btrfs: fix return value check of btrfs_join_transaction()
[linux-3.10.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 #include "free-space-cache.h"
33
34 /*
35  * backref_node, mapping_node and tree_block start with this
36  */
37 struct tree_entry {
38         struct rb_node rb_node;
39         u64 bytenr;
40 };
41
42 /*
43  * present a tree block in the backref cache
44  */
45 struct backref_node {
46         struct rb_node rb_node;
47         u64 bytenr;
48
49         u64 new_bytenr;
50         /* objectid of tree block owner, can be not uptodate */
51         u64 owner;
52         /* link to pending, changed or detached list */
53         struct list_head list;
54         /* list of upper level blocks reference this block */
55         struct list_head upper;
56         /* list of child blocks in the cache */
57         struct list_head lower;
58         /* NULL if this node is not tree root */
59         struct btrfs_root *root;
60         /* extent buffer got by COW the block */
61         struct extent_buffer *eb;
62         /* level of tree block */
63         unsigned int level:8;
64         /* is the block in non-reference counted tree */
65         unsigned int cowonly:1;
66         /* 1 if no child node in the cache */
67         unsigned int lowest:1;
68         /* is the extent buffer locked */
69         unsigned int locked:1;
70         /* has the block been processed */
71         unsigned int processed:1;
72         /* have backrefs of this block been checked */
73         unsigned int checked:1;
74         /*
75          * 1 if corresponding block has been cowed but some upper
76          * level block pointers may not point to the new location
77          */
78         unsigned int pending:1;
79         /*
80          * 1 if the backref node isn't connected to any other
81          * backref node.
82          */
83         unsigned int detached:1;
84 };
85
86 /*
87  * present a block pointer in the backref cache
88  */
89 struct backref_edge {
90         struct list_head list[2];
91         struct backref_node *node[2];
92 };
93
94 #define LOWER   0
95 #define UPPER   1
96
97 struct backref_cache {
98         /* red black tree of all backref nodes in the cache */
99         struct rb_root rb_root;
100         /* for passing backref nodes to btrfs_reloc_cow_block */
101         struct backref_node *path[BTRFS_MAX_LEVEL];
102         /*
103          * list of blocks that have been cowed but some block
104          * pointers in upper level blocks may not reflect the
105          * new location
106          */
107         struct list_head pending[BTRFS_MAX_LEVEL];
108         /* list of backref nodes with no child node */
109         struct list_head leaves;
110         /* list of blocks that have been cowed in current transaction */
111         struct list_head changed;
112         /* list of detached backref node. */
113         struct list_head detached;
114
115         u64 last_trans;
116
117         int nr_nodes;
118         int nr_edges;
119 };
120
121 /*
122  * map address of tree root to tree
123  */
124 struct mapping_node {
125         struct rb_node rb_node;
126         u64 bytenr;
127         void *data;
128 };
129
130 struct mapping_tree {
131         struct rb_root rb_root;
132         spinlock_t lock;
133 };
134
135 /*
136  * present a tree block to process
137  */
138 struct tree_block {
139         struct rb_node rb_node;
140         u64 bytenr;
141         struct btrfs_key key;
142         unsigned int level:8;
143         unsigned int key_ready:1;
144 };
145
146 #define MAX_EXTENTS 128
147
148 struct file_extent_cluster {
149         u64 start;
150         u64 end;
151         u64 boundary[MAX_EXTENTS];
152         unsigned int nr;
153 };
154
155 struct reloc_control {
156         /* block group to relocate */
157         struct btrfs_block_group_cache *block_group;
158         /* extent tree */
159         struct btrfs_root *extent_root;
160         /* inode for moving data */
161         struct inode *data_inode;
162
163         struct btrfs_block_rsv *block_rsv;
164
165         struct backref_cache backref_cache;
166
167         struct file_extent_cluster cluster;
168         /* tree blocks have been processed */
169         struct extent_io_tree processed_blocks;
170         /* map start of tree root to corresponding reloc tree */
171         struct mapping_tree reloc_root_tree;
172         /* list of reloc trees */
173         struct list_head reloc_roots;
174         /* size of metadata reservation for merging reloc trees */
175         u64 merging_rsv_size;
176         /* size of relocated tree nodes */
177         u64 nodes_relocated;
178
179         u64 search_start;
180         u64 extents_found;
181
182         unsigned int stage:8;
183         unsigned int create_reloc_tree:1;
184         unsigned int merge_reloc_tree:1;
185         unsigned int found_file_extent:1;
186         unsigned int commit_transaction:1;
187 };
188
189 /* stages of data relocation */
190 #define MOVE_DATA_EXTENTS       0
191 #define UPDATE_DATA_PTRS        1
192
193 static void remove_backref_node(struct backref_cache *cache,
194                                 struct backref_node *node);
195 static void __mark_block_processed(struct reloc_control *rc,
196                                    struct backref_node *node);
197
198 static void mapping_tree_init(struct mapping_tree *tree)
199 {
200         tree->rb_root = RB_ROOT;
201         spin_lock_init(&tree->lock);
202 }
203
204 static void backref_cache_init(struct backref_cache *cache)
205 {
206         int i;
207         cache->rb_root = RB_ROOT;
208         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
209                 INIT_LIST_HEAD(&cache->pending[i]);
210         INIT_LIST_HEAD(&cache->changed);
211         INIT_LIST_HEAD(&cache->detached);
212         INIT_LIST_HEAD(&cache->leaves);
213 }
214
215 static void backref_cache_cleanup(struct backref_cache *cache)
216 {
217         struct backref_node *node;
218         int i;
219
220         while (!list_empty(&cache->detached)) {
221                 node = list_entry(cache->detached.next,
222                                   struct backref_node, list);
223                 remove_backref_node(cache, node);
224         }
225
226         while (!list_empty(&cache->leaves)) {
227                 node = list_entry(cache->leaves.next,
228                                   struct backref_node, lower);
229                 remove_backref_node(cache, node);
230         }
231
232         cache->last_trans = 0;
233
234         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
235                 BUG_ON(!list_empty(&cache->pending[i]));
236         BUG_ON(!list_empty(&cache->changed));
237         BUG_ON(!list_empty(&cache->detached));
238         BUG_ON(!RB_EMPTY_ROOT(&cache->rb_root));
239         BUG_ON(cache->nr_nodes);
240         BUG_ON(cache->nr_edges);
241 }
242
243 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
244 {
245         struct backref_node *node;
246
247         node = kzalloc(sizeof(*node), GFP_NOFS);
248         if (node) {
249                 INIT_LIST_HEAD(&node->list);
250                 INIT_LIST_HEAD(&node->upper);
251                 INIT_LIST_HEAD(&node->lower);
252                 RB_CLEAR_NODE(&node->rb_node);
253                 cache->nr_nodes++;
254         }
255         return node;
256 }
257
258 static void free_backref_node(struct backref_cache *cache,
259                               struct backref_node *node)
260 {
261         if (node) {
262                 cache->nr_nodes--;
263                 kfree(node);
264         }
265 }
266
267 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
268 {
269         struct backref_edge *edge;
270
271         edge = kzalloc(sizeof(*edge), GFP_NOFS);
272         if (edge)
273                 cache->nr_edges++;
274         return edge;
275 }
276
277 static void free_backref_edge(struct backref_cache *cache,
278                               struct backref_edge *edge)
279 {
280         if (edge) {
281                 cache->nr_edges--;
282                 kfree(edge);
283         }
284 }
285
286 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
287                                    struct rb_node *node)
288 {
289         struct rb_node **p = &root->rb_node;
290         struct rb_node *parent = NULL;
291         struct tree_entry *entry;
292
293         while (*p) {
294                 parent = *p;
295                 entry = rb_entry(parent, struct tree_entry, rb_node);
296
297                 if (bytenr < entry->bytenr)
298                         p = &(*p)->rb_left;
299                 else if (bytenr > entry->bytenr)
300                         p = &(*p)->rb_right;
301                 else
302                         return parent;
303         }
304
305         rb_link_node(node, parent, p);
306         rb_insert_color(node, root);
307         return NULL;
308 }
309
310 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
311 {
312         struct rb_node *n = root->rb_node;
313         struct tree_entry *entry;
314
315         while (n) {
316                 entry = rb_entry(n, struct tree_entry, rb_node);
317
318                 if (bytenr < entry->bytenr)
319                         n = n->rb_left;
320                 else if (bytenr > entry->bytenr)
321                         n = n->rb_right;
322                 else
323                         return n;
324         }
325         return NULL;
326 }
327
328 /*
329  * walk up backref nodes until reach node presents tree root
330  */
331 static struct backref_node *walk_up_backref(struct backref_node *node,
332                                             struct backref_edge *edges[],
333                                             int *index)
334 {
335         struct backref_edge *edge;
336         int idx = *index;
337
338         while (!list_empty(&node->upper)) {
339                 edge = list_entry(node->upper.next,
340                                   struct backref_edge, list[LOWER]);
341                 edges[idx++] = edge;
342                 node = edge->node[UPPER];
343         }
344         BUG_ON(node->detached);
345         *index = idx;
346         return node;
347 }
348
349 /*
350  * walk down backref nodes to find start of next reference path
351  */
352 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
353                                               int *index)
354 {
355         struct backref_edge *edge;
356         struct backref_node *lower;
357         int idx = *index;
358
359         while (idx > 0) {
360                 edge = edges[idx - 1];
361                 lower = edge->node[LOWER];
362                 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
363                         idx--;
364                         continue;
365                 }
366                 edge = list_entry(edge->list[LOWER].next,
367                                   struct backref_edge, list[LOWER]);
368                 edges[idx - 1] = edge;
369                 *index = idx;
370                 return edge->node[UPPER];
371         }
372         *index = 0;
373         return NULL;
374 }
375
376 static void unlock_node_buffer(struct backref_node *node)
377 {
378         if (node->locked) {
379                 btrfs_tree_unlock(node->eb);
380                 node->locked = 0;
381         }
382 }
383
384 static void drop_node_buffer(struct backref_node *node)
385 {
386         if (node->eb) {
387                 unlock_node_buffer(node);
388                 free_extent_buffer(node->eb);
389                 node->eb = NULL;
390         }
391 }
392
393 static void drop_backref_node(struct backref_cache *tree,
394                               struct backref_node *node)
395 {
396         BUG_ON(!list_empty(&node->upper));
397
398         drop_node_buffer(node);
399         list_del(&node->list);
400         list_del(&node->lower);
401         if (!RB_EMPTY_NODE(&node->rb_node))
402                 rb_erase(&node->rb_node, &tree->rb_root);
403         free_backref_node(tree, node);
404 }
405
406 /*
407  * remove a backref node from the backref cache
408  */
409 static void remove_backref_node(struct backref_cache *cache,
410                                 struct backref_node *node)
411 {
412         struct backref_node *upper;
413         struct backref_edge *edge;
414
415         if (!node)
416                 return;
417
418         BUG_ON(!node->lowest && !node->detached);
419         while (!list_empty(&node->upper)) {
420                 edge = list_entry(node->upper.next, struct backref_edge,
421                                   list[LOWER]);
422                 upper = edge->node[UPPER];
423                 list_del(&edge->list[LOWER]);
424                 list_del(&edge->list[UPPER]);
425                 free_backref_edge(cache, edge);
426
427                 if (RB_EMPTY_NODE(&upper->rb_node)) {
428                         BUG_ON(!list_empty(&node->upper));
429                         drop_backref_node(cache, node);
430                         node = upper;
431                         node->lowest = 1;
432                         continue;
433                 }
434                 /*
435                  * add the node to leaf node list if no other
436                  * child block cached.
