Btrfs: Metadata ENOSPC handling for balance
[linux-2.6.git] / fs / btrfs / relocation.c
1 /*
2  * Copyright (C) 2009 Oracle.  All rights reserved.
3  *
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.
7  *
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.
12  *
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.
17  */
18
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>
25 #include "ctree.h"
26 #include "disk-io.h"
27 #include "transaction.h"
28 #include "volumes.h"
29 #include "locking.h"
30 #include "btrfs_inode.h"
31 #include "async-thread.h"
32
33 /*
34  * backref_node, mapping_node and tree_block start with this
35  */
36 struct tree_entry {
37         struct rb_node rb_node;
38         u64 bytenr;
39 };
40
41 /*
42  * present a tree block in the backref cache
43  */
44 struct backref_node {
45         struct rb_node rb_node;
46         u64 bytenr;
47
48         u64 new_bytenr;
49         /* objectid of tree block owner, can be not uptodate */
50         u64 owner;
51         /* link to pending, changed or detached list */
52         struct list_head list;
53         /* list of upper level blocks reference this block */
54         struct list_head upper;
55         /* list of child blocks in the cache */
56         struct list_head lower;
57         /* NULL if this node is not tree root */
58         struct btrfs_root *root;
59         /* extent buffer got by COW the block */
60         struct extent_buffer *eb;
61         /* level of tree block */
62         unsigned int level:8;
63         /* is the block in non-reference counted tree */
64         unsigned int cowonly:1;
65         /* 1 if no child node in the cache */
66         unsigned int lowest:1;
67         /* is the extent buffer locked */
68         unsigned int locked:1;
69         /* has the block been processed */
70         unsigned int processed:1;
71         /* have backrefs of this block been checked */
72         unsigned int checked:1;
73         /*
74          * 1 if corresponding block has been cowed but some upper
75          * level block pointers may not point to the new location
76          */
77         unsigned int pending:1;
78         /*
79          * 1 if the backref node isn't connected to any other
80          * backref node.
81          */
82         unsigned int detached:1;
83 };
84
85 /*
86  * present a block pointer in the backref cache
87  */
88 struct backref_edge {
89         struct list_head list[2];
90         struct backref_node *node[2];
91 };
92
93 #define LOWER   0
94 #define UPPER   1
95
96 struct backref_cache {
97         /* red black tree of all backref nodes in the cache */
98         struct rb_root rb_root;
99         /* for passing backref nodes to btrfs_reloc_cow_block */
100         struct backref_node *path[BTRFS_MAX_LEVEL];
101         /*
102          * list of blocks that have been cowed but some block
103          * pointers in upper level blocks may not reflect the
104          * new location
105          */
106         struct list_head pending[BTRFS_MAX_LEVEL];
107         /* list of backref nodes with no child node */
108         struct list_head leaves;
109         /* list of blocks that have been cowed in current transaction */
110         struct list_head changed;
111         /* list of detached backref node. */
112         struct list_head detached;
113
114         u64 last_trans;
115
116         int nr_nodes;
117         int nr_edges;
118 };
119
120 /*
121  * map address of tree root to tree
122  */
123 struct mapping_node {
124         struct rb_node rb_node;
125         u64 bytenr;
126         void *data;
127 };
128
129 struct mapping_tree {
130         struct rb_root rb_root;
131         spinlock_t lock;
132 };
133
134 /*
135  * present a tree block to process
136  */
137 struct tree_block {
138         struct rb_node rb_node;
139         u64 bytenr;
140         struct btrfs_key key;
141         unsigned int level:8;
142         unsigned int key_ready:1;
143 };
144
145 #define MAX_EXTENTS 128
146
147 struct file_extent_cluster {
148         u64 start;
149         u64 end;
150         u64 boundary[MAX_EXTENTS];
151         unsigned int nr;
152 };
153
154 struct reloc_control {
155         /* block group to relocate */
156         struct btrfs_block_group_cache *block_group;
157         /* extent tree */
158         struct btrfs_root *extent_root;
159         /* inode for moving data */
160         struct inode *data_inode;
161
162         struct btrfs_block_rsv *block_rsv;
163
164         struct backref_cache backref_cache;
165
166         struct file_extent_cluster cluster;
167         /* tree blocks have been processed */
168         struct extent_io_tree processed_blocks;
169         /* map start of tree root to corresponding reloc tree */
170         struct mapping_tree reloc_root_tree;
171         /* list of reloc trees */
172         struct list_head reloc_roots;
173         /* size of metadata reservation for merging reloc trees */
174         u64 merging_rsv_size;
175         /* size of relocated tree nodes */
176         u64 nodes_relocated;
177
178         u64 search_start;
179         u64 extents_found;
180
181         int block_rsv_retries;
182
183         unsigned int stage:8;
184         unsigned int create_reloc_tree:1;
185         unsigned int merge_reloc_tree:1;
186         unsigned int found_file_extent:1;
187         unsigned int commit_transaction:1;
188 };
189
190 /* stages of data relocation */
191 #define MOVE_DATA_EXTENTS       0
192 #define UPDATE_DATA_PTRS        1
193
194 static void remove_backref_node(struct backref_cache *cache,
195                                 struct backref_node *node);
196 static void __mark_block_processed(struct reloc_control *rc,
197                                    struct backref_node *node);
198
199 static void mapping_tree_init(struct mapping_tree *tree)
200 {
201         tree->rb_root = RB_ROOT;
202         spin_lock_init(&tree->lock);
203 }
204
205 static void backref_cache_init(struct backref_cache *cache)
206 {
207         int i;
208         cache->rb_root = RB_ROOT;
209         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
210                 INIT_LIST_HEAD(&cache->pending[i]);
211         INIT_LIST_HEAD(&cache->changed);
212         INIT_LIST_HEAD(&cache->detached);
213         INIT_LIST_HEAD(&cache->leaves);
214 }
215
216 static void backref_cache_cleanup(struct backref_cache *cache)
217 {
218         struct backref_node *node;
219         int i;
220
221         while (!list_empty(&cache->detached)) {
222                 node = list_entry(cache->detached.next,
223                                   struct backref_node, list);
224                 remove_backref_node(cache, node);
225         }
226
227         while (!list_empty(&cache->leaves)) {
228                 node = list_entry(cache->leaves.next,
229                                   struct backref_node, lower);
230                 remove_backref_node(cache, node);
231         }
232
233         cache->last_trans = 0;
234
235         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
236                 BUG_ON(!list_empty(&cache->pending[i]));
237         BUG_ON(!list_empty(&cache->changed));
238         BUG_ON(!list_empty(&cache->detached));
239         BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
240         BUG_ON(cache->nr_nodes);
241         BUG_ON(cache->nr_edges);
242 }
243
244 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
245 {
246         struct backref_node *node;
247
248         node = kzalloc(sizeof(*node), GFP_NOFS);
249         if (node) {
250                 INIT_LIST_HEAD(&node->list);
251                 INIT_LIST_HEAD(&node->upper);
252                 INIT_LIST_HEAD(&node->lower);
253                 RB_CLEAR_NODE(&node->rb_node);
254                 cache->nr_nodes++;
255         }
256         return node;
257 }
258
259 static void free_backref_node(struct backref_cache *cache,
260                               struct backref_node *node)
261 {
262         if (node) {
263                 cache->nr_nodes--;
264                 kfree(node);
265         }
266 }
267
268 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
269 {
270         struct backref_edge *edge;
271
272         edge = kzalloc(sizeof(*edge), GFP_NOFS);
273         if (edge)
274                 cache->nr_edges++;
275         return edge;
276 }
277
278 static void free_backref_edge(struct backref_cache *cache,
279                               struct backref_edge *edge)
280 {
281         if (edge) {
282                 cache->nr_edges--;
283                 kfree(edge);
284         }
285 }
286
287 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
288                                    struct rb_node *node)
289 {
290         struct rb_node **p = &root->rb_node;
291         struct rb_node *parent = NULL;
292         struct tree_entry *entry;
293
294         while (*p) {
295                 parent = *p;
296                 entry = rb_entry(parent, struct tree_entry, rb_node);
297
298                 if (bytenr < entry->bytenr)
299                         p = &(*p)->rb_left;
300                 else if (bytenr > entry->bytenr)
301                         p = &(*p)->rb_right;
302                 else
303                         return parent;
304         }
305
306         rb_link_node(node, parent, p);
307         rb_insert_color(node, root);
308         return NULL;
309 }
310
311 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
312 {
313         struct rb_node *n = root->rb_node;
314         struct tree_entry *entry;
315
316         while (n) {
317                 entry = rb_entry(n, struct tree_entry, rb_node);
318
319                 if (bytenr < entry->bytenr)
320                         n = n->rb_left;
321                 else if (bytenr > entry->bytenr)
322                         n = n->rb_right;
323                 else
324                         return n;
325         }
326         return NULL;
327 }
328
329 /*
330  * walk up backref nodes until reach node presents tree root
331  */
332 static struct backref_node *walk_up_backref(struct backref_node *node,
333                                             struct backref_edge *edges[],
334                                             int *index)
335 {
336         struct backref_edge *edge;
337         int idx = *index;
338
339         while (!list_empty(&node->upper)) {
340                 edge = list_entry(node->upper.next,
341                                   struct backref_edge, list[LOWER]);
342                 edges[idx++] = edge;
343                 node = edge->node[UPPER];
344         }
345         BUG_ON(node->detached);
346         *index = idx;
347         return node;
348 }
349
350 /*
351  * walk down backref nodes to find start of next reference path
352  */
353 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
354                                               int *index)
355 {
356         struct backref_edge *edge;
357         struct backref_node *lower;
358         int idx = *index;
359
360         while (idx > 0) {
361                 edge = edges[idx - 1];
362                 lower = edge->node[LOWER];
363                 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
364                         idx--;
365                         continue;
366                 }
367                 edge = list_entry(edge->list[LOWER].next,
368                                   struct backref_edge, list[LOWER]);
369                 edges[idx - 1] = edge;
370                 *index = idx;
371                 return edge->node[UPPER];
372         }
373         *index = 0;
374         return NULL;
375 }
376
377 static void unlock_node_buffer(struct backref_node *node)
378 {
379         if (node->locked) {
380                 btrfs_tree_unlock(node->eb);
381                 node->locked = 0;
382         }
383 }
384
385 static void drop_node_buffer(struct backref_node *node)
386 {
387         if (node->eb) {
388                 unlock_node_buffer(node);
389                 free_extent_buffer(node->eb);
390                 node->eb = NULL;
391         }
392 }
393
394 static void drop_backref_node(struct backref_cache *tree,
395                               struct backref_node *node)
396 {
397         BUG_ON(!list_empty(&node->upper));
398
399         drop_node_buffer(node);
400         list_del(&node->list);
401         list_del(&node->lower);
402         if (!RB_EMPTY_NODE(&node->rb_node))
403                 rb_erase(&node->rb_node, &tree->rb_root);
404         free_backref_node(tree, node);
405 }
406
407 /*
408  * remove a backref node from the backref cache
409  */
410 static void remove_backref_node(struct backref_cache *cache,
411                                 struct backref_node *node)
412 {
413         struct backref_node *upper;
414         struct backref_edge *edge;
415
416         if (!node)
417                 return;
418
419         BUG_ON(!node->lowest && !node->detached);
420         while (!list_empty(&node->upper)) {
421                 edge = list_entry(node->upper.next, struct backref_edge,
422                                   list[LOWER]);
423                 upper = edge->node[UPPER];
424                 list_del(&edge->list[LOWER]);
425                 list_del(&edge->list[UPPER]);
426                 free_backref_edge(cache, edge);
427
428                 if (RB_EMPTY_NODE(&upper->rb_node)) {
429                         BUG_ON(!list_empty(&node->upper));
430                         drop_backref_node(cache, node);
431                         node = upper;
432                         node->lowest = 1;
433                         continue;
434                 }
435                 /*
436                  * add the node to leaf node list if no other
437                  * child block cached.
438                  */
439                 if (list_empty(&upper->lower)) {
440                         list_add_tail(&upper->lower, &cache->leaves);
441                         upper->lowest = 1;
442                 }
443         }
444
445         drop_backref_node(cache, node);
446 }
447
448 static void update_backref_node(struct backref_cache *cache,
449                                 struct backref_node *node, u64 bytenr)
450 {
451         struct rb_node *rb_node;
452         rb_erase(&node->rb_node, &cache->rb_root);
453         node->bytenr = bytenr;
454         rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
455         BUG_ON(rb_node);
456 }
457
458 /*
459  * update backref cache after a transaction commit
460  */
461 static int update_backref_cache(struct btrfs_trans_handle *trans,
462                                 struct backref_cache *cache)
463 {
464         struct backref_node *node;
465         int level = 0;
466
467         if (cache->last_trans == 0) {
468                 cache->last_trans = trans->transid;
469                 return 0;
470         }
471
472         if (cache->last_trans == trans->transid)
473                 return 0;
474
475         /*
476          * detached nodes are used to avoid unnecessary backref
477          * lookup. transaction commit changes the extent tree.
478          * so the detached nodes are no longer useful.
479          */
480         while (!list_empty(&cache->detached)) {
481                 node = list_entry(cache->detached.next,
482                                   struct backref_node, list);
483                 remove_backref_node(cache, node);
484         }
485
486         while (!list_empty(&cache->changed)) {
487                 node = list_entry(cache->changed.next,
488                                   struct backref_node, list);
489                 list_del_init(&node->list);
490                 BUG_ON(node->pending);
491                 update_backref_node(cache, node, node->new_bytenr);
492         }
493
494         /*
495          * some nodes can be left in the pending list if there were
496          * errors during processing the pending nodes.
497          */
498         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
499                 list_for_each_entry(node, &cache->pending[level], list) {
500                         BUG_ON(!node->pending);
501                         if (node->bytenr == node->new_bytenr)
502                                 continue;
503                         update_backref_node(cache, node, node->new_bytenr);
504                 }
505         }
506
507         cache->last_trans = 0;
508         return 1;
509 }
510
511 static int should_ignore_root(struct btrfs_root *root)
512 {
513         struct btrfs_root *reloc_root;
514
515         if (!root->ref_cows)
516                 return 0;
517
518         reloc_root = root->reloc_root;
519         if (!reloc_root)
520                 return 0;
521
522         if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
523             root->fs_info->running_transaction->transid - 1)
524                 return 0;
525         /*
526          * if there is reloc tree and it was created in previous
527          * transaction backref lookup can find the reloc tree,
528          * so backref node for the fs tree root is useless for
529          * relocation.
530          */
531         return 1;
532 }
533
534 /*
535  * find reloc tree by address of tree root
536  */
537 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
538                                           u64 bytenr)
539 {
540         struct rb_node *rb_node;
541         struct mapping_node *node;
542         struct btrfs_root *root = NULL;
543
544         spin_lock(&rc->reloc_root_tree.lock);
545         rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
546         if (rb_node) {
547                 node = rb_entry(rb_node, struct mapping_node, rb_node);
548                 root = (struct btrfs_root *)node->data;
549         }
550         spin_unlock(&rc->reloc_root_tree.lock);
551         return root;
552 }
553
554 static int is_cowonly_root(u64 root_objectid)
555 {
556         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
557             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
558             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
559             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
560             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
561             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
562                 return 1;
563         return 0;
564 }
565
566 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
567                                         u64 root_objectid)
568 {
569         struct btrfs_key key;
570
571         key.objectid = root_objectid;
572         key.type = BTRFS_ROOT_ITEM_KEY;
573         if (is_cowonly_root(root_objectid))
574                 key.offset = 0;
575         else
576                 key.offset = (u64)-1;
577
578         return btrfs_read_fs_root_no_name(fs_info, &key);
579 }
580
581 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
582 static noinline_for_stack
583 struct btrfs_root *find_tree_root(struct reloc_control *rc,
584                                   struct extent_buffer *leaf,
585                                   struct btrfs_extent_ref_v0 *ref0)
586 {
587         struct btrfs_root *root;
588         u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
589         u64 generation = btrfs_ref_generation_v0(leaf, ref0);
590
591         BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
592
593         root = read_fs_root(rc->extent_root->fs_info, root_objectid);
594         BUG_ON(IS_ERR(root));
595
596         if (root->ref_cows &&
597             generation != btrfs_root_generation(&root->root_item))
598                 return NULL;
599
600         return root;
601 }
602 #endif
603
604 static noinline_for_stack
605 int find_inline_backref(struct extent_buffer *leaf, int slot,
606                         unsigned long *ptr, unsigned long *end)
607 {
608         struct btrfs_extent_item *ei;
609         struct btrfs_tree_block_info *bi;
610         u32 item_size;
611
612         item_size = btrfs_item_size_nr(leaf, slot);
613 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
614         if (item_size < sizeof(*ei)) {
615                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
616                 return 1;
617         }
618 #endif
619         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
620         WARN_ON(!(btrfs_extent_flags(leaf, ei) &
621                   BTRFS_EXTENT_FLAG_TREE_BLOCK));
622
623         if (item_size <= sizeof(*ei) + sizeof(*bi)) {
624                 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
625                 return 1;
626         }
627
628         bi = (struct btrfs_tree_block_info *)(ei + 1);
629         *ptr = (unsigned long)(bi + 1);
630         *end = (unsigned long)ei + item_size;
631         return 0;
632 }
633
634 /*
635  * build backref tree for a given tree block. root of the backref tree
636  * corresponds the tree block, leaves of the backref tree correspond
637  * roots of b-trees that reference the tree block.
