I(inode)->index_cnt;
3135 unsigned long index;
3136 unsigned long last_index;
3138 struct file_ra_state *ra;
3139 gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3146 ra = kzalloc(sizeof(*ra), GFP_NOFS);
3150 ret = prealloc_file_extent_cluster(inode, cluster);
3154 file_ra_state_init(ra, inode->i_mapping);
3156 ret = setup_extent_mapping(inode, cluster->start - offset,
3157 cluster->end - offset, cluster->start);
3161 index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
3162 last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
3163 while (index <= last_index) {
3164 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
3168 page = find_lock_page(inode->i_mapping, index);
3170 page_cache_sync_readahead(inode->i_mapping,
3172 last_index + 1 - index);
3173 page = find_or_create_page(inode->i_mapping, index,
3176 btrfs_delalloc_release_metadata(inode,
3183 if (PageReadahead(page)) {
3184 page_cache_async_readahead(inode->i_mapping,
3185 ra, NULL, page, index,
3186 last_index + 1 - index);
3189 if (!PageUptodate(page)) {
3190 btrfs_readpage(NULL, page);
3192 if (!PageUptodate(page)) {
3194 page_cache_release(page);
3195 btrfs_delalloc_release_metadata(inode,
3202 page_start = page_offset(page);
3203 page_end = page_start + PAGE_CACHE_SIZE - 1;
3205 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3207 set_page_extent_mapped(page);
3209 if (nr < cluster->nr &&
3210 page_start + offset == cluster->boundary[nr]) {
3211 set_extent_bits(&BTRFS_I(inode)->io_tree,
3212 page_start, page_end,
3213 EXTENT_BOUNDARY, GFP_NOFS);
3217 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
3218 set_page_dirty(page);
3220 unlock_extent(&BTRFS_I(inode)->io_tree,
3221 page_start, page_end);
3223 page_cache_release(page);
3226 balance_dirty_pages_ratelimited(inode->i_mapping);
3227 btrfs_throttle(BTRFS_I(inode)->root);
3229 WARN_ON(nr != cluster->nr);
3235 static noinline_for_stack
3236 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3237 struct file_extent_cluster *cluster)
3241 if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3242 ret = relocate_file_extent_cluster(inode, cluster);
3249 cluster->start = extent_key->objectid;
3251 BUG_ON(cluster->nr >= MAX_EXTENTS);
3252 cluster->end = extent_key->objectid + extent_key->offset - 1;
3253 cluster->boundary[cluster->nr] = extent_key->objectid;
3256 if (cluster->nr >= MAX_EXTENTS) {
3257 ret = relocate_file_extent_cluster(inode, cluster);
3265 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3266 static int get_ref_objectid_v0(struct reloc_control *rc,
3267 struct btrfs_path *path,
3268 struct btrfs_key *extent_key,
3269 u64 *ref_objectid, int *path_change)
3271 struct btrfs_key key;
3272 struct extent_buffer *leaf;
3273 struct btrfs_extent_ref_v0 *ref0;
3277 leaf = path->nodes[0];
3278 slot = path->slots[0];
3280 if (slot >= btrfs_header_nritems(leaf)) {
3281 ret = btrfs_next_leaf(rc->extent_root, path);
3285 leaf = path->nodes[0];
3286 slot = path->slots[0];
3290 btrfs_item_key_to_cpu(leaf, &key, slot);
3291 if (key.objectid != extent_key->objectid)
3294 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3298 ref0 = btrfs_item_ptr(leaf, slot,
3299 struct btrfs_extent_ref_v0);
3300 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3308 * helper to add a tree block to the list.