437                  */
438                 if (list_empty(&upper->lower)) {
439                         list_add_tail(&upper->lower, &cache->leaves);
440                         upper->lowest = 1;
441                 }
442         }
443
444         drop_backref_node(cache, node);
445 }
446
447 static void update_backref_node(struct backref_cache *cache,
448                                 struct backref_node *node, u64 bytenr)
449 {
450         struct rb_node *rb_node;
451         rb_erase(&node->rb_node, &cache->rb_root);
452         node->bytenr = bytenr;
453         rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
454         BUG_ON(rb_node);
455 }
456
457 /*
458  * update backref cache after a transaction commit
459  */
460 static int update_backref_cache(struct btrfs_trans_handle *trans,
461                                 struct backref_cache *cache)
462 {
463         struct backref_node *node;
464         int level = 0;
465
466         if (cache->last_trans == 0) {
467                 cache->last_trans = trans->transid;
468                 return 0;
469         }
470
471         if (cache->last_trans == trans->transid)
472                 return 0;
473
474         /*
475          * detached nodes are used to avoid unnecessary backref
476          * lookup. transaction commit changes the extent tree.
477          * so the detached nodes are no longer useful.
478          */
479         while (!list_empty(&cache->detached)) {
480                 node = list_entry(cache->detached.next,
481                                   struct backref_node, list);
482                 remove_backref_node(cache, node);
483         }
484
485         while (!list_empty(&cache->changed)) {
486                 node = list_entry(cache->changed.next,
487                                   struct backref_node, list);
488                 list_del_init(&node->list);
489                 BUG_ON(node->pending);
490                 update_backref_node(cache, node, node->new_bytenr);
491         }
492
493         /*
494          * some nodes can be left in the pending list if there were
495          * errors during processing the pending nodes.
496          */
497         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
498                 list_for_each_entry(node, &cache->pending[level], list) {
499                         BUG_ON(!node->pending);
500                         if (node->bytenr == node->new_bytenr)
501                                 continue;
502                         update_backref_node(cache, node, node->new_bytenr);
503                 }
504         }
505
506         cache->last_trans = 0;
507         return 1;
508 }
509
510 static int should_ignore_root(struct btrfs_root *root)
511 {
512         struct btrfs_root *reloc_root;
513
514         if (!root->ref_cows)
515                 return 0;
516
517         reloc_root = root->reloc_root;
518         if (!reloc_root)
519                 return 0;
520
521         if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
522             root->fs_info->running_transaction->transid - 1)
523                 return 0;
524         /*
525          * if there is reloc tree and it was created in previous
526          * transaction backref lookup can find the reloc tree,
527          * so backref node for the fs tree root is useless for
528          * relocation.
529          */
530         return 1;
531 }
532
533 /*
534  * find reloc tree by address of tree root
535  */
536 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
537                                           u64 bytenr)
538 {
539         struct rb_node *rb_node;
540         struct mapping_node *node;
541         struct btrfs_root *root = NULL;
542
543         spin_lock(&rc->reloc_root_tree.lock);
544         rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
545         if (rb_node) {
546                 node = rb_entry(rb_node, struct mapping_node, rb_node);
547                 root = (struct btrfs_root *)node->data;
548         }
549         spin_unlock(&rc->reloc_root_tree.lock);
550         return root;
551 }
552
553 static int is_cowonly_root(u64 root_objectid)
554 {
555         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
556             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
557             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
558             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
559             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
560             root_objectid == BTRFS_CSUM_TREE_OBJECTID)
561                 return 1;
562         return 0;
563 }
564
565 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
566                                         u64 root_objectid)
567 {
568         struct btrfs_key key;
569
570         key.objectid = root_objectid;
571         key.type = BTRFS_ROOT_ITEM_KEY;
572         if (is_cowonly_root(root_objectid))
573                 key.offset = 0;
574         else
575                 key.offset = (u64)-1;
576
577         return btrfs_read_fs_root_no_name(fs_info, &key);
578 }
579
580 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
581 static noinline_for_stack
582 struct btrfs_root *find_tree_root(struct reloc_control *rc,
583                                   struct extent_buffer *leaf,
584                                   struct btrfs_extent_ref_v0 *ref0)
585 {
586         struct btrfs_root *root;
587         u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
588         u64 generation = btrfs_ref_generation_v0(leaf, ref0);
589
590         BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
591
592         root = read_fs_root(rc->extent_root->fs_info, root_objectid);
593         BUG_ON(IS_ERR(root));
594
595         if (root->ref_cows &&
596             generation != btrfs_root_generation(&root->root_item))
597                 return NULL;
598
599         return root;
600 }
601 #endif
602
603 static noinline_for_stack
604 int find_inline_backref(struct extent_buffer *leaf, int slot,
605                         unsigned long *ptr, unsigned long *end)
606 {
607         struct btrfs_extent_item *ei;
608         struct btrfs_tree_block_info *bi;
609         u32 item_size;
610
611         item_size = btrfs_item_size_nr(leaf, slot);
612 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
613         if (item_size < sizeof(*ei)) {
614                 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
615                 return 1;
616         }
617 #endif
618         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
619         WARN_ON(!(btrfs_extent_flags(leaf, ei) &
620                   BTRFS_EXTENT_FLAG_TREE_BLOCK));
621
622         if (item_size <= sizeof(*ei) + sizeof(*bi)) {
623                 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
624                 return 1;
625         }
626
627         bi = (struct btrfs_tree_block_info *)(ei + 1);
628         *ptr = (unsigned long)(bi + 1);
629         *end = (unsigned long)ei + item_size;
630         return 0;
631 }
632
633 /*
634  * build backref tree for a given tree block. root of the backref tree
635  * corresponds the tree block, leaves of the backref tree correspond
636  * roots of b-trees that reference the tree block.
637  *
638  * the basic idea of this function is check backrefs of a given block
639  * to find upper level blocks that refernece the block, and then check
640  * bakcrefs of these upper level blocks recursively. the recursion stop
641  * when tree root is reached or backrefs for the block is cached.
642  *
643  * NOTE: if we find backrefs for a block are cached, we know backrefs
644  * for all upper level blocks that directly/indirectly reference the
645  * block are also cached.
646  */
647 static noinline_for_stack
648 struct backref_node *build_backref_tree(struct reloc_control *rc,
649                                         struct btrfs_key *node_key,
650                                         int level, u64 bytenr)
651 {
652         struct backref_cache *cache = &rc->backref_cache;
653         struct btrfs_path *path1;
654         struct btrfs_path *path2;
655         struct extent_buffer *eb;
656         struct btrfs_root *root;
657         struct backref_node *cur;
658         struct backref_node *upper;
659         struct backref_node *lower;
660         struct backref_node *node = NULL;
661         struct backref_node *exist = NULL;
662         struct backref_edge *edge;
663         struct rb_node *rb_node;
664         struct btrfs_key key;
665         unsigned long end;
666         unsigned long ptr;
667         LIST_HEAD(list);
668         LIST_HEAD(useless);
669         int cowonly;
670         int ret;
671         int err = 0;
672
673         path1 = btrfs_alloc_path();
674         path2 = btrfs_alloc_path();
675         if (!path1 || !path2) {
676                 err = -ENOMEM;
677                 goto out;
678         }
679
680         node = alloc_backref_node(cache);
681         if (!node) {
682                 err = -ENOMEM;
683                 goto out;
684         }
685
686         node->bytenr = bytenr;
687         node->level = level;
688         node->lowest = 1;
689         cur = node;
690 again:
691         end = 0;
692         ptr = 0;
693         key.objectid = cur->bytenr;
694         key.type = BTRFS_EXTENT_ITEM_KEY;
695         key.offset = (u64)-1;
696
697         path1->search_commit_root = 1;
698         path1->skip_locking = 1;
699         ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
700                                 0, 0);
701         if (ret < 0) {
702                 err = ret;
703                 goto out;
704         }
705         BUG_ON(!ret || !path1->slots[0]);
706
707         path1->slots[0]--;
708
709         WARN_ON(cur->checked);
710         if (!list_empty(&cur->upper)) {
711                 /*
712                  * the backref was added previously when processsing
713                  * backref of type BTRFS_TREE_BLOCK_REF_KEY
714                  */
715                 BUG_ON(!list_is_singular(&cur->upper));
716                 edge = list_entry(cur->upper.next, struct backref_edge,
717                                   list[LOWER]);
718                 BUG_ON(!list_empty(&edge->list[UPPER]));
719                 exist = edge->node[UPPER];
720                 /*
721                  * add the upper level block to pending list if we need
722                  * check its backrefs
723                  */
724                 if (!exist->checked)
725                         list_add_tail(&edge->list[UPPER], &list);
726         } else {
727                 exist = NULL;
728         }
729
730         while (1) {
731                 cond_resched();
732                 eb = path1->nodes[0];
733
734                 if (ptr >= end) {
735                         if (path1->slots[0] >= btrfs_header_nritems(eb)) {
736                                 ret = btrfs_next_leaf(rc->extent_root, path1);
737                                 if (ret < 0) {
738                                         err = ret;
739                                         goto out;
740                                 }
741                                 if (ret > 0)
742                                         break;
743                                 eb = path1->nodes[0];
744                         }
745
746                         btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
747                         if (key.objectid != cur->bytenr) {
748                                 WARN_ON(exist);
749                                 break;
750                         }
751
752                         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
753                                 ret = find_inline_backref(eb, path1->slots[0],
754                                                           &ptr, &end);
755                                 if (ret)
756                                         goto next;
757                         }
758                 }
759
760                 if (ptr < end) {
761                         /* update key for inline back ref */
762                         struct btrfs_extent_inline_ref *iref;
763                         iref = (struct btrfs_extent_inline_ref *)ptr;
764                         key.type = btrfs_extent_inline_ref_type(eb, iref);
765                         key.offset = btrfs_extent_inline_ref_offset(eb, iref);
766                         WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
767                                 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
768                 }
769
770                 if (exist &&
771                     ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
772                       exist->owner == key.offset) ||
773                      (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
774                       exist->bytenr == key.offset))) {
775                         exist = NULL;
776                         goto next;
777                 }
778
779 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
780                 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
781                     key.type == BTRFS_EXTENT_REF_V0_KEY) {
782                         if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
783                                 struct btrfs_extent_ref_v0 *ref0;
784                                 ref0 = btrfs_item_ptr(eb, path1->slots[0],
785                                                 struct btrfs_extent_ref_v0);
786                                 if (key.objectid == key.offset) {
787                                         root = find_tree_root(rc, eb, ref0);
788                                         if (root && !should_ignore_root(root))
789                                                 cur->root = root;
790                                         else
791                                                 list_add(&cur->list, &useless);
792                                         break;
793                                 }
794                                 if (is_cowonly_root(btrfs_ref_root_v0(eb,
795                                                                       ref0)))
796                                         cur->cowonly = 1;
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
2136         mutex_lock(&root->fs_info->trans_mutex);
2137         rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2138         rc->merging_rsv_size += rc->nodes_relocated * 2;
2139         mutex_unlock(&root->fs_info->trans_mutex);
2140 again:
2141         if (!err) {
2142                 num_bytes = rc->merging_rsv_size;
2143                 ret = btrfs_block_rsv_add(NULL, root, rc->block_rsv,
2144                                           num_bytes);
2145                 if (ret)
2146                         err = ret;
2147         }
2148
2149         trans = btrfs_join_transaction(rc->extent_root, 1);
2150         if (IS_ERR(trans)) {
2151                 if (!err)
2152                         btrfs_block_rsv_release(rc->extent_root,
2153                                                 rc->block_rsv, num_bytes);
2154                 return PTR_ERR(trans);
2155         }
2156
2157         if (!err) {
2158                 if (num_bytes != rc->merging_rsv_size) {
2159                         btrfs_end_transaction(trans, rc->extent_root);
2160                         btrfs_block_rsv_release(rc->extent_root,
2161                                                 rc->block_rsv, num_bytes);
2162                         goto again;
2163                 }
2164         }
2165
2166         rc->merge_reloc_tree = 1;
2167
2168         while (!list_empty(&rc->reloc_roots)) {
2169                 reloc_root = list_entry(rc->reloc_roots.next,
2170                                         struct btrfs_root, root_list);
2171                 list_del_init(&reloc_root->root_list);
2172
2173                 root = read_fs_root(reloc_root->fs_info,
2174                                     reloc_root->root_key.offset);
2175                 BUG_ON(IS_ERR(root));
2176                 BUG_ON(root->reloc_root != reloc_root);
2177
2178                 /*
2179                  * set reference count to 1, so btrfs_recover_relocation
2180                  * knows it should resumes merging
2181                  */
2182                 if (!err)
2183                         btrfs_set_root_refs(&reloc_root->root_item, 1);
2184                 btrfs_update_reloc_root(trans, root);
2185
2186                 list_add(&reloc_root->root_list, &reloc_roots);
2187         }
2188
2189         list_splice(&reloc_roots, &rc->reloc_roots);
2190
2191         if (!err)
2192                 btrfs_commit_transaction(trans, rc->extent_root);
2193         else
2194                 btrfs_end_transaction(trans, rc->extent_root);
2195         return err;
2196 }
2197
2198 static noinline_for_stack
2199 int merge_reloc_roots(struct reloc_control *rc)
2200 {
