btrfs-progs: extent-tree: Enhance btrfs_record_file_extent
[platform/upstream/btrfs-progs.git] / extent-tree.c
1 /*
2  * Copyright (C) 2007 Oracle.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18
19 #include <stdio.h>
20 #include <stdlib.h>
21 #include <stdint.h>
22 #include <math.h>
23 #include "kerncompat.h"
24 #include "radix-tree.h"
25 #include "ctree.h"
26 #include "disk-io.h"
27 #include "print-tree.h"
28 #include "transaction.h"
29 #include "crc32c.h"
30 #include "volumes.h"
31 #include "free-space-cache.h"
32 #include "utils.h"
33
34 #define PENDING_EXTENT_INSERT 0
35 #define PENDING_EXTENT_DELETE 1
36 #define PENDING_BACKREF_UPDATE 2
37
38 struct pending_extent_op {
39         int type;
40         u64 bytenr;
41         u64 num_bytes;
42         u64 flags;
43         struct btrfs_disk_key key;
44         int level;
45 };
46
47 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
48                                      struct btrfs_root *root,
49                                      u64 root_objectid, u64 generation,
50                                      u64 flags, struct btrfs_disk_key *key,
51                                      int level, struct btrfs_key *ins);
52 static int __free_extent(struct btrfs_trans_handle *trans,
53                          struct btrfs_root *root,
54                          u64 bytenr, u64 num_bytes, u64 parent,
55                          u64 root_objectid, u64 owner_objectid,
56                          u64 owner_offset, int refs_to_drop);
57 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
58                                  btrfs_root *extent_root);
59 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
60                                btrfs_root *extent_root);
61 static struct btrfs_block_group_cache *
62 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
63                        *hint, u64 search_start, int data, int owner);
64
65 static int remove_sb_from_cache(struct btrfs_root *root,
66                                 struct btrfs_block_group_cache *cache)
67 {
68         u64 bytenr;
69         u64 *logical;
70         int stripe_len;
71         int i, nr, ret;
72         struct extent_io_tree *free_space_cache;
73
74         free_space_cache = &root->fs_info->free_space_cache;
75         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
76                 bytenr = btrfs_sb_offset(i);
77                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
78                                        cache->key.objectid, bytenr, 0,
79                                        &logical, &nr, &stripe_len);
80                 BUG_ON(ret);
81                 while (nr--) {
82                         clear_extent_dirty(free_space_cache, logical[nr],
83                                 logical[nr] + stripe_len - 1, GFP_NOFS);
84                 }
85                 kfree(logical);
86         }
87         return 0;
88 }
89
90 static int cache_block_group(struct btrfs_root *root,
91                              struct btrfs_block_group_cache *block_group)
92 {
93         struct btrfs_path *path;
94         int ret;
95         struct btrfs_key key;
96         struct extent_buffer *leaf;
97         struct extent_io_tree *free_space_cache;
98         int slot;
99         u64 last;
100         u64 hole_size;
101
102         if (!block_group)
103                 return 0;
104
105         root = root->fs_info->extent_root;
106         free_space_cache = &root->fs_info->free_space_cache;
107
108         if (block_group->cached)
109                 return 0;
110
111         path = btrfs_alloc_path();
112         if (!path)
113                 return -ENOMEM;
114
115         path->reada = 2;
116         last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
117         key.objectid = last;
118         key.offset = 0;
119         key.type = 0;
120
121         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
122         if (ret < 0)
123                 goto err;
124
125         while(1) {
126                 leaf = path->nodes[0];
127                 slot = path->slots[0];
128                 if (slot >= btrfs_header_nritems(leaf)) {
129                         ret = btrfs_next_leaf(root, path);
130                         if (ret < 0)
131                                 goto err;
132                         if (ret == 0) {
133                                 continue;
134                         } else {
135                                 break;
136                         }
137                 }
138                 btrfs_item_key_to_cpu(leaf, &key, slot);
139                 if (key.objectid < block_group->key.objectid) {
140                         goto next;
141                 }
142                 if (key.objectid >= block_group->key.objectid +
143                     block_group->key.offset) {
144                         break;
145                 }
146
147                 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
148                     key.type == BTRFS_METADATA_ITEM_KEY) {
149                         if (key.objectid > last) {
150                                 hole_size = key.objectid - last;
151                                 set_extent_dirty(free_space_cache, last,
152                                                  last + hole_size - 1,
153                                                  GFP_NOFS);
154                         }
155                         if (key.type == BTRFS_METADATA_ITEM_KEY)
156                                 last = key.objectid + root->nodesize;
157                         else
158                                 last = key.objectid + key.offset;
159                 }
160 next:
161                 path->slots[0]++;
162         }
163
164         if (block_group->key.objectid +
165             block_group->key.offset > last) {
166                 hole_size = block_group->key.objectid +
167                         block_group->key.offset - last;
168                 set_extent_dirty(free_space_cache, last,
169                                  last + hole_size - 1, GFP_NOFS);
170         }
171         remove_sb_from_cache(root, block_group);
172         block_group->cached = 1;
173 err:
174         btrfs_free_path(path);
175         return 0;
176 }
177
178 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
179                                                        btrfs_fs_info *info,
180                                                        u64 bytenr)
181 {
182         struct extent_io_tree *block_group_cache;
183         struct btrfs_block_group_cache *block_group = NULL;
184         u64 ptr;
185         u64 start;
186         u64 end;
187         int ret;
188
189         bytenr = max_t(u64, bytenr,
190                        BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
191         block_group_cache = &info->block_group_cache;
192         ret = find_first_extent_bit(block_group_cache,
193                                     bytenr, &start, &end,
194                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
195                                     BLOCK_GROUP_SYSTEM);
196         if (ret) {
197                 return NULL;
198         }
199         ret = get_state_private(block_group_cache, start, &ptr);
200         if (ret)
201                 return NULL;
202
203         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
204         return block_group;
205 }
206
207 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
208                                                          btrfs_fs_info *info,
209                                                          u64 bytenr)
210 {
211         struct extent_io_tree *block_group_cache;
212         struct btrfs_block_group_cache *block_group = NULL;
213         u64 ptr;
214         u64 start;
215         u64 end;
216         int ret;
217
218         block_group_cache = &info->block_group_cache;
219         ret = find_first_extent_bit(block_group_cache,
220                                     bytenr, &start, &end,
221                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
222                                     BLOCK_GROUP_SYSTEM);
223         if (ret) {
224                 return NULL;
225         }
226         ret = get_state_private(block_group_cache, start, &ptr);
227         if (ret)
228                 return NULL;
229
230         block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
231         if (block_group->key.objectid <= bytenr && bytenr <
232             block_group->key.objectid + block_group->key.offset)
233                 return block_group;
234         return NULL;
235 }
236
237 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
238 {
239         return (cache->flags & bits) == bits;
240 }
241
242 static int noinline find_search_start(struct btrfs_root *root,
243                               struct btrfs_block_group_cache **cache_ret,
244                               u64 *start_ret, int num, int data)
245 {
246         int ret;
247         struct btrfs_block_group_cache *cache = *cache_ret;
248         u64 last = *start_ret;
249         u64 start = 0;
250         u64 end = 0;
251         u64 search_start = *start_ret;
252         int wrapped = 0;
253
254         if (!cache)
255                 goto out;
256 again:
257         ret = cache_block_group(root, cache);
258         if (ret)
259                 goto out;
260
261         last = max(search_start, cache->key.objectid);
262         if (cache->ro || !block_group_bits(cache, data))
263                 goto new_group;
264