3309 * the major work is getting the generation and level of the block
3311 static int add_tree_block(struct reloc_control *rc,
3312 struct btrfs_key *extent_key,
3313 struct btrfs_path *path,
3314 struct rb_root *blocks)
3316 struct extent_buffer *eb;
3317 struct btrfs_extent_item *ei;
3318 struct btrfs_tree_block_info *bi;
3319 struct tree_block *block;
3320 struct rb_node *rb_node;
3325 eb = path->nodes[0];
3326 item_size = btrfs_item_size_nr(eb, path->slots[0]);
3328 if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3329 item_size >= sizeof(*ei) + sizeof(*bi)) {
3330 ei = btrfs_item_ptr(eb, path->slots[0],
3331 struct btrfs_extent_item);
3332 if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3333 bi = (struct btrfs_tree_block_info *)(ei + 1);
3334 level = btrfs_tree_block_level(eb, bi);
3336 level = (int)extent_key->offset;
3338 generation = btrfs_extent_generation(eb, ei);
3340 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3344 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3345 ret = get_ref_objectid_v0(rc, path, extent_key,
3349 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3350 level = (int)ref_owner;
3351 /* FIXME: get real generation */
3358 btrfs_release_path(path);
3360 BUG_ON(level == -1);
3362 block = kmalloc(sizeof(*block), GFP_NOFS);
3366 block->bytenr = extent_key->objectid;
3367 block->key.objectid = rc->extent_root->nodesize;
3368 block->key.offset = generation;
3369 block->level = level;
3370 block->key_ready = 0;
3372 rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3374 backref_tree_panic(rb_node, -EEXIST, block->bytenr);
3380 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3382 static int __add_tree_block(struct reloc_control *rc,
3383 u64 bytenr, u32 blocksize,
3384 struct rb_root *blocks)
3386 struct btrfs_path *path;
3387 struct btrfs_key key;
3389 bool skinny = btrfs_fs_incompat(rc->extent_root->fs_info,
3392 if (tree_block_processed(bytenr, rc))
3395 if (tree_search(blocks, bytenr))
3398 path = btrfs_alloc_path();
3402 key.objectid = bytenr;
3404 key.type = BTRFS_METADATA_ITEM_KEY;
3405 key.offset = (u64)-1;
3407 key.type = BTRFS_EXTENT_ITEM_KEY;
3408 key.offset = blocksize;
3411 path->search_commit_root = 1;
3412 path->skip_locking = 1;
3413 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3417 if (ret > 0 && skinny) {
3418 if (path->slots[0]) {
3420 btrfs_item_key_to_cpu(path->nodes[0], &key,
3422 if (key.objectid == bytenr &&
3423 (key.type == BTRFS_METADATA_ITEM_KEY ||
3424 (key.type == BTRFS_EXTENT_ITEM_KEY &&
3425 key.offset == blocksize)))
3431 btrfs_release_path(path);
3437 ret = add_tree_block(rc, &key, path, blocks);
3439 btrfs_free_path(path);
3444 * helper to check if the block use full backrefs for pointers in it
3446 static int block_use_full_backref(struct reloc_control *rc,
3447 struct extent_buffer *eb)
3452 if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3453 btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3456 ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3457 eb->start, btrfs_header_level(eb), 1,
3461 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3468 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3469 struct btrfs_block_group_cache *block_group,
3470 struct inode *inode,
3473 struct btrfs_key key;
3474 struct btrfs_root *root = fs_info->tree_root;
3475 struct btrfs_trans_handle *trans;
3482 key.type = BTRFS_INODE_ITEM_KEY;
3485 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3486 if (IS_ERR(inode) || is_bad_inode(inode)) {
3493 ret = btrfs_check_trunc_cache_free_space(root,
3494 &fs_info->global_block_rsv);
3498 trans = btrfs_join_transaction(root);
3499 if (IS_ERR(trans)) {
3500 ret = PTR_ERR(trans);
3504 ret = btrfs_truncate_free_space_cache(root, trans, block_group, inode);
3506 btrfs_end_transaction(trans, root);
3507 btrfs_btree_balance_dirty(root);
3514 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3515 * this function scans fs tree to find blocks reference the data extent
3517 static int find_data_references(struct reloc_control *rc,
3518 struct btrfs_key *extent_key,
3519 struct extent_buffer *leaf,
3520 struct btrfs_extent_data_ref *ref,
3521 struct rb_root *blocks)
3523 struct btrfs_path *path;
3524 struct tree_block *block;
3525 struct btrfs_root *root;
3526 struct btrfs_file_extent_item *fi;
3527 struct rb_node *rb_node;
3528 struct btrfs_key key;
3539 ref_root = btrfs_extent_data_ref_root(leaf, ref);
3540 ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3541 ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3542 ref_count = btrfs_extent_data_ref_count(leaf, ref);
3545 * This is an extent belonging to the free space cache, lets just delete
3546 * it and redo the search.