2201         struct btrfs_root *root;
2202         struct btrfs_root *reloc_root;
2203         LIST_HEAD(reloc_roots);
2204         int found = 0;
2205         int ret;
2206 again:
2207         root = rc->extent_root;
2208         mutex_lock(&root->fs_info->trans_mutex);
2209         list_splice_init(&rc->reloc_roots, &reloc_roots);
2210         mutex_unlock(&root->fs_info->trans_mutex);
2211
2212         while (!list_empty(&reloc_roots)) {
2213                 found = 1;
2214                 reloc_root = list_entry(reloc_roots.next,
2215                                         struct btrfs_root, root_list);
2216
2217                 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2218                         root = read_fs_root(reloc_root->fs_info,
2219                                             reloc_root->root_key.offset);
2220                         BUG_ON(IS_ERR(root));
2221                         BUG_ON(root->reloc_root != reloc_root);
2222
2223                         ret = merge_reloc_root(rc, root);
2224                         BUG_ON(ret);
2225                 } else {
2226                         list_del_init(&reloc_root->root_list);
2227                 }
2228                 btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0);
2229         }
2230
2231         if (found) {
2232                 found = 0;
2233                 goto again;
2234         }
2235         BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2236         return 0;
2237 }
2238
2239 static void free_block_list(struct rb_root *blocks)
2240 {
2241         struct tree_block *block;
2242         struct rb_node *rb_node;
2243         while ((rb_node = rb_first(blocks))) {
2244                 block = rb_entry(rb_node, struct tree_block, rb_node);
2245                 rb_erase(rb_node, blocks);
2246                 kfree(block);
2247         }
2248 }
2249
2250 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2251                                       struct btrfs_root *reloc_root)
2252 {
2253         struct btrfs_root *root;
2254
2255         if (reloc_root->last_trans == trans->transid)
2256                 return 0;
2257
2258         root = read_fs_root(reloc_root->fs_info, reloc_root->root_key.offset);
2259         BUG_ON(IS_ERR(root));
2260         BUG_ON(root->reloc_root != reloc_root);
2261
2262         return btrfs_record_root_in_trans(trans, root);
2263 }
2264
2265 static noinline_for_stack
2266 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2267                                      struct reloc_control *rc,
2268                                      struct backref_node *node,
2269                                      struct backref_edge *edges[], int *nr)
2270 {
2271         struct backref_node *next;
2272         struct btrfs_root *root;
2273         int index = 0;
2274
2275         next = node;
2276         while (1) {
2277                 cond_resched();
2278                 next = walk_up_backref(next, edges, &index);
2279                 root = next->root;
2280                 BUG_ON(!root);
2281                 BUG_ON(!root->ref_cows);
2282
2283                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2284                         record_reloc_root_in_trans(trans, root);
2285                         break;
2286                 }
2287
2288                 btrfs_record_root_in_trans(trans, root);
2289                 root = root->reloc_root;
2290
2291                 if (next->new_bytenr != root->node->start) {
2292                         BUG_ON(next->new_bytenr);
2293                         BUG_ON(!list_empty(&next->list));
2294                         next->new_bytenr = root->node->start;
2295                         next->root = root;
2296                         list_add_tail(&next->list,
2297                                       &rc->backref_cache.changed);
2298                         __mark_block_processed(rc, next);
2299                         break;
2300                 }
2301
2302                 WARN_ON(1);
2303                 root = NULL;
2304                 next = walk_down_backref(edges, &index);
2305                 if (!next || next->level <= node->level)
2306                         break;
2307         }
2308         if (!root)
2309                 return NULL;
2310
2311         *nr = index;
2312         next = node;
2313         /* setup backref node path for btrfs_reloc_cow_block */
2314         while (1) {
2315                 rc->backref_cache.path[next->level] = next;
2316                 if (--index < 0)
2317                         break;
2318                 next = edges[index]->node[UPPER];
2319         }
2320         return root;
2321 }
2322
2323 /*
2324  * select a tree root for relocation. return NULL if the block
2325  * is reference counted. we should use do_relocation() in this
2326  * case. return a tree root pointer if the block isn't reference
2327  * counted. return -ENOENT if the block is root of reloc tree.
2328  */
2329 static noinline_for_stack
2330 struct btrfs_root *select_one_root(struct btrfs_trans_handle *trans,
2331                                    struct backref_node *node)
2332 {
2333         struct backref_node *next;
2334         struct btrfs_root *root;
2335         struct btrfs_root *fs_root = NULL;
2336         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2337         int index = 0;
2338
2339         next = node;
2340         while (1) {
2341                 cond_resched();
2342                 next = walk_up_backref(next, edges, &index);
2343                 root = next->root;
2344                 BUG_ON(!root);
2345
2346                 /* no other choice for non-refernce counted tree */
2347                 if (!root->ref_cows)
2348                         return root;
2349
2350                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2351                         fs_root = root;
2352
2353                 if (next != node)
2354                         return NULL;
2355
2356                 next = walk_down_backref(edges, &index);
2357                 if (!next || next->level <= node->level)
2358                         break;
2359         }
2360
2361         if (!fs_root)
2362                 return ERR_PTR(-ENOENT);
2363         return fs_root;
2364 }
2365
2366 static noinline_for_stack
2367 u64 calcu_metadata_size(struct reloc_control *rc,
2368                         struct backref_node *node, int reserve)
2369 {
2370         struct backref_node *next = node;
2371         struct backref_edge *edge;
2372         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2373         u64 num_bytes = 0;
2374         int index = 0;
2375
2376         BUG_ON(reserve && node->processed);
2377
2378         while (next) {
2379                 cond_resched();
2380                 while (1) {
2381                         if (next->processed && (reserve || next != node))
2382                                 break;
2383
2384                         num_bytes += btrfs_level_size(rc->extent_root,
2385                                                       next->level);
2386
2387                         if (list_empty(&next->upper))
2388                                 break;
2389
2390                         edge = list_entry(next->upper.next,
2391                                           struct backref_edge, list[LOWER]);
2392                         edges[index++] = edge;
2393                         next = edge->node[UPPER];
2394                 }
2395                 next = walk_down_backref(edges, &index);
2396         }
2397         return num_bytes;
2398 }
2399
2400 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2401                                   struct reloc_control *rc,
2402                                   struct backref_node *node)
2403 {
2404         struct btrfs_root *root = rc->extent_root;
2405         u64 num_bytes;
2406         int ret;
2407
2408         num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2409
2410         trans->block_rsv = rc->block_rsv;
2411         ret = btrfs_block_rsv_add(trans, root, rc->block_rsv, num_bytes);
2412         if (ret) {
2413                 if (ret == -EAGAIN)
2414                         rc->commit_transaction = 1;
2415                 return ret;
2416         }
2417
2418         return 0;
2419 }
2420
2421 static void release_metadata_space(struct reloc_control *rc,
2422                                    struct backref_node *node)
2423 {
2424         u64 num_bytes = calcu_metadata_size(rc, node, 0) * 2;
2425         btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, num_bytes);
2426 }
2427
2428 /*
2429  * relocate a block tree, and then update pointers in upper level
2430  * blocks that reference the block to point to the new location.
2431  *
2432  * if called by link_to_upper, the block has already been relocated.
2433  * in that case this function just updates pointers.
2434  */
2435 static int do_relocation(struct btrfs_trans_handle *trans,
2436                          struct reloc_control *rc,
2437                          struct backref_node *node,
2438                          struct btrfs_key *key,
2439                          struct btrfs_path *path, int lowest)
2440 {
2441         struct backref_node *upper;
2442         struct backref_edge *edge;
2443         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2444         struct btrfs_root *root;
2445         struct extent_buffer *eb;
2446         u32 blocksize;
2447         u64 bytenr;
2448         u64 generation;
2449         int nr;
2450         int slot;
2451         int ret;
2452         int err = 0;
2453
2454         BUG_ON(lowest && node->eb);
2455
2456         path->lowest_level = node->level + 1;
2457         rc->backref_cache.path[node->level] = node;
2458         list_for_each_entry(edge, &node->upper, list[LOWER]) {
2459                 cond_resched();
2460
2461                 upper = edge->node[UPPER];
2462                 root = select_reloc_root(trans, rc, upper, edges, &nr);
2463                 BUG_ON(!root);
2464
2465                 if (upper->eb && !upper->locked) {
2466                         if (!lowest) {
2467                                 ret = btrfs_bin_search(upper->eb, key,
2468                                                        upper->level, &slot);
2469                                 BUG_ON(ret);
2470                                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2471                                 if (node->eb->start == bytenr)
2472                                         goto next;
2473                         }
2474                         drop_node_buffer(upper);
2475                 }
2476
2477                 if (!upper->eb) {
2478                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2479                         if (ret < 0) {
2480                                 err = ret;
2481                                 break;
2482                         }
2483                         BUG_ON(ret > 0);
2484
2485                         if (!upper->eb) {
2486                                 upper->eb = path->nodes[upper->level];
2487                                 path->nodes[upper->level] = NULL;
2488                         } else {
2489                                 BUG_ON(upper->eb != path->nodes[upper->level]);
2490                         }
2491
2492                         upper->locked = 1;
2493                         path->locks[upper->level] = 0;
2494
2495                         slot = path->slots[upper->level];
2496                         btrfs_release_path(NULL, path);
2497                 } else {
2498                         ret = btrfs_bin_search(upper->eb, key, upper->level,
2499                                                &slot);
2500                         BUG_ON(ret);
2501                 }
2502
2503                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2504                 if (lowest) {
2505                         BUG_ON(bytenr != node->bytenr);
2506                 } else {
2507                         if (node->eb->start == bytenr)
2508                                 goto next;
2509                 }
2510
2511                 blocksize = btrfs_level_size(root, node->level);
2512                 generation = btrfs_node_ptr_generation(upper->eb, slot);
2513                 eb = read_tree_block(root, bytenr, blocksize, generation);
2514                 btrfs_tree_lock(eb);
2515                 btrfs_set_lock_blocking(eb);
2516
2517                 if (!node->eb) {
2518                         ret = btrfs_cow_block(trans, root, eb, upper->eb,
2519                                               slot, &eb);
2520                         btrfs_tree_unlock(eb);
2521                         free_extent_buffer(eb);
2522                         if (ret < 0) {
2523                                 err = ret;
2524                                 goto next;
2525                         }
2526                         BUG_ON(node->eb != eb);
2527                 } else {
2528                         btrfs_set_node_blockptr(upper->eb, slot,
2529                                                 node->eb->start);
2530                         btrfs_set_node_ptr_generation(upper->eb, slot,
2531                                                       trans->transid);
2532                         btrfs_mark_buffer_dirty(upper->eb);
2533
2534                         ret = btrfs_inc_extent_ref(trans, root,
2535                                                 node->eb->start, blocksize,
2536                                                 upper->eb->start,
2537                                                 btrfs_header_owner(upper->eb),
2538                                                 node->level, 0);
2539                         BUG_ON(ret);
2540
2541                         ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2542                         BUG_ON(ret);
2543                 }
2544 next:
2545                 if (!upper->pending)
2546                         drop_node_buffer(upper);
2547                 else
2548                         unlock_node_buffer(upper);
2549                 if (err)
2550                         break;
2551         }
2552
2553         if (!err && node->pending) {
2554                 drop_node_buffer(node);
2555                 list_move_tail(&node->list, &rc->backref_cache.changed);
2556                 node->pending = 0;
2557         }
2558
2559         path->lowest_level = 0;
2560         BUG_ON(err == -ENOSPC);
2561         return err;
2562 }
2563
2564 static int link_to_upper(struct btrfs_trans_handle *trans,
2565                          struct reloc_control *rc,
2566                          struct backref_node *node,
2567                          struct btrfs_path *path)
2568 {
2569         struct btrfs_key key;
2570
2571         btrfs_node_key_to_cpu(node->eb, &key, 0);
2572         return do_relocation(trans, rc, node, &key, path, 0);
2573 }
2574
2575 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2576                                 struct reloc_control *rc,
2577                                 struct btrfs_path *path, int err)
2578 {
2579         LIST_HEAD(list);
2580         struct backref_cache *cache = &rc->backref_cache;
2581         struct backref_node *node;
2582         int level;
2583         int ret;
2584
2585         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2586                 while (!list_empty(&cache->pending[level])) {
2587                         node = list_entry(cache->pending[level].next,
2588                                           struct backref_node, list);
2589                         list_move_tail(&node->list, &list);
2590                         BUG_ON(!node->pending);
2591
2592                         if (!err) {
2593                                 ret = link_to_upper(trans, rc, node, path);
2594                                 if (ret < 0)
2595                                         err = ret;
2596                         }
2597                 }
2598                 list_splice_init(&list, &cache->pending[level]);
2599         }
2600         return err;
2601 }
2602
2603 static void mark_block_processed(struct reloc_control *rc,
2604                                  u64 bytenr, u32 blocksize)
2605 {
2606         set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2607                         EXTENT_DIRTY, GFP_NOFS);
2608 }
2609
2610 static void __mark_block_processed(struct reloc_control *rc,
2611                                    struct backref_node *node)
2612 {
2613         u32 blocksize;
2614         if (node->level == 0 ||
2615             in_block_group(node->bytenr, rc->block_group)) {
2616                 blocksize = btrfs_level_size(rc->extent_root, node->level);
2617                 mark_block_processed(rc, node->bytenr, blocksize);
2618         }
2619         node->processed = 1;
2620 }
2621
2622 /*
2623  * mark a block and all blocks directly/indirectly reference the block
2624  * as processed.