638  *
639  * the basic idea of this function is check backrefs of a given block
640  * to find upper level blocks that refernece the block, and then check
641  * bakcrefs of these upper level blocks recursively. the recursion stop
642  * when tree root is reached or backrefs for the block is cached.
643  *
644  * NOTE: if we find backrefs for a block are cached, we know backrefs
645  * for all upper level blocks that directly/indirectly reference the
646  * block are also cached.
647  */
648 static noinline_for_stack
649 struct backref_node *build_backref_tree(struct reloc_control *rc,
650                                         struct btrfs_key *node_key,
651                                         int level, u64 bytenr)
652 {
653         struct backref_cache *cache = &rc->backref_cache;
654         struct btrfs_path *path1;
655         struct btrfs_path *path2;
656         struct extent_buffer *eb;
657         struct btrfs_root *root;
658         struct backref_node *cur;
659         struct backref_node *upper;
660         struct backref_node *lower;
661         struct backref_node *node = NULL;
662         struct backref_node *exist = NULL;
663         struct backref_edge *edge;
664         struct rb_node *rb_node;
665         struct btrfs_key key;
666         unsigned long end;
667         unsigned long ptr;
668         LIST_HEAD(list);
669         LIST_HEAD(useless);
670         int cowonly;
671         int ret;
672         int err = 0;
673
674         path1 = btrfs_alloc_path();
675         path2 = btrfs_alloc_path();
676         if (!path1 || !path2) {
677                 err = -ENOMEM;
678                 goto out;
679         }
680
681         node = alloc_backref_node(cache);
682         if (!node) {
683                 err = -ENOMEM;
684                 goto out;
685         }
686
687         node->bytenr = bytenr;
688         node->level = level;
689         node->lowest = 1;
690         cur = node;
691 again:
692         end = 0;
693         ptr = 0;
694         key.objectid = cur->bytenr;
695         key.type = BTRFS_EXTENT_ITEM_KEY;
696         key.offset = (u64)-1;
697
698         path1->search_commit_root = 1;
699         path1->skip_locking = 1;
700         ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
701                                 0, 0);
702         if (ret < 0) {
703                 err = ret;
704                 goto out;
705         }
706         BUG_ON(!ret || !path1->slots[0]);
707
708         path1->slots[0]--;
709
710         WARN_ON(cur->checked);
711         if (!list_empty(&cur->upper)) {
712                 /*
713                  * the backref was added previously when processsing
714                  * backref of type BTRFS_TREE_BLOCK_REF_KEY
715                  */
716                 BUG_ON(!list_is_singular(&cur->upper));
717                 edge = list_entry(cur->upper.next, struct backref_edge,
718                                   list[LOWER]);
719                 BUG_ON(!list_empty(&edge->list[UPPER]));
720                 exist = edge->node[UPPER];
721                 /*
722                  * add the upper level block to pending list if we need
723                  * check its backrefs
724                  */
725                 if (!exist->checked)
726                         list_add_tail(&edge->list[UPPER], &list);
727         } else {
728                 exist = NULL;
729         }
730
731         while (1) {
732                 cond_resched();
733                 eb = path1->nodes[0];
734
735                 if (ptr >= end) {
736                         if (path1->slots[0] >= btrfs_header_nritems(eb)) {
737                                 ret = btrfs_next_leaf(rc->extent_root, path1);
738                                 if (ret < 0) {
739                                         err = ret;
740                                         goto out;
741                                 }
742                                 if (ret > 0)
743                                         break;
744                                 eb = path1->nodes[0];
745                         }
746
747                         btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
748                         if (key.objectid != cur->bytenr) {
749                                 WARN_ON(exist);
750                                 break;
751                         }
752
753                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
754                                 ret = find_inline_backref(eb, path1->slots[0],
755                                                           &ptr, &end);
756                                 if (ret)
757                                         goto next;
758                         }
759                 }
760
761                 if (ptr < end) {
762                         /* update key for inline back ref */
763                         struct btrfs_extent_inline_ref *iref;
764                         iref = (struct btrfs_extent_inline_ref *)ptr;
765                         key.type = btrfs_extent_inline_ref_type(eb, iref);
766                         key.offset = btrfs_extent_inline_ref_offset(eb, iref);
767                         WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
768                                 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
769                 }
770
771                 if (exist &&
772                     ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
773                       exist->owner == key.offset) ||
774                      (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
775                       exist->bytenr == key.offset))) {
776                         exist = NULL;
777                         goto next;
778                 }
779
780 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
781                 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
782                     key.type == BTRFS_EXTENT_REF_V0_KEY) {
783                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
784                                 struct btrfs_extent_ref_v0 *ref0;
785                                 ref0 = btrfs_item_ptr(eb, path1->slots[0],
786                                                 struct btrfs_extent_ref_v0);
787                                 root = find_tree_root(rc, eb, ref0);
788                                 if (!root->ref_cows)
789                                         cur->cowonly = 1;
790                                 if (key.objectid == key.offset) {
791                                         if (root && !should_ignore_root(root))
792                                                 cur->root = root;
793                                         else
794                                                 list_add(&cur->list, &useless);
795                                         break;
796                                 }
797                         }
798 #else
799                 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
800                 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
801 #endif
802                         if (key.objectid == key.offset) {
803                                 /*
804                                  * only root blocks of reloc trees use
805                                  * backref of this type.
806                                  */
807                                 root = find_reloc_root(rc, cur->bytenr);
808                                 BUG_ON(!root);
809                                 cur->root = root;
810                                 break;
811                         }
812
813                         edge = alloc_backref_edge(cache);
814                         if (!edge) {
815                                 err = -ENOMEM;
816                                 goto out;
817                         }
818                         rb_node = tree_search(&cache->rb_root, key.offset);
819                         if (!rb_node) {
820                                 upper = alloc_backref_node(cache);
821                                 if (!upper) {
822                                         free_backref_edge(cache, edge);
823                                         err = -ENOMEM;
824                                         goto out;
825                                 }
826                                 upper->bytenr = key.offset;
827                                 upper->level = cur->level + 1;
828                                 /*
829                                  *  backrefs for the upper level block isn't
830                                  *  cached, add the block to pending list
831                                  */
832                                 list_add_tail(&edge->list[UPPER], &list);
833                         } else {
834                                 upper = rb_entry(rb_node, struct backref_node,
835                                                  rb_node);
836                                 BUG_ON(!upper->checked);
837                                 INIT_LIST_HEAD(&edge->list[UPPER]);
838                         }
839                         list_add_tail(&edge->list[LOWER], &cur->upper);
840                         edge->node[LOWER] = cur;
841                         edge->node[UPPER] = upper;
842
843                         goto next;
844                 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
845                         goto next;
846                 }
847
848                 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
849                 root = read_fs_root(rc->extent_root->fs_info, key.offset);
850                 if (IS_ERR(root)) {
851                         err = PTR_ERR(root);
852                         goto out;
853                 }
854
855                 if (!root->ref_cows)
856                         cur->cowonly = 1;
857
858                 if (btrfs_root_level(&root->root_item) == cur->level) {
859                         /* tree root */
860                         BUG_ON(btrfs_root_bytenr(&root->root_item) !=
861                                cur->bytenr);
862                         if (should_ignore_root(root))
863                                 list_add(&cur->list, &useless);
864                         else
865                                 cur->root = root;
866                         break;
867                 }
868
869                 level = cur->level + 1;
870
871                 /*
872                  * searching the tree to find upper level blocks
873                  * reference the block.
874                  */
875                 path2->search_commit_root = 1;
876                 path2->skip_locking = 1;
877                 path2->lowest_level = level;
878                 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
879                 path2->lowest_level = 0;
880                 if (ret < 0) {
881                         err = ret;
882                         goto out;
883                 }
884                 if (ret > 0 && path2->slots[level] > 0)
885                         path2->slots[level]--;
886
887                 eb = path2->nodes[level];
888                 WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) !=
889                         cur->bytenr);
890
891                 lower = cur;
892                 for (; level < BTRFS_MAX_LEVEL; level++) {
893                         if (!path2->nodes[level]) {
894                                 BUG_ON(btrfs_root_bytenr(&root->root_item) !=
895                                        lower->bytenr);
896                                 if (should_ignore_root(root))
897                                         list_add(&lower->list, &useless);
898                                 else
899                                         lower->root = root;
900                                 break;
901                         }
902
903                         edge = alloc_backref_edge(cache);
904                         if (!edge) {
905                                 err = -ENOMEM;
906                                 goto out;
907                         }
908
909                         eb = path2->nodes[level];
910                         rb_node = tree_search(&cache->rb_root, eb->start);
911                         if (!rb_node) {
912                                 upper = alloc_backref_node(cache);
913                                 if (!upper) {
914                                         free_backref_edge(cache, edge);
915                                         err = -ENOMEM;
916                                         goto out;
917                                 }
918                                 upper->bytenr = eb->start;
919                                 upper->owner = btrfs_header_owner(eb);
920                                 upper->level = lower->level + 1;
921                                 if (!root->ref_cows)
922                                         upper->cowonly = 1;
923
924                                 /*
925                                  * if we know the block isn't shared
926                                  * we can void checking its backrefs.
927                                  */
928                                 if (btrfs_block_can_be_shared(root, eb))
929                                         upper->checked = 0;
930                                 else
931                                         upper->checked = 1;
932
933                                 /*
934                                  * add the block to pending list if we
935                                  * need check its backrefs. only block
936                                  * at 'cur->level + 1' is added to the
937                                  * tail of pending list. this guarantees
938                                  * we check backrefs from lower level
939                                  * blocks to upper level blocks.
940                                  */
941                                 if (!upper->checked &&
942                                     level == cur->level + 1) {
943                                         list_add_tail(&edge->list[UPPER],
944                                                       &list);
945                                 } else
946                                         INIT_LIST_HEAD(&edge->list[UPPER]);
947                         } else {
948                                 upper = rb_entry(rb_node, struct backref_node,
949                                                  rb_node);
950                                 BUG_ON(!upper->checked);
951                                 INIT_LIST_HEAD(&edge->list[UPPER]);
952                                 if (!upper->owner)
953                                         upper->owner = btrfs_header_owner(eb);
954                         }
955                         list_add_tail(&edge->list[LOWER], &lower->upper);
956                         edge->node[LOWER] = lower;
957                         edge->node[UPPER] = upper;
958
959                         if (rb_node)
960                                 break;
961                         lower = upper;
962                         upper = NULL;
963                 }
964                 btrfs_release_path(root, path2);
965 next:
966                 if (ptr < end) {
967                         ptr += btrfs_extent_inline_ref_size(key.type);
968                         if (ptr >= end) {
969                                 WARN_ON(ptr > end);
970                                 ptr = 0;
971                                 end = 0;
972                         }
973                 }
974                 if (ptr >= end)
975                         path1->slots[0]++;
976         }
977         btrfs_release_path(rc->extent_root, path1);
978
979         cur->checked = 1;
980         WARN_ON(exist);
981
982         /* the pending list isn't empty, take the first block to process */
983         if (!list_empty(&list)) {
984                 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
985                 list_del_init(&edge->list[UPPER]);
986                 cur = edge->node[UPPER];
987                 goto again;
988         }
989
990         /*
991          * everything goes well, connect backref nodes and insert backref nodes
992          * into the cache.
993          */
994         BUG_ON(!node->checked);
995         cowonly = node->cowonly;
996         if (!cowonly) {
997                 rb_node = tree_insert(&cache->rb_root, node->bytenr,
998                                       &node->rb_node);
999                 BUG_ON(rb_node);
1000                 list_add_tail(&node->lower, &cache->leaves);
1001         }
1002
1003         list_for_each_entry(edge, &node->upper, list[LOWER])
1004                 list_add_tail(&edge->list[UPPER], &list);
1005
1006         while (!list_empty(&list)) {
1007                 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1008                 list_del_init(&edge->list[UPPER]);
1009                 upper = edge->node[UPPER];
1010                 if (upper->detached) {
1011                         list_del(&edge->list[LOWER]);
1012                         lower = edge->node[LOWER];
1013                         free_backref_edge(cache, edge);
1014                         if (list_empty(&lower->upper))
1015                                 list_add(&lower->list, &useless);
1016                         continue;
1017                 }
1018
1019                 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1020                         if (upper->lowest) {
1021                                 list_del_init(&upper->lower);
1022                                 upper->lowest = 0;
1023                         }
1024
1025                         list_add_tail(&edge->list[UPPER], &upper->lower);
1026                         continue;
1027                 }
1028
1029                 BUG_ON(!upper->checked);
1030                 BUG_ON(cowonly != upper->cowonly);
1031                 if (!cowonly) {
1032                         rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1033                                               &upper->rb_node);
1034                         BUG_ON(rb_node);
1035                 }
1036
1037                 list_add_tail(&edge->list[UPPER], &upper->lower);
1038
1039                 list_for_each_entry(edge, &upper->upper, list[LOWER])
1040                         list_add_tail(&edge->list[UPPER], &list);
1041         }
1042         /*
1043          * process useless backref nodes. backref nodes for tree leaves
1044          * are deleted from the cache. backref nodes for upper level
1045          * tree blocks are left in the cache to avoid unnecessary backref
1046          * lookup.
1047          */
1048         while (!list_empty(&useless)) {
1049                 upper = list_entry(useless.next, struct backref_node, list);
1050                 list_del_init(&upper->list);
1051                 BUG_ON(!list_empty(&upper->upper));
1052                 if (upper == node)
1053                         node = NULL;
1054                 if (upper->lowest) {
1055                         list_del_init(&upper->lower);
1056                         upper->lowest = 0;
1057                 }
1058                 while (!list_empty(&upper->lower)) {
1059                         edge = list_entry(upper->lower.next,
1060                                           struct backref_edge, list[UPPER]);
1061                         list_del(&edge->list[UPPER]);
1062                         list_del(&edge->list[LOWER]);
1063                         lower = edge->node[LOWER];
1064                         free_backref_edge(cache, edge);
1065
1066                         if (list_empty(&lower->upper))
1067                                 list_add(&lower->list, &useless);
1068                 }
1069                 __mark_block_processed(rc, upper);
1070                 if (upper->level > 0) {
1071                         list_add(&upper->list, &cache->detached);
1072                         upper->detached = 1;
1073                 } else {
1074                         rb_erase(&upper->rb_node, &cache->rb_root);
1075                         free_backref_node(cache, upper);
1076                 }
1077         }
1078 out:
1079         btrfs_free_path(path1);
1080         btrfs_free_path(path2);
1081         if (err) {
1082                 while (!list_empty(&useless)) {
1083                         lower = list_entry(useless.next,
1084                                            struct backref_node, upper);
1085                         list_del_init(&lower->upper);
1086                 }
1087                 upper = node;
1088                 INIT_LIST_HEAD(&list);
1089                 while (upper) {
1090                         if (RB_EMPTY_NODE(&upper->rb_node)) {
1091                                 list_splice_tail(&upper->upper, &list);
1092                                 free_backref_node(cache, upper);
1093                         }
1094
1095                         if (list_empty(&list))
1096                                 break;
1097
1098                         edge = list_entry(list.next, struct backref_edge,
1099                                           list[LOWER]);
1100                         list_del(&edge->list[LOWER]);
1101                         upper = edge->node[UPPER];
1102                         free_backref_edge(cache, edge);
1103                 }
1104                 return ERR_PTR(err);
1105         }
1106         BUG_ON(node && node->detached);
1107         return node;
1108 }
1109
1110 /*
1111  * helper to add backref node for the newly created snapshot.