3548 if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3549 ret = delete_block_group_cache(rc->extent_root->fs_info,
3551 NULL, ref_objectid);
3557 path = btrfs_alloc_path();
3562 root = read_fs_root(rc->extent_root->fs_info, ref_root);
3564 err = PTR_ERR(root);
3568 key.objectid = ref_objectid;
3569 key.type = BTRFS_EXTENT_DATA_KEY;
3570 if (ref_offset > ((u64)-1 << 32))
3573 key.offset = ref_offset;
3575 path->search_commit_root = 1;
3576 path->skip_locking = 1;
3577 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3583 leaf = path->nodes[0];
3584 nritems = btrfs_header_nritems(leaf);
3586 * the references in tree blocks that use full backrefs
3587 * are not counted in
3589 if (block_use_full_backref(rc, leaf))
3593 rb_node = tree_search(blocks, leaf->start);
3598 path->slots[0] = nritems;
3601 while (ref_count > 0) {
3602 while (path->slots[0] >= nritems) {
3603 ret = btrfs_next_leaf(root, path);
3608 if (WARN_ON(ret > 0))
3611 leaf = path->nodes[0];
3612 nritems = btrfs_header_nritems(leaf);
3615 if (block_use_full_backref(rc, leaf))
3619 rb_node = tree_search(blocks, leaf->start);
3624 path->slots[0] = nritems;
3628 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3629 if (WARN_ON(key.objectid != ref_objectid ||
3630 key.type != BTRFS_EXTENT_DATA_KEY))
3633 fi = btrfs_item_ptr(leaf, path->slots[0],
3634 struct btrfs_file_extent_item);
3636 if (btrfs_file_extent_type(leaf, fi) ==
3637 BTRFS_FILE_EXTENT_INLINE)
3640 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3641 extent_key->objectid)
3644 key.offset -= btrfs_file_extent_offset(leaf, fi);
3645 if (key.offset != ref_offset)
3653 if (!tree_block_processed(leaf->start, rc)) {
3654 block = kmalloc(sizeof(*block), GFP_NOFS);
3659 block->bytenr = leaf->start;
3660 btrfs_item_key_to_cpu(leaf, &block->key, 0);
3662 block->key_ready = 1;
3663 rb_node = tree_insert(blocks, block->bytenr,
3666 backref_tree_panic(rb_node, -EEXIST,
3672 path->slots[0] = nritems;
3678 btrfs_free_path(path);
3683 * helper to find all tree blocks that reference a given data extent
3685 static noinline_for_stack
3686 int add_data_references(struct reloc_control *rc,
3687 struct btrfs_key *extent_key,
3688 struct btrfs_path *path,
3689 struct rb_root *blocks)
3691 struct btrfs_key key;
3692 struct extent_buffer *eb;
3693 struct btrfs_extent_data_ref *dref;
3694 struct btrfs_extent_inline_ref *iref;
3697 u32 blocksize = rc->extent_root->nodesize;
3701 eb = path->nodes[0];
3702 ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3703 end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3704 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3705 if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3709 ptr += sizeof(struct btrfs_extent_item);
3712 iref = (struct btrfs_extent_inline_ref *)ptr;
3713 key.type = btrfs_extent_inline_ref_type(eb, iref);
3714 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3715 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3716 ret = __add_tree_block(rc, key.offset, blocksize,
3718 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3719 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3720 ret = find_data_references(rc, extent_key,
3729 ptr += btrfs_extent_inline_ref_size(key.type);
3735 eb = path->nodes[0];
3736 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3737 ret = btrfs_next_leaf(rc->extent_root, path);
3744 eb = path->nodes[0];
3747 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3748 if (key.objectid != extent_key->objectid)
3751 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3752 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3753 key.type == BTRFS_EXTENT_REF_V0_KEY) {
3755 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3756 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3758 ret = __add_tree_block(rc, key.offset, blocksize,
3760 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3761 dref = btrfs_item_ptr(eb, path->slots[0],
3762 struct btrfs_extent_data_ref);
3763 ret = find_data_references(rc, extent_key,
3775 btrfs_release_path(path);
3777 free_block_list(blocks);
3782 * helper to find next unprocessed extent
3784 static noinline_for_stack
3785 int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3786 struct btrfs_key *extent_key)
3788 struct btrfs_key key;
3789 struct extent_buffer *leaf;
3790 u64 start, end, last;
3793 last = rc->block_group->key.objectid + rc->block_group->key.offset;
3796 if (rc->search_start >= last) {
3801 key.objectid = rc->search_start;
3802 key.type = BTRFS_EXTENT_ITEM_KEY;
3805 path->search_commit_root = 1;
3806 path->skip_locking = 1;
3807 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3812 leaf = path->nodes[0];
3813 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3814 ret = btrfs_next_leaf(rc->extent_root, path);