2625  */
2626 static void update_processed_blocks(struct reloc_control *rc,
2627                                     struct backref_node *node)
2628 {
2629         struct backref_node *next = node;
2630         struct backref_edge *edge;
2631         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2632         int index = 0;
2633
2634         while (next) {
2635                 cond_resched();
2636                 while (1) {
2637                         if (next->processed)
2638                                 break;
2639
2640                         __mark_block_processed(rc, next);
2641
2642                         if (list_empty(&next->upper))
2643                                 break;
2644
2645                         edge = list_entry(next->upper.next,
2646                                           struct backref_edge, list[LOWER]);
2647                         edges[index++] = edge;
2648                         next = edge->node[UPPER];
2649                 }
2650                 next = walk_down_backref(edges, &index);
2651         }
2652 }
2653
2654 static int tree_block_processed(u64 bytenr, u32 blocksize,
2655                                 struct reloc_control *rc)
2656 {
2657         if (test_range_bit(&rc->processed_blocks, bytenr,
2658                            bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2659                 return 1;
2660         return 0;
2661 }
2662
2663 static int get_tree_block_key(struct reloc_control *rc,
2664                               struct tree_block *block)
2665 {
2666         struct extent_buffer *eb;
2667
2668         BUG_ON(block->key_ready);
2669         eb = read_tree_block(rc->extent_root, block->bytenr,
2670                              block->key.objectid, block->key.offset);
2671         WARN_ON(btrfs_header_level(eb) != block->level);
2672         if (block->level == 0)
2673                 btrfs_item_key_to_cpu(eb, &block->key, 0);
2674         else
2675                 btrfs_node_key_to_cpu(eb, &block->key, 0);
2676         free_extent_buffer(eb);
2677         block->key_ready = 1;
2678         return 0;
2679 }
2680
2681 static int reada_tree_block(struct reloc_control *rc,
2682                             struct tree_block *block)
2683 {
2684         BUG_ON(block->key_ready);
2685         readahead_tree_block(rc->extent_root, block->bytenr,
2686                              block->key.objectid, block->key.offset);
2687         return 0;
2688 }
2689
2690 /*
2691  * helper function to relocate a tree block
2692  */
2693 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2694                                 struct reloc_control *rc,
2695                                 struct backref_node *node,
2696                                 struct btrfs_key *key,
2697                                 struct btrfs_path *path)
2698 {
2699         struct btrfs_root *root;
2700         int release = 0;
2701         int ret = 0;
2702
2703         if (!node)
2704                 return 0;
2705
2706         BUG_ON(node->processed);
2707         root = select_one_root(trans, node);
2708         if (root == ERR_PTR(-ENOENT)) {
2709                 update_processed_blocks(rc, node);
2710                 goto out;
2711         }
2712
2713         if (!root || root->ref_cows) {
2714                 ret = reserve_metadata_space(trans, rc, node);
2715                 if (ret)
2716                         goto out;
2717                 release = 1;
2718         }
2719
2720         if (root) {
2721                 if (root->ref_cows) {
2722                         BUG_ON(node->new_bytenr);
2723                         BUG_ON(!list_empty(&node->list));
2724                         btrfs_record_root_in_trans(trans, root);
2725                         root = root->reloc_root;
2726                         node->new_bytenr = root->node->start;
2727                         node->root = root;
2728                         list_add_tail(&node->list, &rc->backref_cache.changed);
2729                 } else {
2730                         path->lowest_level = node->level;
2731                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2732                         btrfs_release_path(root, path);
2733                         if (ret > 0)
2734                                 ret = 0;
2735                 }
2736                 if (!ret)
2737                         update_processed_blocks(rc, node);
2738         } else {
2739                 ret = do_relocation(trans, rc, node, key, path, 1);
2740         }
2741 out:
2742         if (ret || node->level == 0 || node->cowonly) {
2743                 if (release)
2744                         release_metadata_space(rc, node);
2745                 remove_backref_node(&rc->backref_cache, node);
2746         }
2747         return ret;
2748 }
2749
2750 /*
2751  * relocate a list of blocks
2752  */
2753 static noinline_for_stack
2754 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2755                          struct reloc_control *rc, struct rb_root *blocks)
2756 {
2757         struct backref_node *node;
2758         struct btrfs_path *path;
2759         struct tree_block *block;
2760         struct rb_node *rb_node;
2761         int ret;
2762         int err = 0;
2763
2764         path = btrfs_alloc_path();
2765         if (!path)
2766                 return -ENOMEM;
2767
2768         rb_node = rb_first(blocks);
2769         while (rb_node) {
2770                 block = rb_entry(rb_node, struct tree_block, rb_node);
2771                 if (!block->key_ready)
2772                         reada_tree_block(rc, block);
2773                 rb_node = rb_next(rb_node);
2774         }
2775
2776         rb_node = rb_first(blocks);
2777         while (rb_node) {
2778                 block = rb_entry(rb_node, struct tree_block, rb_node);
2779                 if (!block->key_ready)
2780                         get_tree_block_key(rc, block);
2781                 rb_node = rb_next(rb_node);
2782         }
2783
2784         rb_node = rb_first(blocks);
2785         while (rb_node) {
2786                 block = rb_entry(rb_node, struct tree_block, rb_node);
2787
2788                 node = build_backref_tree(rc, &block->key,
2789                                           block->level, block->bytenr);
2790                 if (IS_ERR(node)) {
2791                         err = PTR_ERR(node);
2792                         goto out;
2793                 }
2794
2795                 ret = relocate_tree_block(trans, rc, node, &block->key,
2796                                           path);
2797                 if (ret < 0) {
2798                         if (ret != -EAGAIN || rb_node == rb_first(blocks))
2799                                 err = ret;
2800                         goto out;
2801                 }
2802                 rb_node = rb_next(rb_node);
2803         }
2804 out:
2805         free_block_list(blocks);
2806         err = finish_pending_nodes(trans, rc, path, err);
2807
2808         btrfs_free_path(path);
2809         return err;
2810 }
2811
2812 static noinline_for_stack
2813 int prealloc_file_extent_cluster(struct inode *inode,
2814                                  struct file_extent_cluster *cluster)
2815 {
2816         u64 alloc_hint = 0;
2817         u64 start;
2818         u64 end;
2819         u64 offset = BTRFS_I(inode)->index_cnt;
2820         u64 num_bytes;
2821         int nr = 0;
2822         int ret = 0;
2823
2824         BUG_ON(cluster->start != cluster->boundary[0]);
2825         mutex_lock(&inode->i_mutex);
2826
2827         ret = btrfs_check_data_free_space(inode, cluster->end +
2828                                           1 - cluster->start);
2829         if (ret)
2830                 goto out;
2831
2832         while (nr < cluster->nr) {
2833                 start = cluster->boundary[nr] - offset;
2834                 if (nr + 1 < cluster->nr)
2835                         end = cluster->boundary[nr + 1] - 1 - offset;
2836                 else
2837                         end = cluster->end - offset;
2838
2839                 lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2840                 num_bytes = end + 1 - start;
2841                 ret = btrfs_prealloc_file_range(inode, 0, start,
2842                                                 num_bytes, num_bytes,
2843                                                 end + 1, &alloc_hint);
2844                 unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2845                 if (ret)
2846                         break;
2847                 nr++;
2848         }
2849         btrfs_free_reserved_data_space(inode, cluster->end +
2850                                        1 - cluster->start);
2851 out:
2852         mutex_unlock(&inode->i_mutex);
2853         return ret;
2854 }
2855
2856 static noinline_for_stack
2857 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
2858                          u64 block_start)
2859 {
2860         struct btrfs_root *root = BTRFS_I(inode)->root;
2861         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2862         struct extent_map *em;
2863         int ret = 0;
2864
2865         em = alloc_extent_map(GFP_NOFS);
2866         if (!em)
2867                 return -ENOMEM;
2868
2869         em->start = start;
2870         em->len = end + 1 - start;
2871         em->block_len = em->len;
2872         em->block_start = block_start;
2873         em->bdev = root->fs_info->fs_devices->latest_bdev;
2874         set_bit(EXTENT_FLAG_PINNED, &em->flags);
2875
2876         lock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2877         while (1) {
2878                 write_lock(&em_tree->lock);
2879                 ret = add_extent_mapping(em_tree, em);
2880                 write_unlock(&em_tree->lock);
2881                 if (ret != -EEXIST) {
2882                         free_extent_map(em);
2883                         break;
2884                 }
2885                 btrfs_drop_extent_cache(inode, start, end, 0);
2886         }
2887         unlock_extent(&BTRFS_I(inode)->io_tree, start, end, GFP_NOFS);
2888         return ret;
2889 }
2890
2891 static int relocate_file_extent_cluster(struct inode *inode,
2892                                         struct file_extent_cluster *cluster)
2893 {
2894         u64 page_start;
2895         u64 page_end;
2896         u64 offset = BTRFS_I(inode)->index_cnt;
2897         unsigned long index;
2898         unsigned long last_index;
2899         struct page *page;
2900         struct file_ra_state *ra;
2901         int nr = 0;
2902         int ret = 0;
2903
2904         if (!cluster->nr)
2905                 return 0;
2906
2907         ra = kzalloc(sizeof(*ra), GFP_NOFS);
2908         if (!ra)
2909                 return -ENOMEM;
2910
2911         ret = prealloc_file_extent_cluster(inode, cluster);
2912         if (ret)
2913                 goto out;
2914
2915         file_ra_state_init(ra, inode->i_mapping);
2916
2917         ret = setup_extent_mapping(inode, cluster->start - offset,
2918                                    cluster->end - offset, cluster->start);
2919         if (ret)
2920                 goto out;
2921
2922         index = (cluster->start - offset) >> PAGE_CACHE_SHIFT;
2923         last_index = (cluster->end - offset) >> PAGE_CACHE_SHIFT;
2924         while (index <= last_index) {
2925                 ret = btrfs_delalloc_reserve_metadata(inode, PAGE_CACHE_SIZE);
2926                 if (ret)
2927                         goto out;
2928
2929                 page = find_lock_page(inode->i_mapping, index);
2930                 if (!page) {
2931                         page_cache_sync_readahead(inode->i_mapping,
2932                                                   ra, NULL, index,
2933                                                   last_index + 1 - index);
2934                         page = grab_cache_page(inode->i_mapping, index);
2935                         if (!page) {
2936                                 btrfs_delalloc_release_metadata(inode,
2937                                                         PAGE_CACHE_SIZE);
2938                                 ret = -ENOMEM;
2939                                 goto out;
2940                         }
2941                 }
2942
2943                 if (PageReadahead(page)) {
2944                         page_cache_async_readahead(inode->i_mapping,