1112  * the backref node is created by cloning backref node that
1113  * corresponds to root of source tree
1114  */
1115 static int clone_backref_node(struct btrfs_trans_handle *trans,
1116                               struct reloc_control *rc,
1117                               struct btrfs_root *src,
1118                               struct btrfs_root *dest)
1119 {
1120         struct btrfs_root *reloc_root = src->reloc_root;
1121         struct backref_cache *cache = &rc->backref_cache;
1122         struct backref_node *node = NULL;
1123         struct backref_node *new_node;
1124         struct backref_edge *edge;
1125         struct backref_edge *new_edge;
1126         struct rb_node *rb_node;
1127
1128         if (cache->last_trans > 0)
1129                 update_backref_cache(trans, cache);
1130
1131         rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1132         if (rb_node) {
1133                 node = rb_entry(rb_node, struct backref_node, rb_node);
1134                 if (node->detached)
1135                         node = NULL;
1136                 else
1137                         BUG_ON(node->new_bytenr != reloc_root->node->start);
1138         }
1139
1140         if (!node) {
1141                 rb_node = tree_search(&cache->rb_root,
1142                                       reloc_root->commit_root->start);
1143                 if (rb_node) {
1144                         node = rb_entry(rb_node, struct backref_node,
1145                                         rb_node);
1146                         BUG_ON(node->detached);
1147                 }
1148         }
1149
1150         if (!node)
1151                 return 0;
1152
1153         new_node = alloc_backref_node(cache);
1154         if (!new_node)
1155                 return -ENOMEM;
1156
1157         new_node->bytenr = dest->node->start;
1158         new_node->level = node->level;
1159         new_node->lowest = node->lowest;
1160         new_node->root = dest;
1161
1162         if (!node->lowest) {
1163                 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1164                         new_edge = alloc_backref_edge(cache);
1165                         if (!new_edge)
1166                                 goto fail;
1167
1168                         new_edge->node[UPPER] = new_node;
1169                         new_edge->node[LOWER] = edge->node[LOWER];
1170                         list_add_tail(&new_edge->list[UPPER],
1171                                       &new_node->lower);
1172                 }
1173         }
1174
1175         rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1176                               &new_node->rb_node);
1177         BUG_ON(rb_node);
1178
1179         if (!new_node->lowest) {
1180                 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1181                         list_add_tail(&new_edge->list[LOWER],
1182                                       &new_edge->node[LOWER]->upper);
1183                 }
1184         }
1185         return 0;
1186 fail:
1187         while (!list_empty(&new_node->lower)) {
1188                 new_edge = list_entry(new_node->lower.next,
1189                                       struct backref_edge, list[UPPER]);
1190                 list_del(&new_edge->list[UPPER]);
1191                 free_backref_edge(cache, new_edge);
1192         }
1193         free_backref_node(cache, new_node);
1194         return -ENOMEM;
1195 }
1196
1197 /*
1198  * helper to add 'address of tree root -> reloc tree' mapping
1199  */
1200 static int __add_reloc_root(struct btrfs_root *root)
1201 {
1202         struct rb_node *rb_node;
1203         struct mapping_node *node;
1204         struct reloc_control *rc = root->fs_info->reloc_ctl;
1205
1206         node = kmalloc(sizeof(*node), GFP_NOFS);
1207         BUG_ON(!node);
1208
1209         node->bytenr = root->node->start;
1210         node->data = root;
1211
1212         spin_lock(&rc->reloc_root_tree.lock);
1213         rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1214                               node->bytenr, &node->rb_node);
1215         spin_unlock(&rc->reloc_root_tree.lock);
1216         BUG_ON(rb_node);
1217
1218         list_add_tail(&root->root_list, &rc->reloc_roots);
1219         return 0;
1220 }
1221
1222 /*
1223  * helper to update/delete the 'address of tree root -> reloc tree'
1224  * mapping
1225  */
1226 static int __update_reloc_root(struct btrfs_root *root, int del)
1227 {
1228         struct rb_node *rb_node;
1229         struct mapping_node *node = NULL;
1230         struct reloc_control *rc = root->fs_info->reloc_ctl;
1231
1232         spin_lock(&rc->reloc_root_tree.lock);
1233         rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1234                               root->commit_root->start);
1235         if (rb_node) {
1236                 node = rb_entry(rb_node, struct mapping_node, rb_node);
1237                 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1238         }
1239         spin_unlock(&rc->reloc_root_tree.lock);
1240
1241         BUG_ON((struct btrfs_root *)node->data != root);
1242
1243         if (!del) {
1244                 spin_lock(&rc->reloc_root_tree.lock);
1245                 node->bytenr = root->node->start;
1246                 rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1247                                       node->bytenr, &node->rb_node);
1248                 spin_unlock(&rc->reloc_root_tree.lock);
1249                 BUG_ON(rb_node);
1250         } else {
1251                 list_del_init(&root->root_list);
1252                 kfree(node);
1253         }
1254         return 0;
1255 }
1256
1257 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1258                                         struct btrfs_root *root, u64 objectid)
1259 {
1260         struct btrfs_root *reloc_root;
1261         struct extent_buffer *eb;
1262         struct btrfs_root_item *root_item;
1263         struct btrfs_key root_key;
1264         int ret;
1265
1266         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1267         BUG_ON(!root_item);
1268
1269         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1270         root_key.type = BTRFS_ROOT_ITEM_KEY;
1271         root_key.offset = objectid;
1272
1273         if (root->root_key.objectid == objectid) {
1274                 /* called by btrfs_init_reloc_root */
1275                 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1276                                       BTRFS_TREE_RELOC_OBJECTID);
1277                 BUG_ON(ret);
1278
1279                 btrfs_set_root_last_snapshot(&root->root_item,
1280                                              trans->transid - 1);
1281         } else {
1282                 /*
1283                  * called by btrfs_reloc_post_snapshot_hook.
1284                  * the source tree is a reloc tree, all tree blocks
1285                  * modified after it was created have RELOC flag
1286                  * set in their headers. so it's OK to not update
1287                  * the 'last_snapshot'.
1288                  */
1289                 ret = btrfs_copy_root(trans, root, root->node, &eb,
1290                                       BTRFS_TREE_RELOC_OBJECTID);
1291                 BUG_ON(ret);
1292         }
1293
1294         memcpy(root_item, &root->root_item, sizeof(*root_item));
1295         btrfs_set_root_bytenr(root_item, eb->start);
1296         btrfs_set_root_level(root_item, btrfs_header_level(eb));
1297         btrfs_set_root_generation(root_item, trans->transid);
1298
1299         if (root->root_key.objectid == objectid) {
1300                 btrfs_set_root_refs(root_item, 0);
1301                 memset(&root_item->drop_progress, 0,
1302                        sizeof(struct btrfs_disk_key));
1303                 root_item->drop_level = 0;
1304         }
1305
1306         btrfs_tree_unlock(eb);
1307         free_extent_buffer(eb);
1308
1309         ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1310                                 &root_key, root_item);
1311         BUG_ON(ret);
1312         kfree(root_item);
1313
1314         reloc_root = btrfs_read_fs_root_no_radix(root->fs_info->tree_root,
1315                                                  &root_key);
1316         BUG_ON(IS_ERR(reloc_root));
1317         reloc_root->last_trans = trans->transid;
1318         return reloc_root;
1319 }
1320
1321 /*
1322  * create reloc tree for a given fs tree. reloc tree is just a
1323  * snapshot of the fs tree with special root objectid.
1324  */
1325 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1326                           struct btrfs_root *root)
1327 {
1328         struct btrfs_root *reloc_root;
1329         struct reloc_control *rc = root->fs_info->reloc_ctl;
1330         int clear_rsv = 0;
1331
1332         if (root->reloc_root) {
1333                 reloc_root = root->reloc_root;
1334                 reloc_root->last_trans = trans->transid;
1335                 return 0;
1336         }
1337
1338         if (!rc || !rc->create_reloc_tree ||
1339             root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1340                 return 0;
1341
1342         if (!trans->block_rsv) {
1343                 trans->block_rsv = rc->block_rsv;
1344                 clear_rsv = 1;
1345         }
1346         reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1347         if (clear_rsv)
1348                 trans->block_rsv = NULL;
1349
1350         __add_reloc_root(reloc_root);
1351         root->reloc_root = reloc_root;
1352         return 0;
1353 }
1354
1355 /*
1356  * update root item of reloc tree
1357  */
1358 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1359                             struct btrfs_root *root)
1360 {
1361         struct btrfs_root *reloc_root;
1362         struct btrfs_root_item *root_item;
1363         int del = 0;
1364         int ret;
1365
1366         if (!root->reloc_root)
1367                 return 0;
1368
1369         reloc_root = root->reloc_root;
1370         root_item = &reloc_root->root_item;
1371
1372         if (root->fs_info->reloc_ctl->merge_reloc_tree &&
1373             btrfs_root_refs(root_item) == 0) {
1374                 root->reloc_root = NULL;
1375                 del = 1;
1376         }
1377
1378         __update_reloc_root(reloc_root, del);
1379
1380         if (reloc_root->commit_root != reloc_root->node) {
1381                 btrfs_set_root_node(root_item, reloc_root->node);
1382                 free_extent_buffer(reloc_root->commit_root);
1383                 reloc_root->commit_root = btrfs_root_node(reloc_root);
1384         }
1385
1386         ret = btrfs_update_root(trans, root->fs_info->tree_root,
1387                                 &reloc_root->root_key, root_item);
1388         BUG_ON(ret);
1389         return 0;
1390 }
1391
1392 /*
1393  * helper to find first cached inode with inode number >= objectid
1394  * in a subvolume
1395  */
1396 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1397 {
1398         struct rb_node *node;
1399         struct rb_node *prev;
1400         struct btrfs_inode *entry;
1401         struct inode *inode;
1402
1403         spin_lock(&root->inode_lock);
1404 again:
1405         node = root->inode_tree.rb_node;
1406         prev = NULL;
1407         while (node) {
1408                 prev = node;
1409                 entry = rb_entry(node, struct btrfs_inode, rb_node);
1410
1411                 if (objectid < entry->vfs_inode.i_ino)
1412                         node = node->rb_left;
1413                 else if (objectid > entry->vfs_inode.i_ino)
1414                         node = node->rb_right;
1415                 else
1416                         break;
1417         }
1418         if (!node) {
1419                 while (prev) {
1420                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
1421                         if (objectid <= entry->vfs_inode.i_ino) {
1422                                 node = prev;
1423                                 break;
1424                         }
1425                         prev = rb_next(prev);
1426                 }
1427         }
1428         while (node) {
1429                 entry = rb_entry(node, struct btrfs_inode, rb_node);
1430                 inode = igrab(&entry->vfs_inode);
1431                 if (inode) {
1432                         spin_unlock(&root->inode_lock);
1433                         return inode;
1434                 }
1435
1436                 objectid = entry->vfs_inode.i_ino + 1;
1437                 if (cond_resched_lock(&root->inode_lock))
1438                         goto again;
1439
1440                 node = rb_next(node);
1441         }
1442         spin_unlock(&root->inode_lock);
1443         return NULL;
1444 }
1445
1446 static int in_block_group(u64 bytenr,
1447                           struct btrfs_block_group_cache *block_group)
1448 {
1449         if (bytenr >= block_group->key.objectid &&
1450             bytenr < block_group->key.objectid + block_group->key.offset)
1451                 return 1;
1452         return 0;
1453 }
1454
1455 /*
1456  * get new location of data
1457  */
1458 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1459                             u64 bytenr, u64 num_bytes)
1460 {
1461         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1462         struct btrfs_path *path;
1463         struct btrfs_file_extent_item *fi;
1464         struct extent_buffer *leaf;
1465         int ret;
1466
1467         path = btrfs_alloc_path();
1468         if (!path)
1469                 return -ENOMEM;
1470
1471         bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1472         ret = btrfs_lookup_file_extent(NULL, root, path, reloc_inode->i_ino,
1473                                        bytenr, 0);
1474         if (ret < 0)
1475                 goto out;
1476         if (ret > 0) {
1477                 ret = -ENOENT;
1478                 goto out;
1479         }
1480
1481         leaf = path->nodes[0];
1482         fi = btrfs_item_ptr(leaf, path->slots[0],
1483                             struct btrfs_file_extent_item);
1484
1485         BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1486                btrfs_file_extent_compression(leaf, fi) ||
1487                btrfs_file_extent_encryption(leaf, fi) ||
1488                btrfs_file_extent_other_encoding(leaf, fi));
1489
1490         if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1491                 ret = 1;
1492                 goto out;
1493         }
1494
1495         *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1496         ret = 0;
1497 out:
1498         btrfs_free_path(path);
1499         return ret;
1500 }
1501
1502 /*
1503  * update file extent items in the tree leaf to point to
1504  * the new locations.
1505  */
1506 static noinline_for_stack
1507 int replace_file_extents(struct btrfs_trans_handle *trans,
1508                          struct reloc_control *rc,
1509                          struct btrfs_root *root,
1510                          struct extent_buffer *leaf)
1511 {
1512         struct btrfs_key key;
1513         struct btrfs_file_extent_item *fi;
1514         struct inode *inode = NULL;
1515         u64 parent;
1516         u64 bytenr;
1517         u64 new_bytenr = 0;
1518         u64 num_bytes;
1519         u64 end;
1520         u32 nritems;
1521         u32 i;
1522         int ret;
1523         int first = 1;
1524         int dirty = 0;
1525
1526         if (rc->stage != UPDATE_DATA_PTRS)
1527                 return 0;
1528
1529         /* reloc trees always use full backref */
1530         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1531                 parent = leaf->start;
1532         else
1533                 parent = 0;
1534
1535         nritems = btrfs_header_nritems(leaf);
1536         for (i = 0; i < nritems; i++) {
1537                 cond_resched();
1538                 btrfs_item_key_to_cpu(leaf, &key, i);
1539                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1540                         continue;
1541                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1542                 if (btrfs_file_extent_type(leaf, fi) ==
1543                     BTRFS_FILE_EXTENT_INLINE)
1544                         continue;
1545                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1546                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1547                 if (bytenr == 0)
1548                         continue;
1549                 if (!in_block_group(bytenr, rc->block_group))
1550                         continue;
1551
1552                 /*
1553                  * if we are modifying block in fs tree, wait for readpage
1554                  * to complete and drop the extent cache
1555                  */
1556                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1557                         if (first) {
1558                                 inode = find_next_inode(root, key.objectid);
1559                                 first = 0;
1560                         } else if (inode && inode->i_ino < key.objectid) {
1561                                 btrfs_add_delayed_iput(inode);
1562                                 inode = find_next_inode(root, key.objectid);
1563                         }
1564                         if (inode && inode->i_ino == key.objectid) {
1565                                 end = key.offset +
1566                                       btrfs_file_extent_num_bytes(leaf, fi);
1567                                 WARN_ON(!IS_ALIGNED(key.offset,
1568                                                     root->sectorsize));
1569                                 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1570                                 end--;
1571                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1572                                                       key.offset, end,
1573                                                       GFP_NOFS);
1574                                 if (!ret)
1575                                         continue;
1576
1577                                 btrfs_drop_extent_cache(inode, key.offset, end,
1578                                                         1);
1579                                 unlock_extent(&BTRFS_I(inode)->io_tree,
1580                                               key.offset, end, GFP_NOFS);
1581                         }
1582                 }
1583
1584                 ret = get_new_location(rc->data_inode, &new_bytenr,
1585                                        bytenr, num_bytes);
1586                 if (ret > 0) {
1587                         WARN_ON(1);
1588                         continue;
1589                 }
1590                 BUG_ON(ret < 0);
1591
1592                 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1593                 dirty = 1;
1594
1595                 key.offset -= btrfs_file_extent_offset(leaf, fi);
1596                 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1597                                            num_bytes, parent,
1598                                            btrfs_header_owner(leaf),
1599                                            key.objectid, key.offset);
1600                 BUG_ON(ret);
1601
1602                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1603                                         parent, btrfs_header_owner(leaf),
1604                                         key.objectid, key.offset);
1605                 BUG_ON(ret);
1606         }
1607         if (dirty)
1608                 btrfs_mark_buffer_dirty(leaf);
1609         if (inode)
1610                 btrfs_add_delayed_iput(inode);
1611         return 0;
1612 }
1613
1614 static noinline_for_stack
1615 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1616                      struct btrfs_path *path, int level)
1617 {
1618         struct btrfs_disk_key key1;
1619         struct btrfs_disk_key key2;
1620         btrfs_node_key(eb, &key1, slot);
1621         btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1622         return memcmp(&key1, &key2, sizeof(key1));
1623 }
1624
1625 /*
1626  * try to replace tree blocks in fs tree with the new blocks
1627  * in reloc tree. tree blocks haven't been modified since the
1628  * reloc tree was create can be replaced.
