btrfs: btrfs_iget never returns an is_bad_inode inode
[muen/linux.git] / fs / btrfs / relocation.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2009 Oracle.  All rights reserved.
4  */
5
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include "ctree.h"
13 #include "disk-io.h"
14 #include "transaction.h"
15 #include "volumes.h"
16 #include "locking.h"
17 #include "btrfs_inode.h"
18 #include "async-thread.h"
19 #include "free-space-cache.h"
20 #include "inode-map.h"
21 #include "qgroup.h"
22 #include "print-tree.h"
23
24 /*
25  * backref_node, mapping_node and tree_block start with this
26  */
27 struct tree_entry {
28         struct rb_node rb_node;
29         u64 bytenr;
30 };
31
32 /*
33  * present a tree block in the backref cache
34  */
35 struct backref_node {
36         struct rb_node rb_node;
37         u64 bytenr;
38
39         u64 new_bytenr;
40         /* objectid of tree block owner, can be not uptodate */
41         u64 owner;
42         /* link to pending, changed or detached list */
43         struct list_head list;
44         /* list of upper level blocks reference this block */
45         struct list_head upper;
46         /* list of child blocks in the cache */
47         struct list_head lower;
48         /* NULL if this node is not tree root */
49         struct btrfs_root *root;
50         /* extent buffer got by COW the block */
51         struct extent_buffer *eb;
52         /* level of tree block */
53         unsigned int level:8;
54         /* is the block in non-reference counted tree */
55         unsigned int cowonly:1;
56         /* 1 if no child node in the cache */
57         unsigned int lowest:1;
58         /* is the extent buffer locked */
59         unsigned int locked:1;
60         /* has the block been processed */
61         unsigned int processed:1;
62         /* have backrefs of this block been checked */
63         unsigned int checked:1;
64         /*
65          * 1 if corresponding block has been cowed but some upper
66          * level block pointers may not point to the new location
67          */
68         unsigned int pending:1;
69         /*
70          * 1 if the backref node isn't connected to any other
71          * backref node.
72          */
73         unsigned int detached:1;
74 };
75
76 /*
77  * present a block pointer in the backref cache
78  */
79 struct backref_edge {
80         struct list_head list[2];
81         struct backref_node *node[2];
82 };
83
84 #define LOWER   0
85 #define UPPER   1
86 #define RELOCATION_RESERVED_NODES       256
87
88 struct backref_cache {
89         /* red black tree of all backref nodes in the cache */
90         struct rb_root rb_root;
91         /* for passing backref nodes to btrfs_reloc_cow_block */
92         struct backref_node *path[BTRFS_MAX_LEVEL];
93         /*
94          * list of blocks that have been cowed but some block
95          * pointers in upper level blocks may not reflect the
96          * new location
97          */
98         struct list_head pending[BTRFS_MAX_LEVEL];
99         /* list of backref nodes with no child node */
100         struct list_head leaves;
101         /* list of blocks that have been cowed in current transaction */
102         struct list_head changed;
103         /* list of detached backref node. */
104         struct list_head detached;
105
106         u64 last_trans;
107
108         int nr_nodes;
109         int nr_edges;
110 };
111
112 /*
113  * map address of tree root to tree
114  */
115 struct mapping_node {
116         struct rb_node rb_node;
117         u64 bytenr;
118         void *data;
119 };
120
121 struct mapping_tree {
122         struct rb_root rb_root;
123         spinlock_t lock;
124 };
125
126 /*
127  * present a tree block to process
128  */
129 struct tree_block {
130         struct rb_node rb_node;
131         u64 bytenr;
132         struct btrfs_key key;
133         unsigned int level:8;
134         unsigned int key_ready:1;
135 };
136
137 #define MAX_EXTENTS 128
138
139 struct file_extent_cluster {
140         u64 start;
141         u64 end;
142         u64 boundary[MAX_EXTENTS];
143         unsigned int nr;
144 };
145
146 struct reloc_control {
147         /* block group to relocate */
148         struct btrfs_block_group_cache *block_group;
149         /* extent tree */
150         struct btrfs_root *extent_root;
151         /* inode for moving data */
152         struct inode *data_inode;
153
154         struct btrfs_block_rsv *block_rsv;
155
156         struct backref_cache backref_cache;
157
158         struct file_extent_cluster cluster;
159         /* tree blocks have been processed */
160         struct extent_io_tree processed_blocks;
161         /* map start of tree root to corresponding reloc tree */
162         struct mapping_tree reloc_root_tree;
163         /* list of reloc trees */
164         struct list_head reloc_roots;
165         /* size of metadata reservation for merging reloc trees */
166         u64 merging_rsv_size;
167         /* size of relocated tree nodes */
168         u64 nodes_relocated;
169         /* reserved size for block group relocation*/
170         u64 reserved_bytes;
171
172         u64 search_start;
173         u64 extents_found;
174
175         unsigned int stage:8;
176         unsigned int create_reloc_tree:1;
177         unsigned int merge_reloc_tree:1;
178         unsigned int found_file_extent:1;
179 };
180
181 /* stages of data relocation */
182 #define MOVE_DATA_EXTENTS       0
183 #define UPDATE_DATA_PTRS        1
184
185 static void remove_backref_node(struct backref_cache *cache,
186                                 struct backref_node *node);
187 static void __mark_block_processed(struct reloc_control *rc,
188                                    struct backref_node *node);
189
190 static void mapping_tree_init(struct mapping_tree *tree)
191 {
192         tree->rb_root = RB_ROOT;
193         spin_lock_init(&tree->lock);
194 }
195
196 static void backref_cache_init(struct backref_cache *cache)
197 {
198         int i;
199         cache->rb_root = RB_ROOT;
200         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
201                 INIT_LIST_HEAD(&cache->pending[i]);
202         INIT_LIST_HEAD(&cache->changed);
203         INIT_LIST_HEAD(&cache->detached);
204         INIT_LIST_HEAD(&cache->leaves);
205 }
206
207 static void backref_cache_cleanup(struct backref_cache *cache)
208 {
209         struct backref_node *node;
210         int i;
211
212         while (!list_empty(&cache->detached)) {
213                 node = list_entry(cache->detached.next,
214                                   struct backref_node, list);
215                 remove_backref_node(cache, node);
216         }
217
218         while (!list_empty(&cache->leaves)) {
219                 node = list_entry(cache->leaves.next,
220                                   struct backref_node, lower);
221                 remove_backref_node(cache, node);
222         }
223
224         cache->last_trans = 0;
225
226         for (i = 0; i < BTRFS_MAX_LEVEL; i++)
227                 ASSERT(list_empty(&cache->pending[i]));
228         ASSERT(list_empty(&cache->changed));
229         ASSERT(list_empty(&cache->detached));
230         ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
231         ASSERT(!cache->nr_nodes);
232         ASSERT(!cache->nr_edges);
233 }
234
235 static struct backref_node *alloc_backref_node(struct backref_cache *cache)
236 {
237         struct backref_node *node;
238
239         node = kzalloc(sizeof(*node), GFP_NOFS);
240         if (node) {
241                 INIT_LIST_HEAD(&node->list);
242                 INIT_LIST_HEAD(&node->upper);
243                 INIT_LIST_HEAD(&node->lower);
244                 RB_CLEAR_NODE(&node->rb_node);
245                 cache->nr_nodes++;
246         }
247         return node;
248 }
249
250 static void free_backref_node(struct backref_cache *cache,
251                               struct backref_node *node)
252 {
253         if (node) {
254                 cache->nr_nodes--;
255                 kfree(node);
256         }
257 }
258
259 static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
260 {
261         struct backref_edge *edge;
262
263         edge = kzalloc(sizeof(*edge), GFP_NOFS);
264         if (edge)
265                 cache->nr_edges++;
266         return edge;
267 }
268
269 static void free_backref_edge(struct backref_cache *cache,
270                               struct backref_edge *edge)
271 {
272         if (edge) {
273                 cache->nr_edges--;
274                 kfree(edge);
275         }
276 }
277
278 static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
279                                    struct rb_node *node)
280 {
281         struct rb_node **p = &root->rb_node;
282         struct rb_node *parent = NULL;
283         struct tree_entry *entry;
284
285         while (*p) {
286                 parent = *p;
287                 entry = rb_entry(parent, struct tree_entry, rb_node);
288
289                 if (bytenr < entry->bytenr)
290                         p = &(*p)->rb_left;
291                 else if (bytenr > entry->bytenr)
292                         p = &(*p)->rb_right;
293                 else
294                         return parent;
295         }
296
297         rb_link_node(node, parent, p);
298         rb_insert_color(node, root);
299         return NULL;
300 }
301
302 static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
303 {
304         struct rb_node *n = root->rb_node;
305         struct tree_entry *entry;
306
307         while (n) {
308                 entry = rb_entry(n, struct tree_entry, rb_node);
309
310                 if (bytenr < entry->bytenr)
311                         n = n->rb_left;
312                 else if (bytenr > entry->bytenr)
313                         n = n->rb_right;
314                 else
315                         return n;
316         }
317         return NULL;
318 }
319
320 static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
321 {
322
323         struct btrfs_fs_info *fs_info = NULL;
324         struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
325                                               rb_node);
326         if (bnode->root)
327                 fs_info = bnode->root->fs_info;
328         btrfs_panic(fs_info, errno,
329                     "Inconsistency in backref cache found at offset %llu",
330                     bytenr);
331 }
332
333 /*
334  * walk up backref nodes until reach node presents tree root
335  */
336 static struct backref_node *walk_up_backref(struct backref_node *node,
337                                             struct backref_edge *edges[],
338                                             int *index)
339 {
340         struct backref_edge *edge;
341         int idx = *index;
342
343         while (!list_empty(&node->upper)) {
344                 edge = list_entry(node->upper.next,
345                                   struct backref_edge, list[LOWER]);
346                 edges[idx++] = edge;
347                 node = edge->node[UPPER];
348         }
349         BUG_ON(node->detached);
350         *index = idx;
351         return node;
352 }
353
354 /*
355  * walk down backref nodes to find start of next reference path
356  */
357 static struct backref_node *walk_down_backref(struct backref_edge *edges[],
358                                               int *index)
359 {
360         struct backref_edge *edge;
361         struct backref_node *lower;
362         int idx = *index;
363
364         while (idx > 0) {
365                 edge = edges[idx - 1];
366                 lower = edge->node[LOWER];
367                 if (list_is_last(&edge->list[LOWER], &lower->upper)) {
368                         idx--;
369                         continue;
370                 }
371                 edge = list_entry(edge->list[LOWER].next,
372                                   struct backref_edge, list[LOWER]);
373                 edges[idx - 1] = edge;
374                 *index = idx;
375                 return edge->node[UPPER];
376         }
377         *index = 0;
378         return NULL;
379 }
380
381 static void unlock_node_buffer(struct backref_node *node)
382 {
383         if (node->locked) {
384                 btrfs_tree_unlock(node->eb);
385                 node->locked = 0;
386         }
387 }
388
389 static void drop_node_buffer(struct backref_node *node)
390 {
391         if (node->eb) {
392                 unlock_node_buffer(node);
393                 free_extent_buffer(node->eb);
394                 node->eb = NULL;
395         }
396 }
397
398 static void drop_backref_node(struct backref_cache *tree,
399                               struct backref_node *node)
400 {
401         BUG_ON(!list_empty(&node->upper));
402
403         drop_node_buffer(node);
404         list_del(&node->list);
405         list_del(&node->lower);
406         if (!RB_EMPTY_NODE(&node->rb_node))
407                 rb_erase(&node->rb_node, &tree->rb_root);
408         free_backref_node(tree, node);
409 }
410
411 /*
412  * remove a backref node from the backref cache
413  */
414 static void remove_backref_node(struct backref_cache *cache,
415                                 struct backref_node *node)
416 {
417         struct backref_node *upper;
418         struct backref_edge *edge;
419
420         if (!node)
421                 return;
422
423         BUG_ON(!node->lowest && !node->detached);
424         while (!list_empty(&node->upper)) {
425                 edge = list_entry(node->upper.next, struct backref_edge,
426                                   list[LOWER]);
427                 upper = edge->node[UPPER];
428                 list_del(&edge->list[LOWER]);
429                 list_del(&edge->list[UPPER]);
430                 free_backref_edge(cache, edge);
431
432                 if (RB_EMPTY_NODE(&upper->rb_node)) {
433                         BUG_ON(!list_empty(&node->upper));
434                         drop_backref_node(cache, node);
435                         node = upper;
436                         node->lowest = 1;
437                         continue;
438                 }
439                 /*
440                  * add the node to leaf node list if no other
441                  * child block cached.