2945                                                    ra, NULL, page, index,
2946                                                    last_index + 1 - index);
2947                 }
2948
2949                 if (!PageUptodate(page)) {
2950                         btrfs_readpage(NULL, page);
2951                         lock_page(page);
2952                         if (!PageUptodate(page)) {
2953                                 unlock_page(page);
2954                                 page_cache_release(page);
2955                                 btrfs_delalloc_release_metadata(inode,
2956                                                         PAGE_CACHE_SIZE);
2957                                 ret = -EIO;
2958                                 goto out;
2959                         }
2960                 }
2961
2962                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2963                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2964
2965                 lock_extent(&BTRFS_I(inode)->io_tree,
2966                             page_start, page_end, GFP_NOFS);
2967
2968                 set_page_extent_mapped(page);
2969
2970                 if (nr < cluster->nr &&
2971                     page_start + offset == cluster->boundary[nr]) {
2972                         set_extent_bits(&BTRFS_I(inode)->io_tree,
2973                                         page_start, page_end,
2974                                         EXTENT_BOUNDARY, GFP_NOFS);
2975                         nr++;
2976                 }
2977
2978                 btrfs_set_extent_delalloc(inode, page_start, page_end, NULL);
2979                 set_page_dirty(page);
2980
2981                 unlock_extent(&BTRFS_I(inode)->io_tree,
2982                               page_start, page_end, GFP_NOFS);
2983                 unlock_page(page);
2984                 page_cache_release(page);
2985
2986                 index++;
2987                 balance_dirty_pages_ratelimited(inode->i_mapping);
2988                 btrfs_throttle(BTRFS_I(inode)->root);
2989         }
2990         WARN_ON(nr != cluster->nr);
2991 out:
2992         kfree(ra);
2993         return ret;
2994 }
2995
2996 static noinline_for_stack
2997 int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
2998                          struct file_extent_cluster *cluster)
2999 {
3000         int ret;
3001
3002         if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3003                 ret = relocate_file_extent_cluster(inode, cluster);
3004                 if (ret)
3005                         return ret;
3006                 cluster->nr = 0;
3007         }
3008
3009         if (!cluster->nr)
3010                 cluster->start = extent_key->objectid;
3011         else
3012                 BUG_ON(cluster->nr >= MAX_EXTENTS);
3013         cluster->end = extent_key->objectid + extent_key->offset - 1;
3014         cluster->boundary[cluster->nr] = extent_key->objectid;
3015         cluster->nr++;
3016
3017         if (cluster->nr >= MAX_EXTENTS) {
3018                 ret = relocate_file_extent_cluster(inode, cluster);
3019                 if (ret)
3020                         return ret;
3021                 cluster->nr = 0;
3022         }
3023         return 0;
3024 }
3025
3026 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3027 static int get_ref_objectid_v0(struct reloc_control *rc,
3028                                struct btrfs_path *path,
3029                                struct btrfs_key *extent_key,
3030                                u64 *ref_objectid, int *path_change)
3031 {
3032         struct btrfs_key key;
3033         struct extent_buffer *leaf;
3034         struct btrfs_extent_ref_v0 *ref0;
3035         int ret;
3036         int slot;
3037
3038         leaf = path->nodes[0];
3039         slot = path->slots[0];
3040         while (1) {
3041                 if (slot >= btrfs_header_nritems(leaf)) {
3042                         ret = btrfs_next_leaf(rc->extent_root, path);
3043                         if (ret < 0)
3044                                 return ret;
3045                         BUG_ON(ret > 0);
3046                         leaf = path->nodes[0];
3047                         slot = path->slots[0];
3048                         if (path_change)
3049                                 *path_change = 1;
3050                 }
3051                 btrfs_item_key_to_cpu(leaf, &key, slot);
3052                 if (key.objectid != extent_key->objectid)
3053                         return -ENOENT;
3054
3055                 if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3056                         slot++;
3057                         continue;
3058                 }
3059                 ref0 = btrfs_item_ptr(leaf, slot,
3060                                 struct btrfs_extent_ref_v0);
3061                 *ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3062                 break;
3063         }
3064         return 0;
3065 }
3066 #endif
3067
3068 /*
3069  * helper to add a tree block to the list.
3070  * the major work is getting the generation and level of the block
3071  */
3072 static int add_tree_block(struct reloc_control *rc,
3073                           struct btrfs_key *extent_key,
3074                           struct btrfs_path *path,
3075                           struct rb_root *blocks)
3076 {
3077         struct extent_buffer *eb;
3078         struct btrfs_extent_item *ei;
3079         struct btrfs_tree_block_info *bi;
3080         struct tree_block *block;
3081         struct rb_node *rb_node;
3082         u32 item_size;
3083         int level = -1;
3084         int generation;
3085
3086         eb =  path->nodes[0];
3087         item_size = btrfs_item_size_nr(eb, path->slots[0]);
3088
3089         if (item_size >= sizeof(*ei) + sizeof(*bi)) {
3090                 ei = btrfs_item_ptr(eb, path->slots[0],
3091                                 struct btrfs_extent_item);
3092                 bi = (struct btrfs_tree_block_info *)(ei + 1);
3093                 generation = btrfs_extent_generation(eb, ei);
3094                 level = btrfs_tree_block_level(eb, bi);
3095         } else {
3096 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3097                 u64 ref_owner;
3098                 int ret;
3099
3100                 BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3101                 ret = get_ref_objectid_v0(rc, path, extent_key,
3102                                           &ref_owner, NULL);
3103                 if (ret < 0)
3104                         return ret;
3105                 BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3106                 level = (int)ref_owner;
3107                 /* FIXME: get real generation */
3108                 generation = 0;
3109 #else
3110                 BUG();
3111 #endif
3112         }
3113
3114         btrfs_release_path(rc->extent_root, path);
3115
3116         BUG_ON(level == -1);
3117
3118         block = kmalloc(sizeof(*block), GFP_NOFS);
3119         if (!block)
3120                 return -ENOMEM;
3121
3122         block->bytenr = extent_key->objectid;
3123         block->key.objectid = extent_key->offset;
3124         block->key.offset = generation;
3125         block->level = level;
3126         block->key_ready = 0;
3127
3128         rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3129         BUG_ON(rb_node);
3130
3131         return 0;
3132 }
3133
3134 /*
3135  * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3136  */
3137 static int __add_tree_block(struct reloc_control *rc,
3138                             u64 bytenr, u32 blocksize,
3139                             struct rb_root *blocks)
3140 {
3141         struct btrfs_path *path;
3142         struct btrfs_key key;
3143         int ret;
3144
3145         if (tree_block_processed(bytenr, blocksize, rc))
3146                 return 0;
3147
3148         if (tree_search(blocks, bytenr))
3149                 return 0;
3150
3151         path = btrfs_alloc_path();
3152         if (!path)
3153                 return -ENOMEM;
3154
3155         key.objectid = bytenr;
3156         key.type = BTRFS_EXTENT_ITEM_KEY;
3157         key.offset = blocksize;
3158
3159         path->search_commit_root = 1;
3160         path->skip_locking = 1;
3161         ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3162         if (ret < 0)
3163                 goto out;
3164         BUG_ON(ret);
3165
3166         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3167         ret = add_tree_block(rc, &key, path, blocks);
3168 out:
3169         btrfs_free_path(path);
3170         return ret;
3171 }
3172
3173 /*
3174  * helper to check if the block use full backrefs for pointers in it
3175  */
3176 static int block_use_full_backref(struct reloc_control *rc,
3177                                   struct extent_buffer *eb)
3178 {
3179         u64 flags;
3180         int ret;
3181
3182         if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3183             btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3184                 return 1;
3185
3186         ret = btrfs_lookup_extent_info(NULL, rc->extent_root,
3187                                        eb->start, eb->len, NULL, &flags);
3188         BUG_ON(ret);
3189
3190         if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3191                 ret = 1;
3192         else
3193                 ret = 0;
3194         return ret;
3195 }
3196
3197 static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3198                                     struct inode *inode, u64 ino)
3199 {
3200         struct btrfs_key key;
3201         struct btrfs_path *path;
3202         struct btrfs_root *root = fs_info->tree_root;
3203         struct btrfs_trans_handle *trans;
3204         unsigned long nr;
3205         int ret = 0;
3206
3207         if (inode)
3208                 goto truncate;
3209
3210         key.objectid = ino;
3211         key.type = BTRFS_INODE_ITEM_KEY;
3212         key.offset = 0;
3213
3214         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3215         if (!inode || IS_ERR(inode) || is_bad_inode(inode)) {
3216                 if (inode && !IS_ERR(inode))
3217                         iput(inode);
3218                 return -ENOENT;
3219         }
3220
3221 truncate:
3222         path = btrfs_alloc_path();
3223         if (!path) {
3224                 ret = -ENOMEM;
3225                 goto out;
3226         }
3227
3228         trans = btrfs_join_transaction(root, 0);
3229         if (IS_ERR(trans)) {
3230                 btrfs_free_path(path);
3231                 ret = PTR_ERR(trans);
3232                 goto out;
3233         }
3234
3235         ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
3236
3237         btrfs_free_path(path);
3238         nr = trans->blocks_used;
3239         btrfs_end_transaction(trans, root);
3240         btrfs_btree_balance_dirty(root, nr);
3241 out:
3242         iput(inode);
3243         return ret;
3244 }
3245
3246 /*
3247  * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3248  * this function scans fs tree to find blocks reference the data extent
3249  */
3250 static int find_data_references(struct reloc_control *rc,
3251                                 struct btrfs_key *extent_key,
3252                                 struct extent_buffer *leaf,
3253                                 struct btrfs_extent_data_ref *ref,
3254                                 struct rb_root *blocks)
3255 {
3256         struct btrfs_path *path;
3257         struct tree_block *block;
3258         struct btrfs_root *root;
3259         struct btrfs_file_extent_item *fi;
3260         struct rb_node *rb_node;
3261         struct btrfs_key key;
3262         u64 ref_root;
3263         u64 ref_objectid;
3264         u64 ref_offset;
3265         u32 ref_count;
3266         u32 nritems;
3267         int err = 0;
3268         int added = 0;
3269         int counted;
3270         int ret;
3271
3272         ref_root = btrfs_extent_data_ref_root(leaf, ref);
3273         ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3274         ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3275         ref_count = btrfs_extent_data_ref_count(leaf, ref);
3276
3277         /*
3278          * This is an extent belonging to the free space cache, lets just delete
3279          * it and redo the search.