1629  *
1630  * if a block was replaced, level of the block + 1 is returned.
1631  * if no block got replaced, 0 is returned. if there are other
1632  * errors, a negative error number is returned.
1633  */
1634 static noinline_for_stack
1635 int replace_path(struct btrfs_trans_handle *trans,
1636                  struct btrfs_root *dest, struct btrfs_root *src,
1637                  struct btrfs_path *path, struct btrfs_key *next_key,
1638                  int lowest_level, int max_level)
1639 {
1640         struct extent_buffer *eb;
1641         struct extent_buffer *parent;
1642         struct btrfs_key key;
1643         u64 old_bytenr;
1644         u64 new_bytenr;
1645         u64 old_ptr_gen;
1646         u64 new_ptr_gen;
1647         u64 last_snapshot;
1648         u32 blocksize;
1649         int cow = 0;
1650         int level;
1651         int ret;
1652         int slot;
1653
1654         BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1655         BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1656
1657         last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1658 again:
1659         slot = path->slots[lowest_level];
1660         btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1661
1662         eb = btrfs_lock_root_node(dest);
1663         btrfs_set_lock_blocking(eb);
1664         level = btrfs_header_level(eb);
1665
1666         if (level < lowest_level) {
1667                 btrfs_tree_unlock(eb);
1668                 free_extent_buffer(eb);
1669                 return 0;
1670         }
1671
1672         if (cow) {
1673                 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1674                 BUG_ON(ret);
1675         }
1676         btrfs_set_lock_blocking(eb);
1677
1678         if (next_key) {
1679                 next_key->objectid = (u64)-1;
1680                 next_key->type = (u8)-1;
1681                 next_key->offset = (u64)-1;
1682         }
1683
1684         parent = eb;
1685         while (1) {
1686                 level = btrfs_header_level(parent);
1687                 BUG_ON(level < lowest_level);
1688
1689                 ret = btrfs_bin_search(parent, &key, level, &slot);
1690                 if (ret && slot > 0)
1691                         slot--;
1692
1693                 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1694                         btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1695
1696                 old_bytenr = btrfs_node_blockptr(parent, slot);
1697                 blocksize = btrfs_level_size(dest, level - 1);
1698                 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1699
1700                 if (level <= max_level) {
1701                         eb = path->nodes[level];
1702                         new_bytenr = btrfs_node_blockptr(eb,
1703                                                         path->slots[level]);
1704                         new_ptr_gen = btrfs_node_ptr_generation(eb,
1705                                                         path->slots[level]);
1706                 } else {
1707                         new_bytenr = 0;
1708                         new_ptr_gen = 0;
1709                 }
1710
1711                 if (new_bytenr > 0 && new_bytenr == old_bytenr) {
1712                         WARN_ON(1);
1713                         ret = level;
1714                         break;
1715                 }
1716
1717                 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1718                     memcmp_node_keys(parent, slot, path, level)) {
1719                         if (level <= lowest_level) {
1720                                 ret = 0;
1721                                 break;
1722                         }
1723
1724                         eb = read_tree_block(dest, old_bytenr, blocksize,
1725                                              old_ptr_gen);
1726                         btrfs_tree_lock(eb);
1727                         if (cow) {
1728                                 ret = btrfs_cow_block(trans, dest, eb, parent,
1729                                                       slot, &eb);
1730                                 BUG_ON(ret);
1731                         }
1732                         btrfs_set_lock_blocking(eb);
1733
1734                         btrfs_tree_unlock(parent);
1735                         free_extent_buffer(parent);
1736
1737                         parent = eb;
1738                         continue;
1739                 }
1740
1741                 if (!cow) {
1742                         btrfs_tree_unlock(parent);
1743                         free_extent_buffer(parent);
1744                         cow = 1;
1745                         goto again;
1746                 }
1747
1748                 btrfs_node_key_to_cpu(path->nodes[level], &key,
1749                                       path->slots[level]);
1750                 btrfs_release_path(src, path);
1751
1752                 path->lowest_level = level;
1753                 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1754                 path->lowest_level = 0;
1755                 BUG_ON(ret);
1756
1757                 /*
1758                  * swap blocks in fs tree and reloc tree.
1759                  */
1760                 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1761                 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1762                 btrfs_mark_buffer_dirty(parent);
1763
1764                 btrfs_set_node_blockptr(path->nodes[level],
1765                                         path->slots[level], old_bytenr);
1766                 btrfs_set_node_ptr_generation(path->nodes[level],
1767                                               path->slots[level], old_ptr_gen);
1768                 btrfs_mark_buffer_dirty(path->nodes[level]);
1769
1770                 ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
1771                                         path->nodes[level]->start,
1772                                         src->root_key.objectid, level - 1, 0);
1773                 BUG_ON(ret);
1774                 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
1775                                         0, dest->root_key.objectid, level - 1,
1776                                         0);
1777                 BUG_ON(ret);
1778
1779                 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1780                                         path->nodes[level]->start,
1781                                         src->root_key.objectid, level - 1, 0);
1782                 BUG_ON(ret);
1783
1784                 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1785                                         0, dest->root_key.objectid, level - 1,
1786                                         0);
1787                 BUG_ON(ret);
1788
1789                 btrfs_unlock_up_safe(path, 0);
1790
1791                 ret = level;
1792                 break;
1793         }
1794         btrfs_tree_unlock(parent);
1795         free_extent_buffer(parent);
1796         return ret;
1797 }
1798
1799 /*
1800  * helper to find next relocated block in reloc tree
1801  */
1802 static noinline_for_stack
1803 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1804                        int *level)
1805 {
1806         struct extent_buffer *eb;
1807         int i;
1808         u64 last_snapshot;
1809         u32 nritems;
1810
1811         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1812
1813         for (i = 0; i < *level; i++) {
1814                 free_extent_buffer(path->nodes[i]);
1815                 path->nodes[i] = NULL;
1816         }
1817
1818         for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1819                 eb = path->nodes[i];
1820                 nritems = btrfs_header_nritems(eb);
1821                 while (path->slots[i] + 1 < nritems) {
1822                         path->slots[i]++;
1823                         if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1824                             last_snapshot)
1825                                 continue;
1826
1827                         *level = i;
1828                         return 0;
1829                 }
1830                 free_extent_buffer(path->nodes[i]);
1831                 path->nodes[i] = NULL;
1832         }
1833         return 1;
1834 }
1835
1836 /*
1837  * walk down reloc tree to find relocated block of lowest level
1838  */
1839 static noinline_for_stack
1840 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1841                          int *level)
1842 {
1843         struct extent_buffer *eb = NULL;
1844         int i;
1845         u64 bytenr;
1846         u64 ptr_gen = 0;
1847         u64 last_snapshot;
1848         u32 blocksize;
1849         u32 nritems;
1850
1851         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1852
1853         for (i = *level; i > 0; i--) {
1854                 eb = path->nodes[i];
1855                 nritems = btrfs_header_nritems(eb);
1856                 while (path->slots[i] < nritems) {
1857                         ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1858                         if (ptr_gen > last_snapshot)
1859                                 break;
1860                         path->slots[i]++;
1861                 }
1862                 if (path->slots[i] >= nritems) {
1863                         if (i == *level)
1864                                 break;
1865                         *level = i + 1;
1866                         return 0;
1867                 }
1868                 if (i == 1) {
1869                         *level = i;
1870                         return 0;
1871                 }
1872
1873                 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
1874                 blocksize = btrfs_level_size(root, i - 1);
1875                 eb = read_tree_block(root, bytenr, blocksize, ptr_gen);
1876                 BUG_ON(btrfs_header_level(eb) != i - 1);
1877                 path->nodes[i - 1] = eb;
1878                 path->slots[i - 1] = 0;
1879         }
1880         return 1;
1881 }
1882
1883 /*
1884  * invalidate extent cache for file extents whose key in range of
1885  * [min_key, max_key)
1886  */
1887 static int invalidate_extent_cache(struct btrfs_root *root,
1888                                    struct btrfs_key *min_key,
1889                                    struct btrfs_key *max_key)
1890 {
1891         struct inode *inode = NULL;
1892         u64 objectid;
1893         u64 start, end;
1894
1895         objectid = min_key->objectid;
1896         while (1) {
1897                 cond_resched();
1898                 iput(inode);
1899
1900                 if (objectid > max_key->objectid)
1901                         break;
1902
1903                 inode = find_next_inode(root, objectid);
1904                 if (!inode)
1905                         break;
1906
1907                 if (inode->i_ino > max_key->objectid) {
1908                         iput(inode);
1909                         break;
1910                 }
1911
1912                 objectid = inode->i_ino + 1;
1913                 if (!S_ISREG(inode->i_mode))
1914                         continue;
1915
1916                 if (unlikely(min_key->objectid == inode->i_ino)) {
1917                         if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1918                                 continue;
1919                         if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1920                                 start = 0;
1921                         else {
1922                                 start = min_key->offset;
1923                                 WARN_ON(!IS_ALIGNED(start, root->sectorsize));
1924                         }
1925                 } else {
1926                         start = 0;
1927                 }
1928
1929                 if (unlikely(max_key->objectid == inode->i_ino)) {
1930                         if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1931                                 continue;
1932                         if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1933                                 end = (u64)-1;
1934                         } else {
1935                                 if (max_key->offset == 0)
1936                                         continue;
1937                                 end = max_key->offset;
1938                                 WARN_ON(!IS_ALIGNED(end, root->sectorsize));
1939                                 end--;
1940                         }
1941                 } else {
1942                         end = (u64)-1;
1943                 }
1944
1945                 /* the lock_extent waits for readpage to complete */
1946                 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
1947                 btrfs_drop_extent_cache(inode, start, end, 1);
1948                 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
1949         }
1950         return 0;
1951 }
1952
1953 static int find_next_key(struct btrfs_path *path, int level,
1954                          struct btrfs_key *key)
1955
1956 {
1957         while (level < BTRFS_MAX_LEVEL) {
1958                 if (!path->nodes[level])
1959                         break;
1960                 if (path->slots[level] + 1 <
1961                     btrfs_header_nritems(path->nodes[level])) {
1962                         btrfs_node_key_to_cpu(path->nodes[level], key,
1963                                               path->slots[level] + 1);
1964                         return 0;
1965                 }
1966                 level++;
1967         }
1968         return 1;
1969 }
1970
1971 /*
1972  * merge the relocated tree blocks in reloc tree with corresponding
1973  * fs tree.
1974  */
1975 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1976                                                struct btrfs_root *root)
1977 {
1978         LIST_HEAD(inode_list);
1979         struct btrfs_key key;
1980         struct btrfs_key next_key;
1981         struct btrfs_trans_handle *trans;
1982         struct btrfs_root *reloc_root;
1983         struct btrfs_root_item *root_item;
1984         struct btrfs_path *path;
1985         struct extent_buffer *leaf;
1986         unsigned long nr;
1987         int level;
1988         int max_level;
1989         int replaced = 0;
1990         int ret;
1991         int err = 0;
1992         u32 min_reserved;
1993
1994         path = btrfs_alloc_path();
1995         if (!path)
1996                 return -ENOMEM;
1997
1998         reloc_root = root->reloc_root;
1999         root_item = &reloc_root->root_item;
2000
2001         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2002                 level = btrfs_root_level(root_item);
2003                 extent_buffer_get(reloc_root->node);
2004                 path->nodes[level] = reloc_root->node;
2005                 path->slots[level] = 0;
2006         } else {
2007                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2008
2009                 level = root_item->drop_level;
2010                 BUG_ON(level == 0);
2011                 path->lowest_level = level;
2012                 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2013                 path->lowest_level = 0;
2014                 if (ret < 0) {
2015                         btrfs_free_path(path);
2016                         return ret;
2017                 }
2018
2019                 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2020                                       path->slots[level]);
2021                 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2022
2023                 btrfs_unlock_up_safe(path, 0);
2024         }
2025
2026         min_reserved = root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2027         memset(&next_key, 0, sizeof(next_key));
2028
2029         while (1) {
2030                 trans = btrfs_start_transaction(root, 0);
2031                 trans->block_rsv = rc->block_rsv;
2032
2033                 ret = btrfs_block_rsv_check(trans, root, rc->block_rsv,
2034                                             min_reserved, 0);
2035                 if (ret) {
2036                         BUG_ON(ret != -EAGAIN);
2037                         ret = btrfs_commit_transaction(trans, root);
2038                         BUG_ON(ret);
2039                         continue;
2040                 }
2041
2042                 replaced = 0;
2043                 max_level = level;
2044
2045                 ret = walk_down_reloc_tree(reloc_root, path, &level);
2046                 if (ret < 0) {
2047                         err = ret;
2048                         goto out;
2049                 }
2050                 if (ret > 0)
2051                         break;
2052
2053                 if (!find_next_key(path, level, &key) &&
2054                     btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2055                         ret = 0;
2056                 } else {
2057                         ret = replace_path(trans, root, reloc_root, path,
2058                                            &next_key, level, max_level);
2059                 }
2060                 if (ret < 0) {
2061                         err = ret;
2062                         goto out;
2063                 }
2064
2065                 if (ret > 0) {
2066                         level = ret;
2067                         btrfs_node_key_to_cpu(path->nodes[level], &key,
2068                                               path->slots[level]);
2069                         replaced = 1;
2070                 }
2071
2072                 ret = walk_up_reloc_tree(reloc_root, path, &level);
2073                 if (ret > 0)
2074                         break;
2075
2076                 BUG_ON(level == 0);
2077                 /*
2078                  * save the merging progress in the drop_progress.
2079                  * this is OK since root refs == 1 in this case.
2080                  */
2081                 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2082                                path->slots[level]);
2083                 root_item->drop_level = level;
2084
2085                 nr = trans->blocks_used;
2086                 btrfs_end_transaction_throttle(trans, root);
2087
2088                 btrfs_btree_balance_dirty(root, nr);
2089
2090                 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2091                         invalidate_extent_cache(root, &key, &next_key);
2092         }
2093
2094         /*
2095          * handle the case only one block in the fs tree need to be
2096          * relocated and the block is tree root.