265         while(1) {
266                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
267                                             last, &start, &end, EXTENT_DIRTY);
268                 if (ret) {
269                         goto new_group;
270                 }
271
272                 start = max(last, start);
273                 last = end + 1;
274                 if (last - start < num) {
275                         continue;
276                 }
277                 if (start + num > cache->key.objectid + cache->key.offset) {
278                         goto new_group;
279                 }
280                 *start_ret = start;
281                 return 0;
282         }
283 out:
284         *start_ret = last;
285         cache = btrfs_lookup_block_group(root->fs_info, search_start);
286         if (!cache) {
287                 printk("Unable to find block group for %llu\n",
288                         (unsigned long long)search_start);
289                 WARN_ON(1);
290         }
291         return -ENOSPC;
292
293 new_group:
294         last = cache->key.objectid + cache->key.offset;
295 wrapped:
296         cache = btrfs_lookup_first_block_group(root->fs_info, last);
297         if (!cache) {
298                 if (!wrapped) {
299                         wrapped = 1;
300                         last = search_start;
301                         goto wrapped;
302                 }
303                 goto out;
304         }
305         *cache_ret = cache;
306         goto again;
307 }
308
309 static int block_group_state_bits(u64 flags)
310 {
311         int bits = 0;
312         if (flags & BTRFS_BLOCK_GROUP_DATA)
313                 bits |= BLOCK_GROUP_DATA;
314         if (flags & BTRFS_BLOCK_GROUP_METADATA)
315                 bits |= BLOCK_GROUP_METADATA;
316         if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
317                 bits |= BLOCK_GROUP_SYSTEM;
318         return bits;
319 }
320
321 static struct btrfs_block_group_cache *
322 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
323                        *hint, u64 search_start, int data, int owner)
324 {
325         struct btrfs_block_group_cache *cache;
326         struct extent_io_tree *block_group_cache;
327         struct btrfs_block_group_cache *found_group = NULL;
328         struct btrfs_fs_info *info = root->fs_info;
329         u64 used;
330         u64 last = 0;
331         u64 hint_last;
332         u64 start;
333         u64 end;
334         u64 free_check;
335         u64 ptr;
336         int bit;
337         int ret;
338         int full_search = 0;
339         int factor = 10;
340
341         block_group_cache = &info->block_group_cache;
342
343         if (!owner)
344                 factor = 10;
345
346         bit = block_group_state_bits(data);
347
348         if (search_start) {
349                 struct btrfs_block_group_cache *shint;
350                 shint = btrfs_lookup_block_group(info, search_start);
351                 if (shint && !shint->ro && block_group_bits(shint, data)) {
352                         used = btrfs_block_group_used(&shint->item);
353                         if (used + shint->pinned <
354                             div_factor(shint->key.offset, factor)) {
355                                 return shint;
356                         }
357                 }
358         }
359         if (hint && !hint->ro && block_group_bits(hint, data)) {
360                 used = btrfs_block_group_used(&hint->item);
361                 if (used + hint->pinned <
362                     div_factor(hint->key.offset, factor)) {
363                         return hint;
364                 }
365                 last = hint->key.objectid + hint->key.offset;
366                 hint_last = last;
367         } else {
368                 if (hint)
369                         hint_last = max(hint->key.objectid, search_start);
370                 else
371                         hint_last = search_start;
372
373                 last = hint_last;
374         }
375 again:
376         while(1) {
377                 ret = find_first_extent_bit(block_group_cache, last,
378                                             &start, &end, bit);
379                 if (ret)
380                         break;
381
382                 ret = get_state_private(block_group_cache, start, &ptr);
383                 if (ret)
384                         break;
385
386                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
387                 last = cache->key.objectid + cache->key.offset;
388                 used = btrfs_block_group_used(&cache->item);
389
390                 if (!cache->ro && block_group_bits(cache, data)) {
391                         if (full_search)
392                                 free_check = cache->key.offset;
393                         else
394                                 free_check = div_factor(cache->key.offset,
395                                                         factor);
396
397                         if (used + cache->pinned < free_check) {
398                                 found_group = cache;
399                                 goto found;
400                         }
401                 }
402                 cond_resched();
403         }
404         if (!full_search) {
405                 last = search_start;
406                 full_search = 1;
407                 goto again;
408         }
409 found:
410         return found_group;
411 }
412
413 /*
414  * Back reference rules.  Back refs have three main goals:
415  *
416  * 1) differentiate between all holders of references to an extent so that
417  *    when a reference is dropped we can make sure it was a valid reference
418  *    before freeing the extent.
419  *
420  * 2) Provide enough information to quickly find the holders of an extent
421  *    if we notice a given block is corrupted or bad.
422  *
423  * 3) Make it easy to migrate blocks for FS shrinking or storage pool
424  *    maintenance.  This is actually the same as #2, but with a slightly
425  *    different use case.
426  *
427  * There are two kinds of back refs. The implicit back refs is optimized
428  * for pointers in non-shared tree blocks. For a given pointer in a block,
429  * back refs of this kind provide information about the block's owner tree
430  * and the pointer's key. These information allow us to find the block by
431  * b-tree searching. The full back refs is for pointers in tree blocks not
432  * referenced by their owner trees. The location of tree block is recorded
433  * in the back refs. Actually the full back refs is generic, and can be
434  * used in all cases the implicit back refs is used. The major shortcoming
435  * of the full back refs is its overhead. Every time a tree block gets
436  * COWed, we have to update back refs entry for all pointers in it.
437  *
438  * For a newly allocated tree block, we use implicit back refs for
439  * pointers in it. This means most tree related operations only involve
440  * implicit back refs. For a tree block created in old transaction, the
441  * only way to drop a reference to it is COW it. So we can detect the
442  * event that tree block loses its owner tree's reference and do the
443  * back refs conversion.
444  *
445  * When a tree block is COW'd through a tree, there are four cases:
446  *
447  * The reference count of the block is one and the tree is the block's
448  * owner tree. Nothing to do in this case.
449  *
450  * The reference count of the block is one and the tree is not the
451  * block's owner tree. In this case, full back refs is used for pointers
452  * in the block. Remove these full back refs, add implicit back refs for
453  * every pointers in the new block.
454  *
455  * The reference count of the block is greater than one and the tree is
456  * the block's owner tree. In this case, implicit back refs is used for
457  * pointers in the block. Add full back refs for every pointers in the
458  * block, increase lower level extents' reference counts. The original
459  * implicit back refs are entailed to the new block.
460  *
461  * The reference count of the block is greater than one and the tree is
462  * not the block's owner tree. Add implicit back refs for every pointer in
463  * the new block, increase lower level extents' reference count.
464  *
465  * Back Reference Key composing:
466  *
467  * The key objectid corresponds to the first byte in the extent,
468  * The key type is used to differentiate between types of back refs.
469  * There are different meanings of the key offset for different types
470  * of back refs.
471  *
472  * File extents can be referenced by:
473  *
474  * - multiple snapshots, subvolumes, or different generations in one subvol
475  * - different files inside a single subvolume
476  * - different offsets inside a file (bookend extents in file.c)
477  *
478  * The extent ref structure for the implicit back refs has fields for:
479  *
480  * - Objectid of the subvolume root
481  * - objectid of the file holding the reference
482  * - original offset in the file
483  * - how many bookend extents
484  *
485  * The key offset for the implicit back refs is hash of the first
486  * three fields.
487  *
488  * The extent ref structure for the full back refs has field for:
489  *
490  * - number of pointers in the tree leaf
491  *
492  * The key offset for the implicit back refs is the first byte of
493  * the tree leaf
494  *
495  * When a file extent is allocated, The implicit back refs is used.