3817 leaf = path->nodes[0];
3820 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3821 if (key.objectid >= last) {
3826 if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3827 key.type != BTRFS_METADATA_ITEM_KEY) {
3832 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3833 key.objectid + key.offset <= rc->search_start) {
3838 if (key.type == BTRFS_METADATA_ITEM_KEY &&
3839 key.objectid + rc->extent_root->nodesize <=
3845 ret = find_first_extent_bit(&rc->processed_blocks,
3846 key.objectid, &start, &end,
3847 EXTENT_DIRTY, NULL);
3849 if (ret == 0 && start <= key.objectid) {
3850 btrfs_release_path(path);
3851 rc->search_start = end + 1;
3853 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3854 rc->search_start = key.objectid + key.offset;
3856 rc->search_start = key.objectid +
3857 rc->extent_root->nodesize;
3858 memcpy(extent_key, &key, sizeof(key));
3862 btrfs_release_path(path);
3866 static void set_reloc_control(struct reloc_control *rc)
3868 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3870 mutex_lock(&fs_info->reloc_mutex);
3871 fs_info->reloc_ctl = rc;
3872 mutex_unlock(&fs_info->reloc_mutex);
3875 static void unset_reloc_control(struct reloc_control *rc)
3877 struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3879 mutex_lock(&fs_info->reloc_mutex);
3880 fs_info->reloc_ctl = NULL;
3881 mutex_unlock(&fs_info->reloc_mutex);
3884 static int check_extent_flags(u64 flags)
3886 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3887 (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3889 if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3890 !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3892 if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3893 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3898 static noinline_for_stack
3899 int prepare_to_relocate(struct reloc_control *rc)
3901 struct btrfs_trans_handle *trans;
3903 rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root,
3904 BTRFS_BLOCK_RSV_TEMP);
3908 memset(&rc->cluster, 0, sizeof(rc->cluster));
3909 rc->search_start = rc->block_group->key.objectid;
3910 rc->extents_found = 0;
3911 rc->nodes_relocated = 0;
3912 rc->merging_rsv_size = 0;
3913 rc->reserved_bytes = 0;
3914 rc->block_rsv->size = rc->extent_root->nodesize *
3915 RELOCATION_RESERVED_NODES;
3917 rc->create_reloc_tree = 1;
3918 set_reloc_control(rc);
3920 trans = btrfs_join_transaction(rc->extent_root);
3921 if (IS_ERR(trans)) {
3922 unset_reloc_control(rc);
3924 * extent tree is not a ref_cow tree and has no reloc_root to
3925 * cleanup. And callers are responsible to free the above
3928 return PTR_ERR(trans);
3930 btrfs_commit_transaction(trans, rc->extent_root);
3934 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3936 struct rb_root blocks = RB_ROOT;
3937 struct btrfs_key key;
3938 struct btrfs_trans_handle *trans = NULL;
3939 struct btrfs_path *path;
3940 struct btrfs_extent_item *ei;
3947 path = btrfs_alloc_path();
3952 ret = prepare_to_relocate(rc);
3959 rc->reserved_bytes = 0;
3960 ret = btrfs_block_rsv_refill(rc->extent_root,
3961 rc->block_rsv, rc->block_rsv->size,
3962 BTRFS_RESERVE_FLUSH_ALL);
3968 trans = btrfs_start_transaction(rc->extent_root, 0);
3969 if (IS_ERR(trans)) {
3970 err = PTR_ERR(trans);
3975 if (update_backref_cache(trans, &rc->backref_cache)) {
3976 btrfs_end_transaction(trans, rc->extent_root);
3980 ret = find_next_extent(rc, path, &key);
3986 rc->extents_found++;
3988 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3989 struct btrfs_extent_item);
3990 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3991 if (item_size >= sizeof(*ei)) {
3992 flags = btrfs_extent_flags(path->nodes[0], ei);
3993 ret = check_extent_flags(flags);
3997 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3999 int path_change = 0;
4002 sizeof(struct btrfs_extent_item_v0));
4003 ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
4009 if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
4010 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
4012 flags = BTRFS_EXTENT_FLAG_DATA;
4015 btrfs_release_path(path);
4017 path->search_commit_root = 1;
4018 path->skip_locking = 1;
4019 ret = btrfs_search_slot(NULL, rc->extent_root,
4032 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4033 ret = add_tree_block(rc, &key, path, &blocks);
4034 } else if (rc->stage == UPDATE_DATA_PTRS &&
4035 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4036 ret = add_data_references(rc, &key, path, &blocks);
4038 btrfs_release_path(path);
4046 if (!RB_EMPTY_ROOT(&blocks)) {
4047 ret = relocate_tree_blocks(trans, rc, &blocks);
4050 * if we fail to relocate tree blocks, force to update
4051 * backref cache when committing transaction.