2097          */
2098         leaf = btrfs_lock_root_node(root);
2099         ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2100         btrfs_tree_unlock(leaf);
2101         free_extent_buffer(leaf);
2102         if (ret < 0)
2103                 err = ret;
2104 out:
2105         btrfs_free_path(path);
2106
2107         if (err == 0) {
2108                 memset(&root_item->drop_progress, 0,
2109                        sizeof(root_item->drop_progress));
2110                 root_item->drop_level = 0;
2111                 btrfs_set_root_refs(root_item, 0);
2112                 btrfs_update_reloc_root(trans, root);
2113         }
2114
2115         nr = trans->blocks_used;
2116         btrfs_end_transaction_throttle(trans, root);
2117
2118         btrfs_btree_balance_dirty(root, nr);
2119
2120         if (replaced && rc->stage == UPDATE_DATA_PTRS)
2121                 invalidate_extent_cache(root, &key, &next_key);
2122
2123         return err;
2124 }
2125
2126 static noinline_for_stack
2127 int prepare_to_merge(struct reloc_control *rc, int err)
2128 {
2129         struct btrfs_root *root = rc->extent_root;
2130         struct btrfs_root *reloc_root;
2131         struct btrfs_trans_handle *trans;
2132         LIST_HEAD(reloc_roots);
2133         u64 num_bytes = 0;
2134         int ret;
2135         int retries = 0;
2136
2137         mutex_lock(&root->fs_info->trans_mutex);
2138         rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2139         rc->merging_rsv_size += rc->nodes_relocated * 2;
2140         mutex_unlock(&root->fs_info->trans_mutex);
2141 again:
2142         if (!err) {
2143                 num_bytes = rc->merging_rsv_size;
2144                 ret = btrfs_block_rsv_add(NULL, root, rc->block_rsv,
2145                                           num_bytes, &retries);
2146                 if (ret)
2147                         err = ret;
2148         }
2149
2150         trans = btrfs_join_transaction(rc->extent_root, 1);
2151
2152         if (!err) {
2153                 if (num_bytes != rc->merging_rsv_size) {
2154                         btrfs_end_transaction(trans, rc->extent_root);
2155                         btrfs_block_rsv_release(rc->extent_root,
2156                                                 rc->block_rsv, num_bytes);
2157                         retries = 0;
2158                         goto again;
2159                 }
2160         }
2161
2162         rc->merge_reloc_tree = 1;
2163
2164         while (!list_empty(&rc->reloc_roots)) {
2165                 reloc_root = list_entry(rc->reloc_roots.next,
2166                                         struct btrfs_root, root_list);
2167                 list_del_init(&reloc_root->root_list);
2168
2169                 root = read_fs_root(reloc_root->fs_info,
2170                                     reloc_root->root_key.offset);
2171                 BUG_ON(IS_ERR(root));
2172                 BUG_ON(root->reloc_root != reloc_root);
2173
2174                 /*
2175                  * set reference count to 1, so btrfs_recover_relocation
2176                  * knows it should resumes merging
2177                  */
2178                 if (!err)
2179                         btrfs_set_root_refs(&reloc_root->root_item, 1);
2180                 btrfs_update_reloc_root(trans, root);
2181
2182                 list_add(&reloc_root->root_list, &reloc_roots);
2183         }
2184
2185         list_splice(&reloc_roots, &rc->reloc_roots);
2186
2187         if (!err)
2188                 btrfs_commit_transaction(trans, rc->extent_root);
2189         else
2190                 btrfs_end_transaction(trans, rc->extent_root);
2191         return err;
2192 }
2193
2194 static noinline_for_stack
2195 int merge_reloc_roots(struct reloc_control *rc)
2196 {
2197         struct btrfs_root *root;
2198         struct btrfs_root *reloc_root;
2199         LIST_HEAD(reloc_roots);
2200         int found = 0;
2201         int ret;
2202 again:
2203         root = rc->extent_root;
2204         mutex_lock(&root->fs_info->trans_mutex);
2205         list_splice_init(&rc->reloc_roots, &reloc_roots);
2206         mutex_unlock(&root->fs_info->trans_mutex);
2207
2208         while (!list_empty(&reloc_roots)) {
2209                 found = 1;
2210                 reloc_root = list_entry(reloc_roots.next,
2211                                         struct btrfs_root, root_list);
2212
2213                 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2214                         root = read_fs_root(reloc_root->fs_info,
2215                                             reloc_root->root_key.offset);
2216                         BUG_ON(IS_ERR(root));
2217                         BUG_ON(root->reloc_root != reloc_root);
2218
2219                         ret = merge_reloc_root(rc, root);
2220                         BUG_ON(ret);
2221                 } else {
2222                         list_del_init(&reloc_root->root_list);
2223                 }
2224                 btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0);
2225         }
2226
2227         if (found) {
2228                 found = 0;
2229                 goto again;
2230         }
2231         BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2232         return 0;
2233 }
2234
2235 static void free_block_list(struct rb_root *blocks)
2236 {
2237         struct tree_block *block;
2238         struct rb_node *rb_node;
2239         while ((rb_node = rb_first(blocks))) {
2240                 block = rb_entry(rb_node, struct tree_block, rb_node);
2241                 rb_erase(rb_node, blocks);
2242                 kfree(block);
2243         }
2244 }
2245
2246 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2247                                       struct btrfs_root *reloc_root)
2248 {
2249         struct btrfs_root *root;
2250
2251         if (reloc_root->last_trans == trans->transid)
2252                 return 0;
2253
2254         root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset);
2255         BUG_ON(IS_ERR(root));
2256         BUG_ON(root->reloc_root != reloc_root);
2257
2258         return btrfs_record_root_in_trans(trans, root);
2259 }
2260
2261 static noinline_for_stack
2262 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2263                                      struct reloc_control *rc,
2264                                      struct backref_node *node,
2265                                      struct backref_edge *edges[], int *nr)
2266 {
2267         struct backref_node *next;
2268         struct btrfs_root *root;
2269         int index = 0;
2270
2271         next = node;
2272         while (1) {
2273                 cond_resched();
2274                 next = walk_up_backref(next, edges, &index);
2275                 root = next->root;
2276                 BUG_ON(!root);
2277                 BUG_ON(!root->ref_cows);
2278
2279                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2280                         record_reloc_root_in_trans(trans, root);
2281                         break;
2282                 }
2283
2284                 btrfs_record_root_in_trans(trans, root);
2285                 root = root->reloc_root;
2286
2287                 if (next->new_bytenr != root->node->start) {
2288                         BUG_ON(next->new_bytenr);
2289                         BUG_ON(!list_empty(&next->list));
2290                         next->new_bytenr = root->node->start;
2291                         next->root = root;
2292                         list_add_tail(&next->list,
2293                                       &rc->backref_cache.changed);
2294                         __mark_block_processed(rc, next);
2295                         break;
2296                 }
2297
2298                 WARN_ON(1);
2299                 root = NULL;
2300                 next = walk_down_backref(edges, &index);
2301                 if (!next || next->level <= node->level)
2302                         break;
2303         }
2304         if (!root)
2305                 return NULL;
2306
2307         *nr = index;
2308         next = node;
2309         /* setup backref node path for btrfs_reloc_cow_block */
2310         while (1) {
2311                 rc->backref_cache.path[next->level] = next;
2312                 if (--index < 0)
2313                         break;
2314                 next = edges[index]->node[UPPER];
2315         }
2316         return root;
2317 }
2318
2319 /*
2320  * select a tree root for relocation. return NULL if the block
2321  * is reference counted. we should use do_relocation() in this
2322  * case. return a tree root pointer if the block isn't reference
2323  * counted. return -ENOENT if the block is root of reloc tree.
2324  */
2325 static noinline_for_stack
2326 struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans,
2327                                    struct backref_node *node)
2328 {
2329         struct backref_node *next;
2330         struct btrfs_root *root;
2331         struct btrfs_root *fs_root = NULL;
2332         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2333         int index = 0;
2334
2335         next = node;
2336         while (1) {
2337                 cond_resched();
2338                 next = walk_up_backref(next, edges, &index);
2339                 root = next->root;
2340                 BUG_ON(!root);
2341
2342                 /* no other choice for non-refernce counted tree */
2343                 if (!root->ref_cows)
2344                         return root;
2345
2346                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2347                         fs_root = root;
2348
2349                 if (next != node)
2350                         return NULL;
2351
2352                 next = walk_down_backref(edges, &index);
2353                 if (!next || next->level <= node->level)
2354                         break;
2355         }
2356
2357         if (!fs_root)
2358                 return ERR_PTR(-ENOENT);
2359         return fs_root;
2360 }
2361
2362 static noinline_for_stack
2363 u64 calcu_metadata_size(struct reloc_control *rc,
2364                         struct backref_node *node, int reserve)
2365 {
2366         struct backref_node *next = node;
2367         struct backref_edge *edge;
2368         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2369         u64 num_bytes = 0;
2370         int index = 0;
2371
2372         BUG_ON(reserve && node->processed);
2373
2374         while (next) {
2375                 cond_resched();
2376                 while (1) {
2377                         if (next->processed && (reserve || next != node))
2378                                 break;
2379
2380                         num_bytes += btrfs_level_size(rc->extent_root,
2381                                                       next->level);
2382
2383                         if (list_empty(&next->upper))
2384                                 break;
2385
2386                         edge = list_entry(next->upper.next,
2387                                           struct backref_edge, list[LOWER]);
2388                         edges[index++] = edge;
2389                         next = edge->node[UPPER];
2390                 }
2391                 next = walk_down_backref(edges, &index);
2392         }
2393         return num_bytes;
2394 }
2395
2396 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2397                                   struct reloc_control *rc,
2398                                   struct backref_node *node)
2399 {
2400         struct btrfs_root *root = rc->extent_root;
2401         u64 num_bytes;
2402         int ret;
2403
2404         num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2405
2406         trans->block_rsv = rc->block_rsv;
2407         ret = btrfs_block_rsv_add(trans, root, rc->block_rsv, num_bytes,
2408                                   &rc->block_rsv_retries);
2409         if (ret) {
2410                 if (ret == -EAGAIN)
2411                         rc->commit_transaction = 1;
2412                 return ret;
2413         }
2414
2415         rc->block_rsv_retries = 0;
2416         return 0;
2417 }
2418
2419 static void release_metadata_space(struct reloc_control *rc,
2420                                    struct backref_node *node)
2421 {
2422         u64 num_bytes = calcu_metadata_size(rc, node, 0) * 2;
2423         btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, num_bytes);
2424 }
2425
2426 /*
2427  * relocate a block tree, and then update pointers in upper level
2428  * blocks that reference the block to point to the new location.
2429  *
2430  * if called by link_to_upper, the block has already been relocated.
2431  * in that case this function just updates pointers.
2432  */
2433 static int do_relocation(struct btrfs_trans_handle *trans,
2434                          struct reloc_control *rc,
2435                          struct backref_node *node,
2436                          struct btrfs_key *key,
2437                          struct btrfs_path *path, int lowest)
2438 {
2439         struct backref_node *upper;
2440         struct backref_edge *edge;
2441         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2442         struct btrfs_root *root;
2443         struct extent_buffer *eb;
2444         u32 blocksize;
2445         u64 bytenr;
2446         u64 generation;
2447         int nr;
2448         int slot;
2449         int ret;
2450         int err = 0;
2451
2452         BUG_ON(lowest && node->eb);
2453
2454         path->lowest_level = node->level + 1;
2455         rc->backref_cache.path[node->level] = node;
2456         list_for_each_entry(edge, &node->upper, list[LOWER]) {
2457                 cond_resched();
2458
2459                 upper = edge->node[UPPER];
2460                 root = select_reloc_root(trans, rc, upper, edges, &nr);
2461                 BUG_ON(!root);
2462
2463                 if (upper->eb && !upper->locked) {
2464                         if (!lowest) {
2465                                 ret = btrfs_bin_search(upper->eb, key,
2466                                                        upper->level, &slot);
2467                                 BUG_ON(ret);
2468                                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2469                                 if (node->eb->start == bytenr)
2470                                         goto next;
2471                         }
2472                         drop_node_buffer(upper);
2473                 }
2474
2475                 if (!upper->eb) {
2476                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2477                         if (ret < 0) {
2478                                 err = ret;
2479                                 break;
2480                         }
2481                         BUG_ON(ret > 0);
2482
2483                         if (!upper->eb) {
2484                                 upper->eb = path->nodes[upper->level];
2485                                 path->nodes[upper->level] = NULL;
2486                         } else {
2487                                 BUG_ON(upper->eb != path->nodes[upper->level]);
2488                         }
2489
2490                         upper->locked = 1;
2491                         path->locks[upper->level] = 0;
2492
2493                         slot = path->slots[upper->level];
2494                         btrfs_release_path(NULL, path);
2495                 } else {
2496                         ret = btrfs_bin_search(upper->eb, key, upper->level,
2497                                                &slot);
2498                         BUG_ON(ret);
2499                 }
2500
2501                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2502                 if (lowest) {
2503                         BUG_ON(bytenr != node->bytenr);
2504                 } else {
2505                         if (node->eb->start == bytenr)
2506                                 goto next;
2507                 }
2508
2509                 blocksize = btrfs_level_size(root, node->level);
2510                 generation = btrfs_node_ptr_generation(upper->eb, slot);
2511                 eb = read_tree_block(root, bytenr, blocksize, generation);
2512                 btrfs_tree_lock(eb);
2513                 btrfs_set_lock_blocking(eb);
2514
2515                 if (!node->eb) {
2516                         ret = btrfs_cow_block(trans, root, eb, upper->eb,
2517                                               slot, &eb);
2518                         btrfs_tree_unlock(eb);
2519                         free_extent_buffer(eb);
2520                         if (ret < 0) {
2521                                 err = ret;
2522                                 goto next;
2523                         }
2524                         BUG_ON(node->eb != eb);
2525                 } else {
2526                         btrfs_set_node_blockptr(upper->eb, slot,
2527                                                 node->eb->start);
2528                         btrfs_set_node_ptr_generation(upper->eb, slot,
2529                                                       trans->transid);
2530                         btrfs_mark_buffer_dirty(upper->eb);
2531
2532                         ret = btrfs_inc_extent_ref(trans, root,
2533                                                 node->eb->start, blocksize,
2534                                                 upper->eb->start,
2535                                                 btrfs_header_owner(upper->eb),
2536                                                 node->level, 0);
2537                         BUG_ON(ret);
2538
2539                         ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2540                         BUG_ON(ret);
2541                 }
2542 next:
2543                 if (!upper->pending)
2544                         drop_node_buffer(upper);
2545                 else
2546                         unlock_node_buffer(upper);
2547                 if (err)
2548                         break;
2549         }
2550
2551         if (!err && node->pending) {
2552                 drop_node_buffer(node);
2553                 list_move_tail(&node->list, &rc->backref_cache.changed);
2554                 node->pending = 0;
2555         }
2556
2557         path->lowest_level = 0;
2558         BUG_ON(err == -ENOSPC);
2559         return err;
2560 }
2561
2562 static int link_to_upper(struct btrfs_trans_handle *trans,
2563                          struct reloc_control *rc,
2564                          struct backref_node *node,
2565                          struct btrfs_path *path)
2566 {
2567         struct btrfs_key key;
2568
2569         btrfs_node_key_to_cpu(node->eb, &key, 0);
2570         return do_relocation(trans, rc, node, &key, path, 0);
2571 }
2572
2573 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2574                                 struct reloc_control *rc,
2575                                 struct btrfs_path *path, int err)
2576 {
2577         LIST_HEAD(list);
2578         struct backref_cache *cache = &rc->backref_cache;
2579         struct backref_node *node;
2580         int level;
2581         int ret;
2582
2583         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2584                 while (!list_empty(&cache->pending[level])) {
2585                         node = list_entry(cache->pending[level].next,
2586                                           struct backref_node, list);
2587                         list_move_tail(&node->list, &list);
2588                         BUG_ON(!node->pending);
2589
2590                         if (!err) {
2591                                 ret = link_to_upper(trans, rc, node, path);
2592                                 if (ret < 0)
2593                                         err = ret;
2594                         }
2595                 }
2596                 list_splice_init(&list, &cache->pending[level]);
2597         }
2598         return err;
2599 }
2600
2601 static void mark_block_processed(struct reloc_control *rc,
2602                                  u64 bytenr, u32 blocksize)
2603 {
2604         set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2605                         EXTENT_DIRTY, GFP_NOFS);
2606 }
2607
2608 static void __mark_block_processed(struct reloc_control *rc,
2609                                    struct backref_node *node)
2610 {
2611         u32 blocksize;
2612         if (node->level == 0 ||
2613             in_block_group(node->bytenr, rc->block_group)) {
2614                 blocksize = btrfs_level_size(rc->extent_root, node->level);
2615                 mark_block_processed(rc, node->bytenr, blocksize);
2616         }
2617         node->processed = 1;
2618 }
2619
2620 /*
2621  * mark a block and all blocks directly/indirectly reference the block
2622  * as processed.