3280          */
3281         if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3282                 ret = delete_block_group_cache(rc->extent_root->fs_info,
3283                                                NULL, ref_objectid);
3284                 if (ret != -ENOENT)
3285                         return ret;
3286                 ret = 0;
3287         }
3288
3289         path = btrfs_alloc_path();
3290         if (!path)
3291                 return -ENOMEM;
3292
3293         root = read_fs_root(rc->extent_root->fs_info, ref_root);
3294         if (IS_ERR(root)) {
3295                 err = PTR_ERR(root);
3296                 goto out;
3297         }
3298
3299         key.objectid = ref_objectid;
3300         key.offset = ref_offset;
3301         key.type = BTRFS_EXTENT_DATA_KEY;
3302
3303         path->search_commit_root = 1;
3304         path->skip_locking = 1;
3305         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3306         if (ret < 0) {
3307                 err = ret;
3308                 goto out;
3309         }
3310
3311         leaf = path->nodes[0];
3312         nritems = btrfs_header_nritems(leaf);
3313         /*
3314          * the references in tree blocks that use full backrefs
3315          * are not counted in
3316          */
3317         if (block_use_full_backref(rc, leaf))
3318                 counted = 0;
3319         else
3320                 counted = 1;
3321         rb_node = tree_search(blocks, leaf->start);
3322         if (rb_node) {
3323                 if (counted)
3324                         added = 1;
3325                 else
3326                         path->slots[0] = nritems;
3327         }
3328
3329         while (ref_count > 0) {
3330                 while (path->slots[0] >= nritems) {
3331                         ret = btrfs_next_leaf(root, path);
3332                         if (ret < 0) {
3333                                 err = ret;
3334                                 goto out;
3335                         }
3336                         if (ret > 0) {
3337                                 WARN_ON(1);
3338                                 goto out;
3339                         }
3340
3341                         leaf = path->nodes[0];
3342                         nritems = btrfs_header_nritems(leaf);
3343                         added = 0;
3344
3345                         if (block_use_full_backref(rc, leaf))
3346                                 counted = 0;
3347                         else
3348                                 counted = 1;
3349                         rb_node = tree_search(blocks, leaf->start);
3350                         if (rb_node) {
3351                                 if (counted)
3352                                         added = 1;
3353                                 else
3354                                         path->slots[0] = nritems;
3355                         }
3356                 }
3357
3358                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3359                 if (key.objectid != ref_objectid ||
3360                     key.type != BTRFS_EXTENT_DATA_KEY) {
3361                         WARN_ON(1);
3362                         break;
3363                 }
3364
3365                 fi = btrfs_item_ptr(leaf, path->slots[0],
3366                                     struct btrfs_file_extent_item);
3367
3368                 if (btrfs_file_extent_type(leaf, fi) ==
3369                     BTRFS_FILE_EXTENT_INLINE)
3370                         goto next;
3371
3372                 if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3373                     extent_key->objectid)
3374                         goto next;
3375
3376                 key.offset -= btrfs_file_extent_offset(leaf, fi);
3377                 if (key.offset != ref_offset)
3378                         goto next;
3379
3380                 if (counted)
3381                         ref_count--;
3382                 if (added)
3383                         goto next;
3384
3385                 if (!tree_block_processed(leaf->start, leaf->len, rc)) {
3386                         block = kmalloc(sizeof(*block), GFP_NOFS);
3387                         if (!block) {
3388                                 err = -ENOMEM;
3389                                 break;
3390                         }
3391                         block->bytenr = leaf->start;
3392                         btrfs_item_key_to_cpu(leaf, &block->key, 0);
3393                         block->level = 0;
3394                         block->key_ready = 1;
3395                         rb_node = tree_insert(blocks, block->bytenr,
3396                                               &block->rb_node);
3397                         BUG_ON(rb_node);
3398                 }
3399                 if (counted)
3400                         added = 1;
3401                 else
3402                         path->slots[0] = nritems;
3403 next:
3404                 path->slots[0]++;
3405
3406         }
3407 out:
3408         btrfs_free_path(path);
3409         return err;
3410 }
3411
3412 /*
3413  * hepler to find all tree blocks that reference a given data extent
3414  */
3415 static noinline_for_stack
3416 int add_data_references(struct reloc_control *rc,
3417                         struct btrfs_key *extent_key,
3418                         struct btrfs_path *path,
3419                         struct rb_root *blocks)
3420 {
3421         struct btrfs_key key;
3422         struct extent_buffer *eb;
3423         struct btrfs_extent_data_ref *dref;
3424         struct btrfs_extent_inline_ref *iref;
3425         unsigned long ptr;
3426         unsigned long end;
3427         u32 blocksize = btrfs_level_size(rc->extent_root, 0);
3428         int ret;
3429         int err = 0;
3430
3431         eb = path->nodes[0];
3432         ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3433         end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3434 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3435         if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3436                 ptr = end;
3437         else
3438 #endif
3439                 ptr += sizeof(struct btrfs_extent_item);
3440
3441         while (ptr < end) {
3442                 iref = (struct btrfs_extent_inline_ref *)ptr;
3443                 key.type = btrfs_extent_inline_ref_type(eb, iref);
3444                 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3445                         key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3446                         ret = __add_tree_block(rc, key.offset, blocksize,
3447                                                blocks);
3448                 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3449                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3450                         ret = find_data_references(rc, extent_key,
3451                                                    eb, dref, blocks);
3452                 } else {
3453                         BUG();
3454                 }
3455                 ptr += btrfs_extent_inline_ref_size(key.type);
3456         }
3457         WARN_ON(ptr > end);
3458
3459         while (1) {
3460                 cond_resched();
3461                 eb = path->nodes[0];
3462                 if (path->slots[0] >= btrfs_header_nritems(eb)) {
3463                         ret = btrfs_next_leaf(rc->extent_root, path);
3464                         if (ret < 0) {
3465                                 err = ret;
3466                                 break;
3467                         }
3468                         if (ret > 0)
3469                                 break;
3470                         eb = path->nodes[0];
3471                 }
3472
3473                 btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3474                 if (key.objectid != extent_key->objectid)
3475                         break;
3476
3477 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3478                 if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3479                     key.type == BTRFS_EXTENT_REF_V0_KEY) {
3480 #else
3481                 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3482                 if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3483 #endif
3484                         ret = __add_tree_block(rc, key.offset, blocksize,
3485                                                blocks);
3486                 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3487                         dref = btrfs_item_ptr(eb, path->slots[0],
3488                                               struct btrfs_extent_data_ref);
3489                         ret = find_data_references(rc, extent_key,
3490                                                    eb, dref, blocks);
3491                 } else {
3492                         ret = 0;
3493                 }
3494                 if (ret) {
3495                         err = ret;
3496                         break;
3497                 }
3498                 path->slots[0]++;
3499         }
3500         btrfs_release_path(rc->extent_root, path);
3501         if (err)
3502                 free_block_list(blocks);
3503         return err;
3504 }
3505
3506 /*
3507  * hepler to find next unprocessed extent
3508  */
3509 static noinline_for_stack
3510 int find_next_extent(struct btrfs_trans_handle *trans,
3511                      struct reloc_control *rc, struct btrfs_path *path,
3512                      struct btrfs_key *extent_key)
3513 {
3514         struct btrfs_key key;
3515         struct extent_buffer *leaf;
3516         u64 start, end, last;
3517         int ret;
3518
3519         last = rc->block_group->key.objectid + rc->block_group->key.offset;
3520         while (1) {
3521                 cond_resched();
3522                 if (rc->search_start >= last) {
3523                         ret = 1;
3524                         break;
3525                 }
3526
3527                 key.objectid = rc->search_start;
3528                 key.type = BTRFS_EXTENT_ITEM_KEY;
3529                 key.offset = 0;
3530
3531                 path->search_commit_root = 1;
3532                 path->skip_locking = 1;
3533                 ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3534                                         0, 0);
3535                 if (ret < 0)
3536                         break;
3537 next:
3538                 leaf = path->nodes[0];
3539                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3540                         ret = btrfs_next_leaf(rc->extent_root, path);
3541                         if (ret != 0)
3542                                 break;
3543                         leaf = path->nodes[0];
3544                 }
3545
3546                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3547                 if (key.objectid >= last) {
3548                         ret = 1;
3549                         break;
3550                 }
3551
3552                 if (key.type != BTRFS_EXTENT_ITEM_KEY ||
3553                     key.objectid + key.offset <= rc->search_start) {
3554                         path->slots[0]++;
3555                         goto next;
3556                 }
3557
3558                 ret = find_first_extent_bit(&rc->processed_blocks,
3559                                             key.objectid, &start, &end,
3560                                             EXTENT_DIRTY);
3561
3562                 if (ret == 0 && start <= key.objectid) {
3563                         btrfs_release_path(rc->extent_root, path);
3564                         rc->search_start = end + 1;
3565                 } else {
3566                         rc->search_start = key.objectid + key.offset;
3567                         memcpy(extent_key, &key, sizeof(key));
3568                         return 0;
3569                 }
3570         }
3571         btrfs_release_path(rc->extent_root, path);
3572         return ret;
3573 }
3574
3575 static void set_reloc_control(struct reloc_control *rc)
3576 {
3577         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3578         mutex_lock(&fs_info->trans_mutex);
3579         fs_info->reloc_ctl = rc;
3580         mutex_unlock(&fs_info->trans_mutex);
3581 }
3582
3583 static void unset_reloc_control(struct reloc_control *rc)
3584 {
3585         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3586         mutex_lock(&fs_info->trans_mutex);
3587         fs_info->reloc_ctl = NULL;
3588         mutex_unlock(&fs_info->trans_mutex);
3589 }
3590
3591 static int check_extent_flags(u64 flags)
3592 {
3593         if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3594             (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3595                 return 1;
3596         if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3597             !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3598                 return 1;
3599         if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3600             (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3601                 return 1;
3602         return 0;
3603 }
3604
3605 static noinline_for_stack
3606 int prepare_to_relocate(struct reloc_control *rc)
3607 {
3608         struct btrfs_trans_handle *trans;
3609         int ret;
3610
3611         rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root);
3612         if (!rc->block_rsv)
3613                 return -ENOMEM;
3614
3615         /*
3616          * reserve some space for creating reloc trees.
3617          * btrfs_init_reloc_root will use them when there
3618          * is no reservation in transaction handle.