4053 rc->backref_cache.last_trans = trans->transid - 1;
4055 if (ret != -EAGAIN) {
4059 rc->extents_found--;
4060 rc->search_start = key.objectid;
4064 btrfs_end_transaction_throttle(trans, rc->extent_root);
4065 btrfs_btree_balance_dirty(rc->extent_root);
4068 if (rc->stage == MOVE_DATA_EXTENTS &&
4069 (flags & BTRFS_EXTENT_FLAG_DATA)) {
4070 rc->found_file_extent = 1;
4071 ret = relocate_data_extent(rc->data_inode,
4072 &key, &rc->cluster);
4079 if (trans && progress && err == -ENOSPC) {
4080 ret = btrfs_force_chunk_alloc(trans, rc->extent_root,
4081 rc->block_group->flags);
4089 btrfs_release_path(path);
4090 clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
4094 btrfs_end_transaction_throttle(trans, rc->extent_root);
4095 btrfs_btree_balance_dirty(rc->extent_root);
4099 ret = relocate_file_extent_cluster(rc->data_inode,
4105 rc->create_reloc_tree = 0;
4106 set_reloc_control(rc);
4108 backref_cache_cleanup(&rc->backref_cache);
4109 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
4111 err = prepare_to_merge(rc, err);
4113 merge_reloc_roots(rc);
4115 rc->merge_reloc_tree = 0;
4116 unset_reloc_control(rc);
4117 btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
4119 /* get rid of pinned extents */
4120 trans = btrfs_join_transaction(rc->extent_root);
4122 err = PTR_ERR(trans);
4124 btrfs_commit_transaction(trans, rc->extent_root);
4126 btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
4127 btrfs_free_path(path);
4131 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4132 struct btrfs_root *root, u64 objectid)
4134 struct btrfs_path *path;
4135 struct btrfs_inode_item *item;
4136 struct extent_buffer *leaf;
4139 path = btrfs_alloc_path();
4143 ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4147 leaf = path->nodes[0];
4148 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4149 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
4150 btrfs_set_inode_generation(leaf, item, 1);
4151 btrfs_set_inode_size(leaf, item, 0);
4152 btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4153 btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4154 BTRFS_INODE_PREALLOC);
4155 btrfs_mark_buffer_dirty(leaf);
4157 btrfs_free_path(path);
4162 * helper to create inode for data relocation.
4163 * the inode is in data relocation tree and its link count is 0
4165 static noinline_for_stack
4166 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4167 struct btrfs_block_group_cache *group)
4169 struct inode *inode = NULL;
4170 struct btrfs_trans_handle *trans;
4171 struct btrfs_root *root;
4172 struct btrfs_key key;
4176 root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4178 return ERR_CAST(root);
4180 trans = btrfs_start_transaction(root, 6);
4182 return ERR_CAST(trans);
4184 err = btrfs_find_free_objectid(root, &objectid);
4188 err = __insert_orphan_inode(trans, root, objectid);
4191 key.objectid = objectid;
4192 key.type = BTRFS_INODE_ITEM_KEY;
4194 inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
4195 BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
4196 BTRFS_I(inode)->index_cnt = group->key.objectid;
4198 err = btrfs_orphan_add(trans, inode);
4200 btrfs_end_transaction(trans, root);
4201 btrfs_btree_balance_dirty(root);
4205 inode = ERR_PTR(err);
4210 static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4212 struct reloc_control *rc;
4214 rc = kzalloc(sizeof(*rc), GFP_NOFS);
4218 INIT_LIST_HEAD(&rc->reloc_roots);
4219 backref_cache_init(&rc->backref_cache);
4220 mapping_tree_init(&rc->reloc_root_tree);
4221 extent_io_tree_init(&rc->processed_blocks,
4222 fs_info->btree_inode->i_mapping);
4227 * function to relocate all extents in a block group.