496  * the fields are filled in:
497  *
498  *     (root_key.objectid, inode objectid, offset in file, 1)
499  *
500  * When a file extent is removed file truncation, we find the
501  * corresponding implicit back refs and check the following fields:
502  *
503  *     (btrfs_header_owner(leaf), inode objectid, offset in file)
504  *
505  * Btree extents can be referenced by:
506  *
507  * - Different subvolumes
508  *
509  * Both the implicit back refs and the full back refs for tree blocks
510  * only consist of key. The key offset for the implicit back refs is
511  * objectid of block's owner tree. The key offset for the full back refs
512  * is the first byte of parent block.
513  *
514  * When implicit back refs is used, information about the lowest key and
515  * level of the tree block are required. These information are stored in
516  * tree block info structure.
517  */
518
519 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
520 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
521                                   struct btrfs_root *root,
522                                   struct btrfs_path *path,
523                                   u64 owner, u32 extra_size)
524 {
525         struct btrfs_extent_item *item;
526         struct btrfs_extent_item_v0 *ei0;
527         struct btrfs_extent_ref_v0 *ref0;
528         struct btrfs_tree_block_info *bi;
529         struct extent_buffer *leaf;
530         struct btrfs_key key;
531         struct btrfs_key found_key;
532         u32 new_size = sizeof(*item);
533         u64 refs;
534         int ret;
535
536         leaf = path->nodes[0];
537         BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
538
539         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
540         ei0 = btrfs_item_ptr(leaf, path->slots[0],
541                              struct btrfs_extent_item_v0);
542         refs = btrfs_extent_refs_v0(leaf, ei0);
543
544         if (owner == (u64)-1) {
545                 while (1) {
546                         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
547                                 ret = btrfs_next_leaf(root, path);
548                                 if (ret < 0)
549                                         return ret;
550                                 BUG_ON(ret > 0);
551                                 leaf = path->nodes[0];
552                         }
553                         btrfs_item_key_to_cpu(leaf, &found_key,
554                                               path->slots[0]);
555                         BUG_ON(key.objectid != found_key.objectid);
556                         if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
557                                 path->slots[0]++;
558                                 continue;
559                         }
560                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
561                                               struct btrfs_extent_ref_v0);
562                         owner = btrfs_ref_objectid_v0(leaf, ref0);
563                         break;
564                 }
565         }
566         btrfs_release_path(path);
567
568         if (owner < BTRFS_FIRST_FREE_OBJECTID)
569                 new_size += sizeof(*bi);
570
571         new_size -= sizeof(*ei0);
572         ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
573         if (ret < 0)
574                 return ret;
575         BUG_ON(ret);
576
577         ret = btrfs_extend_item(trans, root, path, new_size);
578         BUG_ON(ret);
579
580         leaf = path->nodes[0];
581         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
582         btrfs_set_extent_refs(leaf, item, refs);
583         /* FIXME: get real generation */
584         btrfs_set_extent_generation(leaf, item, 0);
585         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
586                 btrfs_set_extent_flags(leaf, item,
587                                        BTRFS_EXTENT_FLAG_TREE_BLOCK |
588                                        BTRFS_BLOCK_FLAG_FULL_BACKREF);
589                 bi = (struct btrfs_tree_block_info *)(item + 1);
590                 /* FIXME: get first key of the block */
591                 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
592                 btrfs_set_tree_block_level(leaf, bi, (int)owner);
593         } else {
594                 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
595         }
596         btrfs_mark_buffer_dirty(leaf);
597         return 0;
598 }
599 #endif
600
601 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
602 {
603         u32 high_crc = ~(u32)0;
604         u32 low_crc = ~(u32)0;
605         __le64 lenum;
606
607         lenum = cpu_to_le64(root_objectid);
608         high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
609         lenum = cpu_to_le64(owner);
610         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
611         lenum = cpu_to_le64(offset);
612         low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
613
614         return ((u64)high_crc << 31) ^ (u64)low_crc;
615 }
616
617 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
618                                      struct btrfs_extent_data_ref *ref)
619 {
620         return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
621                                     btrfs_extent_data_ref_objectid(leaf, ref),
622                                     btrfs_extent_data_ref_offset(leaf, ref));
623 }
624
625 static int match_extent_data_ref(struct extent_buffer *leaf,
626                                  struct btrfs_extent_data_ref *ref,
627                                  u64 root_objectid, u64 owner, u64 offset)
628 {
629         if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
630             btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
631             btrfs_extent_data_ref_offset(leaf, ref) != offset)
632                 return 0;
633         return 1;
634 }
635
636 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
637                                            struct btrfs_root *root,
638                                            struct btrfs_path *path,
639                                            u64 bytenr, u64 parent,
640                                            u64 root_objectid,
641                                            u64 owner, u64 offset)
642 {
643         struct btrfs_key key;
644         struct btrfs_extent_data_ref *ref;
645         struct extent_buffer *leaf;
646         u32 nritems;
647         int ret;
648         int recow;
649         int err = -ENOENT;
650
651         key.objectid = bytenr;
652         if (parent) {
653                 key.type = BTRFS_SHARED_DATA_REF_KEY;
654                 key.offset = parent;
655         } else {
656                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
657                 key.offset = hash_extent_data_ref(root_objectid,
658                                                   owner, offset);
659         }
660 again:
661         recow = 0;
662         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
663         if (ret < 0) {
664                 err = ret;
665                 goto fail;
666         }
667
668         if (parent) {
669                 if (!ret)
670                         return 0;
671 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
672                 key.type = BTRFS_EXTENT_REF_V0_KEY;
673                 btrfs_release_path(path);
674                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
675                 if (ret < 0) {
676                         err = ret;
677                         goto fail;
678                 }
679                 if (!ret)
680                         return 0;
681 #endif
682                 goto fail;
683         }
684
685         leaf = path->nodes[0];
686         nritems = btrfs_header_nritems(leaf);
687         while (1) {
688                 if (path->slots[0] >= nritems) {
689                         ret = btrfs_next_leaf(root, path);
690                         if (ret < 0)
691                                 err = ret;
692                         if (ret)
693                                 goto fail;
694
695                         leaf = path->nodes[0];
696                         nritems = btrfs_header_nritems(leaf);
697                         recow = 1;
698                 }
699
700                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
701                 if (key.objectid != bytenr ||
702                     key.type != BTRFS_EXTENT_DATA_REF_KEY)
703                         goto fail;
704                 
705                 ref = btrfs_item_ptr(leaf, path->slots[0],
706                                      struct btrfs_extent_data_ref);
707
708                 if (match_extent_data_ref(leaf, ref, root_objectid,
709                                           owner, offset)) {
710                         if (recow) {
711                                 btrfs_release_path(path);
712                                 goto again;
713                         }
714                         err = 0;
715                         break;
716                 }
717                 path->slots[0]++;
718         }
719 fail:
720         return err;
721 }
722
723 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
724                                            struct btrfs_root *root,
725                                            struct btrfs_path *path,
726                                            u64 bytenr, u64 parent,
727                                            u64 root_objectid, u64 owner,
728                                            u64 offset, int refs_to_add)
729 {
730         struct btrfs_key key;
731         struct extent_buffer *leaf;
732         u32 size;
733         u32 num_refs;
734         int ret;
735
736         key.objectid = bytenr;
737         if (parent) {
738                 key.type = BTRFS_SHARED_DATA_REF_KEY;
739                 key.offset = parent;
740                 size = sizeof(struct btrfs_shared_data_ref);
741         } else {
742                 key.type = BTRFS_EXTENT_DATA_REF_KEY;
743                 key.offset = hash_extent_data_ref(root_objectid,
744                                                   owner, offset);
745                 size = sizeof(struct btrfs_extent_data_ref);
746         }
747
748         ret = btrfs_insert_empty_item(trans, root, path, &key, size);
749         if (ret && ret != -EEXIST)
750                 goto fail;
751
752         leaf = path->nodes[0];
753         if (parent) {
754                 struct btrfs_shared_data_ref *ref;
755                 ref = btrfs_item_ptr(leaf, path->slots[0],
756                                      struct btrfs_shared_data_ref);
757                 if (ret == 0) {
758                         btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
759                 } else {
760                         num_refs = btrfs_shared_data_ref_count(leaf, ref);
761                         num_refs += refs_to_add;
762                         btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
763                 }
764         } else {
765                 struct btrfs_extent_data_ref *ref;
766                 while (ret == -EEXIST) {