442                  */
443                 if (list_empty(&upper->lower)) {
444                         list_add_tail(&upper->lower, &cache->leaves);
445                         upper->lowest = 1;
446                 }
447         }
448
449         drop_backref_node(cache, node);
450 }
451
452 static void update_backref_node(struct backref_cache *cache,
453                                 struct backref_node *node, u64 bytenr)
454 {
455         struct rb_node *rb_node;
456         rb_erase(&node->rb_node, &cache->rb_root);
457         node->bytenr = bytenr;
458         rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
459         if (rb_node)
460                 backref_tree_panic(rb_node, -EEXIST, bytenr);
461 }
462
463 /*
464  * update backref cache after a transaction commit
465  */
466 static int update_backref_cache(struct btrfs_trans_handle *trans,
467                                 struct backref_cache *cache)
468 {
469         struct backref_node *node;
470         int level = 0;
471
472         if (cache->last_trans == 0) {
473                 cache->last_trans = trans->transid;
474                 return 0;
475         }
476
477         if (cache->last_trans == trans->transid)
478                 return 0;
479
480         /*
481          * detached nodes are used to avoid unnecessary backref
482          * lookup. transaction commit changes the extent tree.
483          * so the detached nodes are no longer useful.
484          */
485         while (!list_empty(&cache->detached)) {
486                 node = list_entry(cache->detached.next,
487                                   struct backref_node, list);
488                 remove_backref_node(cache, node);
489         }
490
491         while (!list_empty(&cache->changed)) {
492                 node = list_entry(cache->changed.next,
493                                   struct backref_node, list);
494                 list_del_init(&node->list);
495                 BUG_ON(node->pending);
496                 update_backref_node(cache, node, node->new_bytenr);
497         }
498
499         /*
500          * some nodes can be left in the pending list if there were
501          * errors during processing the pending nodes.
502          */
503         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
504                 list_for_each_entry(node, &cache->pending[level], list) {
505                         BUG_ON(!node->pending);
506                         if (node->bytenr == node->new_bytenr)
507                                 continue;
508                         update_backref_node(cache, node, node->new_bytenr);
509                 }
510         }
511
512         cache->last_trans = 0;
513         return 1;
514 }
515
516
517 static int should_ignore_root(struct btrfs_root *root)
518 {
519         struct btrfs_root *reloc_root;
520
521         if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
522                 return 0;
523
524         reloc_root = root->reloc_root;
525         if (!reloc_root)
526                 return 0;
527
528         if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
529             root->fs_info->running_transaction->transid - 1)
530                 return 0;
531         /*
532          * if there is reloc tree and it was created in previous
533          * transaction backref lookup can find the reloc tree,
534          * so backref node for the fs tree root is useless for
535          * relocation.
536          */
537         return 1;
538 }
539 /*
540  * find reloc tree by address of tree root
541  */
542 static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
543                                           u64 bytenr)
544 {
545         struct rb_node *rb_node;
546         struct mapping_node *node;
547         struct btrfs_root *root = NULL;
548
549         spin_lock(&rc->reloc_root_tree.lock);
550         rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
551         if (rb_node) {
552                 node = rb_entry(rb_node, struct mapping_node, rb_node);
553                 root = (struct btrfs_root *)node->data;
554         }
555         spin_unlock(&rc->reloc_root_tree.lock);
556         return root;
557 }
558
559 static int is_cowonly_root(u64 root_objectid)
560 {
561         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
562             root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
563             root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
564             root_objectid == BTRFS_DEV_TREE_OBJECTID ||
565             root_objectid == BTRFS_TREE_LOG_OBJECTID ||
566             root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
567             root_objectid == BTRFS_UUID_TREE_OBJECTID ||
568             root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
569             root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
570                 return 1;
571         return 0;
572 }
573
574 static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
575                                         u64 root_objectid)
576 {
577         struct btrfs_key key;
578
579         key.objectid = root_objectid;
580         key.type = BTRFS_ROOT_ITEM_KEY;
581         if (is_cowonly_root(root_objectid))
582                 key.offset = 0;
583         else
584                 key.offset = (u64)-1;
585
586         return btrfs_get_fs_root(fs_info, &key, false);
587 }
588
589 static noinline_for_stack
590 int find_inline_backref(struct extent_buffer *leaf, int slot,
591                         unsigned long *ptr, unsigned long *end)
592 {
593         struct btrfs_key key;
594         struct btrfs_extent_item *ei;
595         struct btrfs_tree_block_info *bi;
596         u32 item_size;
597
598         btrfs_item_key_to_cpu(leaf, &key, slot);
599
600         item_size = btrfs_item_size_nr(leaf, slot);
601         if (item_size < sizeof(*ei)) {
602                 btrfs_print_v0_err(leaf->fs_info);
603                 btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL);
604                 return 1;
605         }
606         ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
607         WARN_ON(!(btrfs_extent_flags(leaf, ei) &
608                   BTRFS_EXTENT_FLAG_TREE_BLOCK));
609
610         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
611             item_size <= sizeof(*ei) + sizeof(*bi)) {
612                 WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
613                 return 1;
614         }
615         if (key.type == BTRFS_METADATA_ITEM_KEY &&
616             item_size <= sizeof(*ei)) {
617                 WARN_ON(item_size < sizeof(*ei));
618                 return 1;
619         }
620
621         if (key.type == BTRFS_EXTENT_ITEM_KEY) {
622                 bi = (struct btrfs_tree_block_info *)(ei + 1);
623                 *ptr = (unsigned long)(bi + 1);
624         } else {
625                 *ptr = (unsigned long)(ei + 1);
626         }
627         *end = (unsigned long)ei + item_size;
628         return 0;
629 }
630
631 /*
632  * build backref tree for a given tree block. root of the backref tree
633  * corresponds the tree block, leaves of the backref tree correspond
634  * roots of b-trees that reference the tree block.
635  *
636  * the basic idea of this function is check backrefs of a given block
637  * to find upper level blocks that reference the block, and then check
638  * backrefs of these upper level blocks recursively. the recursion stop
639  * when tree root is reached or backrefs for the block is cached.
640  *
641  * NOTE: if we find backrefs for a block are cached, we know backrefs
642  * for all upper level blocks that directly/indirectly reference the
643  * block are also cached.
644  */
645 static noinline_for_stack
646 struct backref_node *build_backref_tree(struct reloc_control *rc,
647                                         struct btrfs_key *node_key,
648                                         int level, u64 bytenr)
649 {
650         struct backref_cache *cache = &rc->backref_cache;
651         struct btrfs_path *path1;
652         struct btrfs_path *path2;
653         struct extent_buffer *eb;
654         struct btrfs_root *root;
655         struct backref_node *cur;
656         struct backref_node *upper;
657         struct backref_node *lower;
658         struct backref_node *node = NULL;
659         struct backref_node *exist = NULL;
660         struct backref_edge *edge;
661         struct rb_node *rb_node;
662         struct btrfs_key key;
663         unsigned long end;
664         unsigned long ptr;
665         LIST_HEAD(list);
666         LIST_HEAD(useless);
667         int cowonly;
668         int ret;
669         int err = 0;
670         bool need_check = true;
671
672         path1 = btrfs_alloc_path();
673         path2 = btrfs_alloc_path();
674         if (!path1 || !path2) {
675                 err = -ENOMEM;
676                 goto out;
677         }
678         path1->reada = READA_FORWARD;
679         path2->reada = READA_FORWARD;
680
681         node = alloc_backref_node(cache);
682         if (!node) {
683                 err = -ENOMEM;
684                 goto out;
685         }
686
687         node->bytenr = bytenr;
688         node->level = level;
689         node->lowest = 1;
690         cur = node;
691 again:
692         end = 0;
693         ptr = 0;
694         key.objectid = cur->bytenr;
695         key.type = BTRFS_METADATA_ITEM_KEY;
696         key.offset = (u64)-1;
697
698         path1->search_commit_root = 1;
699         path1->skip_locking = 1;
700         ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
701                                 0, 0);
702         if (ret < 0) {
703                 err = ret;
704                 goto out;
705         }
706         ASSERT(ret);
707         ASSERT(path1->slots[0]);
708
709         path1->slots[0]--;
710
711         WARN_ON(cur->checked);
712         if (!list_empty(&cur->upper)) {
713                 /*
714                  * the backref was added previously when processing
715                  * backref of type BTRFS_TREE_BLOCK_REF_KEY
716                  */
717                 ASSERT(list_is_singular(&cur->upper));
718                 edge = list_entry(cur->upper.next, struct backref_edge,
719                                   list[LOWER]);
720                 ASSERT(list_empty(&edge->list[UPPER]));
721                 exist = edge->node[UPPER];
722                 /*
723                  * add the upper level block to pending list if we need
724                  * check its backrefs
725                  */
726                 if (!exist->checked)
727                         list_add_tail(&edge->list[UPPER], &list);
728         } else {
729                 exist = NULL;
730         }
731
732         while (1) {
733                 cond_resched();
734                 eb = path1->nodes[0];
735
736                 if (ptr >= end) {
737                         if (path1->slots[0] >= btrfs_header_nritems(eb)) {
738                                 ret = btrfs_next_leaf(rc->extent_root, path1);
739                                 if (ret < 0) {
740                                         err = ret;
741                                         goto out;
742                                 }
743                                 if (ret > 0)
744                                         break;
745                                 eb = path1->nodes[0];
746                         }
747
748                         btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
749                         if (key.objectid != cur->bytenr) {
750                                 WARN_ON(exist);
751                                 break;
752                         }
753
754                         if (key.type == BTRFS_EXTENT_ITEM_KEY ||
755                             key.type == BTRFS_METADATA_ITEM_KEY) {
756                                 ret = find_inline_backref(eb, path1->slots[0],
757                                                           &ptr, &end);
758                                 if (ret)
759                                         goto next;
760                         }
761                 }
762
763                 if (ptr < end) {
764                         /* update key for inline back ref */
765                         struct btrfs_extent_inline_ref *iref;
766                         int type;
767                         iref = (struct btrfs_extent_inline_ref *)ptr;
768                         type = btrfs_get_extent_inline_ref_type(eb, iref,
769                                                         BTRFS_REF_TYPE_BLOCK);
770                         if (type == BTRFS_REF_TYPE_INVALID) {
771                                 err = -EUCLEAN;
772                                 goto out;
773                         }
774                         key.type = type;
775                         key.offset = btrfs_extent_inline_ref_offset(eb, iref);
776
777                         WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
778                                 key.type != BTRFS_SHARED_BLOCK_REF_KEY);
779                 }
780
781                 if (exist &&
782                     ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
783                       exist->owner == key.offset) ||
784                      (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
785                       exist->bytenr == key.offset))) {
786                         exist = NULL;
787                         goto next;
788                 }
789
790                 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
791                         if (key.objectid == key.offset) {
792                                 /*
793                                  * only root blocks of reloc trees use
794                                  * backref of this type.
795                                  */
796                                 root = find_reloc_root(rc, cur->bytenr);
797                                 ASSERT(root);
798                                 cur->root = root;
799                                 break;
800                         }
801
802                         edge = alloc_backref_edge(cache);
803                         if (!edge) {
804                                 err = -ENOMEM;
805                                 goto out;
806                         }
807                         rb_node = tree_search(&cache->rb_root, key.offset);
808                         if (!rb_node) {
809                                 upper = alloc_backref_node(cache);
810                                 if (!upper) {
811                                         free_backref_edge(cache, edge);
812                                         err = -ENOMEM;
813                                         goto out;
814                                 }
815                                 upper->bytenr = key.offset;
816                                 upper->level = cur->level + 1;
817                                 /*
818                                  *  backrefs for the upper level block isn't
819                                  *  cached, add the block to pending list
820                                  */
821                                 list_add_tail(&edge->list[UPPER], &list);
822                         } else {
823                                 upper = rb_entry(rb_node, struct backref_node,
824                                                  rb_node);
825                                 ASSERT(upper->checked);
826                                 INIT_LIST_HEAD(&edge->list[UPPER]);
827                         }
828                         list_add_tail(&edge->list[LOWER], &cur->upper);
829                         edge->node[LOWER] = cur;
830                         edge->node[UPPER] = upper;
831
832                         goto next;
833                 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
834                         err = -EINVAL;
835                         btrfs_print_v0_err(rc->extent_root->fs_info);
836                         btrfs_handle_fs_error(rc->extent_root->fs_info, err,
837                                               NULL);
838                         goto out;
839                 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
840                         goto next;
841                 }
842
843                 /* key.type == BTRFS_TREE_BLOCK_REF_KEY */
844                 root = read_fs_root(rc->extent_root->fs_info, key.offset);
845                 if (IS_ERR(root)) {
846                         err = PTR_ERR(root);
847                         goto out;
848                 }
849
850                 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
851                         cur->cowonly = 1;
852
853                 if (btrfs_root_level(&root->root_item) == cur->level) {
854                         /* tree root */
855                         ASSERT(btrfs_root_bytenr(&root->root_item) ==
856                                cur->bytenr);
857                         if (should_ignore_root(root))
858                                 list_add(&cur->list, &useless);
859                         else
860                                 cur->root = root;
861                         break;
862                 }
863
864                 level = cur->level + 1;
865
866                 /*
867                  * searching the tree to find upper level blocks
868                  * reference the block.