4229 int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
4231 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4232 struct reloc_control *rc;
4233 struct inode *inode;
4234 struct btrfs_path *path;
4239 rc = alloc_reloc_control(fs_info);
4243 rc->extent_root = extent_root;
4245 rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4246 BUG_ON(!rc->block_group);
4248 ret = btrfs_inc_block_group_ro(extent_root, rc->block_group);
4255 path = btrfs_alloc_path();
4261 inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
4263 btrfs_free_path(path);
4266 ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4268 ret = PTR_ERR(inode);
4270 if (ret && ret != -ENOENT) {
4275 rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4276 if (IS_ERR(rc->data_inode)) {
4277 err = PTR_ERR(rc->data_inode);
4278 rc->data_inode = NULL;
4282 btrfs_info(extent_root->fs_info, "relocating block group %llu flags %llu",
4283 rc->block_group->key.objectid, rc->block_group->flags);
4285 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
4290 btrfs_wait_ordered_roots(fs_info, -1);
4293 mutex_lock(&fs_info->cleaner_mutex);
4294 ret = relocate_block_group(rc);
4295 mutex_unlock(&fs_info->cleaner_mutex);
4301 if (rc->extents_found == 0)
4304 btrfs_info(extent_root->fs_info, "found %llu extents",
4307 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4308 ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4314 invalidate_mapping_pages(rc->data_inode->i_mapping,
4316 rc->stage = UPDATE_DATA_PTRS;
4320 WARN_ON(rc->block_group->pinned > 0);
4321 WARN_ON(rc->block_group->reserved > 0);
4322 WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4325 btrfs_dec_block_group_ro(extent_root, rc->block_group);
4326 iput(rc->data_inode);
4327 btrfs_put_block_group(rc->block_group);
4332 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4334 struct btrfs_trans_handle *trans;
4337 trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
4339 return PTR_ERR(trans);
4341 memset(&root->root_item.drop_progress, 0,
4342 sizeof(root->root_item.drop_progress));
4343 root->root_item.drop_level = 0;
4344 btrfs_set_root_refs(&root->root_item, 0);
4345 ret = btrfs_update_root(trans, root->fs_info->tree_root,
4346 &root->root_key, &root->root_item);
4348 err = btrfs_end_transaction(trans, root->fs_info->tree_root);
4355 * recover relocation interrupted by system crash.
4357 * this function resumes merging reloc trees with corresponding fs trees.
4358 * this is important for keeping the sharing of tree blocks
4360 int btrfs_recover_relocation(struct btrfs_root *root)
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();
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, root->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(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(root->fs_info);
4443 rc->extent_root = 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(root->fs_info,
4468 reloc_root->root_key.offset);
4469 if (IS_ERR(fs_root)) {
4470 err = PTR_ERR(fs_root);
4471 list_add_tail(&reloc_root->root_list, &reloc_roots);
4475 err = __add_reloc_root(reloc_root);
4476 BUG_ON(err < 0); /* -ENOMEM or logic error */
4477 fs_root->reloc_root = reloc_root;
4480 err = btrfs_commit_transaction(trans, rc->extent_root);
4484 merge_reloc_roots(rc);
4486 unset_reloc_control(rc);
4488 trans = btrfs_join_transaction(rc->extent_root);
4490 err = PTR_ERR(trans);
4492 err = btrfs_commit_transaction(trans, rc->extent_root);
4496 if (!list_empty(&reloc_roots))
4497 free_reloc_roots(&reloc_roots);
4499 btrfs_free_path(path);
4502 /* cleanup orphan inode in data relocation tree */
4503 fs_root = read_fs_root(root->fs_info,
4504 BTRFS_DATA_RELOC_TREE_OBJECTID);
4505 if (IS_ERR(fs_root))
4506 err = PTR_ERR(fs_root);
4508 err = btrfs_orphan_cleanup(fs_root);
4514 * helper to add ordered checksum for data relocation.