767                         ref = btrfs_item_ptr(leaf, path->slots[0],
768                                              struct btrfs_extent_data_ref);
769                         if (match_extent_data_ref(leaf, ref, root_objectid,
770                                                   owner, offset))
771                                 break;
772                         btrfs_release_path(path);
773
774                         key.offset++;
775                         ret = btrfs_insert_empty_item(trans, root, path, &key,
776                                                       size);
777                         if (ret && ret != -EEXIST)
778                                 goto fail;
779
780                         leaf = path->nodes[0];
781                 }
782                 ref = btrfs_item_ptr(leaf, path->slots[0],
783                                      struct btrfs_extent_data_ref);
784                 if (ret == 0) {
785                         btrfs_set_extent_data_ref_root(leaf, ref,
786                                                        root_objectid);
787                         btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
788                         btrfs_set_extent_data_ref_offset(leaf, ref, offset);
789                         btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
790                 } else {
791                         num_refs = btrfs_extent_data_ref_count(leaf, ref);
792                         num_refs += refs_to_add;
793                         btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
794                 }
795         }
796         btrfs_mark_buffer_dirty(leaf);
797         ret = 0;
798 fail:
799         btrfs_release_path(path);
800         return ret;
801 }
802
803 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
804                                            struct btrfs_root *root,
805                                            struct btrfs_path *path,
806                                            int refs_to_drop)
807 {
808         struct btrfs_key key;
809         struct btrfs_extent_data_ref *ref1 = NULL;
810         struct btrfs_shared_data_ref *ref2 = NULL;
811         struct extent_buffer *leaf;
812         u32 num_refs = 0;
813         int ret = 0;
814
815         leaf = path->nodes[0];
816         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
817
818         if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
819                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
820                                       struct btrfs_extent_data_ref);
821                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
822         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
823                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
824                                       struct btrfs_shared_data_ref);
825                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
826 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
827         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
828                 struct btrfs_extent_ref_v0 *ref0;
829                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
830                                       struct btrfs_extent_ref_v0);
831                 num_refs = btrfs_ref_count_v0(leaf, ref0);
832 #endif
833         } else {
834                 BUG();
835         }
836
837         BUG_ON(num_refs < refs_to_drop);
838         num_refs -= refs_to_drop;
839
840         if (num_refs == 0) {
841                 ret = btrfs_del_item(trans, root, path);
842         } else {
843                 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
844                         btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
845                 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
846                         btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
847 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
848                 else {
849                         struct btrfs_extent_ref_v0 *ref0;
850                         ref0 = btrfs_item_ptr(leaf, path->slots[0],
851                                         struct btrfs_extent_ref_v0);
852                         btrfs_set_ref_count_v0(leaf, ref0, num_refs);
853                 }
854 #endif
855                 btrfs_mark_buffer_dirty(leaf);
856         }
857         return ret;
858 }
859
860 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
861                                           struct btrfs_path *path,
862                                           struct btrfs_extent_inline_ref *iref)
863 {
864         struct btrfs_key key;
865         struct extent_buffer *leaf;
866         struct btrfs_extent_data_ref *ref1;
867         struct btrfs_shared_data_ref *ref2;
868         u32 num_refs = 0;
869
870         leaf = path->nodes[0];
871         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
872         if (iref) {
873                 if (btrfs_extent_inline_ref_type(leaf, iref) ==
874                     BTRFS_EXTENT_DATA_REF_KEY) {
875                         ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
876                         num_refs = btrfs_extent_data_ref_count(leaf, ref1);
877                 } else {
878                         ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
879                         num_refs = btrfs_shared_data_ref_count(leaf, ref2);
880                 }
881         } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
882                 ref1 = btrfs_item_ptr(leaf, path->slots[0],
883                                       struct btrfs_extent_data_ref);
884                 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
885         } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
886                 ref2 = btrfs_item_ptr(leaf, path->slots[0],
887                                       struct btrfs_shared_data_ref);
888                 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
889 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
890         } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
891                 struct btrfs_extent_ref_v0 *ref0;
892                 ref0 = btrfs_item_ptr(leaf, path->slots[0],
893                                       struct btrfs_extent_ref_v0);
894                 num_refs = btrfs_ref_count_v0(leaf, ref0);
895 #endif
896         } else {
897                 BUG();
898         }
899         return num_refs;
900 }
901
902 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
903                                           struct btrfs_root *root,
904                                           struct btrfs_path *path,
905                                           u64 bytenr, u64 parent,
906                                           u64 root_objectid)
907 {
908         struct btrfs_key key;
909         int ret;
910
911         key.objectid = bytenr;
912         if (parent) {
913                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
914                 key.offset = parent;
915         } else {
916                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
917                 key.offset = root_objectid;
918         }
919
920         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
921         if (ret > 0)
922                 ret = -ENOENT;
923 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
924         if (ret == -ENOENT && parent) {
925                 btrfs_release_path(path);
926                 key.type = BTRFS_EXTENT_REF_V0_KEY;
927                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
928                 if (ret > 0)
929                         ret = -ENOENT;
930         }
931 #endif
932         return ret;
933 }
934
935 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
936                                           struct btrfs_root *root,
937                                           struct btrfs_path *path,
938                                           u64 bytenr, u64 parent,
939                                           u64 root_objectid)
940 {
941         struct btrfs_key key;
942         int ret;
943
944         key.objectid = bytenr;
945         if (parent) {
946                 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
947                 key.offset = parent;
948         } else {
949                 key.type = BTRFS_TREE_BLOCK_REF_KEY;
950                 key.offset = root_objectid;
951         }
952
953         ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
954
955         btrfs_release_path(path);
956         return ret;
957 }
958
959 static inline int extent_ref_type(u64 parent, u64 owner)
960 {
961         int type;
962         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
963                 if (parent > 0)
964                         type = BTRFS_SHARED_BLOCK_REF_KEY;
965                 else
966                         type = BTRFS_TREE_BLOCK_REF_KEY;
967         } else {
968                 if (parent > 0)
969                         type = BTRFS_SHARED_DATA_REF_KEY;
970                 else
971                         type = BTRFS_EXTENT_DATA_REF_KEY;
972         }
973         return type;
974 }
975
976 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
977                                  struct btrfs_root *root,
978                                  struct btrfs_path *path,
979                                  struct btrfs_extent_inline_ref **ref_ret,
980                                  u64 bytenr, u64 num_bytes,
981                                  u64 parent, u64 root_objectid,
982                                  u64 owner, u64 offset, int insert)
983 {
984         struct btrfs_key key;
985         struct extent_buffer *leaf;
986         struct btrfs_extent_item *ei;
987         struct btrfs_extent_inline_ref *iref;
988         u64 flags;
989         u32 item_size;
990         unsigned long ptr;
991         unsigned long end;
992         int extra_size;
993         int type;
994         int want;
995         int ret;
996         int err = 0;
997         int skinny_metadata =
998                 btrfs_fs_incompat(root->fs_info,
999                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1000
1001         key.objectid = bytenr;
1002         key.type = BTRFS_EXTENT_ITEM_KEY;
1003         key.offset = num_bytes;
1004
1005         want = extent_ref_type(parent, owner);
1006         if (insert)
1007                 extra_size = btrfs_extent_inline_ref_size(want);
1008         else
1009                 extra_size = -1;
1010
1011         if (owner < BTRFS_FIRST_FREE_OBJECTID && skinny_metadata) {
1012                 skinny_metadata = 1;
1013                 key.type = BTRFS_METADATA_ITEM_KEY;
1014                 key.offset = owner;
1015         } else if (skinny_metadata) {
1016                 skinny_metadata = 0;
1017         }
1018
1019 again:
1020         ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1021         if (ret < 0) {
1022                 err = ret;
1023                 goto out;
1024         }
1025
1026         /*
1027          * We may be a newly converted file system which still has the old fat
1028          * extent entries for metadata, so try and see if we have one of those.