2623  */
2624 static void update_processed_blocks(struct reloc_control *rc,
2625                                     struct backref_node *node)
2626 {
2627         struct backref_node *next = node;
2628         struct backref_edge *edge;
2629         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2630         int index = 0;
2631
2632         while (next) {
2633                 cond_resched();
2634                 while (1) {
2635                         if (next->processed)
2636                                 break;
2637
2638                         __mark_block_processed(rc, next);
2639
2640                         if (list_empty(&next->upper))
2641                                 break;
2642
2643                         edge = list_entry(next->upper.next,
2644                                           struct backref_edge, list[LOWER]);
2645                         edges[index++] = edge;
2646                         next = edge->node[UPPER];
2647                 }
2648                 next = walk_down_backref(edges, &index);
2649         }
2650 }
2651
2652 static int tree_block_processed(u64 bytenr, u32 blocksize,
2653                                 struct reloc_control *rc)
2654 {
2655         if (test_range_bit(&rc->processed_blocks, bytenr,
2656                            bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2657                 return 1;
2658         return 0;
2659 }
2660
2661 static int get_tree_block_key(struct reloc_control *rc,
2662                               struct tree_block *block)
2663 {
2664         struct extent_buffer *eb;
2665
2666         BUG_ON(block->key_ready);
2667         eb = read_tree_block(rc->extent_root, block->bytenr,
2668                              block->key.objectid, block->key.offset);
2669         WARN_ON(btrfs_header_level(eb) != block->level);
2670         if (block->level == 0)
2671                 btrfs_item_key_to_cpu(eb, &block->key, 0);
2672         else
2673                 btrfs_node_key_to_cpu(eb, &block->key, 0);
2674         free_extent_buffer(eb);
2675         block->key_ready = 1;
2676         return 0;
2677 }
2678
2679 static int reada_tree_block(struct reloc_control *rc,
2680                             struct tree_block *block)
2681 {
2682         BUG_ON(block->key_ready);
2683         readahead_tree_block(rc->extent_root, block->bytenr,
2684                              block->key.objectid, block->key.offset);
2685         return 0;
2686 }
2687
2688 /*
2689  * helper function to relocate a tree block
2690  */
2691 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2692                                 struct reloc_control *rc,
2693                                 struct backref_node *node,
2694                                 struct btrfs_key *key,
2695                                 struct btrfs_path *path)
2696 {
2697         struct btrfs_root *root;
2698         int release = 0;
2699         int ret = 0;
2700
2701         if (!node)
2702                 return 0;
2703
2704         BUG_ON(node->processed);
2705         root = select_one_root(trans, node);
2706         if (root == ERR_PTR(-ENOENT)) {
2707                 update_processed_blocks(rc, node);
2708                 goto out;
2709         }
2710
2711         if (!root || root->ref_cows) {
2712                 ret = reserve_metadata_space(trans, rc, node);
2713                 if (ret)
2714                         goto out;
2715                 release = 1;
2716         }
2717
2718         if (root) {
2719                 if (root->ref_cows) {
2720                         BUG_ON(node->new_bytenr);
2721                         BUG_ON(!list_empty(&node->list));
2722                         btrfs_record_root_in_trans(trans, root);
2723                         root = root->reloc_root;
2724                         node->new_bytenr = root->node->start;
2725                         node->root = root;
2726                         list_add_tail(&node->list, &rc->backref_cache.changed);
2727                 } else {
2728                         path->lowest_level = node->level;
2729                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2730                         btrfs_release_path(root, path);
2731                         if (ret > 0)
2732                                 ret = 0;
2733                 }
2734                 if (!ret)
2735                         update_processed_blocks(rc, node);
2736         } else {
2737                 ret = do_relocation(trans, rc, node, key, path, 1);
2738         }
2739 out:
2740         if (ret || node->level == 0 || node->cowonly) {
2741                 if (release)
2742                         release_metadata_space(rc, node);
2743                 remove_backref_node(&rc->backref_cache, node);
2744         }
2745         return ret;
2746 }
2747
2748 /*
2749  * relocate a list of blocks
2750  */
2751 static noinline_for_stack
2752 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2753                          struct reloc_control *rc, struct rb_root *blocks)
2754 {
2755         struct backref_node *node;
2756         struct btrfs_path *path;
2757         struct tree_block *block;
2758         struct rb_node *rb_node;
2759         int ret;
2760         int err = 0;
2761
2762         path = btrfs_alloc_path();
2763         if (!path)
2764                 return -ENOMEM;
2765
2766         rb_node = rb_first(blocks);
2767         while (rb_node) {
2768                 block = rb_entry(rb_node, struct tree_block, rb_node);
2769                 if (!block->key_ready)
2770                         reada_tree_block(rc, block);
2771                 rb_node = rb_next(rb_node);
2772         }
2773
2774         rb_node = rb_first(blocks);
2775         while (rb_node) {
2776                 block = rb_entry(rb_node, struct tree_block, rb_node);
2777                 if (!block->key_ready)
2778                         get_tree_block_key(rc, block);
2779                 rb_node = rb_next(rb_node);
2780         }
2781
2782         rb_node = rb_first(blocks);
2783         while (rb_node) {
2784                 block = rb_entry(rb_node, struct tree_block, rb_node);
2785
2786                 node = build_backref_tree(rc, &block->key,
2787                                           block->level, block->bytenr);
2788                 if (IS_ERR(node)) {
2789                         err = PTR_ERR(node);
2790                         goto out;
2791                 }
2792
2793                 ret = relocate_tree_block(trans, rc, node, &block->key,
2794                                           path);
2795                 if (ret < 0) {
2796                         if (ret != -EAGAIN || rb_node == rb_first(blocks))
2797                                 err = ret;
2798                         goto out;
2799                 }
2800                 rb_node = rb_next(rb_node);
2801         }
2802 out:
2803         free_block_list(blocks);
2804         err = finish_pending_nodes(trans, rc, path, err);
2805
2806         btrfs_free_path(path);
2807         return err;
2808 }
2809
2810 static noinline_for_stack
2811 int prealloc_file_extent_cluster(struct inode *inode,
2812                                  struct file_extent_cluster *cluster)
2813 {
2814         u64 alloc_hint = 0;
2815         u64 start;
2816         u64 end;
2817         u64 offset = BTRFS_I(inode)->index_cnt;
2818         u64 num_bytes;
2819         int nr = 0;
2820         int ret = 0;
2821
2822         BUG_ON(cluster->start != cluster->boundary[0]);
2823         mutex_lock(&inode->i_mutex);
2824
2825         ret = btrfs_check_data_free_space(inode, cluster->end +
2826                                           1 - cluster->start);
2827         if (ret)
2828                 goto out;
2829
2830         while (nr < cluster->nr) {
2831                 start = cluster->boundary[nr] - offset;
2832                 if (nr + 1 < cluster->nr)
2833                         end = cluster->boundary[nr + 1] - 1 - offset;
2834                 else
2835                         end = cluster->end - offset;
2836
2837                 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2838                 num_bytes = end + 1 - start;
2839                 ret = btrfs_prealloc_file_range(inode, 0, start,
2840                                                 num_bytes, num_bytes,
2841                                                 end + 1, &alloc_hint);
2842                 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2843                 if (ret)
2844                         break;
2845                 nr++;
2846         }
2847         btrfs_free_reserved_data_space(inode, cluster->end +
2848                                        1 - cluster->start);
2849 out:
2850         mutex_unlock(&inode->i_mutex);
2851         return ret;
2852 }
2853
2854 static noinline_for_stack
2855 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2856                          u64 block_start)
2857 {
2858         struct btrfs_root *root = BTRFS_I(inode)->root;
2859         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2860         struct extent_map *em;
2861         int ret = 0;
2862
2863         em = alloc_extent_map(GFP_NOFS);
2864         if (!em)
2865                 return -ENOMEM;
2866
2867         em->start = start;
2868         em->len = end + 1 - start;
2869         em->block_len = em->len;
2870         em->block_start = block_start;
2871         em->bdev = root->fs_info->fs_devices->latest_bdev;
2872         set_bit(EXTENT_FLAG_PINNED, &em->flags);
2873
2874         lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2875         while (1) {
2876                 write_lock(&em_tree->lock);
2877                 ret = add_extent_mapping(em_tree, em);
2878                 write_unlock(&em_tree->lock);
2879                 if (ret != -EEXIST) {
2880                         free_extent_map(em);
2881                         break;
2882                 }
2883                 btrfs_drop_extent_cache(inode, start, end, 0);
2884         }
2885         unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2886         return ret;
2887 }
2888
2889 static int relocate_file_extent_cluster(struct inode *inode,
2890                                         struct file_extent_cluster *cluster)
2891 {
2892         u64 page_start;
2893         u64 page_end;
2894         u64 offset = BTRFS_I(inode)->index_cnt;
2895         unsigned long index;
2896         unsigned long last_index;
2897         struct page *page;
2898         struct file_ra_state *ra;
2899         int nr = 0;
2900         int ret = 0;
2901
2902         if (!cluster->nr)
2903                 return 0;
2904
2905         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2906         if (!ra)
2907                 return -ENOMEM;
2908
2909         ret = prealloc_file_extent_cluster(inode, cluster);
2910         if (ret)
2911                 goto out;
2912
2913         file_ra_state_init(ra, inode->i_mapping);
2914
2915         ret = setup_extent_mapping(inode, cluster->start - offset,
2916                                    cluster->end - offset, cluster->start);
2917         if (ret)
2918                 goto out;
2919
2920         index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
2921         last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
2922         while (index <= last_index) {
2923                 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
2924                 if (ret)
2925                         goto out;
2926
2927                 page = find_lock_page(inode->i_mapping, index);
2928                 if (!page) {
2929                         page_cache_sync_readahead(inode->i_mapping,
2930                                                   ra, NULL, index,
2931                                                   last_index + 1 - index);
2932                         page = grab_cache_page(inode->i_mapping, index);
2933                         if (!page) {
2934                                 btrfs_delalloc_release_metadata(inode,
2935                                                         PAGE_CACHE_SIZE);
2936                                 ret = -ENOMEM;
2937                                 goto out;
2938                         }
2939                 }
2940
2941                 if (PageReadahead(page)) {
2942                         page_cache_async_readahead(inode->i_mapping,
2943                                                    ra, NULL, page, index,
2944                                                    last_index + 1 - index);
2945                 }
2946
2947                 if (!PageUptodate(page)) {
2948                         btrfs_readpage(NULL, page);
2949                         lock_page(page);
2950                         if (!PageUptodate(page)) {
2951                                 unlock_page(page);
2952                                 page_cache_release(page);
2953                                 btrfs_delalloc_release_metadata(inode,
2954                                                         PAGE_CACHE_SIZE);
2955                                 ret = -EIO;
2956                                 goto out;
2957                         }
2958                 }
2959
2960                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2961                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2962
2963                 lock_extent(&BTRFS_I(inode)->io_tree,
2964                             page_start, page_end, GFP_NOFS);
2965
2966                 set_page_extent_mapped(page);
2967
2968                 if (nr < cluster->nr &&
2969                     page_start + offset == cluster->boundary[nr]) {
2970                         set_extent_bits(&BTRFS_I(inode)->io_tree,
2971                                         page_start, page_end,
2972                                         EXTENT_BOUNDARY, GFP_NOFS);
2973                         nr++;
2974                 }
2975
2976                 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
2977                 set_page_dirty(page);
2978
2979                 unlock_extent(&BTRFS_I(inode)->io_tree,
2980                               page_start, page_end, GFP_NOFS);
2981                 unlock_page(page);
2982                 page_cache_release(page);
2983
2984                 index++;
2985                 balance_dirty_pages_ratelimited(inode->i_mapping);
2986                 btrfs_throttle(BTRFS_I(inode)->root);
2987         }
2988         WARN_ON(nr != cluster->nr);
2989 out:
2990         kfree(ra);
2991         return ret;
2992 }
2993
2994 static noinline_for_stack
2995 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2996                          struct file_extent_cluster *cluster)
2997 {
2998         int ret;
2999
3000         if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3001                 ret = relocate_file_extent_cluster(inode, cluster);
3002                 if (ret)
3003                         return ret;
3004                 cluster->nr = 0;
3005         }
3006
3007         if (!cluster->nr)
3008                 cluster->start = extent_key->objectid;
3009         else
3010                 BUG_ON(cluster->nr >= MAX_EXTENTS);
3011         cluster->end = extent_key->objectid + extent_key->offset - 1;
3012         cluster->boundary[cluster->nr] = extent_key->objectid;
3013         cluster->nr++;
3014
3015         if (cluster->nr >= MAX_EXTENTS) {
3016                 ret = relocate_file_extent_cluster(inode, cluster);
3017                 if (ret)
3018                         return ret;
3019                 cluster->nr = 0;
3020         }
3021         return 0;
3022 }
3023
3024 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3025 static int get_ref_objectid_v0(struct reloc_control *rc,
3026                                struct btrfs_path *path,
3027                                struct btrfs_key *extent_key,
3028                                u64 *ref_objectid, int *path_change)
3029 {
3030         struct btrfs_key key;
3031         struct extent_buffer *leaf;
3032         struct btrfs_extent_ref_v0 *ref0;
3033         int ret;
3034         int slot;
3035
3036         leaf = path->nodes[0];
3037         slot = path->slots[0];
3038         while (1) {
3039                 if (slot >= btrfs_header_nritems(leaf)) {
3040                         ret = btrfs_next_leaf(rc->extent_root, path);
3041                         if (ret < 0)
3042                                 return ret;
3043                         BUG_ON(ret > 0);
3044                         leaf = path->nodes[0];
3045                         slot = path->slots[0];
3046                         if (path_change)
3047                                 *path_change = 1;
3048                 }
3049                 btrfs_item_key_to_cpu(leaf, &key, slot);
3050                 if (key.objectid != extent_key->objectid)
3051                         return -ENOENT;
3052
3053                 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3054                         slot++;
3055                         continue;
3056                 }
3057                 ref0 = btrfs_item_ptr(leaf, slot,
3058                                 struct btrfs_extent_ref_v0);
3059                 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3060                 break;
3061         }
3062         return 0;
3063 }
3064 #endif
3065
3066 /*
3067  * helper to add a tree block to the list.