869                  */
870                 path2->search_commit_root = 1;
871                 path2->skip_locking = 1;
872                 path2->lowest_level = level;
873                 ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
874                 path2->lowest_level = 0;
875                 if (ret < 0) {
876                         err = ret;
877                         goto out;
878                 }
879                 if (ret > 0 && path2->slots[level] > 0)
880                         path2->slots[level]--;
881
882                 eb = path2->nodes[level];
883                 if (btrfs_node_blockptr(eb, path2->slots[level]) !=
884                     cur->bytenr) {
885                         btrfs_err(root->fs_info,
886         "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
887                                   cur->bytenr, level - 1, root->objectid,
888                                   node_key->objectid, node_key->type,
889                                   node_key->offset);
890                         err = -ENOENT;
891                         goto out;
892                 }
893                 lower = cur;
894                 need_check = true;
895                 for (; level < BTRFS_MAX_LEVEL; level++) {
896                         if (!path2->nodes[level]) {
897                                 ASSERT(btrfs_root_bytenr(&root->root_item) ==
898                                        lower->bytenr);
899                                 if (should_ignore_root(root))
900                                         list_add(&lower->list, &useless);
901                                 else
902                                         lower->root = root;
903                                 break;
904                         }
905
906                         edge = alloc_backref_edge(cache);
907                         if (!edge) {
908                                 err = -ENOMEM;
909                                 goto out;
910                         }
911
912                         eb = path2->nodes[level];
913                         rb_node = tree_search(&cache->rb_root, eb->start);
914                         if (!rb_node) {
915                                 upper = alloc_backref_node(cache);
916                                 if (!upper) {
917                                         free_backref_edge(cache, edge);
918                                         err = -ENOMEM;
919                                         goto out;
920                                 }
921                                 upper->bytenr = eb->start;
922                                 upper->owner = btrfs_header_owner(eb);
923                                 upper->level = lower->level + 1;
924                                 if (!test_bit(BTRFS_ROOT_REF_COWS,
925                                               &root->state))
926                                         upper->cowonly = 1;
927
928                                 /*
929                                  * if we know the block isn't shared
930                                  * we can void checking its backrefs.
931                                  */
932                                 if (btrfs_block_can_be_shared(root, eb))
933                                         upper->checked = 0;
934                                 else
935                                         upper->checked = 1;
936
937                                 /*
938                                  * add the block to pending list if we
939                                  * need check its backrefs, we only do this once
940                                  * while walking up a tree as we will catch
941                                  * anything else later on.
942                                  */
943                                 if (!upper->checked && need_check) {
944                                         need_check = false;
945                                         list_add_tail(&edge->list[UPPER],
946                                                       &list);
947                                 } else {
948                                         if (upper->checked)
949                                                 need_check = true;
950                                         INIT_LIST_HEAD(&edge->list[UPPER]);
951                                 }
952                         } else {
953                                 upper = rb_entry(rb_node, struct backref_node,
954                                                  rb_node);
955                                 ASSERT(upper->checked);
956                                 INIT_LIST_HEAD(&edge->list[UPPER]);
957                                 if (!upper->owner)
958                                         upper->owner = btrfs_header_owner(eb);
959                         }
960                         list_add_tail(&edge->list[LOWER], &lower->upper);
961                         edge->node[LOWER] = lower;
962                         edge->node[UPPER] = upper;
963
964                         if (rb_node)
965                                 break;
966                         lower = upper;
967                         upper = NULL;
968                 }
969                 btrfs_release_path(path2);
970 next:
971                 if (ptr < end) {
972                         ptr += btrfs_extent_inline_ref_size(key.type);
973                         if (ptr >= end) {
974                                 WARN_ON(ptr > end);
975                                 ptr = 0;
976                                 end = 0;
977                         }
978                 }
979                 if (ptr >= end)
980                         path1->slots[0]++;
981         }
982         btrfs_release_path(path1);
983
984         cur->checked = 1;
985         WARN_ON(exist);
986
987         /* the pending list isn't empty, take the first block to process */
988         if (!list_empty(&list)) {
989                 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
990                 list_del_init(&edge->list[UPPER]);
991                 cur = edge->node[UPPER];
992                 goto again;
993         }
994
995         /*
996          * everything goes well, connect backref nodes and insert backref nodes
997          * into the cache.
998          */
999         ASSERT(node->checked);
1000         cowonly = node->cowonly;
1001         if (!cowonly) {
1002                 rb_node = tree_insert(&cache->rb_root, node->bytenr,
1003                                       &node->rb_node);
1004                 if (rb_node)
1005                         backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1006                 list_add_tail(&node->lower, &cache->leaves);
1007         }
1008
1009         list_for_each_entry(edge, &node->upper, list[LOWER])
1010                 list_add_tail(&edge->list[UPPER], &list);
1011
1012         while (!list_empty(&list)) {
1013                 edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1014                 list_del_init(&edge->list[UPPER]);
1015                 upper = edge->node[UPPER];
1016                 if (upper->detached) {
1017                         list_del(&edge->list[LOWER]);
1018                         lower = edge->node[LOWER];
1019                         free_backref_edge(cache, edge);
1020                         if (list_empty(&lower->upper))
1021                                 list_add(&lower->list, &useless);
1022                         continue;
1023                 }
1024
1025                 if (!RB_EMPTY_NODE(&upper->rb_node)) {
1026                         if (upper->lowest) {
1027                                 list_del_init(&upper->lower);
1028                                 upper->lowest = 0;
1029                         }
1030
1031                         list_add_tail(&edge->list[UPPER], &upper->lower);
1032                         continue;
1033                 }
1034
1035                 if (!upper->checked) {
1036                         /*
1037                          * Still want to blow up for developers since this is a
1038                          * logic bug.
1039                          */
1040                         ASSERT(0);
1041                         err = -EINVAL;
1042                         goto out;
1043                 }
1044                 if (cowonly != upper->cowonly) {
1045                         ASSERT(0);
1046                         err = -EINVAL;
1047                         goto out;
1048                 }
1049
1050                 if (!cowonly) {
1051                         rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1052                                               &upper->rb_node);
1053                         if (rb_node)
1054                                 backref_tree_panic(rb_node, -EEXIST,
1055                                                    upper->bytenr);
1056                 }
1057
1058                 list_add_tail(&edge->list[UPPER], &upper->lower);
1059
1060                 list_for_each_entry(edge, &upper->upper, list[LOWER])
1061                         list_add_tail(&edge->list[UPPER], &list);
1062         }
1063         /*
1064          * process useless backref nodes. backref nodes for tree leaves
1065          * are deleted from the cache. backref nodes for upper level
1066          * tree blocks are left in the cache to avoid unnecessary backref
1067          * lookup.
1068          */
1069         while (!list_empty(&useless)) {
1070                 upper = list_entry(useless.next, struct backref_node, list);
1071                 list_del_init(&upper->list);
1072                 ASSERT(list_empty(&upper->upper));
1073                 if (upper == node)
1074                         node = NULL;
1075                 if (upper->lowest) {
1076                         list_del_init(&upper->lower);
1077                         upper->lowest = 0;
1078                 }
1079                 while (!list_empty(&upper->lower)) {
1080                         edge = list_entry(upper->lower.next,
1081                                           struct backref_edge, list[UPPER]);
1082                         list_del(&edge->list[UPPER]);
1083                         list_del(&edge->list[LOWER]);
1084                         lower = edge->node[LOWER];
1085                         free_backref_edge(cache, edge);
1086
1087                         if (list_empty(&lower->upper))
1088                                 list_add(&lower->list, &useless);
1089                 }
1090                 __mark_block_processed(rc, upper);
1091                 if (upper->level > 0) {
1092                         list_add(&upper->list, &cache->detached);
1093                         upper->detached = 1;
1094                 } else {
1095                         rb_erase(&upper->rb_node, &cache->rb_root);
1096                         free_backref_node(cache, upper);
1097                 }
1098         }
1099 out:
1100         btrfs_free_path(path1);
1101         btrfs_free_path(path2);
1102         if (err) {
1103                 while (!list_empty(&useless)) {
1104                         lower = list_entry(useless.next,
1105                                            struct backref_node, list);
1106                         list_del_init(&lower->list);
1107                 }
1108                 while (!list_empty(&list)) {
1109                         edge = list_first_entry(&list, struct backref_edge,
1110                                                 list[UPPER]);
1111                         list_del(&edge->list[UPPER]);
1112                         list_del(&edge->list[LOWER]);
1113                         lower = edge->node[LOWER];
1114                         upper = edge->node[UPPER];
1115                         free_backref_edge(cache, edge);
1116
1117                         /*
1118                          * Lower is no longer linked to any upper backref nodes
1119                          * and isn't in the cache, we can free it ourselves.
1120                          */
1121                         if (list_empty(&lower->upper) &&
1122                             RB_EMPTY_NODE(&lower->rb_node))
1123                                 list_add(&lower->list, &useless);
1124
1125                         if (!RB_EMPTY_NODE(&upper->rb_node))
1126                                 continue;
1127
1128                         /* Add this guy's upper edges to the list to process */
1129                         list_for_each_entry(edge, &upper->upper, list[LOWER])
1130                                 list_add_tail(&edge->list[UPPER], &list);
1131                         if (list_empty(&upper->upper))
1132                                 list_add(&upper->list, &useless);
1133                 }
1134
1135                 while (!list_empty(&useless)) {
1136                         lower = list_entry(useless.next,
1137                                            struct backref_node, list);
1138                         list_del_init(&lower->list);
1139                         if (lower == node)
1140                                 node = NULL;
1141                         free_backref_node(cache, lower);
1142                 }
1143
1144                 free_backref_node(cache, node);
1145                 return ERR_PTR(err);
1146         }
1147         ASSERT(!node || !node->detached);
1148         return node;
1149 }
1150
1151 /*
1152  * helper to add backref node for the newly created snapshot.