1029          */
1030         if (ret > 0 && skinny_metadata) {
1031                 skinny_metadata = 0;
1032                 if (path->slots[0]) {
1033                         path->slots[0]--;
1034                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1035                                               path->slots[0]);
1036                         if (key.objectid == bytenr &&
1037                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1038                             key.offset == num_bytes)
1039                                 ret = 0;
1040                 }
1041                 if (ret) {
1042                         key.type = BTRFS_EXTENT_ITEM_KEY;
1043                         key.offset = num_bytes;
1044                         btrfs_release_path(path);
1045                         goto again;
1046                 }
1047         }
1048
1049         if (ret) {
1050                 printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
1051                 return -ENOENT;
1052         }
1053
1054         BUG_ON(ret);
1055
1056         leaf = path->nodes[0];
1057         item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1058 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1059         if (item_size < sizeof(*ei)) {
1060                 if (!insert) {
1061                         err = -ENOENT;
1062                         goto out;
1063                 }
1064                 ret = convert_extent_item_v0(trans, root, path, owner,
1065                                              extra_size);
1066                 if (ret < 0) {
1067                         err = ret;
1068                         goto out;
1069                 }
1070                 leaf = path->nodes[0];
1071                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1072         }
1073 #endif
1074         if (item_size < sizeof(*ei)) {
1075                 printf("Size is %u, needs to be %u, slot %d\n",
1076                        (unsigned)item_size,
1077                        (unsigned)sizeof(*ei), path->slots[0]);
1078                 btrfs_print_leaf(root, leaf);
1079                 return -EINVAL;
1080         }
1081         BUG_ON(item_size < sizeof(*ei));
1082
1083         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1084         flags = btrfs_extent_flags(leaf, ei);
1085
1086         ptr = (unsigned long)(ei + 1);
1087         end = (unsigned long)ei + item_size;
1088
1089         if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1090                 ptr += sizeof(struct btrfs_tree_block_info);
1091                 BUG_ON(ptr > end);
1092         } else if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1093                 if (!(flags & BTRFS_EXTENT_FLAG_DATA)) {
1094                         return -EIO;
1095                 }
1096         }
1097
1098         err = -ENOENT;
1099         while (1) {
1100                 if (ptr >= end) {
1101                         WARN_ON(ptr > end);
1102                         break;
1103                 }
1104                 iref = (struct btrfs_extent_inline_ref *)ptr;
1105                 type = btrfs_extent_inline_ref_type(leaf, iref);
1106                 if (want < type)
1107                         break;
1108                 if (want > type) {
1109                         ptr += btrfs_extent_inline_ref_size(type);
1110                         continue;
1111                 }
1112
1113                 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1114                         struct btrfs_extent_data_ref *dref;
1115                         dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1116                         if (match_extent_data_ref(leaf, dref, root_objectid,
1117                                                   owner, offset)) {
1118                                 err = 0;
1119                                 break;
1120                         }
1121                         if (hash_extent_data_ref_item(leaf, dref) <
1122                             hash_extent_data_ref(root_objectid, owner, offset))
1123                                 break;
1124                 } else {
1125                         u64 ref_offset;
1126                         ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1127                         if (parent > 0) {
1128                                 if (parent == ref_offset) {
1129                                         err = 0;
1130                                         break;
1131                                 }
1132                                 if (ref_offset < parent)
1133                                         break;
1134                         } else {
1135                                 if (root_objectid == ref_offset) {
1136                                         err = 0;
1137                                         break;
1138                                 }
1139                                 if (ref_offset < root_objectid)
1140                                         break;
1141                         }
1142                 }
1143                 ptr += btrfs_extent_inline_ref_size(type);
1144         }
1145         if (err == -ENOENT && insert) {
1146                 if (item_size + extra_size >=
1147                     BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1148                         err = -EAGAIN;
1149                         goto out;
1150                 }
1151                 /*
1152                  * To add new inline back ref, we have to make sure
1153                  * there is no corresponding back ref item.
1154                  * For simplicity, we just do not add new inline back
1155                  * ref if there is any back ref item.