4516 * cloning checksum properly handles the nodatasum extents.
4517 * it also saves CPU time to re-calculate the checksum.
4519 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4521 struct btrfs_ordered_sum *sums;
4522 struct btrfs_ordered_extent *ordered;
4523 struct btrfs_root *root = BTRFS_I(inode)->root;
4529 ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4530 BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4532 disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4533 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
4534 disk_bytenr + len - 1, &list, 0);
4538 while (!list_empty(&list)) {
4539 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4540 list_del_init(&sums->list);
4543 * We need to offset the new_bytenr based on where the csum is.
4544 * We need to do this because we will read in entire prealloc
4545 * extents but we may have written to say the middle of the
4546 * prealloc extent, so we need to make sure the csum goes with
4547 * the right disk offset.
4549 * We can do this because the data reloc inode refers strictly
4550 * to the on disk bytes, so we don't have to worry about
4551 * disk_len vs real len like with real inodes since it's all
4554 new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4555 sums->bytenr = new_bytenr;
4557 btrfs_add_ordered_sum(inode, ordered, sums);
4560 btrfs_put_ordered_extent(ordered);
4564 int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4565 struct btrfs_root *root, struct extent_buffer *buf,
4566 struct extent_buffer *cow)
4568 struct reloc_control *rc;
4569 struct backref_node *node;
4574 rc = root->fs_info->reloc_ctl;
4578 BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4579 root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4581 level = btrfs_header_level(buf);
4582 if (btrfs_header_generation(buf) <=
4583 btrfs_root_last_snapshot(&root->root_item))
4586 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4587 rc->create_reloc_tree) {
4588 WARN_ON(!first_cow && level == 0);
4590 node = rc->backref_cache.path[level];
4591 BUG_ON(node->bytenr != buf->start &&
4592 node->new_bytenr != buf->start);
4594 drop_node_buffer(node);
4595 extent_buffer_get(cow);
4597 node->new_bytenr = cow->start;
4599 if (!node->pending) {
4600 list_move_tail(&node->list,
4601 &rc->backref_cache.pending[level]);
4606 __mark_block_processed(rc, node);
4608 if (first_cow && level > 0)
4609 rc->nodes_relocated += buf->len;
4612 if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4613 ret = replace_file_extents(trans, rc, root, cow);
4618 * called before creating snapshot. it calculates metadata reservation
4619 * requried for relocating tree blocks in the snapshot
4621 void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4622 u64 *bytes_to_reserve)
4624 struct btrfs_root *root;
4625 struct reloc_control *rc;
4627 root = pending->root;
4628 if (!root->reloc_root)
4631 rc = root->fs_info->reloc_ctl;
4632 if (!rc->merge_reloc_tree)
4635 root = root->reloc_root;
4636 BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4638 * relocation is in the stage of merging trees. the space
4639 * used by merging a reloc tree is twice the size of
4640 * relocated tree nodes in the worst case. half for cowing
4641 * the reloc tree, half for cowing the fs tree. the space
4642 * used by cowing the reloc tree will be freed after the
4643 * tree is dropped. if we create snapshot, cowing the fs
4644 * tree may use more space than it frees. so we need
4645 * reserve extra space.
4647 *bytes_to_reserve += rc->nodes_relocated;
4651 * called after snapshot is created. migrate block reservation
4652 * and create reloc root for the newly created snapshot
4654 int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4655 struct btrfs_pending_snapshot *pending)
4657 struct btrfs_root *root = pending->root;
4658 struct btrfs_root *reloc_root;
4659 struct btrfs_root *new_root;
4660 struct reloc_control *rc;
4663 if (!root->reloc_root)
4666 rc = root->fs_info->reloc_ctl;
4667 rc->merging_rsv_size += rc->nodes_relocated;
4669 if (rc->merge_reloc_tree) {
4670 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4672 rc->nodes_relocated);
4677 new_root = pending->snap;
4678 reloc_root = create_reloc_root(trans, root->reloc_root,
4679 new_root->root_key.objectid);
4680 if (IS_ERR(reloc_root))
4681 return PTR_ERR(reloc_root);
4683 ret = __add_reloc_root(reloc_root);
4685 new_root->reloc_root = reloc_root;
4687 if (rc->create_reloc_tree)
4688 ret = clone_backref_node(trans, rc, root, reloc_root);