1153  * the backref node is created by cloning backref node that
1154  * corresponds to root of source tree
1155  */
1156 static int clone_backref_node(struct btrfs_trans_handle *trans,
1157                               struct reloc_control *rc,
1158                               struct btrfs_root *src,
1159                               struct btrfs_root *dest)
1160 {
1161         struct btrfs_root *reloc_root = src->reloc_root;
1162         struct backref_cache *cache = &rc->backref_cache;
1163         struct backref_node *node = NULL;
1164         struct backref_node *new_node;
1165         struct backref_edge *edge;
1166         struct backref_edge *new_edge;
1167         struct rb_node *rb_node;
1168
1169         if (cache->last_trans > 0)
1170                 update_backref_cache(trans, cache);
1171
1172         rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1173         if (rb_node) {
1174                 node = rb_entry(rb_node, struct backref_node, rb_node);
1175                 if (node->detached)
1176                         node = NULL;
1177                 else
1178                         BUG_ON(node->new_bytenr != reloc_root->node->start);
1179         }
1180
1181         if (!node) {
1182                 rb_node = tree_search(&cache->rb_root,
1183                                       reloc_root->commit_root->start);
1184                 if (rb_node) {
1185                         node = rb_entry(rb_node, struct backref_node,
1186                                         rb_node);
1187                         BUG_ON(node->detached);
1188                 }
1189         }
1190
1191         if (!node)
1192                 return 0;
1193
1194         new_node = alloc_backref_node(cache);
1195         if (!new_node)
1196                 return -ENOMEM;
1197
1198         new_node->bytenr = dest->node->start;
1199         new_node->level = node->level;
1200         new_node->lowest = node->lowest;
1201         new_node->checked = 1;
1202         new_node->root = dest;
1203
1204         if (!node->lowest) {
1205                 list_for_each_entry(edge, &node->lower, list[UPPER]) {
1206                         new_edge = alloc_backref_edge(cache);
1207                         if (!new_edge)
1208                                 goto fail;
1209
1210                         new_edge->node[UPPER] = new_node;
1211                         new_edge->node[LOWER] = edge->node[LOWER];
1212                         list_add_tail(&new_edge->list[UPPER],
1213                                       &new_node->lower);
1214                 }
1215         } else {
1216                 list_add_tail(&new_node->lower, &cache->leaves);
1217         }
1218
1219         rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1220                               &new_node->rb_node);
1221         if (rb_node)
1222                 backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1223
1224         if (!new_node->lowest) {
1225                 list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1226                         list_add_tail(&new_edge->list[LOWER],
1227                                       &new_edge->node[LOWER]->upper);
1228                 }
1229         }
1230         return 0;
1231 fail:
1232         while (!list_empty(&new_node->lower)) {
1233                 new_edge = list_entry(new_node->lower.next,
1234                                       struct backref_edge, list[UPPER]);
1235                 list_del(&new_edge->list[UPPER]);
1236                 free_backref_edge(cache, new_edge);
1237         }
1238         free_backref_node(cache, new_node);
1239         return -ENOMEM;
1240 }
1241
1242 /*
1243  * helper to add 'address of tree root -> reloc tree' mapping
1244  */
1245 static int __must_check __add_reloc_root(struct btrfs_root *root)
1246 {
1247         struct btrfs_fs_info *fs_info = root->fs_info;
1248         struct rb_node *rb_node;
1249         struct mapping_node *node;
1250         struct reloc_control *rc = fs_info->reloc_ctl;
1251
1252         node = kmalloc(sizeof(*node), GFP_NOFS);
1253         if (!node)
1254                 return -ENOMEM;
1255
1256         node->bytenr = root->node->start;
1257         node->data = root;
1258
1259         spin_lock(&rc->reloc_root_tree.lock);
1260         rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1261                               node->bytenr, &node->rb_node);
1262         spin_unlock(&rc->reloc_root_tree.lock);
1263         if (rb_node) {
1264                 btrfs_panic(fs_info, -EEXIST,
1265                             "Duplicate root found for start=%llu while inserting into relocation tree",
1266                             node->bytenr);
1267         }
1268
1269         list_add_tail(&root->root_list, &rc->reloc_roots);
1270         return 0;
1271 }
1272
1273 /*
1274  * helper to delete the 'address of tree root -> reloc tree'
1275  * mapping
1276  */
1277 static void __del_reloc_root(struct btrfs_root *root)
1278 {
1279         struct btrfs_fs_info *fs_info = root->fs_info;
1280         struct rb_node *rb_node;
1281         struct mapping_node *node = NULL;
1282         struct reloc_control *rc = fs_info->reloc_ctl;
1283
1284         if (rc) {
1285                 spin_lock(&rc->reloc_root_tree.lock);
1286                 rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1287                                       root->node->start);
1288                 if (rb_node) {
1289                         node = rb_entry(rb_node, struct mapping_node, rb_node);
1290                         rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1291                 }
1292                 spin_unlock(&rc->reloc_root_tree.lock);
1293                 if (!node)
1294                         return;
1295                 BUG_ON((struct btrfs_root *)node->data != root);
1296         }
1297
1298         spin_lock(&fs_info->trans_lock);
1299         list_del_init(&root->root_list);
1300         spin_unlock(&fs_info->trans_lock);
1301         kfree(node);
1302 }
1303
1304 /*
1305  * helper to update the 'address of tree root -> reloc tree'
1306  * mapping
1307  */
1308 static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1309 {
1310         struct btrfs_fs_info *fs_info = root->fs_info;
1311         struct rb_node *rb_node;
1312         struct mapping_node *node = NULL;
1313         struct reloc_control *rc = fs_info->reloc_ctl;
1314
1315         spin_lock(&rc->reloc_root_tree.lock);
1316         rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1317                               root->node->start);
1318         if (rb_node) {
1319                 node = rb_entry(rb_node, struct mapping_node, rb_node);
1320                 rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1321         }
1322         spin_unlock(&rc->reloc_root_tree.lock);
1323
1324         if (!node)
1325                 return 0;
1326         BUG_ON((struct btrfs_root *)node->data != root);
1327
1328         spin_lock(&rc->reloc_root_tree.lock);
1329         node->bytenr = new_bytenr;
1330         rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1331                               node->bytenr, &node->rb_node);
1332         spin_unlock(&rc->reloc_root_tree.lock);
1333         if (rb_node)
1334                 backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1335         return 0;
1336 }
1337
1338 static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1339                                         struct btrfs_root *root, u64 objectid)
1340 {
1341         struct btrfs_fs_info *fs_info = root->fs_info;
1342         struct btrfs_root *reloc_root;
1343         struct extent_buffer *eb;
1344         struct btrfs_root_item *root_item;
1345         struct btrfs_key root_key;
1346         int ret;
1347
1348         root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1349         BUG_ON(!root_item);
1350
1351         root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1352         root_key.type = BTRFS_ROOT_ITEM_KEY;
1353         root_key.offset = objectid;
1354
1355         if (root->root_key.objectid == objectid) {
1356                 u64 commit_root_gen;
1357
1358                 /* called by btrfs_init_reloc_root */
1359                 ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1360                                       BTRFS_TREE_RELOC_OBJECTID);
1361                 BUG_ON(ret);
1362                 /*
1363                  * Set the last_snapshot field to the generation of the commit
1364                  * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1365                  * correctly (returns true) when the relocation root is created
1366                  * either inside the critical section of a transaction commit
1367                  * (through transaction.c:qgroup_account_snapshot()) and when
1368                  * it's created before the transaction commit is started.
1369                  */
1370                 commit_root_gen = btrfs_header_generation(root->commit_root);
1371                 btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1372         } else {
1373                 /*
1374                  * called by btrfs_reloc_post_snapshot_hook.
1375                  * the source tree is a reloc tree, all tree blocks
1376                  * modified after it was created have RELOC flag
1377                  * set in their headers. so it's OK to not update
1378                  * the 'last_snapshot'.
1379                  */
1380                 ret = btrfs_copy_root(trans, root, root->node, &eb,
1381                                       BTRFS_TREE_RELOC_OBJECTID);
1382                 BUG_ON(ret);
1383         }
1384
1385         memcpy(root_item, &root->root_item, sizeof(*root_item));
1386         btrfs_set_root_bytenr(root_item, eb->start);
1387         btrfs_set_root_level(root_item, btrfs_header_level(eb));
1388         btrfs_set_root_generation(root_item, trans->transid);
1389
1390         if (root->root_key.objectid == objectid) {
1391                 btrfs_set_root_refs(root_item, 0);
1392                 memset(&root_item->drop_progress, 0,
1393                        sizeof(struct btrfs_disk_key));
1394                 root_item->drop_level = 0;
1395         }
1396
1397         btrfs_tree_unlock(eb);
1398         free_extent_buffer(eb);
1399
1400         ret = btrfs_insert_root(trans, fs_info->tree_root,
1401                                 &root_key, root_item);
1402         BUG_ON(ret);
1403         kfree(root_item);
1404
1405         reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1406         BUG_ON(IS_ERR(reloc_root));
1407         reloc_root->last_trans = trans->transid;
1408         return reloc_root;
1409 }
1410
1411 /*
1412  * create reloc tree for a given fs tree. reloc tree is just a
1413  * snapshot of the fs tree with special root objectid.
1414  */
1415 int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1416                           struct btrfs_root *root)
1417 {
1418         struct btrfs_fs_info *fs_info = root->fs_info;
1419         struct btrfs_root *reloc_root;
1420         struct reloc_control *rc = fs_info->reloc_ctl;
1421         struct btrfs_block_rsv *rsv;
1422         int clear_rsv = 0;
1423         int ret;
1424
1425         if (root->reloc_root) {
1426                 reloc_root = root->reloc_root;
1427                 reloc_root->last_trans = trans->transid;
1428                 return 0;
1429         }
1430
1431         if (!rc || !rc->create_reloc_tree ||
1432             root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1433                 return 0;
1434
1435         if (!trans->reloc_reserved) {
1436                 rsv = trans->block_rsv;
1437                 trans->block_rsv = rc->block_rsv;
1438                 clear_rsv = 1;
1439         }
1440         reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1441         if (clear_rsv)
1442                 trans->block_rsv = rsv;
1443
1444         ret = __add_reloc_root(reloc_root);
1445         BUG_ON(ret < 0);
1446         root->reloc_root = reloc_root;
1447         return 0;
1448 }
1449
1450 /*
1451  * update root item of reloc tree
1452  */
1453 int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1454                             struct btrfs_root *root)
1455 {
1456         struct btrfs_fs_info *fs_info = root->fs_info;
1457         struct btrfs_root *reloc_root;
1458         struct btrfs_root_item *root_item;
1459         int ret;
1460
1461         if (!root->reloc_root)
1462                 goto out;
1463
1464         reloc_root = root->reloc_root;
1465         root_item = &reloc_root->root_item;
1466
1467         if (fs_info->reloc_ctl->merge_reloc_tree &&
1468             btrfs_root_refs(root_item) == 0) {
1469                 root->reloc_root = NULL;
1470                 __del_reloc_root(reloc_root);
1471         }
1472
1473         if (reloc_root->commit_root != reloc_root->node) {
1474                 btrfs_set_root_node(root_item, reloc_root->node);
1475                 free_extent_buffer(reloc_root->commit_root);
1476                 reloc_root->commit_root = btrfs_root_node(reloc_root);
1477         }
1478
1479         ret = btrfs_update_root(trans, fs_info->tree_root,
1480                                 &reloc_root->root_key, root_item);
1481         BUG_ON(ret);
1482
1483 out:
1484         return 0;
1485 }
1486
1487 /*
1488  * helper to find first cached inode with inode number >= objectid
1489  * in a subvolume
1490  */
1491 static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1492 {
1493         struct rb_node *node;
1494         struct rb_node *prev;
1495         struct btrfs_inode *entry;
1496         struct inode *inode;
1497
1498         spin_lock(&root->inode_lock);
1499 again:
1500         node = root->inode_tree.rb_node;
1501         prev = NULL;
1502         while (node) {
1503                 prev = node;
1504                 entry = rb_entry(node, struct btrfs_inode, rb_node);
1505
1506                 if (objectid < btrfs_ino(entry))
1507                         node = node->rb_left;
1508                 else if (objectid > btrfs_ino(entry))
1509                         node = node->rb_right;
1510                 else
1511                         break;
1512         }
1513         if (!node) {
1514                 while (prev) {
1515                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
1516                         if (objectid <= btrfs_ino(entry)) {
1517                                 node = prev;
1518                                 break;
1519                         }
1520                         prev = rb_next(prev);
1521                 }
1522         }
1523         while (node) {
1524                 entry = rb_entry(node, struct btrfs_inode, rb_node);
1525                 inode = igrab(&entry->vfs_inode);
1526                 if (inode) {
1527                         spin_unlock(&root->inode_lock);
1528                         return inode;
1529                 }
1530
1531                 objectid = btrfs_ino(entry) + 1;
1532                 if (cond_resched_lock(&root->inode_lock))
1533                         goto again;
1534
1535                 node = rb_next(node);
1536         }
1537         spin_unlock(&root->inode_lock);
1538         return NULL;
1539 }
1540
1541 static int in_block_group(u64 bytenr,
1542                           struct btrfs_block_group_cache *block_group)
1543 {
1544         if (bytenr >= block_group->key.objectid &&
1545             bytenr < block_group->key.objectid + block_group->key.offset)
1546                 return 1;
1547         return 0;
1548 }
1549
1550 /*
1551  * get new location of data
1552  */
1553 static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1554                             u64 bytenr, u64 num_bytes)
1555 {
1556         struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1557         struct btrfs_path *path;
1558         struct btrfs_file_extent_item *fi;
1559         struct extent_buffer *leaf;
1560         int ret;
1561
1562         path = btrfs_alloc_path();
1563         if (!path)
1564                 return -ENOMEM;
1565
1566         bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1567         ret = btrfs_lookup_file_extent(NULL, root, path,
1568                         btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1569         if (ret < 0)
1570                 goto out;
1571         if (ret > 0) {
1572                 ret = -ENOENT;
1573                 goto out;
1574         }
1575
1576         leaf = path->nodes[0];
1577         fi = btrfs_item_ptr(leaf, path->slots[0],
1578                             struct btrfs_file_extent_item);
1579
1580         BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1581                btrfs_file_extent_compression(leaf, fi) ||
1582                btrfs_file_extent_encryption(leaf, fi) ||
1583                btrfs_file_extent_other_encoding(leaf, fi));
1584
1585         if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1586                 ret = -EINVAL;
1587                 goto out;
1588         }
1589
1590         *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1591         ret = 0;
1592 out:
1593         btrfs_free_path(path);
1594         return ret;
1595 }
1596
1597 /*
1598  * update file extent items in the tree leaf to point to
1599  * the new locations.