3619          */
3620         ret = btrfs_block_rsv_add(NULL, rc->extent_root, rc->block_rsv,
3621                                   rc->extent_root->nodesize * 256);
3622         if (ret)
3623                 return ret;
3624
3625         rc->block_rsv->refill_used = 1;
3626         btrfs_add_durable_block_rsv(rc->extent_root->fs_info, rc->block_rsv);
3627
3628         memset(&rc->cluster, 0, sizeof(rc->cluster));
3629         rc->search_start = rc->block_group->key.objectid;
3630         rc->extents_found = 0;
3631         rc->nodes_relocated = 0;
3632         rc->merging_rsv_size = 0;
3633
3634         rc->create_reloc_tree = 1;
3635         set_reloc_control(rc);
3636
3637         trans = btrfs_join_transaction(rc->extent_root, 1);
3638         BUG_ON(IS_ERR(trans));
3639         btrfs_commit_transaction(trans, rc->extent_root);
3640         return 0;
3641 }
3642
3643 static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3644 {
3645         struct rb_root blocks = RB_ROOT;
3646         struct btrfs_key key;
3647         struct btrfs_trans_handle *trans = NULL;
3648         struct btrfs_path *path;
3649         struct btrfs_extent_item *ei;
3650         unsigned long nr;
3651         u64 flags;
3652         u32 item_size;
3653         int ret;
3654         int err = 0;
3655
3656         path = btrfs_alloc_path();
3657         if (!path)
3658                 return -ENOMEM;
3659
3660         ret = prepare_to_relocate(rc);
3661         if (ret) {
3662                 err = ret;
3663                 goto out_free;
3664         }
3665
3666         while (1) {
3667                 trans = btrfs_start_transaction(rc->extent_root, 0);
3668
3669                 if (update_backref_cache(trans, &rc->backref_cache)) {
3670                         btrfs_end_transaction(trans, rc->extent_root);
3671                         continue;
3672                 }
3673
3674                 ret = find_next_extent(trans, rc, path, &key);
3675                 if (ret < 0)
3676                         err = ret;
3677                 if (ret != 0)
3678                         break;
3679
3680                 rc->extents_found++;
3681
3682                 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3683                                     struct btrfs_extent_item);
3684                 item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
3685                 if (item_size >= sizeof(*ei)) {
3686                         flags = btrfs_extent_flags(path->nodes[0], ei);
3687                         ret = check_extent_flags(flags);
3688                         BUG_ON(ret);
3689
3690                 } else {
3691 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3692                         u64 ref_owner;
3693                         int path_change = 0;
3694
3695                         BUG_ON(item_size !=
3696                                sizeof(struct btrfs_extent_item_v0));
3697                         ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
3698                                                   &path_change);
3699                         if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
3700                                 flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
3701                         else
3702                                 flags = BTRFS_EXTENT_FLAG_DATA;
3703
3704                         if (path_change) {
3705                                 btrfs_release_path(rc->extent_root, path);
3706
3707                                 path->search_commit_root = 1;
3708                                 path->skip_locking = 1;
3709                                 ret = btrfs_search_slot(NULL, rc->extent_root,
3710                                                         &key, path, 0, 0);
3711                                 if (ret < 0) {
3712                                         err = ret;
3713                                         break;
3714                                 }
3715                                 BUG_ON(ret > 0);
3716                         }
3717 #else
3718                         BUG();
3719 #endif
3720                 }
3721
3722                 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3723                         ret = add_tree_block(rc, &key, path, &blocks);
3724                 } else if (rc->stage == UPDATE_DATA_PTRS &&
3725                            (flags & BTRFS_EXTENT_FLAG_DATA)) {
3726                         ret = add_data_references(rc, &key, path, &blocks);
3727                 } else {
3728                         btrfs_release_path(rc->extent_root, path);
3729                         ret = 0;
3730                 }
3731                 if (ret < 0) {
3732                         err = ret;
3733                         break;
3734                 }
3735
3736                 if (!RB_EMPTY_ROOT(&blocks)) {
3737                         ret = relocate_tree_blocks(trans, rc, &blocks);
3738                         if (ret < 0) {
3739                                 if (ret != -EAGAIN) {
3740                                         err = ret;
3741                                         break;
3742                                 }
3743                                 rc->extents_found--;
3744                                 rc->search_start = key.objectid;
3745                         }
3746                 }
3747
3748                 ret = btrfs_block_rsv_check(trans, rc->extent_root,
3749                                             rc->block_rsv, 0, 5);
3750                 if (ret < 0) {
3751                         if (ret != -EAGAIN) {
3752                                 err = ret;
3753                                 WARN_ON(1);
3754                                 break;
3755                         }
3756                         rc->commit_transaction = 1;
3757                 }
3758
3759                 if (rc->commit_transaction) {
3760                         rc->commit_transaction = 0;
3761                         ret = btrfs_commit_transaction(trans, rc->extent_root);
3762                         BUG_ON(ret);
3763                 } else {
3764                         nr = trans->blocks_used;
3765                         btrfs_end_transaction_throttle(trans, rc->extent_root);
3766                         btrfs_btree_balance_dirty(rc->extent_root, nr);
3767                 }
3768                 trans = NULL;
3769
3770                 if (rc->stage == MOVE_DATA_EXTENTS &&
3771                     (flags & BTRFS_EXTENT_FLAG_DATA)) {
3772                         rc->found_file_extent = 1;
3773                         ret = relocate_data_extent(rc->data_inode,
3774                                                    &key, &rc->cluster);
3775                         if (ret < 0) {
3776                                 err = ret;
3777                                 break;
3778                         }
3779                 }
3780         }
3781
3782         btrfs_release_path(rc->extent_root, path);
3783         clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY,
3784                           GFP_NOFS);
3785
3786         if (trans) {
3787                 nr = trans->blocks_used;
3788                 btrfs_end_transaction_throttle(trans, rc->extent_root);
3789                 btrfs_btree_balance_dirty(rc->extent_root, nr);
3790         }
3791
3792         if (!err) {
3793                 ret = relocate_file_extent_cluster(rc->data_inode,
3794                                                    &rc->cluster);
3795                 if (ret < 0)
3796                         err = ret;
3797         }
3798
3799         rc->create_reloc_tree = 0;
3800         set_reloc_control(rc);
3801
3802         backref_cache_cleanup(&rc->backref_cache);
3803         btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3804
3805         err = prepare_to_merge(rc, err);
3806
3807         merge_reloc_roots(rc);
3808
3809         rc->merge_reloc_tree = 0;
3810         unset_reloc_control(rc);
3811         btrfs_block_rsv_release(rc->extent_root, rc->block_rsv, (u64)-1);
3812
3813         /* get rid of pinned extents */
3814         trans = btrfs_join_transaction(rc->extent_root, 1);
3815         if (IS_ERR(trans))
3816                 err = PTR_ERR(trans);
3817         else
3818                 btrfs_commit_transaction(trans, rc->extent_root);
3819 out_free:
3820         btrfs_free_block_rsv(rc->extent_root, rc->block_rsv);
3821         btrfs_free_path(path);
3822         return err;
3823 }
3824
3825 static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3826                                  struct btrfs_root *root, u64 objectid)
3827 {
3828         struct btrfs_path *path;
3829         struct btrfs_inode_item *item;
3830         struct extent_buffer *leaf;
3831         int ret;
3832
3833         path = btrfs_alloc_path();
3834         if (!path)
3835                 return -ENOMEM;
3836
3837         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3838         if (ret)
3839                 goto out;
3840
3841         leaf = path->nodes[0];
3842         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3843         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3844         btrfs_set_inode_generation(leaf, item, 1);
3845         btrfs_set_inode_size(leaf, item, 0);
3846         btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3847         btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3848                                           BTRFS_INODE_PREALLOC);
3849         btrfs_mark_buffer_dirty(leaf);
3850         btrfs_release_path(root, path);
3851 out:
3852         btrfs_free_path(path);
3853         return ret;
3854 }
3855
3856 /*
3857  * helper to create inode for data relocation.
3858  * the inode is in data relocation tree and its link count is 0
3859  */
3860 static noinline_for_stack
3861 struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3862                                  struct btrfs_block_group_cache *group)
3863 {
3864         struct inode *inode = NULL;
3865         struct btrfs_trans_handle *trans;
3866         struct btrfs_root *root;
3867         struct btrfs_key key;
3868         unsigned long nr;
3869         u64 objectid = BTRFS_FIRST_FREE_OBJECTID;
3870         int err = 0;
3871
3872         root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
3873         if (IS_ERR(root))
3874                 return ERR_CAST(root);
3875
3876         trans = btrfs_start_transaction(root, 6);
3877         if (IS_ERR(trans))
3878                 return ERR_CAST(trans);
3879
3880         err = btrfs_find_free_objectid(trans, root, objectid, &objectid);
3881         if (err)
3882                 goto out;
3883
3884         err = __insert_orphan_inode(trans, root, objectid);
3885         BUG_ON(err);
3886
3887         key.objectid = objectid;
3888         key.type = BTRFS_INODE_ITEM_KEY;
3889         key.offset = 0;
3890         inode = btrfs_iget(root->fs_info->sb, &key, root, NULL);
3891         BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
3892         BTRFS_I(inode)->index_cnt = group->key.objectid;
3893
3894         err = btrfs_orphan_add(trans, inode);
3895 out:
3896         nr = trans->blocks_used;
3897         btrfs_end_transaction(trans, root);
3898         btrfs_btree_balance_dirty(root, nr);
3899         if (err) {
3900                 if (inode)
3901                         iput(inode);
3902                 inode = ERR_PTR(err);
3903         }
3904         return inode;
3905 }
3906
3907 static struct reloc_control *alloc_reloc_control(void)
3908 {
3909         struct reloc_control *rc;
3910
3911         rc = kzalloc(sizeof(*rc), GFP_NOFS);
3912         if (!rc)
3913                 return NULL;
3914
3915         INIT_LIST_HEAD(&rc->reloc_roots);
3916         backref_cache_init(&rc->backref_cache);
3917         mapping_tree_init(&rc->reloc_root_tree);
3918         extent_io_tree_init(&rc->processed_blocks, NULL, GFP_NOFS);
3919         return rc;
3920 }
3921
3922 /*
3923  * function to relocate all extents in a block group.
3924  */
3925 int btrfs_relocate_block_group(struct btrfs_root *extent_root, u64 group_start)
3926 {
3927         struct btrfs_fs_info *fs_info = extent_root->fs_info;
3928         struct reloc_control *rc;
3929         struct inode *inode;
3930         struct btrfs_path *path;
3931         int ret;
3932         int rw = 0;
3933         int err = 0;
3934
3935         rc = alloc_reloc_control();
3936         if (!rc)
3937                 return -ENOMEM;
3938
3939         rc->extent_root = extent_root;
3940
3941         rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
3942         BUG_ON(!rc->block_group);
3943
3944         if (!rc->block_group->ro) {
3945                 ret = btrfs_set_block_group_ro(extent_root, rc->block_group);
3946                 if (ret) {
3947                         err = ret;
3948                         goto out;
3949                 }
3950                 rw = 1;
3951         }
3952
3953         path = btrfs_alloc_path();
3954         if (!path) {
3955                 err = -ENOMEM;
3956                 goto out;
3957         }
3958
3959         inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
3960                                         path);
3961         btrfs_free_path(path);
3962
3963         if (!IS_ERR(inode))
3964                 ret = delete_block_group_cache(fs_info, inode, 0);
3965         else
3966                 ret = PTR_ERR(inode);
3967
3968         if (ret && ret != -ENOENT) {
3969                 err = ret;
3970                 goto out;
3971         }
3972
3973         rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
3974         if (IS_ERR(rc->data_inode)) {
3975                 err = PTR_ERR(rc->data_inode);
3976                 rc->data_inode = NULL;
3977                 goto out;
3978         }
3979
3980         printk(KERN_INFO "btrfs: relocating block group %llu flags %llu\n",
3981                (unsigned long long)rc->block_group->key.objectid,
3982                (unsigned long long)rc->block_group->flags);
3983
3984         btrfs_start_delalloc_inodes(fs_info->tree_root, 0);
3985         btrfs_wait_ordered_extents(fs_info->tree_root, 0, 0);
3986
3987         while (1) {
3988                 mutex_lock(&fs_info->cleaner_mutex);
3989
3990                 btrfs_clean_old_snapshots(fs_info->tree_root);
3991                 ret = relocate_block_group(rc);
3992
3993                 mutex_unlock(&fs_info->cleaner_mutex);
3994                 if (ret < 0) {
3995                         err = ret;
3996                         goto out;
3997                 }
3998
3999                 if (rc->extents_found == 0)
4000                         break;
4001
4002                 printk(KERN_INFO "btrfs: found %llu extents\n",
4003                         (unsigned long long)rc->extents_found);
4004
4005                 if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4006                         btrfs_wait_ordered_range(rc->data_inode, 0, (u64)-1);
4007                         invalidate_mapping_pages(rc->data_inode->i_mapping,
4008                                                  0, -1);
4009                         rc->stage = UPDATE_DATA_PTRS;
4010                 }
4011         }
4012
4013         filemap_write_and_wait_range(fs_info->btree_inode->i_mapping,
4014                                      rc->block_group->key.objectid,
4015                                      rc->block_group->key.objectid +
4016                                      rc->block_group->key.offset - 1);
4017
4018         WARN_ON(rc->block_group->pinned > 0);
4019         WARN_ON(rc->block_group->reserved > 0);
4020         WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4021 out:
4022         if (err && rw)
4023                 btrfs_set_block_group_rw(extent_root, rc->block_group);
4024         iput(rc->data_inode);
4025         btrfs_put_block_group(rc->block_group);
4026         kfree(rc);
4027         return err;
4028 }
4029
4030 static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4031 {
4032         struct btrfs_trans_handle *trans;
4033         int ret;
4034
4035         trans = btrfs_start_transaction(root->fs_info->tree_root, 0);
4036
4037         memset(&root->root_item.drop_progress, 0,
4038                 sizeof(root->root_item.drop_progress));
4039         root->root_item.drop_level = 0;
4040         btrfs_set_root_refs(&root->root_item, 0);
4041         ret = btrfs_update_root(trans, root->fs_info->tree_root,
4042                                 &root->root_key, &root->root_item);
4043         BUG_ON(ret);
4044
4045         ret = btrfs_end_transaction(trans, root->fs_info->tree_root);
4046         BUG_ON(ret);
4047         return 0;
4048 }
4049
4050 /*
4051  * recover relocation interrupted by system crash.