3068  * the major work is getting the generation and level of the block
3069  */
3070 static int add_tree_block(struct reloc_control *rc,
3071                           struct btrfs_key *extent_key,
3072                           struct btrfs_path *path,
3073                           struct rb_root *blocks)
3074 {
3075         struct extent_buffer *eb;
3076         struct btrfs_extent_item *ei;
3077         struct btrfs_tree_block_info *bi;
3078         struct tree_block *block;
3079         struct rb_node *rb_node;
3080         u32 item_size;
3081         int level = -1;
3082         int generation;
3083
3084         eb =  path->nodes[0];
3085         item_size = btrfs_item_size_nr(eb, path->slots[0]);
3086
3087         if (item_size >= sizeof(*ei) + sizeof(*bi)) {
3088                 ei = btrfs_item_ptr(eb, path->slots[0],
3089                                 struct btrfs_extent_item);
3090                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3091                 generation = btrfs_extent_generation(eb, ei);
3092                 level = btrfs_tree_block_level(eb, bi);
3093         } else {
3094 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3095                 u64 ref_owner;
3096                 int ret;
3097
3098                 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3099                 ret = get_ref_objectid_v0(rc, path, extent_key,
3100                                           &ref_owner, NULL);
3101                 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3102                 level = (int)ref_owner;
3103                 /* FIXME: get real generation */
3104                 generation = 0;
3105 #else
3106                 BUG();
3107 #endif
3108         }
3109
3110         btrfs_release_path(rc->extent_root, path);
3111
3112         BUG_ON(level == -1);
3113
3114         block = kmalloc(sizeof(*block), GFP_NOFS);
3115         if (!block)
3116                 return -ENOMEM;
3117
3118         block->bytenr = extent_key->objectid;
3119         block->key.objectid = extent_key->offset;
3120         block->key.offset = generation;
3121         block->level = level;
3122         block->key_ready = 0;
3123
3124         rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3125         BUG_ON(rb_node);
3126
3127         return 0;
3128 }
3129
3130 /*
3131  * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3132  */
3133 static int __add_tree_block(struct reloc_control *rc,
3134                             u64 bytenr, u32 blocksize,
3135                             struct rb_root *blocks)
3136 {
3137         struct btrfs_path *path;
3138         struct btrfs_key key;
3139         int ret;
3140
3141         if (tree_block_processed(bytenr, blocksize, rc))
3142                 return 0;
3143
3144         if (tree_search(blocks, bytenr))
3145                 return 0;
3146
3147         path = btrfs_alloc_path();
3148         if (!path)
3149                 return -ENOMEM;
3150
3151         key.objectid = bytenr;
3152         key.type = BTRFS_EXTENT_ITEM_KEY;
3153         key.offset = blocksize;
3154
3155         path->search_commit_root = 1;
3156         path->skip_locking = 1;
3157         ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3158         if (ret < 0)
3159                 goto out;
3160         BUG_ON(ret);
3161
3162         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3163         ret = add_tree_block(rc, &key, path, blocks);
3164 out:
3165         btrfs_free_path(path);
3166         return ret;
3167 }
3168
3169 /*
3170  * helper to check if the block use full backrefs for pointers in it
3171  */
3172 static int block_use_full_backref(struct reloc_control *rc,
3173                                   struct extent_buffer *eb)
3174 {
3175         u64 flags;
3176         int ret;
3177
3178         if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3179             btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3180                 return 1;
3181
3182         ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3183                                        eb->start, eb->len, NULL, &flags);
3184         BUG_ON(ret);
3185
3186         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3187                 ret = 1;
3188         else
3189                 ret = 0;
3190         return ret;
3191 }
3192
3193 /*
3194  * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3195  * this function scans fs tree to find blocks reference the data extent
3196  */
3197 static int find_data_references(struct reloc_control *rc,
3198                                 struct btrfs_key *extent_key,
3199                                 struct extent_buffer *leaf,
3200                                 struct btrfs_extent_data_ref *ref,
3201                                 struct rb_root *blocks)
3202 {
3203         struct btrfs_path *path;
3204         struct tree_block *block;
3205         struct btrfs_root *root;
3206         struct btrfs_file_extent_item *fi;
3207         struct rb_node *rb_node;
3208         struct btrfs_key key;
3209         u64 ref_root;
3210         u64 ref_objectid;
3211         u64 ref_offset;
3212         u32 ref_count;
3213         u32 nritems;
3214         int err = 0;
3215         int added = 0;
3216         int counted;
3217         int ret;
3218
3219         path = btrfs_alloc_path();
3220         if (!path)
3221                 return -ENOMEM;
3222
3223         ref_root = btrfs_extent_data_ref_root(leaf, ref);
3224         ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3225         ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3226         ref_count = btrfs_extent_data_ref_count(leaf, ref);
3227
3228         root = read_fs_root(rc->extent_root->fs_info, ref_root);
3229         if (IS_ERR(root)) {
3230                 err = PTR_ERR(root);
3231                 goto out;
3232         }
3233
3234         key.objectid = ref_objectid;
3235         key.offset = ref_offset;
3236         key.type = BTRFS_EXTENT_DATA_KEY;
3237
3238         path->search_commit_root = 1;
3239         path->skip_locking = 1;
3240         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3241         if (ret < 0) {
3242                 err = ret;
3243                 goto out;
3244         }
3245
3246         leaf = path->nodes[0];
3247         nritems = btrfs_header_nritems(leaf);
3248         /*
3249          * the references in tree blocks that use full backrefs
3250          * are not counted in
3251          */
3252         if (block_use_full_backref(rc, leaf))
3253                 counted = 0;
3254         else
3255                 counted = 1;
3256         rb_node = tree_search(blocks, leaf->start);
3257         if (rb_node) {
3258                 if (counted)
3259                         added = 1;
3260                 else
3261                         path->slots[0] = nritems;
3262         }
3263
3264         while (ref_count > 0) {
3265                 while (path->slots[0] >= nritems) {
3266                         ret = btrfs_next_leaf(root, path);
3267                         if (ret < 0) {
3268                                 err = ret;
3269                                 goto out;
3270                         }
3271                         if (ret > 0) {
3272                                 WARN_ON(1);
3273                                 goto out;
3274                         }
3275
3276                         leaf = path->nodes[0];
3277                         nritems = btrfs_header_nritems(leaf);
3278                         added = 0;
3279
3280                         if (block_use_full_backref(rc, leaf))
3281                                 counted = 0;
3282                         else
3283                                 counted = 1;
3284                         rb_node = tree_search(blocks, leaf->start);
3285                         if (rb_node) {
3286                                 if (counted)
3287                                         added = 1;
3288                                 else
3289                                         path->slots[0] = nritems;
3290                         }
3291                 }
3292
3293                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3294                 if (key.objectid != ref_objectid ||
3295                     key.type != BTRFS_EXTENT_DATA_KEY) {
3296                         WARN_ON(1);
3297                         break;
3298                 }
3299
3300                 fi = btrfs_item_ptr(leaf, path->slots[0],
3301                                     struct btrfs_file_extent_item);
3302
3303                 if (btrfs_file_extent_type(leaf, fi) ==
3304                     BTRFS_FILE_EXTENT_INLINE)
3305                         goto next;
3306
3307                 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3308                     extent_key->objectid)
3309                         goto next;
3310
3311                 key.offset -= btrfs_file_extent_offset(leaf, fi);
3312                 if (key.offset != ref_offset)
3313                         goto next;
3314
3315                 if (counted)
3316                         ref_count--;
3317                 if (added)
3318                         goto next;
3319
3320                 if (!tree_block_processed(leaf->start, leaf->len, rc)) {
3321                         block = kmalloc(sizeof(*block), GFP_NOFS);
3322                         if (!block) {
3323                                 err = -ENOMEM;
3324                                 break;
3325                         }
3326                         block->bytenr = leaf->start;
3327                         btrfs_item_key_to_cpu(leaf, &block->key, 0);
3328                         block->level = 0;
3329                         block->key_ready = 1;
3330                         rb_node = tree_insert(blocks, block->bytenr,
3331                                               &block->rb_node);
3332                         BUG_ON(rb_node);
3333                 }
3334                 if (counted)
3335                         added = 1;
3336                 else
3337                         path->slots[0] = nritems;
3338 next:
3339                 path->slots[0]++;
3340
3341         }
3342 out:
3343         btrfs_free_path(path);
3344         return err;
3345 }
3346
3347 /*
3348  * hepler to find all tree blocks that reference a given data extent
3349  */
3350 static noinline_for_stack
3351 int add_data_references(struct reloc_control *rc,
3352                         struct btrfs_key *extent_key,
3353                         struct btrfs_path *path,
3354                         struct rb_root *blocks)
3355 {
3356         struct btrfs_key key;
3357         struct extent_buffer *eb;
3358         struct btrfs_extent_data_ref *dref;
3359         struct btrfs_extent_inline_ref *iref;
3360         unsigned long ptr;
3361         unsigned long end;
3362         u32 blocksize = btrfs_level_size(rc->extent_root, 0);
3363         int ret;
3364         int err = 0;
3365
3366         eb = path->nodes[0];
3367         ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3368         end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3369 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3370         if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3371                 ptr = end;
3372         else
3373 #endif
3374                 ptr += sizeof(struct btrfs_extent_item);
3375
3376         while (ptr < end) {
3377                 iref = (struct btrfs_extent_inline_ref *)ptr;
3378                 key.type = btrfs_extent_inline_ref_type(eb, iref);
3379                 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3380                         key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3381                         ret = __add_tree_block(rc, key.offset, blocksize,
3382                                                blocks);
3383                 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3384                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3385                         ret = find_data_references(rc, extent_key,
3386                                                    eb, dref, blocks);
3387                 } else {
3388                         BUG();
3389                 }
3390                 ptr += btrfs_extent_inline_ref_size(key.type);
3391         }
3392         WARN_ON(ptr > end);
3393
3394         while (1) {
3395                 cond_resched();
3396                 eb = path->nodes[0];
3397                 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3398                         ret = btrfs_next_leaf(rc->extent_root, path);
3399                         if (ret < 0) {
3400                                 err = ret;
3401                                 break;
3402                         }
3403                         if (ret > 0)
3404                                 break;
3405                         eb = path->nodes[0];
3406                 }
3407
3408                 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3409                 if (key.objectid != extent_key->objectid)
3410                         break;
3411
3412 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3413                 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3414                     key.type == BTRFS_EXTENT_REF_V0_KEY) {
3415 #else
3416                 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3417                 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3418 #endif
3419                         ret = __add_tree_block(rc, key.offset, blocksize,
3420                                                blocks);
3421                 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3422                         dref = btrfs_item_ptr(eb, path->slots[0],
3423                                               struct btrfs_extent_data_ref);
3424                         ret = find_data_references(rc, extent_key,
3425                                                    eb, dref, blocks);
3426                 } else {
3427                         ret = 0;
3428                 }
3429                 if (ret) {
3430                         err = ret;
3431                         break;
3432                 }
3433                 path->slots[0]++;
3434         }
3435         btrfs_release_path(rc->extent_root, path);
3436         if (err)
3437                 free_block_list(blocks);
3438         return err;
3439 }
3440
3441 /*
3442  * hepler to find next unprocessed extent
3443  */
3444 static noinline_for_stack
3445 int find_next_extent(struct btrfs_trans_handle *trans,
3446                      struct reloc_control *rc, struct btrfs_path *path,
3447                      struct btrfs_key *extent_key)
3448 {
3449         struct btrfs_key key;
3450         struct extent_buffer *leaf;
3451         u64 start, end, last;
3452         int ret;
3453
3454         last = rc->block_group->key.objectid + rc->block_group->key.offset;
3455         while (1) {
3456                 cond_resched();
3457                 if (rc->search_start >= last) {
3458                         ret = 1;
3459                         break;
3460                 }
3461
3462                 key.objectid = rc->search_start;
3463                 key.type = BTRFS_EXTENT_ITEM_KEY;
3464                 key.offset = 0;
3465
3466                 path->search_commit_root = 1;
3467                 path->skip_locking = 1;
3468                 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3469                                         0, 0);
3470                 if (ret < 0)
3471                         break;
3472 next:
3473                 leaf = path->nodes[0];
3474                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3475                         ret = btrfs_next_leaf(rc->extent_root, path);
3476                         if (ret != 0)
3477                                 break;
3478                         leaf = path->nodes[0];
3479                 }
3480
3481                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3482                 if (key.objectid >= last) {
3483                         ret = 1;
3484                         break;
3485                 }
3486
3487                 if (key.type != BTRFS_EXTENT_ITEM_KEY ||
3488                     key.objectid + key.offset <= rc->search_start) {
3489                         path->slots[0]++;
3490                         goto next;
3491                 }
3492
3493                 ret = find_first_extent_bit(&rc->processed_blocks,
3494                                             key.objectid, &start, &end,
3495                                             EXTENT_DIRTY);
3496
3497                 if (ret == 0 && start <= key.objectid) {
3498                         btrfs_release_path(rc->extent_root, path);
3499                         rc->search_start = end + 1;
3500                 } else {
3501                         rc->search_start = key.objectid + key.offset;
3502                         memcpy(extent_key, &key, sizeof(key));
3503                         return 0;
3504                 }
3505         }
3506         btrfs_release_path(rc->extent_root, path);
3507         return ret;
3508 }
3509
3510 static void set_reloc_control(struct reloc_control *rc)
3511 {
3512         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3513         mutex_lock(&fs_info->trans_mutex);
3514         fs_info->reloc_ctl = rc;
3515         mutex_unlock(&fs_info->trans_mutex);
3516 }
3517
3518 static void unset_reloc_control(struct reloc_control *rc)
3519 {
3520         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3521         mutex_lock(&fs_info->trans_mutex);
3522         fs_info->reloc_ctl = NULL;
3523         mutex_unlock(&fs_info->trans_mutex);
3524 }
3525
3526 static int check_extent_flags(u64 flags)
3527 {
3528         if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3529             (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3530                 return 1;
3531         if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3532             !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3533                 return 1;
3534         if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3535             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3536                 return 1;
3537         return 0;
3538 }
3539
3540 static noinline_for_stack
3541 int prepare_to_relocate(struct reloc_control *rc)
3542 {
3543         struct btrfs_trans_handle *trans;
3544         int ret;
3545
3546         rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root);
3547         if (!rc->block_rsv)
3548                 return -ENOMEM;
3549
3550         /*
3551          * reserve some space for creating reloc trees.
3552          * btrfs_init_reloc_root will use them when there
3553          * is no reservation in transaction handle.
3554          */
3555         ret = btrfs_block_rsv_add(NULL, rc->extent_root, rc->block_rsv,
3556                                   rc->extent_root->nodesize * 256,
3557                                   &rc->block_rsv_retries);
3558         if (ret)
3559                 return ret;
3560
3561         rc->block_rsv->refill_used = 1;
3562         btrfs_add_durable_block_rsv(rc->extent_root->fs_info, rc->block_rsv);
3563
3564         memset(&rc->cluster, 0, sizeof(rc->cluster));
3565         rc->search_start = rc->block_group->key.objectid;
3566         rc->extents_found = 0;
3567         rc->nodes_relocated = 0;
3568         rc->merging_rsv_size = 0;
3569         rc->block_rsv_retries = 0;
3570
3571         rc->create_reloc_tree = 1;
3572         set_reloc_control(rc);
3573
3574         trans = btrfs_join_transaction(rc->extent_root, 1);
3575         btrfs_commit_transaction(trans, rc->extent_root);
3576         return 0;
3577 }
3578
3579 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3580 {
3581         struct rb_root blocks = RB_ROOT;
3582         struct btrfs_key key;
3583         struct btrfs_trans_handle *trans = NULL;
3584         struct btrfs_path *path;
3585         struct btrfs_extent_item *ei;
3586         unsigned long nr;
3587         u64 flags;
3588         u32 item_size;
3589         int ret;
3590         int err = 0;
3591
3592         path = btrfs_alloc_path();
3593         if (!path)
3594                 return -ENOMEM;
3595
3596         ret = prepare_to_relocate(rc);
3597         if (ret) {
3598                 err = ret;
3599                 goto out_free;
3600         }
3601
3602         while (1) {
3603                 trans = btrfs_start_transaction(rc->extent_root, 0);
3604
3605                 if (update_backref_cache(trans, &rc->backref_cache)) {
3606                         btrfs_end_transaction(trans, rc->extent_root);
3607                         continue;
3608                 }
3609
3610                 ret = find_next_extent(trans, rc, path, &key);
3611                 if (ret < 0)
3612                         err = ret;
3613                 if (ret != 0)
3614                         break;
3615
3616                 rc->extents_found++;
3617
3618                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3619                                     struct btrfs_extent_item);
3620                 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3621                 if (item_size >= sizeof(*ei)) {
3622                         flags = btrfs_extent_flags(path->nodes[0], ei);
3623                         ret = check_extent_flags(flags);
3624                         BUG_ON(ret);
3625
3626                 } else {
3627 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3628                         u64 ref_owner;
3629                         int path_change = 0;
3630
3631                         BUG_ON(item_size !=
3632                                sizeof(struct btrfs_extent_item_v0));
3633                         ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
3634                                                   &path_change);
3635                         if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
3636                                 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
3637                         else
3638                                 flags = BTRFS_EXTENT_FLAG_DATA;
3639
3640                         if (path_change) {
3641                                 btrfs_release_path(rc->extent_root, path);
3642
3643                                 path->search_commit_root = 1;
3644                                 path->skip_locking = 1;
3645                                 ret = btrfs_search_slot(NULL, rc->extent_root,
3646                                                         &key, path, 0, 0);
3647                                 if (ret < 0) {
3648                                         err = ret;
3649                                         break;
3650                                 }
3651                                 BUG_ON(ret > 0);
3652                         }
3653 #else
3654                         BUG();
3655 #endif
3656                 }
3657
3658                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3659                         ret = add_tree_block(rc, &key, path, &blocks);
3660                 } else if (rc->stage == UPDATE_DATA_PTRS &&
3661                            (flags & BTRFS_EXTENT_FLAG_DATA)) {
3662                         ret = add_data_references(rc, &key, path, &blocks);
3663                 } else {
3664                         btrfs_release_path(rc->extent_root, path);
3665                         ret = 0;
3666                 }
3667                 if (ret < 0) {
3668                         err = ret;
3669                         break;
3670                 }
3671
3672                 if (!RB_EMPTY_ROOT(&blocks)) {
3673                         ret = relocate_tree_blocks(trans, rc, &blocks);
3674                         if (ret < 0) {
3675                                 if (ret != -EAGAIN) {
3676                                         err = ret;
3677                                         break;
3678                                 }
3679                                 rc->extents_found--;
3680                                 rc->search_start = key.objectid;
3681                         }
3682                 }
3683
3684                 ret = btrfs_block_rsv_check(trans, rc->extent_root,
3685                                             rc->block_rsv, 0, 5);
3686                 if (ret < 0) {
3687                         if (ret != -EAGAIN) {
3688                                 err = ret;
3689                                 WARN_ON(1);
3690                                 break;
3691                         }
3692                         rc->commit_transaction = 1;
3693                 }
3694
3695                 if (rc->commit_transaction) {
3696                         rc->commit_transaction = 0;
3697                         ret = btrfs_commit_transaction(trans, rc->extent_root);
3698                         BUG_ON(ret);
3699                 } else {
3700                         nr = trans->blocks_used;
3701                         btrfs_end_transaction_throttle(trans, rc->extent_root);
3702                         btrfs_btree_balance_dirty(rc->extent_root, nr);
3703                 }
3704                 trans = NULL;
3705
3706                 if (rc->stage == MOVE_DATA_EXTENTS &&
3707                     (flags & BTRFS_EXTENT_FLAG_DATA)) {
3708                         rc->found_file_extent = 1;
3709                         ret = relocate_data_extent(rc->data_inode,
3710                                                    &key, &rc->cluster);
3711                         if (ret < 0) {
3712                                 err = ret;
3713                                 break;
3714                         }
3715                 }
3716         }
3717
3718         btrfs_release_path(rc->extent_root, path);
3719         clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
3720                           GFP_NOFS);
3721
3722         if (trans) {
3723                 nr = trans->blocks_used;
3724                 btrfs_end_transaction_throttle(trans, rc->extent_root);
3725                 btrfs_btree_balance_dirty(rc->extent_root, nr);
3726         }
3727
3728         if (!err) {
3729                 ret = relocate_file_extent_cluster(rc->data_inode,
3730                                                    &rc->cluster);
3731                 if (ret < 0)
3732                         err = ret;
3733         }
3734
3735         rc->create_reloc_tree = 0;
3736         set_reloc_control(rc);
3737
3738         backref_cache_cleanup(&rc->backref_cache);
3739         btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3740
3741         err = prepare_to_merge(rc, err);
3742
3743         merge_reloc_roots(rc);
3744
3745         rc->merge_reloc_tree = 0;
3746         unset_reloc_control(rc);
3747         btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3748
3749         /* get rid of pinned extents */
3750         trans = btrfs_join_transaction(rc->extent_root, 1);
3751         btrfs_commit_transaction(trans, rc->extent_root);
3752 out_free:
3753         btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
3754         btrfs_free_path(path);
3755         return err;
3756 }
3757
3758 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3759                                  struct btrfs_root *root, u64 objectid)
3760 {
3761         struct btrfs_path *path;
3762         struct btrfs_inode_item *item;
3763         struct extent_buffer *leaf;
3764         int ret;
3765
3766         path = btrfs_alloc_path();
3767         if (!path)
3768                 return -ENOMEM;
3769
3770         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3771         if (ret)
3772                 goto out;
3773
3774         leaf = path->nodes[0];
3775         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3776         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3777         btrfs_set_inode_generation(leaf, item, 1);
3778         btrfs_set_inode_size(leaf, item, 0);
3779         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3780         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3781                                           BTRFS_INODE_PREALLOC);
3782         btrfs_mark_buffer_dirty(leaf);
3783         btrfs_release_path(root, path);
3784 out:
3785         btrfs_free_path(path);
3786         return ret;
3787 }
3788
3789 /*
3790  * helper to create inode for data relocation.