1600  */
1601 static noinline_for_stack
1602 int replace_file_extents(struct btrfs_trans_handle *trans,
1603                          struct reloc_control *rc,
1604                          struct btrfs_root *root,
1605                          struct extent_buffer *leaf)
1606 {
1607         struct btrfs_fs_info *fs_info = root->fs_info;
1608         struct btrfs_key key;
1609         struct btrfs_file_extent_item *fi;
1610         struct inode *inode = NULL;
1611         u64 parent;
1612         u64 bytenr;
1613         u64 new_bytenr = 0;
1614         u64 num_bytes;
1615         u64 end;
1616         u32 nritems;
1617         u32 i;
1618         int ret = 0;
1619         int first = 1;
1620         int dirty = 0;
1621
1622         if (rc->stage != UPDATE_DATA_PTRS)
1623                 return 0;
1624
1625         /* reloc trees always use full backref */
1626         if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1627                 parent = leaf->start;
1628         else
1629                 parent = 0;
1630
1631         nritems = btrfs_header_nritems(leaf);
1632         for (i = 0; i < nritems; i++) {
1633                 cond_resched();
1634                 btrfs_item_key_to_cpu(leaf, &key, i);
1635                 if (key.type != BTRFS_EXTENT_DATA_KEY)
1636                         continue;
1637                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1638                 if (btrfs_file_extent_type(leaf, fi) ==
1639                     BTRFS_FILE_EXTENT_INLINE)
1640                         continue;
1641                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1642                 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1643                 if (bytenr == 0)
1644                         continue;
1645                 if (!in_block_group(bytenr, rc->block_group))
1646                         continue;
1647
1648                 /*
1649                  * if we are modifying block in fs tree, wait for readpage
1650                  * to complete and drop the extent cache
1651                  */
1652                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1653                         if (first) {
1654                                 inode = find_next_inode(root, key.objectid);
1655                                 first = 0;
1656                         } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1657                                 btrfs_add_delayed_iput(inode);
1658                                 inode = find_next_inode(root, key.objectid);
1659                         }
1660                         if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1661                                 end = key.offset +
1662                                       btrfs_file_extent_num_bytes(leaf, fi);
1663                                 WARN_ON(!IS_ALIGNED(key.offset,
1664                                                     fs_info->sectorsize));
1665                                 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1666                                 end--;
1667                                 ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1668                                                       key.offset, end);
1669                                 if (!ret)
1670                                         continue;
1671
1672                                 btrfs_drop_extent_cache(BTRFS_I(inode),
1673                                                 key.offset,     end, 1);
1674                                 unlock_extent(&BTRFS_I(inode)->io_tree,
1675                                               key.offset, end);
1676                         }
1677                 }
1678
1679                 ret = get_new_location(rc->data_inode, &new_bytenr,
1680                                        bytenr, num_bytes);
1681                 if (ret) {
1682                         /*
1683                          * Don't have to abort since we've not changed anything
1684                          * in the file extent yet.
1685                          */
1686                         break;
1687                 }
1688
1689                 btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1690                 dirty = 1;
1691
1692                 key.offset -= btrfs_file_extent_offset(leaf, fi);
1693                 ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1694                                            num_bytes, parent,
1695                                            btrfs_header_owner(leaf),
1696                                            key.objectid, key.offset);
1697                 if (ret) {
1698                         btrfs_abort_transaction(trans, ret);
1699                         break;
1700                 }
1701
1702                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1703                                         parent, btrfs_header_owner(leaf),
1704                                         key.objectid, key.offset);
1705                 if (ret) {
1706                         btrfs_abort_transaction(trans, ret);
1707                         break;
1708                 }
1709         }
1710         if (dirty)
1711                 btrfs_mark_buffer_dirty(leaf);
1712         if (inode)
1713                 btrfs_add_delayed_iput(inode);
1714         return ret;
1715 }
1716
1717 static noinline_for_stack
1718 int memcmp_node_keys(struct extent_buffer *eb, int slot,
1719                      struct btrfs_path *path, int level)
1720 {
1721         struct btrfs_disk_key key1;
1722         struct btrfs_disk_key key2;
1723         btrfs_node_key(eb, &key1, slot);
1724         btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1725         return memcmp(&key1, &key2, sizeof(key1));
1726 }
1727
1728 /*
1729  * try to replace tree blocks in fs tree with the new blocks
1730  * in reloc tree. tree blocks haven't been modified since the
1731  * reloc tree was create can be replaced.
1732  *
1733  * if a block was replaced, level of the block + 1 is returned.
1734  * if no block got replaced, 0 is returned. if there are other
1735  * errors, a negative error number is returned.
1736  */
1737 static noinline_for_stack
1738 int replace_path(struct btrfs_trans_handle *trans,
1739                  struct btrfs_root *dest, struct btrfs_root *src,
1740                  struct btrfs_path *path, struct btrfs_key *next_key,
1741                  int lowest_level, int max_level)
1742 {
1743         struct btrfs_fs_info *fs_info = dest->fs_info;
1744         struct extent_buffer *eb;
1745         struct extent_buffer *parent;
1746         struct btrfs_key key;
1747         u64 old_bytenr;
1748         u64 new_bytenr;
1749         u64 old_ptr_gen;
1750         u64 new_ptr_gen;
1751         u64 last_snapshot;
1752         u32 blocksize;
1753         int cow = 0;
1754         int level;
1755         int ret;
1756         int slot;
1757
1758         BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1759         BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1760
1761         last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1762 again:
1763         slot = path->slots[lowest_level];
1764         btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1765
1766         eb = btrfs_lock_root_node(dest);
1767         btrfs_set_lock_blocking(eb);
1768         level = btrfs_header_level(eb);
1769
1770         if (level < lowest_level) {
1771                 btrfs_tree_unlock(eb);
1772                 free_extent_buffer(eb);
1773                 return 0;
1774         }
1775
1776         if (cow) {
1777                 ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1778                 BUG_ON(ret);
1779         }
1780         btrfs_set_lock_blocking(eb);
1781
1782         if (next_key) {
1783                 next_key->objectid = (u64)-1;
1784                 next_key->type = (u8)-1;
1785                 next_key->offset = (u64)-1;
1786         }
1787
1788         parent = eb;
1789         while (1) {
1790                 struct btrfs_key first_key;
1791
1792                 level = btrfs_header_level(parent);
1793                 BUG_ON(level < lowest_level);
1794
1795                 ret = btrfs_bin_search(parent, &key, level, &slot);
1796                 if (ret && slot > 0)
1797                         slot--;
1798
1799                 if (next_key && slot + 1 < btrfs_header_nritems(parent))
1800                         btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1801
1802                 old_bytenr = btrfs_node_blockptr(parent, slot);
1803                 blocksize = fs_info->nodesize;
1804                 old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1805                 btrfs_node_key_to_cpu(parent, &first_key, slot);
1806
1807                 if (level <= max_level) {
1808                         eb = path->nodes[level];
1809                         new_bytenr = btrfs_node_blockptr(eb,
1810                                                         path->slots[level]);
1811                         new_ptr_gen = btrfs_node_ptr_generation(eb,
1812                                                         path->slots[level]);
1813                 } else {
1814                         new_bytenr = 0;
1815                         new_ptr_gen = 0;
1816                 }
1817
1818                 if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1819                         ret = level;
1820                         break;
1821                 }
1822
1823                 if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1824                     memcmp_node_keys(parent, slot, path, level)) {
1825                         if (level <= lowest_level) {
1826                                 ret = 0;
1827                                 break;
1828                         }
1829
1830                         eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1831                                              level - 1, &first_key);
1832                         if (IS_ERR(eb)) {
1833                                 ret = PTR_ERR(eb);
1834                                 break;
1835                         } else if (!extent_buffer_uptodate(eb)) {
1836                                 ret = -EIO;
1837                                 free_extent_buffer(eb);
1838                                 break;
1839                         }
1840                         btrfs_tree_lock(eb);
1841                         if (cow) {
1842                                 ret = btrfs_cow_block(trans, dest, eb, parent,
1843                                                       slot, &eb);
1844                                 BUG_ON(ret);
1845                         }
1846                         btrfs_set_lock_blocking(eb);
1847
1848                         btrfs_tree_unlock(parent);
1849                         free_extent_buffer(parent);
1850
1851                         parent = eb;
1852                         continue;
1853                 }
1854
1855                 if (!cow) {
1856                         btrfs_tree_unlock(parent);
1857                         free_extent_buffer(parent);
1858                         cow = 1;
1859                         goto again;
1860                 }
1861
1862                 btrfs_node_key_to_cpu(path->nodes[level], &key,
1863                                       path->slots[level]);
1864                 btrfs_release_path(path);
1865
1866                 path->lowest_level = level;
1867                 ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1868                 path->lowest_level = 0;
1869                 BUG_ON(ret);
1870
1871                 /*
1872                  * Info qgroup to trace both subtrees.
1873                  *
1874                  * We must trace both trees.
1875                  * 1) Tree reloc subtree
1876                  *    If not traced, we will leak data numbers
1877                  * 2) Fs subtree
1878                  *    If not traced, we will double count old data
1879                  *    and tree block numbers, if current trans doesn't free
1880                  *    data reloc tree inode.
1881                  */
1882                 ret = btrfs_qgroup_trace_subtree(trans, parent,
1883                                 btrfs_header_generation(parent),
1884                                 btrfs_header_level(parent));
1885                 if (ret < 0)
1886                         break;
1887                 ret = btrfs_qgroup_trace_subtree(trans, path->nodes[level],
1888                                 btrfs_header_generation(path->nodes[level]),
1889                                 btrfs_header_level(path->nodes[level]));
1890                 if (ret < 0)
1891                         break;
1892
1893                 /*
1894                  * swap blocks in fs tree and reloc tree.
1895                  */
1896                 btrfs_set_node_blockptr(parent, slot, new_bytenr);
1897                 btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1898                 btrfs_mark_buffer_dirty(parent);
1899
1900                 btrfs_set_node_blockptr(path->nodes[level],
1901                                         path->slots[level], old_bytenr);
1902                 btrfs_set_node_ptr_generation(path->nodes[level],
1903                                               path->slots[level], old_ptr_gen);
1904                 btrfs_mark_buffer_dirty(path->nodes[level]);
1905
1906                 ret = btrfs_inc_extent_ref(trans, src, old_bytenr,
1907                                         blocksize, path->nodes[level]->start,
1908                                         src->root_key.objectid, level - 1, 0);
1909                 BUG_ON(ret);
1910                 ret = btrfs_inc_extent_ref(trans, dest, new_bytenr,
1911                                         blocksize, 0, dest->root_key.objectid,
1912                                         level - 1, 0);
1913                 BUG_ON(ret);
1914
1915                 ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1916                                         path->nodes[level]->start,
1917                                         src->root_key.objectid, level - 1, 0);
1918                 BUG_ON(ret);
1919
1920                 ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1921                                         0, dest->root_key.objectid, level - 1,
1922                                         0);
1923                 BUG_ON(ret);
1924
1925                 btrfs_unlock_up_safe(path, 0);
1926
1927                 ret = level;
1928                 break;
1929         }
1930         btrfs_tree_unlock(parent);
1931         free_extent_buffer(parent);
1932         return ret;
1933 }
1934
1935 /*
1936  * helper to find next relocated block in reloc tree
1937  */
1938 static noinline_for_stack
1939 int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1940                        int *level)
1941 {
1942         struct extent_buffer *eb;
1943         int i;
1944         u64 last_snapshot;
1945         u32 nritems;
1946
1947         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1948
1949         for (i = 0; i < *level; i++) {
1950                 free_extent_buffer(path->nodes[i]);
1951                 path->nodes[i] = NULL;
1952         }
1953
1954         for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1955                 eb = path->nodes[i];
1956                 nritems = btrfs_header_nritems(eb);
1957                 while (path->slots[i] + 1 < nritems) {
1958                         path->slots[i]++;
1959                         if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1960                             last_snapshot)
1961                                 continue;
1962
1963                         *level = i;
1964                         return 0;
1965                 }
1966                 free_extent_buffer(path->nodes[i]);
1967                 path->nodes[i] = NULL;
1968         }
1969         return 1;
1970 }
1971
1972 /*
1973  * walk down reloc tree to find relocated block of lowest level
1974  */
1975 static noinline_for_stack
1976 int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1977                          int *level)
1978 {
1979         struct btrfs_fs_info *fs_info = root->fs_info;
1980         struct extent_buffer *eb = NULL;
1981         int i;
1982         u64 bytenr;
1983         u64 ptr_gen = 0;
1984         u64 last_snapshot;
1985         u32 nritems;
1986
1987         last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1988
1989         for (i = *level; i > 0; i--) {
1990                 struct btrfs_key first_key;
1991
1992                 eb = path->nodes[i];
1993                 nritems = btrfs_header_nritems(eb);
1994                 while (path->slots[i] < nritems) {
1995                         ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1996                         if (ptr_gen > last_snapshot)
1997                                 break;
1998                         path->slots[i]++;
1999                 }
2000                 if (path->slots[i] >= nritems) {
2001                         if (i == *level)
2002                                 break;
2003                         *level = i + 1;
2004                         return 0;
2005                 }
2006                 if (i == 1) {
2007                         *level = i;
2008                         return 0;
2009                 }
2010
2011                 bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2012                 btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2013                 eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2014                                      &first_key);
2015                 if (IS_ERR(eb)) {
2016                         return PTR_ERR(eb);
2017                 } else if (!extent_buffer_uptodate(eb)) {
2018                         free_extent_buffer(eb);
2019                         return -EIO;
2020                 }
2021                 BUG_ON(btrfs_header_level(eb) != i - 1);
2022                 path->nodes[i - 1] = eb;
2023                 path->slots[i - 1] = 0;
2024         }
2025         return 1;
2026 }
2027
2028 /*
2029  * invalidate extent cache for file extents whose key in range of
2030  * [min_key, max_key)
2031  */
2032 static int invalidate_extent_cache(struct btrfs_root *root,
2033                                    struct btrfs_key *min_key,
2034                                    struct btrfs_key *max_key)
2035 {
2036         struct btrfs_fs_info *fs_info = root->fs_info;
2037         struct inode *inode = NULL;
2038         u64 objectid;
2039         u64 start, end;
2040         u64 ino;
2041
2042         objectid = min_key->objectid;
2043         while (1) {
2044                 cond_resched();
2045                 iput(inode);
2046
2047                 if (objectid > max_key->objectid)
2048                         break;
2049
2050                 inode = find_next_inode(root, objectid);
2051                 if (!inode)
2052                         break;
2053                 ino = btrfs_ino(BTRFS_I(inode));
2054
2055                 if (ino > max_key->objectid) {
2056                         iput(inode);
2057                         break;
2058                 }
2059
2060                 objectid = ino + 1;
2061                 if (!S_ISREG(inode->i_mode))
2062                         continue;
2063
2064                 if (unlikely(min_key->objectid == ino)) {
2065                         if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2066                                 continue;
2067                         if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2068                                 start = 0;
2069                         else {
2070                                 start = min_key->offset;
2071                                 WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2072                         }
2073                 } else {
2074                         start = 0;
2075                 }
2076
2077                 if (unlikely(max_key->objectid == ino)) {
2078                         if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2079                                 continue;
2080                         if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2081                                 end = (u64)-1;
2082                         } else {
2083                                 if (max_key->offset == 0)
2084                                         continue;
2085                                 end = max_key->offset;
2086                                 WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2087                                 end--;
2088                         }
2089                 } else {
2090                         end = (u64)-1;
2091                 }
2092
2093                 /* the lock_extent waits for readpage to complete */
2094                 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2095                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2096                 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2097         }
2098         return 0;
2099 }
2100
2101 static int find_next_key(struct btrfs_path *path, int level,
2102                          struct btrfs_key *key)
2103
2104 {
2105         while (level < BTRFS_MAX_LEVEL) {
2106                 if (!path->nodes[level])
2107                         break;
2108                 if (path->slots[level] + 1 <
2109                     btrfs_header_nritems(path->nodes[level])) {
2110                         btrfs_node_key_to_cpu(path->nodes[level], key,
2111                                               path->slots[level] + 1);
2112                         return 0;
2113                 }
2114                 level++;
2115         }
2116         return 1;
2117 }
2118
2119 /*
2120  * merge the relocated tree blocks in reloc tree with corresponding
2121  * fs tree.