1156                  */
1157                 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1158                     key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1159                         err = -EAGAIN;
1160                         goto out;
1161                 }
1162         }
1163         *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1164 out:
1165         return err;
1166 }
1167
1168 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1169                                 struct btrfs_root *root,
1170                                 struct btrfs_path *path,
1171                                 struct btrfs_extent_inline_ref *iref,
1172                                 u64 parent, u64 root_objectid,
1173                                 u64 owner, u64 offset, int refs_to_add)
1174 {
1175         struct extent_buffer *leaf;
1176         struct btrfs_extent_item *ei;
1177         unsigned long ptr;
1178         unsigned long end;
1179         unsigned long item_offset;
1180         u64 refs;
1181         int size;
1182         int type;
1183         int ret;
1184
1185         leaf = path->nodes[0];
1186         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1187         item_offset = (unsigned long)iref - (unsigned long)ei;
1188
1189         type = extent_ref_type(parent, owner);
1190         size = btrfs_extent_inline_ref_size(type);
1191
1192         ret = btrfs_extend_item(trans, root, path, size);
1193         BUG_ON(ret);
1194
1195         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1196         refs = btrfs_extent_refs(leaf, ei);
1197         refs += refs_to_add;
1198         btrfs_set_extent_refs(leaf, ei, refs);
1199
1200         ptr = (unsigned long)ei + item_offset;
1201         end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1202         if (ptr < end - size)
1203                 memmove_extent_buffer(leaf, ptr + size, ptr,
1204                                       end - size - ptr);
1205
1206         iref = (struct btrfs_extent_inline_ref *)ptr;
1207         btrfs_set_extent_inline_ref_type(leaf, iref, type);
1208         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1209                 struct btrfs_extent_data_ref *dref;
1210                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1211                 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1212                 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1213                 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1214                 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1215         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1216                 struct btrfs_shared_data_ref *sref;
1217                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1218                 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1219                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1220         } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1221                 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1222         } else {
1223                 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1224         }
1225         btrfs_mark_buffer_dirty(leaf);
1226         return 0;
1227 }
1228
1229 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1230                                  struct btrfs_root *root,
1231                                  struct btrfs_path *path,
1232                                  struct btrfs_extent_inline_ref **ref_ret,
1233                                  u64 bytenr, u64 num_bytes, u64 parent,
1234                                  u64 root_objectid, u64 owner, u64 offset)
1235 {
1236         int ret;
1237
1238         ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1239                                            bytenr, num_bytes, parent,
1240                                            root_objectid, owner, offset, 0);
1241         if (ret != -ENOENT)
1242                 return ret;
1243
1244         btrfs_release_path(path);
1245         *ref_ret = NULL;
1246
1247         if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1248                 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1249                                             root_objectid);
1250         } else {
1251                 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1252                                              root_objectid, owner, offset);
1253         }
1254         return ret;
1255 }
1256
1257 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1258                                  struct btrfs_root *root,
1259                                  struct btrfs_path *path,
1260                                  struct btrfs_extent_inline_ref *iref,
1261                                  int refs_to_mod)
1262 {
1263         struct extent_buffer *leaf;
1264         struct btrfs_extent_item *ei;
1265         struct btrfs_extent_data_ref *dref = NULL;
1266         struct btrfs_shared_data_ref *sref = NULL;
1267         unsigned long ptr;
1268         unsigned long end;
1269         u32 item_size;
1270         int size;
1271         int type;
1272         int ret;
1273         u64 refs;
1274
1275         leaf = path->nodes[0];
1276         ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1277         refs = btrfs_extent_refs(leaf, ei);
1278         WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1279         refs += refs_to_mod;
1280         btrfs_set_extent_refs(leaf, ei, refs);
1281
1282         type = btrfs_extent_inline_ref_type(leaf, iref);
1283
1284         if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1285                 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1286                 refs = btrfs_extent_data_ref_count(leaf, dref);
1287         } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1288                 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1289                 refs = btrfs_shared_data_ref_count(leaf, sref);
1290         } else {
1291                 refs = 1;
1292                 BUG_ON(refs_to_mod != -1);
1293         }
1294
1295         BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1296         refs += refs_to_mod;
1297
1298         if (refs > 0) {
1299                 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1300                         btrfs_set_extent_data_ref_count(leaf, dref, refs);
1301                 else
1302                         btrfs_set_shared_data_ref_count(leaf, sref, refs);
1303         } else {
1304                 size =  btrfs_extent_inline_ref_size(type);
1305                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1306                 ptr = (unsigned long)iref;
1307                 end = (unsigned long)ei + item_size;
1308                 if (ptr + size < end)
1309                         memmove_extent_buffer(leaf, ptr, ptr + size,
1310                                               end - ptr - size);
1311                 item_size -= size;
1312                 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1313                 BUG_ON(ret);
1314         }
1315         btrfs_mark_buffer_dirty(leaf);
1316         return 0;
1317 }
1318
1319 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1320                                  struct btrfs_root *root,
1321                                  struct btrfs_path *path,
1322                                  u64 bytenr, u64 num_bytes, u64 parent,
1323                                  u64 root_objectid, u64 owner,
1324                                  u64 offset, int refs_to_add)
1325 {
1326         struct btrfs_extent_inline_ref *iref;
1327         int ret;
1328
1329         ret = lookup_inline_extent_backref(trans, root, path, &iref,
1330                                            bytenr, num_bytes, parent,
1331                                            root_objectid, owner, offset, 1);
1332         if (ret == 0) {
1333                 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1334                 ret = update_inline_extent_backref(trans, root, path, iref,
1335                                                    refs_to_add);
1336         } else if (ret == -ENOENT) {
1337                 ret = setup_inline_extent_backref(trans, root, path, iref,
1338                                                   parent, root_objectid,
1339                                                   owner, offset, refs_to_add);
1340         }
1341         return ret;
1342 }
1343
1344 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1345                                  struct btrfs_root *root,
1346                                  struct btrfs_path *path,
1347                                  u64 bytenr, u64 parent, u64 root_objectid,
1348                                  u64 owner, u64 offset, int refs_to_add)
1349 {
1350         int ret;
1351
1352         if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1353                 ret = insert_extent_data_ref(trans, root, path, bytenr,
1354                                              parent, root_objectid,
1355                                              owner, offset, refs_to_add);
1356         } else {
1357                 BUG_ON(refs_to_add != 1);
1358                 ret = insert_tree_block_ref(trans, root, path, bytenr,
1359                                             parent, root_objectid);
1360         }
1361         return ret;
1362 }
1363
1364 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1365                                  struct btrfs_root *root,
1366                                  struct btrfs_path *path,
1367                                  struct btrfs_extent_inline_ref *iref,
1368                                  int refs_to_drop, int is_data)
1369 {
1370         int ret;
1371
1372         BUG_ON(!is_data && refs_to_drop != 1);
1373         if (iref) {
1374                 ret = update_inline_extent_backref(trans, root, path, iref,
1375                                                    -refs_to_drop);
1376         } else if (is_data) {
1377                 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1378         } else {
1379                 ret = btrfs_del_item(trans, root, path);
1380         }
1381         return ret;
1382 }
1383
1384 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1385                          struct btrfs_root *root,
1386                          u64 bytenr, u64 num_bytes, u64 parent,
1387                          u64 root_objectid, u64 owner, u64 offset)
1388 {
1389         struct btrfs_path *path;
1390         struct extent_buffer *leaf;
1391         struct btrfs_extent_item *item;
1392         u64 refs;
1393         int ret;
1394         int err = 0;
1395
1396         path = btrfs_alloc_path();
1397         if (!path)
1398                 return -ENOMEM;
1399
1400         path->reada = 1;
1401
1402         ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1403                                            path, bytenr, num_bytes, parent,
1404                                            root_objectid, owner, offset, 1);
1405         if (ret == 0)
1406                 goto out;
1407
1408         if (ret != -EAGAIN) {
1409                 err = ret;
1410                 goto out;
1411         }
1412         
1413         leaf = path->nodes[0];
1414         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1415         refs = btrfs_extent_refs(leaf, item);
1416         btrfs_set_extent_refs(leaf, item, refs + 1);
1417
1418         btrfs_mark_buffer_dirty(leaf);
1419         btrfs_release_path(path);
1420
1421         path->reada = 1;
1422
1423         /* now insert the actual backref */
1424         ret = insert_extent_backref(trans, root->fs_info->extent_root,
1425                                     path, bytenr, parent, root_objectid,
1426                                     owner, offset, 1);
1427         if (ret)
1428                 err = ret;
1429 out:
1430         btrfs_free_path(path);
1431         finish_current_insert(trans, root->fs_info->extent_root);
1432         del_pending_extents(trans, root->fs_info->extent_root);
1433         BUG_ON(err);
1434         return err;
1435 }
1436
1437 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1438                          struct btrfs_root *root)
1439 {
1440         finish_current_insert(trans, root->fs_info->extent_root);
1441         del_pending_extents(trans, root->fs_info->extent_root);
1442         return 0;
1443 }
1444
1445 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1446                              struct btrfs_root *root, u64 bytenr,
1447                              u64 offset, int metadata, u64 *refs, u64 *flags)
1448 {
1449         struct btrfs_path *path;
1450         int ret;
1451         struct btrfs_key key;
1452         struct extent_buffer *l;
1453         struct btrfs_extent_item *item;
1454         u32 item_size;
1455         u64 num_refs;
1456         u64 extent_flags;
1457
1458         if (metadata &&
1459             !btrfs_fs_incompat(root->fs_info,
1460                                BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
1461                 offset = root->nodesize;
1462                 metadata = 0;
1463         }
1464
1465         path = btrfs_alloc_path();
1466         if (!path)
1467                 return -ENOMEM;
1468         path->reada = 1;
1469
1470         key.objectid = bytenr;
1471         key.offset = offset;
1472         if (metadata)
1473                 key.type = BTRFS_METADATA_ITEM_KEY;
1474         else
1475                 key.type = BTRFS_EXTENT_ITEM_KEY;
1476
1477 again:
1478         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1479                                 0, 0);
1480         if (ret < 0)
1481                 goto out;
1482
1483         /*
1484          * Deal with the fact that we may have mixed SKINNY and normal refs.  If
1485          * we didn't find what we wanted check and see if we have a normal ref
1486          * right next to us, or re-search if we are on the edge of the leaf just
1487          * to make sure.