3791  * the inode is in data relocation tree and its link count is 0
3792  */
3793 static noinline_for_stack
3794 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3795                                  struct btrfs_block_group_cache *group)
3796 {
3797         struct inode *inode = NULL;
3798         struct btrfs_trans_handle *trans;
3799         struct btrfs_root *root;
3800         struct btrfs_key key;
3801         unsigned long nr;
3802         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
3803         int err = 0;
3804
3805         root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
3806         if (IS_ERR(root))
3807                 return ERR_CAST(root);
3808
3809         trans = btrfs_start_transaction(root, 6);
3810         if (IS_ERR(trans))
3811                 return ERR_CAST(trans);
3812
3813         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
3814         if (err)
3815                 goto out;
3816
3817         err = __insert_orphan_inode(trans, root, objectid);
3818         BUG_ON(err);
3819
3820         key.objectid = objectid;
3821         key.type = BTRFS_INODE_ITEM_KEY;
3822         key.offset = 0;
3823         inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
3824         BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
3825         BTRFS_I(inode)->index_cnt = group->key.objectid;
3826
3827         err = btrfs_orphan_add(trans, inode);
3828 out:
3829         nr = trans->blocks_used;
3830         btrfs_end_transaction(trans, root);
3831         btrfs_btree_balance_dirty(root, nr);
3832         if (err) {
3833                 if (inode)
3834                         iput(inode);
3835                 inode = ERR_PTR(err);
3836         }
3837         return inode;
3838 }
3839
3840 static struct reloc_control *alloc_reloc_control(void)
3841 {
3842         struct reloc_control *rc;
3843
3844         rc = kzalloc(sizeof(*rc), GFP_NOFS);
3845         if (!rc)
3846                 return NULL;
3847
3848         INIT_LIST_HEAD(&rc->reloc_roots);
3849         backref_cache_init(&rc->backref_cache);
3850         mapping_tree_init(&rc->reloc_root_tree);
3851         extent_io_tree_init(&rc->processed_blocks, NULL, GFP_NOFS);
3852         return rc;
3853 }
3854
3855 /*
3856  * function to relocate all extents in a block group.
3857  */
3858 int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
3859 {
3860         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3861         struct reloc_control *rc;
3862         int ret;
3863         int rw = 0;
3864         int err = 0;
3865
3866         rc = alloc_reloc_control();
3867         if (!rc)
3868                 return -ENOMEM;
3869
3870         rc->extent_root = extent_root;
3871
3872         rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
3873         BUG_ON(!rc->block_group);
3874
3875         if (!rc->block_group->ro) {
3876                 ret = btrfs_set_block_group_ro(extent_root, rc->block_group);
3877                 if (ret) {
3878                         err = ret;
3879                         goto out;
3880                 }
3881                 rw = 1;
3882         }
3883
3884         rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3885         if (IS_ERR(rc->data_inode)) {
3886                 err = PTR_ERR(rc->data_inode);
3887                 rc->data_inode = NULL;
3888                 goto out;
3889         }
3890
3891         printk(KERN_INFO "btrfs: relocating block group %llu flags %llu\n",
3892                (unsigned long long)rc->block_group->key.objectid,
3893                (unsigned long long)rc->block_group->flags);
3894
3895         btrfs_start_delalloc_inodes(fs_info->tree_root, 0);
3896         btrfs_wait_ordered_extents(fs_info->tree_root, 0, 0);
3897
3898         while (1) {
3899                 mutex_lock(&fs_info->cleaner_mutex);
3900
3901                 btrfs_clean_old_snapshots(fs_info->tree_root);
3902                 ret = relocate_block_group(rc);
3903
3904                 mutex_unlock(&fs_info->cleaner_mutex);
3905                 if (ret < 0) {
3906                         err = ret;
3907                         goto out;
3908                 }
3909
3910                 if (rc->extents_found == 0)
3911                         break;
3912
3913                 printk(KERN_INFO "btrfs: found %llu extents\n",
3914                         (unsigned long long)rc->extents_found);
3915
3916                 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
3917                         btrfs_wait_ordered_range(rc->data_inode, 0, (u64)-1);
3918                         invalidate_mapping_pages(rc->data_inode->i_mapping,
3919                                                  0, -1);
3920                         rc->stage = UPDATE_DATA_PTRS;
3921                 }
3922         }
3923
3924         filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
3925                                      rc->block_group->key.objectid,
3926                                      rc->block_group->key.objectid +
3927                                      rc->block_group->key.offset - 1);
3928
3929         WARN_ON(rc->block_group->pinned > 0);
3930         WARN_ON(rc->block_group->reserved > 0);
3931         WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
3932 out:
3933         if (err && rw)
3934                 btrfs_set_block_group_rw(extent_root, rc->block_group);
3935         iput(rc->data_inode);
3936         btrfs_put_block_group(rc->block_group);
3937         kfree(rc);
3938         return err;
3939 }
3940
3941 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
3942 {
3943         struct btrfs_trans_handle *trans;
3944         int ret;
3945
3946         trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
3947
3948         memset(&root->root_item.drop_progress, 0,
3949                 sizeof(root->root_item.drop_progress));
3950         root->root_item.drop_level = 0;
3951         btrfs_set_root_refs(&root->root_item, 0);
3952         ret = btrfs_update_root(trans, root->fs_info->tree_root,
3953                                 &root->root_key, &root->root_item);
3954         BUG_ON(ret);
3955
3956         ret = btrfs_end_transaction(trans, root->fs_info->tree_root);
3957         BUG_ON(ret);
3958         return 0;
3959 }
3960
3961 /*
3962  * recover relocation interrupted by system crash.
3963  *
3964  * this function resumes merging reloc trees with corresponding fs trees.
3965  * this is important for keeping the sharing of tree blocks
3966  */
3967 int btrfs_recover_relocation(struct btrfs_root *root)
3968 {
3969         LIST_HEAD(reloc_roots);
3970         struct btrfs_key key;
3971         struct btrfs_root *fs_root;
3972         struct btrfs_root *reloc_root;
3973         struct btrfs_path *path;
3974         struct extent_buffer *leaf;
3975         struct reloc_control *rc = NULL;
3976         struct btrfs_trans_handle *trans;
3977         int ret;
3978         int err = 0;
3979
3980         path = btrfs_alloc_path();
3981         if (!path)
3982                 return -ENOMEM;
3983
3984         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
3985         key.type = BTRFS_ROOT_ITEM_KEY;
3986         key.offset = (u64)-1;
3987
3988         while (1) {
3989                 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key,
3990                                         path, 0, 0);
3991                 if (ret < 0) {
3992                         err = ret;
3993                         goto out;
3994                 }
3995                 if (ret > 0) {
3996                         if (path->slots[0] == 0)
3997                                 break;
3998                         path->slots[0]--;
3999                 }
4000                 leaf = path->nodes[0];
4001                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4002                 btrfs_release_path(root->fs_info->tree_root, path);
4003
4004                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4005                     key.type != BTRFS_ROOT_ITEM_KEY)
4006                         break;
4007
4008                 reloc_root = btrfs_read_fs_root_no_radix(root, &key);
4009                 if (IS_ERR(reloc_root)) {
4010                         err = PTR_ERR(reloc_root);
4011                         goto out;
4012                 }
4013
4014                 list_add(&reloc_root->root_list, &reloc_roots);
4015
4016                 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4017                         fs_root = read_fs_root(root->fs_info,
4018                                                reloc_root->root_key.offset);
4019                         if (IS_ERR(fs_root)) {
4020                                 ret = PTR_ERR(fs_root);
4021                                 if (ret != -ENOENT) {
4022                                         err = ret;
4023                                         goto out;
4024                                 }
4025                                 mark_garbage_root(reloc_root);
4026                         }
4027                 }
4028
4029                 if (key.offset == 0)
4030                         break;
4031
4032                 key.offset--;
4033         }
4034         btrfs_release_path(root->fs_info->tree_root, path);
4035
4036         if (list_empty(&reloc_roots))
4037                 goto out;
4038
4039         rc = alloc_reloc_control();
4040         if (!rc) {
4041                 err = -ENOMEM;
4042                 goto out;
4043         }
4044
4045         rc->extent_root = root->fs_info->extent_root;
4046
4047         set_reloc_control(rc);
4048
4049         trans = btrfs_join_transaction(rc->extent_root, 1);
4050
4051         rc->merge_reloc_tree = 1;
4052
4053         while (!list_empty(&reloc_roots)) {
4054                 reloc_root = list_entry(reloc_roots.next,
4055                                         struct btrfs_root, root_list);
4056                 list_del(&reloc_root->root_list);
4057
4058                 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4059                         list_add_tail(&reloc_root->root_list,
4060                                       &rc->reloc_roots);
4061                         continue;
4062                 }
4063
4064                 fs_root = read_fs_root(root->fs_info,
4065                                        reloc_root->root_key.offset);
4066                 BUG_ON(IS_ERR(fs_root));
4067
4068                 __add_reloc_root(reloc_root);
4069                 fs_root->reloc_root = reloc_root;
4070         }
4071
4072         btrfs_commit_transaction(trans, rc->extent_root);
4073
4074         merge_reloc_roots(rc);
4075
4076         unset_reloc_control(rc);
4077
4078         trans = btrfs_join_transaction(rc->extent_root, 1);
4079         btrfs_commit_transaction(trans, rc->extent_root);
4080 out:
4081         kfree(rc);
4082         while (!list_empty(&reloc_roots)) {
4083                 reloc_root = list_entry(reloc_roots.next,
4084                                         struct btrfs_root, root_list);
4085                 list_del(&reloc_root->root_list);
4086                 free_extent_buffer(reloc_root->node);
4087                 free_extent_buffer(reloc_root->commit_root);
4088                 kfree(reloc_root);
4089         }
4090         btrfs_free_path(path);
4091
4092         if (err == 0) {
4093                 /* cleanup orphan inode in data relocation tree */
4094                 fs_root = read_fs_root(root->fs_info,
4095                                        BTRFS_DATA_RELOC_TREE_OBJECTID);
4096                 if (IS_ERR(fs_root))
4097                         err = PTR_ERR(fs_root);
4098                 else
4099                         btrfs_orphan_cleanup(fs_root);
4100         }
4101         return err;
4102 }
4103
4104 /*
4105  * helper to add ordered checksum for data relocation.
4106  *
4107  * cloning checksum properly handles the nodatasum extents.
4108  * it also saves CPU time to re-calculate the checksum.
4109  */
4110 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4111 {
4112         struct btrfs_ordered_sum *sums;
4113         struct btrfs_sector_sum *sector_sum;
4114         struct btrfs_ordered_extent *ordered;
4115         struct btrfs_root *root = BTRFS_I(inode)->root;
4116         size_t offset;
4117         int ret;
4118         u64 disk_bytenr;
4119         LIST_HEAD(list);
4120
4121         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4122         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4123
4124         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4125         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
4126                                        disk_bytenr + len - 1, &list);
4127
4128         while (!list_empty(&list)) {
4129                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4130                 list_del_init(&sums->list);
4131
4132                 sector_sum = sums->sums;
4133                 sums->bytenr = ordered->start;
4134
4135                 offset = 0;
4136                 while (offset < sums->len) {
4137                         sector_sum->bytenr += ordered->start - disk_bytenr;
4138                         sector_sum++;
4139                         offset += root->sectorsize;
4140                 }
4141
4142                 btrfs_add_ordered_sum(inode, ordered, sums);
4143         }
4144         btrfs_put_ordered_extent(ordered);
4145         return 0;
4146 }
4147
4148 void btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4149                            struct btrfs_root *root, struct extent_buffer *buf,
4150                            struct extent_buffer *cow)
4151 {
4152         struct reloc_control *rc;
4153         struct backref_node *node;
4154         int first_cow = 0;
4155         int level;
4156         int ret;
4157
4158         rc = root->fs_info->reloc_ctl;
4159         if (!rc)
4160                 return;
4161
4162         BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4163                root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4164
4165         level = btrfs_header_level(buf);
4166         if (btrfs_header_generation(buf) <=
4167             btrfs_root_last_snapshot(&root->root_item))
4168                 first_cow = 1;
4169
4170         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4171             rc->create_reloc_tree) {
4172                 WARN_ON(!first_cow && level == 0);
4173
4174                 node = rc->backref_cache.path[level];
4175                 BUG_ON(node->bytenr != buf->start &&
4176                        node->new_bytenr != buf->start);
4177
4178                 drop_node_buffer(node);
4179                 extent_buffer_get(cow);
4180                 node->eb = cow;
4181                 node->new_bytenr = cow->start;
4182
4183                 if (!node->pending) {
4184                         list_move_tail(&node->list,
4185                                        &rc->backref_cache.pending[level]);
4186                         node->pending = 1;
4187                 }
4188
4189                 if (first_cow)
4190                         __mark_block_processed(rc, node);
4191
4192                 if (first_cow && level > 0)
4193                         rc->nodes_relocated += buf->len;
4194         }
4195
4196         if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) {
4197                 ret = replace_file_extents(trans, rc, root, cow);
4198                 BUG_ON(ret);
4199         }
4200 }
4201
4202 /*
4203  * called before creating snapshot. it calculates metadata reservation
4204  * requried for relocating tree blocks in the snapshot
4205  */
4206 void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans,
4207                               struct btrfs_pending_snapshot *pending,
4208                               u64 *bytes_to_reserve)
4209 {
4210         struct btrfs_root *root;
4211         struct reloc_control *rc;
4212
4213         root = pending->root;
4214         if (!root->reloc_root)
4215                 return;
4216
4217         rc = root->fs_info->reloc_ctl;
4218         if (!rc->merge_reloc_tree)
4219                 return;
4220
4221         root = root->reloc_root;
4222         BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4223         /*
4224          * relocation is in the stage of merging trees. the space
4225          * used by merging a reloc tree is twice the size of
4226          * relocated tree nodes in the worst case. half for cowing
4227          * the reloc tree, half for cowing the fs tree. the space
4228          * used by cowing the reloc tree will be freed after the
4229          * tree is dropped. if we create snapshot, cowing the fs
4230          * tree may use more space than it frees. so we need
4231          * reserve extra space.
4232          */
4233         *bytes_to_reserve += rc->nodes_relocated;
4234 }
4235
4236 /*
4237  * called after snapshot is created. migrate block reservation
4238  * and create reloc root for the newly created snapshot
4239  */
4240 void btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4241                                struct btrfs_pending_snapshot *pending)
4242 {
4243         struct btrfs_root *root = pending->root;
4244         struct btrfs_root *reloc_root;
4245         struct btrfs_root *new_root;
4246         struct reloc_control *rc;
4247         int ret;
4248
4249         if (!root->reloc_root)
4250                 return;
4251
4252         rc = root->fs_info->reloc_ctl;
4253         rc->merging_rsv_size += rc->nodes_relocated;
4254
4255         if (rc->merge_reloc_tree) {
4256                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4257                                               rc->block_rsv,
4258                                               rc->nodes_relocated);
4259                 BUG_ON(ret);
4260         }
4261
4262         new_root = pending->snap;
4263         reloc_root = create_reloc_root(trans, root->reloc_root,
4264                                        new_root->root_key.objectid);
4265
4266         __add_reloc_root(reloc_root);
4267         new_root->reloc_root = reloc_root;
4268
4269         if (rc->create_reloc_tree) {
4270                 ret = clone_backref_node(trans, rc, root, reloc_root);
4271                 BUG_ON(ret);
4272         }
4273 }