2122  */
2123 static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2124                                                struct btrfs_root *root)
2125 {
2126         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2127         LIST_HEAD(inode_list);
2128         struct btrfs_key key;
2129         struct btrfs_key next_key;
2130         struct btrfs_trans_handle *trans = NULL;
2131         struct btrfs_root *reloc_root;
2132         struct btrfs_root_item *root_item;
2133         struct btrfs_path *path;
2134         struct extent_buffer *leaf;
2135         int level;
2136         int max_level;
2137         int replaced = 0;
2138         int ret;
2139         int err = 0;
2140         u32 min_reserved;
2141
2142         path = btrfs_alloc_path();
2143         if (!path)
2144                 return -ENOMEM;
2145         path->reada = READA_FORWARD;
2146
2147         reloc_root = root->reloc_root;
2148         root_item = &reloc_root->root_item;
2149
2150         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2151                 level = btrfs_root_level(root_item);
2152                 extent_buffer_get(reloc_root->node);
2153                 path->nodes[level] = reloc_root->node;
2154                 path->slots[level] = 0;
2155         } else {
2156                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2157
2158                 level = root_item->drop_level;
2159                 BUG_ON(level == 0);
2160                 path->lowest_level = level;
2161                 ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2162                 path->lowest_level = 0;
2163                 if (ret < 0) {
2164                         btrfs_free_path(path);
2165                         return ret;
2166                 }
2167
2168                 btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2169                                       path->slots[level]);
2170                 WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2171
2172                 btrfs_unlock_up_safe(path, 0);
2173         }
2174
2175         min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2176         memset(&next_key, 0, sizeof(next_key));
2177
2178         while (1) {
2179                 ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2180                                              BTRFS_RESERVE_FLUSH_ALL);
2181                 if (ret) {
2182                         err = ret;
2183                         goto out;
2184                 }
2185                 trans = btrfs_start_transaction(root, 0);
2186                 if (IS_ERR(trans)) {
2187                         err = PTR_ERR(trans);
2188                         trans = NULL;
2189                         goto out;
2190                 }
2191                 trans->block_rsv = rc->block_rsv;
2192
2193                 replaced = 0;
2194                 max_level = level;
2195
2196                 ret = walk_down_reloc_tree(reloc_root, path, &level);
2197                 if (ret < 0) {
2198                         err = ret;
2199                         goto out;
2200                 }
2201                 if (ret > 0)
2202                         break;
2203
2204                 if (!find_next_key(path, level, &key) &&
2205                     btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2206                         ret = 0;
2207                 } else {
2208                         ret = replace_path(trans, root, reloc_root, path,
2209                                            &next_key, level, max_level);
2210                 }
2211                 if (ret < 0) {
2212                         err = ret;
2213                         goto out;
2214                 }
2215
2216                 if (ret > 0) {
2217                         level = ret;
2218                         btrfs_node_key_to_cpu(path->nodes[level], &key,
2219                                               path->slots[level]);
2220                         replaced = 1;
2221                 }
2222
2223                 ret = walk_up_reloc_tree(reloc_root, path, &level);
2224                 if (ret > 0)
2225                         break;
2226
2227                 BUG_ON(level == 0);
2228                 /*
2229                  * save the merging progress in the drop_progress.
2230                  * this is OK since root refs == 1 in this case.
2231                  */
2232                 btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2233                                path->slots[level]);
2234                 root_item->drop_level = level;
2235
2236                 btrfs_end_transaction_throttle(trans);
2237                 trans = NULL;
2238
2239                 btrfs_btree_balance_dirty(fs_info);
2240
2241                 if (replaced && rc->stage == UPDATE_DATA_PTRS)
2242                         invalidate_extent_cache(root, &key, &next_key);
2243         }
2244
2245         /*
2246          * handle the case only one block in the fs tree need to be
2247          * relocated and the block is tree root.
2248          */
2249         leaf = btrfs_lock_root_node(root);
2250         ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
2251         btrfs_tree_unlock(leaf);
2252         free_extent_buffer(leaf);
2253         if (ret < 0)
2254                 err = ret;
2255 out:
2256         btrfs_free_path(path);
2257
2258         if (err == 0) {
2259                 memset(&root_item->drop_progress, 0,
2260                        sizeof(root_item->drop_progress));
2261                 root_item->drop_level = 0;
2262                 btrfs_set_root_refs(root_item, 0);
2263                 btrfs_update_reloc_root(trans, root);
2264         }
2265
2266         if (trans)
2267                 btrfs_end_transaction_throttle(trans);
2268
2269         btrfs_btree_balance_dirty(fs_info);
2270
2271         if (replaced && rc->stage == UPDATE_DATA_PTRS)
2272                 invalidate_extent_cache(root, &key, &next_key);
2273
2274         return err;
2275 }
2276
2277 static noinline_for_stack
2278 int prepare_to_merge(struct reloc_control *rc, int err)
2279 {
2280         struct btrfs_root *root = rc->extent_root;
2281         struct btrfs_fs_info *fs_info = root->fs_info;
2282         struct btrfs_root *reloc_root;
2283         struct btrfs_trans_handle *trans;
2284         LIST_HEAD(reloc_roots);
2285         u64 num_bytes = 0;
2286         int ret;
2287
2288         mutex_lock(&fs_info->reloc_mutex);
2289         rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2290         rc->merging_rsv_size += rc->nodes_relocated * 2;
2291         mutex_unlock(&fs_info->reloc_mutex);
2292
2293 again:
2294         if (!err) {
2295                 num_bytes = rc->merging_rsv_size;
2296                 ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2297                                           BTRFS_RESERVE_FLUSH_ALL);
2298                 if (ret)
2299                         err = ret;
2300         }
2301
2302         trans = btrfs_join_transaction(rc->extent_root);
2303         if (IS_ERR(trans)) {
2304                 if (!err)
2305                         btrfs_block_rsv_release(fs_info, rc->block_rsv,
2306                                                 num_bytes);
2307                 return PTR_ERR(trans);
2308         }
2309
2310         if (!err) {
2311                 if (num_bytes != rc->merging_rsv_size) {
2312                         btrfs_end_transaction(trans);
2313                         btrfs_block_rsv_release(fs_info, rc->block_rsv,
2314                                                 num_bytes);
2315                         goto again;
2316                 }
2317         }
2318
2319         rc->merge_reloc_tree = 1;
2320
2321         while (!list_empty(&rc->reloc_roots)) {
2322                 reloc_root = list_entry(rc->reloc_roots.next,
2323                                         struct btrfs_root, root_list);
2324                 list_del_init(&reloc_root->root_list);
2325
2326                 root = read_fs_root(fs_info, reloc_root->root_key.offset);
2327                 BUG_ON(IS_ERR(root));
2328                 BUG_ON(root->reloc_root != reloc_root);
2329
2330                 /*
2331                  * set reference count to 1, so btrfs_recover_relocation
2332                  * knows it should resumes merging
2333                  */
2334                 if (!err)
2335                         btrfs_set_root_refs(&reloc_root->root_item, 1);
2336                 btrfs_update_reloc_root(trans, root);
2337
2338                 list_add(&reloc_root->root_list, &reloc_roots);
2339         }
2340
2341         list_splice(&reloc_roots, &rc->reloc_roots);
2342
2343         if (!err)
2344                 btrfs_commit_transaction(trans);
2345         else
2346                 btrfs_end_transaction(trans);
2347         return err;
2348 }
2349
2350 static noinline_for_stack
2351 void free_reloc_roots(struct list_head *list)
2352 {
2353         struct btrfs_root *reloc_root;
2354
2355         while (!list_empty(list)) {
2356                 reloc_root = list_entry(list->next, struct btrfs_root,
2357                                         root_list);
2358                 __del_reloc_root(reloc_root);
2359                 free_extent_buffer(reloc_root->node);
2360                 free_extent_buffer(reloc_root->commit_root);
2361                 reloc_root->node = NULL;
2362                 reloc_root->commit_root = NULL;
2363         }
2364 }
2365
2366 static noinline_for_stack
2367 void merge_reloc_roots(struct reloc_control *rc)
2368 {
2369         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2370         struct btrfs_root *root;
2371         struct btrfs_root *reloc_root;
2372         LIST_HEAD(reloc_roots);
2373         int found = 0;
2374         int ret = 0;
2375 again:
2376         root = rc->extent_root;
2377
2378         /*
2379          * this serializes us with btrfs_record_root_in_transaction,
2380          * we have to make sure nobody is in the middle of
2381          * adding their roots to the list while we are
2382          * doing this splice
2383          */
2384         mutex_lock(&fs_info->reloc_mutex);
2385         list_splice_init(&rc->reloc_roots, &reloc_roots);
2386         mutex_unlock(&fs_info->reloc_mutex);
2387
2388         while (!list_empty(&reloc_roots)) {
2389                 found = 1;
2390                 reloc_root = list_entry(reloc_roots.next,
2391                                         struct btrfs_root, root_list);
2392
2393                 if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2394                         root = read_fs_root(fs_info,
2395                                             reloc_root->root_key.offset);
2396                         BUG_ON(IS_ERR(root));
2397                         BUG_ON(root->reloc_root != reloc_root);
2398
2399                         ret = merge_reloc_root(rc, root);
2400                         if (ret) {
2401                                 if (list_empty(&reloc_root->root_list))
2402                                         list_add_tail(&reloc_root->root_list,
2403                                                       &reloc_roots);
2404                                 goto out;
2405                         }
2406                 } else {
2407                         list_del_init(&reloc_root->root_list);
2408                 }
2409
2410                 ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2411                 if (ret < 0) {
2412                         if (list_empty(&reloc_root->root_list))
2413                                 list_add_tail(&reloc_root->root_list,
2414                                               &reloc_roots);
2415                         goto out;
2416                 }
2417         }
2418
2419         if (found) {
2420                 found = 0;
2421                 goto again;
2422         }
2423 out:
2424         if (ret) {
2425                 btrfs_handle_fs_error(fs_info, ret, NULL);
2426                 if (!list_empty(&reloc_roots))
2427                         free_reloc_roots(&reloc_roots);
2428
2429                 /* new reloc root may be added */
2430                 mutex_lock(&fs_info->reloc_mutex);
2431                 list_splice_init(&rc->reloc_roots, &reloc_roots);
2432                 mutex_unlock(&fs_info->reloc_mutex);
2433                 if (!list_empty(&reloc_roots))
2434                         free_reloc_roots(&reloc_roots);
2435         }
2436
2437         BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2438 }
2439
2440 static void free_block_list(struct rb_root *blocks)
2441 {
2442         struct tree_block *block;
2443         struct rb_node *rb_node;
2444         while ((rb_node = rb_first(blocks))) {
2445                 block = rb_entry(rb_node, struct tree_block, rb_node);
2446                 rb_erase(rb_node, blocks);
2447                 kfree(block);
2448         }
2449 }
2450
2451 static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2452                                       struct btrfs_root *reloc_root)
2453 {
2454         struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2455         struct btrfs_root *root;
2456
2457         if (reloc_root->last_trans == trans->transid)
2458                 return 0;
2459
2460         root = read_fs_root(fs_info, reloc_root->root_key.offset);
2461         BUG_ON(IS_ERR(root));
2462         BUG_ON(root->reloc_root != reloc_root);
2463
2464         return btrfs_record_root_in_trans(trans, root);
2465 }
2466
2467 static noinline_for_stack
2468 struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2469                                      struct reloc_control *rc,
2470                                      struct backref_node *node,
2471                                      struct backref_edge *edges[])
2472 {
2473         struct backref_node *next;
2474         struct btrfs_root *root;
2475         int index = 0;
2476
2477         next = node;
2478         while (1) {
2479                 cond_resched();
2480                 next = walk_up_backref(next, edges, &index);
2481                 root = next->root;
2482                 BUG_ON(!root);
2483                 BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2484
2485                 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2486                         record_reloc_root_in_trans(trans, root);
2487                         break;
2488                 }
2489
2490                 btrfs_record_root_in_trans(trans, root);
2491                 root = root->reloc_root;
2492
2493                 if (next->new_bytenr != root->node->start) {
2494                         BUG_ON(next->new_bytenr);
2495                         BUG_ON(!list_empty(&next->list));
2496                         next->new_bytenr = root->node->start;
2497                         next->root = root;
2498                         list_add_tail(&next->list,
2499                                       &rc->backref_cache.changed);
2500                         __mark_block_processed(rc, next);
2501                         break;
2502                 }
2503
2504                 WARN_ON(1);
2505                 root = NULL;
2506                 next = walk_down_backref(edges, &index);
2507                 if (!next || next->level <= node->level)
2508                         break;
2509         }
2510         if (!root)
2511                 return NULL;
2512
2513         next = node;
2514         /* setup backref node path for btrfs_reloc_cow_block */
2515         while (1) {
2516                 rc->backref_cache.path[next->level] = next;
2517                 if (--index < 0)
2518                         break;
2519                 next = edges[index]->node[UPPER];
2520         }
2521         return root;
2522 }
2523
2524 /*
2525  * select a tree root for relocation. return NULL if the block
2526  * is reference counted. we should use do_relocation() in this
2527  * case. return a tree root pointer if the block isn't reference
2528  * counted. return -ENOENT if the block is root of reloc tree.