4052  *
4053  * this function resumes merging reloc trees with corresponding fs trees.
4054  * this is important for keeping the sharing of tree blocks
4055  */
4056 int btrfs_recover_relocation(struct btrfs_root *root)
4057 {
4058         LIST_HEAD(reloc_roots);
4059         struct btrfs_key key;
4060         struct btrfs_root *fs_root;
4061         struct btrfs_root *reloc_root;
4062         struct btrfs_path *path;
4063         struct extent_buffer *leaf;
4064         struct reloc_control *rc = NULL;
4065         struct btrfs_trans_handle *trans;
4066         int ret;
4067         int err = 0;
4068
4069         path = btrfs_alloc_path();
4070         if (!path)
4071                 return -ENOMEM;
4072
4073         key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4074         key.type = BTRFS_ROOT_ITEM_KEY;
4075         key.offset = (u64)-1;
4076
4077         while (1) {
4078                 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key,
4079                                         path, 0, 0);
4080                 if (ret < 0) {
4081                         err = ret;
4082                         goto out;
4083                 }
4084                 if (ret > 0) {
4085                         if (path->slots[0] == 0)
4086                                 break;
4087                         path->slots[0]--;
4088                 }
4089                 leaf = path->nodes[0];
4090                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4091                 btrfs_release_path(root->fs_info->tree_root, path);
4092
4093                 if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4094                     key.type != BTRFS_ROOT_ITEM_KEY)
4095                         break;
4096
4097                 reloc_root = btrfs_read_fs_root_no_radix(root, &key);
4098                 if (IS_ERR(reloc_root)) {
4099                         err = PTR_ERR(reloc_root);
4100                         goto out;
4101                 }
4102
4103                 list_add(&reloc_root->root_list, &reloc_roots);
4104
4105                 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4106                         fs_root = read_fs_root(root->fs_info,
4107                                                reloc_root->root_key.offset);
4108                         if (IS_ERR(fs_root)) {
4109                                 ret = PTR_ERR(fs_root);
4110                                 if (ret != -ENOENT) {
4111                                         err = ret;
4112                                         goto out;
4113                                 }
4114                                 mark_garbage_root(reloc_root);
4115                         }
4116                 }
4117
4118                 if (key.offset == 0)
4119                         break;
4120
4121                 key.offset--;
4122         }
4123         btrfs_release_path(root->fs_info->tree_root, path);
4124
4125         if (list_empty(&reloc_roots))
4126                 goto out;
4127
4128         rc = alloc_reloc_control();
4129         if (!rc) {
4130                 err = -ENOMEM;
4131                 goto out;
4132         }
4133
4134         rc->extent_root = root->fs_info->extent_root;
4135
4136         set_reloc_control(rc);
4137
4138         trans = btrfs_join_transaction(rc->extent_root, 1);
4139         if (IS_ERR(trans)) {
4140                 unset_reloc_control(rc);
4141                 err = PTR_ERR(trans);
4142                 goto out_free;
4143         }
4144
4145         rc->merge_reloc_tree = 1;
4146
4147         while (!list_empty(&reloc_roots)) {
4148                 reloc_root = list_entry(reloc_roots.next,
4149                                         struct btrfs_root, root_list);
4150                 list_del(&reloc_root->root_list);
4151
4152                 if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4153                         list_add_tail(&reloc_root->root_list,
4154                                       &rc->reloc_roots);
4155                         continue;
4156                 }
4157
4158                 fs_root = read_fs_root(root->fs_info,
4159                                        reloc_root->root_key.offset);
4160                 BUG_ON(IS_ERR(fs_root));
4161
4162                 __add_reloc_root(reloc_root);
4163                 fs_root->reloc_root = reloc_root;
4164         }
4165
4166         btrfs_commit_transaction(trans, rc->extent_root);
4167
4168         merge_reloc_roots(rc);
4169
4170         unset_reloc_control(rc);
4171
4172         trans = btrfs_join_transaction(rc->extent_root, 1);
4173         if (IS_ERR(trans))
4174                 err = PTR_ERR(trans);
4175         else
4176                 btrfs_commit_transaction(trans, rc->extent_root);
4177 out_free:
4178         kfree(rc);
4179 out:
4180         while (!list_empty(&reloc_roots)) {
4181                 reloc_root = list_entry(reloc_roots.next,
4182                                         struct btrfs_root, root_list);
4183                 list_del(&reloc_root->root_list);
4184                 free_extent_buffer(reloc_root->node);
4185                 free_extent_buffer(reloc_root->commit_root);
4186                 kfree(reloc_root);
4187         }
4188         btrfs_free_path(path);
4189
4190         if (err == 0) {
4191                 /* cleanup orphan inode in data relocation tree */
4192                 fs_root = read_fs_root(root->fs_info,
4193                                        BTRFS_DATA_RELOC_TREE_OBJECTID);
4194                 if (IS_ERR(fs_root))
4195                         err = PTR_ERR(fs_root);
4196                 else
4197                         btrfs_orphan_cleanup(fs_root);
4198         }
4199         return err;
4200 }
4201
4202 /*
4203  * helper to add ordered checksum for data relocation.
4204  *
4205  * cloning checksum properly handles the nodatasum extents.
4206  * it also saves CPU time to re-calculate the checksum.
4207  */
4208 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4209 {
4210         struct btrfs_ordered_sum *sums;
4211         struct btrfs_sector_sum *sector_sum;
4212         struct btrfs_ordered_extent *ordered;
4213         struct btrfs_root *root = BTRFS_I(inode)->root;
4214         size_t offset;
4215         int ret;
4216         u64 disk_bytenr;
4217         LIST_HEAD(list);
4218
4219         ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4220         BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4221
4222         disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4223         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, disk_bytenr,
4224                                        disk_bytenr + len - 1, &list);
4225
4226         while (!list_empty(&list)) {
4227                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4228                 list_del_init(&sums->list);
4229
4230                 sector_sum = sums->sums;
4231                 sums->bytenr = ordered->start;
4232
4233                 offset = 0;
4234                 while (offset < sums->len) {
4235                         sector_sum->bytenr += ordered->start - disk_bytenr;
4236                         sector_sum++;
4237                         offset += root->sectorsize;
4238                 }
4239
4240                 btrfs_add_ordered_sum(inode, ordered, sums);
4241         }
4242         btrfs_put_ordered_extent(ordered);
4243         return ret;
4244 }
4245
4246 void btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4247                            struct btrfs_root *root, struct extent_buffer *buf,
4248                            struct extent_buffer *cow)
4249 {
4250         struct reloc_control *rc;
4251         struct backref_node *node;
4252         int first_cow = 0;
4253         int level;
4254         int ret;
4255
4256         rc = root->fs_info->reloc_ctl;
4257         if (!rc)
4258                 return;
4259
4260         BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4261                root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4262
4263         level = btrfs_header_level(buf);
4264         if (btrfs_header_generation(buf) <=
4265             btrfs_root_last_snapshot(&root->root_item))
4266                 first_cow = 1;
4267
4268         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4269             rc->create_reloc_tree) {
4270                 WARN_ON(!first_cow && level == 0);
4271
4272                 node = rc->backref_cache.path[level];
4273                 BUG_ON(node->bytenr != buf->start &&
4274                        node->new_bytenr != buf->start);
4275
4276                 drop_node_buffer(node);
4277                 extent_buffer_get(cow);
4278                 node->eb = cow;
4279                 node->new_bytenr = cow->start;
4280
4281                 if (!node->pending) {
4282                         list_move_tail(&node->list,
4283                                        &rc->backref_cache.pending[level]);
4284                         node->pending = 1;
4285                 }
4286
4287                 if (first_cow)
4288                         __mark_block_processed(rc, node);
4289
4290                 if (first_cow && level > 0)
4291                         rc->nodes_relocated += buf->len;
4292         }
4293
4294         if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) {
4295                 ret = replace_file_extents(trans, rc, root, cow);
4296                 BUG_ON(ret);
4297         }
4298 }
4299
4300 /*
4301  * called before creating snapshot. it calculates metadata reservation
4302  * requried for relocating tree blocks in the snapshot
4303  */
4304 void btrfs_reloc_pre_snapshot(struct btrfs_trans_handle *trans,
4305                               struct btrfs_pending_snapshot *pending,
4306                               u64 *bytes_to_reserve)
4307 {
4308         struct btrfs_root *root;
4309         struct reloc_control *rc;
4310
4311         root = pending->root;
4312         if (!root->reloc_root)
4313                 return;
4314
4315         rc = root->fs_info->reloc_ctl;
4316         if (!rc->merge_reloc_tree)
4317                 return;
4318
4319         root = root->reloc_root;
4320         BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4321         /*
4322          * relocation is in the stage of merging trees. the space
4323          * used by merging a reloc tree is twice the size of
4324          * relocated tree nodes in the worst case. half for cowing
4325          * the reloc tree, half for cowing the fs tree. the space
4326          * used by cowing the reloc tree will be freed after the
4327          * tree is dropped. if we create snapshot, cowing the fs
4328          * tree may use more space than it frees. so we need
4329          * reserve extra space.
4330          */
4331         *bytes_to_reserve += rc->nodes_relocated;
4332 }
4333
4334 /*
4335  * called after snapshot is created. migrate block reservation
4336  * and create reloc root for the newly created snapshot
4337  */
4338 void btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4339                                struct btrfs_pending_snapshot *pending)
4340 {
4341         struct btrfs_root *root = pending->root;
4342         struct btrfs_root *reloc_root;
4343         struct btrfs_root *new_root;
4344         struct reloc_control *rc;
4345         int ret;
4346
4347         if (!root->reloc_root)
4348                 return;
4349
4350         rc = root->fs_info->reloc_ctl;
4351         rc->merging_rsv_size += rc->nodes_relocated;
4352
4353         if (rc->merge_reloc_tree) {
4354                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4355                                               rc->block_rsv,
4356                                               rc->nodes_relocated);
4357                 BUG_ON(ret);
4358         }
4359
4360         new_root = pending->snap;
4361         reloc_root = create_reloc_root(trans, root->reloc_root,
4362                                        new_root->root_key.objectid);
4363
4364         __add_reloc_root(reloc_root);
4365         new_root->reloc_root = reloc_root;
4366
4367         if (rc->create_reloc_tree) {
4368                 ret = clone_backref_node(trans, rc, root, reloc_root);
4369                 BUG_ON(ret);
4370         }
4371 }