1488          */
1489         if (ret > 0 && metadata) {
1490                 if (path->slots[0]) {
1491                         path->slots[0]--;
1492                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1493                                               path->slots[0]);
1494                         if (key.objectid == bytenr &&
1495                             key.type == BTRFS_EXTENT_ITEM_KEY &&
1496                             key.offset == root->nodesize)
1497                                 ret = 0;
1498                 }
1499
1500                 if (ret) {
1501                         btrfs_release_path(path);
1502                         key.type = BTRFS_EXTENT_ITEM_KEY;
1503                         key.offset = root->nodesize;
1504                         metadata = 0;
1505                         goto again;
1506                 }
1507         }
1508
1509         if (ret != 0) {
1510                 ret = -EIO;
1511                 goto out;
1512         }
1513
1514         l = path->nodes[0];
1515         item_size = btrfs_item_size_nr(l, path->slots[0]);
1516         if (item_size >= sizeof(*item)) {
1517                 item = btrfs_item_ptr(l, path->slots[0],
1518                                       struct btrfs_extent_item);
1519                 num_refs = btrfs_extent_refs(l, item);
1520                 extent_flags = btrfs_extent_flags(l, item);
1521         } else {
1522 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1523                         struct btrfs_extent_item_v0 *ei0;
1524                         BUG_ON(item_size != sizeof(*ei0));
1525                         ei0 = btrfs_item_ptr(l, path->slots[0],
1526                                              struct btrfs_extent_item_v0);
1527                         num_refs = btrfs_extent_refs_v0(l, ei0);
1528                         /* FIXME: this isn't correct for data */
1529                         extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1530 #else
1531                         BUG();
1532 #endif
1533         }
1534         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1535         if (refs)
1536                 *refs = num_refs;
1537         if (flags)
1538                 *flags = extent_flags;
1539 out:
1540         btrfs_free_path(path);
1541         return ret;
1542 }
1543
1544 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1545                           struct btrfs_root *root,
1546                           u64 bytenr, int level, u64 flags)
1547 {
1548         struct btrfs_path *path;
1549         int ret;
1550         struct btrfs_key key;
1551         struct extent_buffer *l;
1552         struct btrfs_extent_item *item;
1553         u32 item_size;
1554         int skinny_metadata =
1555                 btrfs_fs_incompat(root->fs_info,
1556                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
1557
1558         path = btrfs_alloc_path();
1559         if (!path)
1560                 return -ENOMEM;
1561         path->reada = 1;
1562
1563         key.objectid = bytenr;
1564         if (skinny_metadata) {
1565                 key.offset = level;
1566                 key.type = BTRFS_METADATA_ITEM_KEY;
1567         } else {
1568                 key.offset = root->nodesize;
1569                 key.type = BTRFS_EXTENT_ITEM_KEY;
1570         }
1571
1572 again:
1573         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1574                                 0, 0);
1575         if (ret < 0)
1576                 goto out;
1577
1578         if (ret > 0 && skinny_metadata) {
1579                 skinny_metadata = 0;
1580                 if (path->slots[0]) {
1581                         path->slots[0]--;
1582                         btrfs_item_key_to_cpu(path->nodes[0], &key,
1583                                               path->slots[0]);
1584                         if (key.objectid == bytenr &&
1585                             key.offset == root->nodesize &&
1586                             key.type == BTRFS_EXTENT_ITEM_KEY)
1587                                 ret = 0;
1588                 }
1589                 if (ret) {
1590                         btrfs_release_path(path);
1591                         key.offset = root->nodesize;
1592                         key.type = BTRFS_EXTENT_ITEM_KEY;
1593                         goto again;
1594                 }
1595         }
1596
1597         if (ret != 0) {
1598                 btrfs_print_leaf(root, path->nodes[0]);
1599                 printk("failed to find block number %Lu\n",
1600                         (unsigned long long)bytenr);
1601                 BUG();
1602         }
1603         l = path->nodes[0];
1604         item_size = btrfs_item_size_nr(l, path->slots[0]);
1605 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1606         if (item_size < sizeof(*item)) {
1607                 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1608                                              path, (u64)-1, 0);
1609                 if (ret < 0)
1610                         goto out;
1611
1612                 l = path->nodes[0];
1613                 item_size = btrfs_item_size_nr(l, path->slots[0]);
1614         }
1615 #endif
1616         BUG_ON(item_size < sizeof(*item));
1617         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1618         flags |= btrfs_extent_flags(l, item);
1619         btrfs_set_extent_flags(l, item, flags);
1620 out:
1621         btrfs_free_path(path);
1622         finish_current_insert(trans, root->fs_info->extent_root);
1623         del_pending_extents(trans, root->fs_info->extent_root);
1624         return ret;
1625 }
1626
1627 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1628                            struct btrfs_root *root,
1629                            struct extent_buffer *buf,
1630                            int record_parent, int inc)
1631 {
1632         u64 bytenr;
1633         u64 num_bytes;
1634         u64 parent;
1635         u64 ref_root;
1636         u32 nritems;
1637         struct btrfs_key key;
1638         struct btrfs_file_extent_item *fi;
1639         int i;
1640         int level;
1641         int ret = 0;
1642         int (*process_func)(struct btrfs_trans_handle *trans,
1643                             struct btrfs_root *root,
1644                             u64, u64, u64, u64, u64, u64);
1645
1646         ref_root = btrfs_header_owner(buf);
1647         nritems = btrfs_header_nritems(buf);
1648         level = btrfs_header_level(buf);
1649
1650         if (!root->ref_cows && level == 0)
1651                 return 0;
1652
1653         if (inc)
1654                 process_func = btrfs_inc_extent_ref;
1655         else
1656                 process_func = btrfs_free_extent;
1657
1658         if (record_parent)
1659                 parent = buf->start;
1660         else
1661                 parent = 0;
1662
1663         for (i = 0; i < nritems; i++) {
1664                 cond_resched();
1665                 if (level == 0) {
1666                         btrfs_item_key_to_cpu(buf, &key, i);
1667                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1668                                 continue;
1669                         fi = btrfs_item_ptr(buf, i,
1670                                             struct btrfs_file_extent_item);
1671                         if (btrfs_file_extent_type(buf, fi) ==
1672                             BTRFS_FILE_EXTENT_INLINE)
1673                                 continue;
1674                         bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1675                         if (bytenr == 0)
1676                                 continue;
1677                         
1678                         num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1679                         key.offset -= btrfs_file_extent_offset(buf, fi);
1680                         ret = process_func(trans, root, bytenr, num_bytes,
1681                                            parent, ref_root, key.objectid,
1682                                            key.offset);
1683                         if (ret) {
1684                                 WARN_ON(1);
1685                                 goto fail;
1686                         }
1687                 } else {
1688                         bytenr = btrfs_node_blockptr(buf, i);
1689                         num_bytes = root->nodesize;
1690                         ret = process_func(trans, root, bytenr, num_bytes,
1691                                            parent, ref_root, level - 1, 0);
1692                         if (ret) {
1693                                 WARN_ON(1);
1694                                 goto fail;
1695                         }
1696                 }
1697         }
1698         return 0;
1699 fail:
1700         WARN_ON(1);