2529  */
2530 static noinline_for_stack
2531 struct btrfs_root *select_one_root(struct backref_node *node)
2532 {
2533         struct backref_node *next;
2534         struct btrfs_root *root;
2535         struct btrfs_root *fs_root = NULL;
2536         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2537         int index = 0;
2538
2539         next = node;
2540         while (1) {
2541                 cond_resched();
2542                 next = walk_up_backref(next, edges, &index);
2543                 root = next->root;
2544                 BUG_ON(!root);
2545
2546                 /* no other choice for non-references counted tree */
2547                 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
2548                         return root;
2549
2550                 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2551                         fs_root = root;
2552
2553                 if (next != node)
2554                         return NULL;
2555
2556                 next = walk_down_backref(edges, &index);
2557                 if (!next || next->level <= node->level)
2558                         break;
2559         }
2560
2561         if (!fs_root)
2562                 return ERR_PTR(-ENOENT);
2563         return fs_root;
2564 }
2565
2566 static noinline_for_stack
2567 u64 calcu_metadata_size(struct reloc_control *rc,
2568                         struct backref_node *node, int reserve)
2569 {
2570         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2571         struct backref_node *next = node;
2572         struct backref_edge *edge;
2573         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2574         u64 num_bytes = 0;
2575         int index = 0;
2576
2577         BUG_ON(reserve && node->processed);
2578
2579         while (next) {
2580                 cond_resched();
2581                 while (1) {
2582                         if (next->processed && (reserve || next != node))
2583                                 break;
2584
2585                         num_bytes += fs_info->nodesize;
2586
2587                         if (list_empty(&next->upper))
2588                                 break;
2589
2590                         edge = list_entry(next->upper.next,
2591                                           struct backref_edge, list[LOWER]);
2592                         edges[index++] = edge;
2593                         next = edge->node[UPPER];
2594                 }
2595                 next = walk_down_backref(edges, &index);
2596         }
2597         return num_bytes;
2598 }
2599
2600 static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2601                                   struct reloc_control *rc,
2602                                   struct backref_node *node)
2603 {
2604         struct btrfs_root *root = rc->extent_root;
2605         struct btrfs_fs_info *fs_info = root->fs_info;
2606         u64 num_bytes;
2607         int ret;
2608         u64 tmp;
2609
2610         num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2611
2612         trans->block_rsv = rc->block_rsv;
2613         rc->reserved_bytes += num_bytes;
2614
2615         /*
2616          * We are under a transaction here so we can only do limited flushing.
2617          * If we get an enospc just kick back -EAGAIN so we know to drop the
2618          * transaction and try to refill when we can flush all the things.
2619          */
2620         ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2621                                 BTRFS_RESERVE_FLUSH_LIMIT);
2622         if (ret) {
2623                 tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2624                 while (tmp <= rc->reserved_bytes)
2625                         tmp <<= 1;
2626                 /*
2627                  * only one thread can access block_rsv at this point,
2628                  * so we don't need hold lock to protect block_rsv.
2629                  * we expand more reservation size here to allow enough
2630                  * space for relocation and we will return eailer in
2631                  * enospc case.
2632                  */
2633                 rc->block_rsv->size = tmp + fs_info->nodesize *
2634                                       RELOCATION_RESERVED_NODES;
2635                 return -EAGAIN;
2636         }
2637
2638         return 0;
2639 }
2640
2641 /*
2642  * relocate a block tree, and then update pointers in upper level
2643  * blocks that reference the block to point to the new location.
2644  *
2645  * if called by link_to_upper, the block has already been relocated.
2646  * in that case this function just updates pointers.
2647  */
2648 static int do_relocation(struct btrfs_trans_handle *trans,
2649                          struct reloc_control *rc,
2650                          struct backref_node *node,
2651                          struct btrfs_key *key,
2652                          struct btrfs_path *path, int lowest)
2653 {
2654         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2655         struct backref_node *upper;
2656         struct backref_edge *edge;
2657         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2658         struct btrfs_root *root;
2659         struct extent_buffer *eb;
2660         u32 blocksize;
2661         u64 bytenr;
2662         u64 generation;
2663         int slot;
2664         int ret;
2665         int err = 0;
2666
2667         BUG_ON(lowest && node->eb);
2668
2669         path->lowest_level = node->level + 1;
2670         rc->backref_cache.path[node->level] = node;
2671         list_for_each_entry(edge, &node->upper, list[LOWER]) {
2672                 struct btrfs_key first_key;
2673
2674                 cond_resched();
2675
2676                 upper = edge->node[UPPER];
2677                 root = select_reloc_root(trans, rc, upper, edges);
2678                 BUG_ON(!root);
2679
2680                 if (upper->eb && !upper->locked) {
2681                         if (!lowest) {
2682                                 ret = btrfs_bin_search(upper->eb, key,
2683                                                        upper->level, &slot);
2684                                 BUG_ON(ret);
2685                                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2686                                 if (node->eb->start == bytenr)
2687                                         goto next;
2688                         }
2689                         drop_node_buffer(upper);
2690                 }
2691
2692                 if (!upper->eb) {
2693                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2694                         if (ret) {
2695                                 if (ret < 0)
2696                                         err = ret;
2697                                 else
2698                                         err = -ENOENT;
2699
2700                                 btrfs_release_path(path);
2701                                 break;
2702                         }
2703
2704                         if (!upper->eb) {
2705                                 upper->eb = path->nodes[upper->level];
2706                                 path->nodes[upper->level] = NULL;
2707                         } else {
2708                                 BUG_ON(upper->eb != path->nodes[upper->level]);
2709                         }
2710
2711                         upper->locked = 1;
2712                         path->locks[upper->level] = 0;
2713
2714                         slot = path->slots[upper->level];
2715                         btrfs_release_path(path);
2716                 } else {
2717                         ret = btrfs_bin_search(upper->eb, key, upper->level,
2718                                                &slot);
2719                         BUG_ON(ret);
2720                 }
2721
2722                 bytenr = btrfs_node_blockptr(upper->eb, slot);
2723                 if (lowest) {
2724                         if (bytenr != node->bytenr) {
2725                                 btrfs_err(root->fs_info,
2726                 "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2727                                           bytenr, node->bytenr, slot,
2728                                           upper->eb->start);
2729                                 err = -EIO;
2730                                 goto next;
2731                         }
2732                 } else {
2733                         if (node->eb->start == bytenr)
2734                                 goto next;
2735                 }
2736
2737                 blocksize = root->fs_info->nodesize;
2738                 generation = btrfs_node_ptr_generation(upper->eb, slot);
2739                 btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2740                 eb = read_tree_block(fs_info, bytenr, generation,
2741                                      upper->level - 1, &first_key);
2742                 if (IS_ERR(eb)) {
2743                         err = PTR_ERR(eb);
2744                         goto next;
2745                 } else if (!extent_buffer_uptodate(eb)) {
2746                         free_extent_buffer(eb);
2747                         err = -EIO;
2748                         goto next;
2749                 }
2750                 btrfs_tree_lock(eb);
2751                 btrfs_set_lock_blocking(eb);
2752
2753                 if (!node->eb) {
2754                         ret = btrfs_cow_block(trans, root, eb, upper->eb,
2755                                               slot, &eb);
2756                         btrfs_tree_unlock(eb);
2757                         free_extent_buffer(eb);
2758                         if (ret < 0) {
2759                                 err = ret;
2760                                 goto next;
2761                         }
2762                         BUG_ON(node->eb != eb);
2763                 } else {
2764                         btrfs_set_node_blockptr(upper->eb, slot,
2765                                                 node->eb->start);
2766                         btrfs_set_node_ptr_generation(upper->eb, slot,
2767                                                       trans->transid);
2768                         btrfs_mark_buffer_dirty(upper->eb);
2769
2770                         ret = btrfs_inc_extent_ref(trans, root,
2771                                                 node->eb->start, blocksize,
2772                                                 upper->eb->start,
2773                                                 btrfs_header_owner(upper->eb),
2774                                                 node->level, 0);
2775                         BUG_ON(ret);
2776
2777                         ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2778                         BUG_ON(ret);
2779                 }
2780 next:
2781                 if (!upper->pending)
2782                         drop_node_buffer(upper);
2783                 else
2784                         unlock_node_buffer(upper);
2785                 if (err)
2786                         break;
2787         }
2788
2789         if (!err && node->pending) {
2790                 drop_node_buffer(node);
2791                 list_move_tail(&node->list, &rc->backref_cache.changed);
2792                 node->pending = 0;
2793         }
2794
2795         path->lowest_level = 0;
2796         BUG_ON(err == -ENOSPC);
2797         return err;
2798 }
2799
2800 static int link_to_upper(struct btrfs_trans_handle *trans,
2801                          struct reloc_control *rc,
2802                          struct backref_node *node,
2803                          struct btrfs_path *path)
2804 {
2805         struct btrfs_key key;
2806
2807         btrfs_node_key_to_cpu(node->eb, &key, 0);
2808         return do_relocation(trans, rc, node, &key, path, 0);
2809 }
2810
2811 static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2812                                 struct reloc_control *rc,
2813                                 struct btrfs_path *path, int err)
2814 {
2815         LIST_HEAD(list);
2816         struct backref_cache *cache = &rc->backref_cache;
2817         struct backref_node *node;
2818         int level;
2819         int ret;
2820
2821         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2822                 while (!list_empty(&cache->pending[level])) {
2823                         node = list_entry(cache->pending[level].next,
2824                                           struct backref_node, list);
2825                         list_move_tail(&node->list, &list);
2826                         BUG_ON(!node->pending);
2827
2828                         if (!err) {
2829                                 ret = link_to_upper(trans, rc, node, path);
2830                                 if (ret < 0)
2831                                         err = ret;
2832                         }
2833                 }
2834                 list_splice_init(&list, &cache->pending[level]);
2835         }
2836         return err;
2837 }
2838
2839 static void mark_block_processed(struct reloc_control *rc,
2840                                  u64 bytenr, u32 blocksize)
2841 {
2842         set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2843                         EXTENT_DIRTY);
2844 }
2845
2846 static void __mark_block_processed(struct reloc_control *rc,
2847                                    struct backref_node *node)
2848 {
2849         u32 blocksize;
2850         if (node->level == 0 ||
2851             in_block_group(node->bytenr, rc->block_group)) {
2852                 blocksize = rc->extent_root->fs_info->nodesize;
2853                 mark_block_processed(rc, node->bytenr, blocksize);
2854         }
2855         node->processed = 1;
2856 }
2857
2858 /*
2859  * mark a block and all blocks directly/indirectly reference the block
2860  * as processed.
2861  */
2862 static void update_processed_blocks(struct reloc_control *rc,
2863                                     struct backref_node *node)
2864 {
2865         struct backref_node *next = node;
2866         struct backref_edge *edge;
2867         struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2868         int index = 0;
2869
2870         while (next) {
2871                 cond_resched();
2872                 while (1) {
2873                         if (next->processed)
2874                                 break;
2875
2876                         __mark_block_processed(rc, next);
2877
2878                         if (list_empty(&next->upper))
2879                                 break;
2880
2881                         edge = list_entry(next->upper.next,
2882                                           struct backref_edge, list[LOWER]);
2883                         edges[index++] = edge;
2884                         next = edge->node[UPPER];
2885                 }
2886                 next = walk_down_backref(edges, &index);
2887         }
2888 }
2889
2890 static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2891 {
2892         u32 blocksize = rc->extent_root->fs_info->nodesize;
2893
2894         if (test_range_bit(&rc->processed_blocks, bytenr,
2895                            bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2896                 return 1;
2897         return 0;
2898 }
2899
2900 static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2901                               struct tree_block *block)
2902 {
2903         struct extent_buffer *eb;
2904
2905         BUG_ON(block->key_ready);
2906         eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2907                              block->level, NULL);
2908         if (IS_ERR(eb)) {
2909                 return PTR_ERR(eb);
2910         } else if (!extent_buffer_uptodate(eb)) {
2911                 free_extent_buffer(eb);
2912                 return -EIO;
2913         }
2914         WARN_ON(btrfs_header_level(eb) != block->level);
2915         if (block->level == 0)
2916                 btrfs_item_key_to_cpu(eb, &block->key, 0);
2917         else
2918                 btrfs_node_key_to_cpu(eb, &block->key, 0);
2919         free_extent_buffer(eb);
2920         block->key_ready = 1;
2921         return 0;
2922 }
2923
2924 /*
2925  * helper function to relocate a tree block
2926  */
2927 static int relocate_tree_block(struct btrfs_trans_handle *trans,
2928                                 struct reloc_control *rc,
2929                                 struct backref_node *node,
2930                                 struct btrfs_key *key,
2931                                 struct btrfs_path *path)
2932 {
2933         struct btrfs_root *root;
2934         int ret = 0;
2935
2936         if (!node)
2937                 return 0;
2938
2939         BUG_ON(node->processed);
2940         root = select_one_root(node);
2941         if (root == ERR_PTR(-ENOENT)) {
2942                 update_processed_blocks(rc, node);
2943                 goto out;
2944         }
2945
2946         if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2947                 ret = reserve_metadata_space(trans, rc, node);
2948                 if (ret)
2949                         goto out;
2950         }
2951
2952         if (root) {
2953                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2954                         BUG_ON(node->new_bytenr);
2955                         BUG_ON(!list_empty(&node->list));
2956                         btrfs_record_root_in_trans(trans, root);
2957                         root = root->reloc_root;
2958                         node->new_bytenr = root->node->start;
2959                         node->root = root;
2960                         list_add_tail(&node->list, &rc->backref_cache.changed);
2961                 } else {
2962                         path->lowest_level = node->level;
2963                         ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2964                         btrfs_release_path(path);
2965                         if (ret > 0)
2966                                 ret = 0;
2967                 }
2968                 if (!ret)
2969                         update_processed_blocks(rc, node);
2970         } else {
2971                 ret = do_relocation(trans, rc, node, key, path, 1);
2972         }
2973 out:
2974         if (ret || node->level == 0 || node->cowonly)
2975                 remove_backref_node(&rc->backref_cache, node);
2976         return ret;
2977 }
2978
2979 /*
2980  * relocate a list of blocks
2981  */
2982 static noinline_for_stack
2983 int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2984                          struct reloc_control *rc, struct rb_root *blocks)
2985 {
2986         struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2987         struct backref_node *node;
2988         struct btrfs_path *path;
2989         struct tree_block *block;
2990         struct rb_node *rb_node;
2991         int ret;
2992         int err = 0;
2993
2994         path = btrfs_alloc_path();
2995         if (!path) {
2996                 err = -ENOMEM;
2997                 goto out_free_blocks;
2998         }
2999
3000         rb_node = rb_first(blocks);
3001         while (rb_node) {
3002                 block = rb_entry(rb_node, struct tree_block, rb_node);
3003                 if (!block->key_ready)
3004                         readahead_tree_block(fs_info, block->bytenr);
3005                 rb_node = rb_next(rb_node);
3006         }
3007
3008         rb_node = rb_first(blocks);
3009         while (rb_node) {
3010                 block = rb_entry(rb_node, struct tree_block, rb_node);
3011                 if (!block->key_ready) {
3012                         err = get_tree_block_key(fs_info, block);
3013                         if (err)
3014                                 goto out_free_path;
3015                 }
3016                 rb_node = rb_next(rb_node);
3017         }
3018
3019         rb_node = rb_first(blocks);
3020         while (rb_node) {
3021                 block = rb_entry(rb_node, struct tree_block, rb_node);
3022
3023                 node = build_backref_tree(rc, &block->key,
3024                                           block->level, block->bytenr);
3025                 if (IS_ERR(node)) {
3026                         err = PTR_ERR(node);
3027                         goto out;
3028                 }
3029
3030                 ret = relocate_tree_block(trans, rc, node, &block->key,
3031                                           path);
3032                 if (ret < 0) {
3033                         if (ret != -EAGAIN || rb_node == rb_first(blocks))
3034                                 err = ret;
3035                         goto out;
3036                 }
3037                 rb_node = rb_next(rb_node);
3038         }
3039 out:
3040         err = finish_pending_nodes(trans, rc, path, err);
3041
3042 out_free_path:
3043         btrfs_free_path(path);
3044 out_free_blocks:
3045         free_block_list(blocks);
3046         return err;
3047 }
3048
3049 static noinline_for_stack
3050 int prealloc_file_extent_cluster(struct inode *inode,
3051                                  struct file_extent_cluster *cluster)
3052 {
3053         u64 alloc_hint = 0;
3054         u64 start;
3055         u64 end;
3056         u64 offset = BTRFS_I(inode)->index_cnt;
3057         u64 num_bytes;
3058         int nr = 0;
3059         int ret = 0;
3060         u64 prealloc_start = cluster->start - offset;
3061         u64 prealloc_end = cluster->end - offset;
3062         u64 cur_offset;
3063         struct extent_changeset *data_reserved = NULL;
3064
3065         BUG_ON(cluster->start != cluster->boundary[0]);
3066         inode_lock(inode);
3067
3068         ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3069                                           prealloc_end + 1 - prealloc_start);
3070         if (ret)
3071                 goto out;
3072
3073         cur_offset = prealloc_start;
3074         while (nr < cluster->nr) {
3075                 start = cluster->boundary[nr] - offset;
3076                 if (nr + 1 < cluster->nr)
3077                         end = cluster->boundary[nr + 1] - 1 - offset;
3078                 else
3079                         end = cluster->end - offset;
3080
3081                 lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3082                 num_bytes = end + 1 - start;
3083                 if (cur_offset < start)
3084                         btrfs_free_reserved_data_space(inode, data_reserved,
3085                                         cur_offset, start - cur_offset);
3086                 ret = btrfs_prealloc_file_range(inode, 0, start,
3087                                                 num_bytes, num_bytes,
3088                                                 end + 1, &alloc_hint);
3089                 cur_offset = end + 1;
3090                 unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3091                 if (ret)
3092                         break;
3093                 nr++;
3094         }
3095         if (cur_offset < prealloc_end)
3096                 btrfs_free_reserved_data_space(inode, data_reserved,
3097                                 cur_offset, prealloc_end + 1 - cur_offset);
3098 out:
3099         inode_unlock(inode);
3100         extent_changeset_free(data_reserved);
3101         return ret;
3102 }
3103
3104 static noinline_for_stack
3105 int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3106                          u64 block_start)
3107 {
3108         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3109         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3110         struct extent_map *em;
3111         int ret = 0;
3112
3113         em = alloc_extent_map();
3114         if (!em)
3115                 return -ENOMEM;
3116
3117         em->start = start;
3118         em->len = end + 1 - start;
3119         em->block_len = em->len;
3120         em->block_start = block_start;
3121         em->bdev = fs_info->fs_devices->latest_bdev;
3122         set_bit(EXTENT_FLAG_PINNED, &em->flags);
3123
3124         lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3125         while (1) {
3126                 write_lock(&em_tree->lock);
3127                 ret = add_extent_mapping(em_tree, em, 0);
3128                 write_unlock(&em_tree->lock);
3129                 if (ret != -EEXIST) {
3130                         free_extent_map(em);
3131                         break;
3132                 }
3133                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3134         }
3135         unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3136         return ret;
3137 }
3138
3139 static int relocate_file_extent_cluster(struct inode *inode,
3140                                         struct file_extent_cluster *cluster)
3141 {
3142         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3143         u64 page_start;
3144         u64 page_end;
3145         u64 offset = BTRFS_I(inode)->index_cnt;
3146         unsigned long index;
3147         unsigned long last_index;
3148         struct page *page;
3149         struct file_ra_state *ra;
3150         gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3151         int nr = 0;
3152         int ret = 0;
3153
3154         if (!cluster->nr)
3155                 return 0;
3156
3157         ra = kzalloc(sizeof(*ra), GFP_NOFS);
3158         if (!ra)
3159                 return -ENOMEM;
3160
3161         ret = prealloc_file_extent_cluster(inode, cluster);
3162         if (ret)
3163                 goto out;
3164
3165         file_ra_state_init(ra, inode->i_mapping);
3166
3167         ret = setup_extent_mapping(inode, cluster->start - offset,
3168                                    cluster->end - offset, cluster->start);
3169         if (ret)
3170                 goto out;
3171
3172         index = (cluster->start - offset) >> PAGE_SHIFT;
3173         last_index = (cluster->end - offset) >> PAGE_SHIFT;
3174         while (index <= last_index) {
3175                 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3176                                 PAGE_SIZE);
3177                 if (ret)
3178                         goto out;
3179
3180                 page = find_lock_page(inode->i_mapping, index);
3181                 if (!page) {
3182                         page_cache_sync_readahead(inode->i_mapping,
3183                                                   ra, NULL, index,
3184                                                   last_index + 1 - index);
3185                         page = find_or_create_page(inode->i_mapping, index,
3186                                                    mask);
3187                         if (!page) {
3188                                 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3189                                                         PAGE_SIZE, true);
3190                                 ret = -ENOMEM;
3191                                 goto out;
3192                         }
3193                 }
3194
3195                 if (PageReadahead(page)) {
3196                         page_cache_async_readahead(inode->i_mapping,
3197                                                    ra, NULL, page, index,
3198                                                    last_index + 1 - index);
3199                 }
3200
3201                 if (!PageUptodate(page)) {
3202                         btrfs_readpage(NULL, page);
3203                         lock_page(page);
3204                         if (!PageUptodate(page)) {
3205                                 unlock_page(page);
3206                                 put_page(page);
3207                                 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3208                                                         PAGE_SIZE, true);
3209                                 btrfs_delalloc_release_extents(BTRFS_I(inode),
3210                                                                PAGE_SIZE, true);
3211                                 ret = -EIO;
3212                                 goto out;
3213                         }
3214                 }
3215
3216                 page_start = page_offset(page);
3217                 page_end = page_start + PAGE_SIZE - 1;
3218
3219                 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3220
3221                 set_page_extent_mapped(page);
3222
3223                 if (nr < cluster->nr &&
3224                     page_start + offset == cluster->boundary[nr]) {
3225                         set_extent_bits(&BTRFS_I(inode)->io_tree,
3226                                         page_start, page_end,
3227                                         EXTENT_BOUNDARY);
3228                         nr++;
3229                 }
3230
3231                 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3232                                                 NULL, 0);
3233                 if (ret) {
3234                         unlock_page(page);
3235                         put_page(page);
3236                         btrfs_delalloc_release_metadata(BTRFS_I(inode),
3237                    &nbs