1701         return ret;
1702 }
1703
1704 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1705                   struct extent_buffer *buf, int record_parent)
1706 {
1707         return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1708 }
1709
1710 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1711                   struct extent_buffer *buf, int record_parent)
1712 {
1713         return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1714 }
1715
1716 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1717                                  struct btrfs_root *root,
1718                                  struct btrfs_path *path,
1719                                  struct btrfs_block_group_cache *cache)
1720 {
1721         int ret;
1722         int pending_ret;
1723         struct btrfs_root *extent_root = root->fs_info->extent_root;
1724         unsigned long bi;
1725         struct extent_buffer *leaf;
1726
1727         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1728         if (ret < 0)
1729                 goto fail;
1730         BUG_ON(ret);
1731
1732         leaf = path->nodes[0];
1733         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1734         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1735         btrfs_mark_buffer_dirty(leaf);
1736         btrfs_release_path(path);
1737 fail:
1738         finish_current_insert(trans, extent_root);
1739         pending_ret = del_pending_extents(trans, extent_root);
1740         if (ret)
1741                 return ret;
1742         if (pending_ret)
1743                 return pending_ret;
1744         return 0;
1745
1746 }
1747
1748 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1749                                    struct btrfs_root *root)
1750 {
1751         struct extent_io_tree *block_group_cache;
1752         struct btrfs_block_group_cache *cache;
1753         int ret;
1754         struct btrfs_path *path;
1755         u64 last = 0;
1756         u64 start;
1757         u64 end;
1758         u64 ptr;
1759
1760         block_group_cache = &root->fs_info->block_group_cache;
1761         path = btrfs_alloc_path();
1762         if (!path)
1763                 return -ENOMEM;
1764
1765         while(1) {
1766                 ret = find_first_extent_bit(block_group_cache, last,
1767                                             &start, &end, BLOCK_GROUP_DIRTY);
1768                 if (ret) {
1769                         if (last == 0)
1770                                 break;
1771                         last = 0;
1772                         continue;
1773                 }
1774
1775                 last = end + 1;
1776                 ret = get_state_private(block_group_cache, start, &ptr);
1777                 BUG_ON(ret);
1778
1779                 clear_extent_bits(block_group_cache, start, end,
1780                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
1781
1782                 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1783                 ret = write_one_cache_group(trans, root, path, cache);
1784         }
1785         btrfs_free_path(path);
1786         return 0;
1787 }
1788
1789 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1790                                                   u64 flags)
1791 {
1792         struct btrfs_space_info *found;
1793
1794         flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
1795
1796         list_for_each_entry(found, &info->space_info, list) {
1797                 if (found->flags & flags)
1798                         return found;
1799         }
1800         return NULL;
1801
1802 }
1803
1804 static int free_space_info(struct btrfs_fs_info *fs_info, u64 flags,
1805                           u64 total_bytes, u64 bytes_used,
1806                           struct btrfs_space_info **space_info)
1807 {
1808         struct btrfs_space_info *found;
1809
1810         /* only support free block group which is empty */
1811         if (bytes_used)
1812                 return -ENOTEMPTY;
1813
1814         found = __find_space_info(fs_info, flags);
1815         if (!found)
1816                 return -ENOENT;
1817         if (found->total_bytes < total_bytes) {
1818                 fprintf(stderr,
1819                         "WARNING: bad space info to free %llu only have %llu\n",
1820                         total_bytes, found->total_bytes);
1821                 return -EINVAL;
1822         }
1823         found->total_bytes -= total_bytes;
1824         if (space_info)
1825                 *space_info = found;
1826         return 0;
1827 }
1828
1829 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1830                              u64 total_bytes, u64 bytes_used,
1831                              struct btrfs_space_info **space_info)
1832 {
1833         struct btrfs_space_info *found;
1834
1835         found = __find_space_info(info, flags);
1836         if (found) {
1837                 found->total_bytes += total_bytes;
1838                 found->bytes_used += bytes_used;
1839                 if (found->total_bytes < found->bytes_used) {
1840                         fprintf(stderr, "warning, bad space info total_bytes "
1841                                 "%llu used %llu\n",
1842                                (unsigned long long)found->total_bytes,
1843                                (unsigned long long)found->bytes_used);
1844                 }
1845                 *space_info = found;
1846                 return 0;
1847         }
1848         found = kmalloc(sizeof(*found), GFP_NOFS);
1849         if (!found)
1850                 return -ENOMEM;
1851
1852         list_add(&found->list, &info->space_info);
1853         found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
1854         found->total_bytes = total_bytes;
1855         found->bytes_used = bytes_used;
1856         found->bytes_pinned = 0;
1857         found->full = 0;
1858         *space_info = found;
1859         return 0;
1860 }
1861
1862
1863 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1864 {
1865         u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1866                                    BTRFS_BLOCK_GROUP_RAID1 |
1867                                    BTRFS_BLOCK_GROUP_RAID10 |
1868                                    BTRFS_BLOCK_GROUP_RAID5 |
1869                                    BTRFS_BLOCK_GROUP_RAID6 |
1870                                    BTRFS_BLOCK_GROUP_DUP);
1871         if (extra_flags) {
1872                 if (flags & BTRFS_BLOCK_GROUP_DATA)
1873                         fs_info->avail_data_alloc_bits |= extra_flags;
1874                 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1875                         fs_info->avail_metadata_alloc_bits |= extra_flags;
1876                 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1877                         fs_info->avail_system_alloc_bits |= extra_flags;
1878         }
1879 }
1880
1881 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1882                           struct btrfs_root *extent_root, u64 alloc_bytes,
1883                           u64 flags)
1884 {
1885         struct btrfs_space_info *space_info;
1886         u64 thresh;
1887         u64 start;
1888         u64 num_bytes;
1889         int ret;
1890
1891         space_info = __find_space_info(extent_root->fs_info, flags);
1892         if (!space_info) {
1893                 ret = update_space_info(extent_root->fs_info, flags,
1894                                         0, 0, &space_info);
1895                 BUG_ON(ret);
1896         }
1897         BUG_ON(!space_info);
1898
1899         if (space_info->full)
1900                 return 0;
1901
1902         thresh = div_factor(space_info->total_bytes, 7);
1903         if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1904             thresh)
1905                 return 0;
1906
1907         ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes,
1908                                 space_info->flags);
1909         if (ret == -ENOSPC) {
1910                 space_info->full = 1;
1911                 return 0;
1912         }
1913
1914         BUG_ON(ret);
1915
1916         ret = btrfs_make_block_group(trans, extent_root, 0, space_info->flags,
1917                      BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1918         BUG_ON(ret);
1919         return 0;
1920 }
1921
1922 static int update_block_group(struct btrfs_trans_handle *trans,
1923                               struct btrfs_root *root,
1924                               u64 bytenr, u64 num_bytes, int alloc,
1925                               int mark_free)
1926 {
1927         struct btrfs_block_group_cache *cache;
1928         struct btrfs_fs_info *info = root->fs_info;
1929         u64 total = num_bytes;
1930         u64 old_val;
1931         u64 byte_in_group;
1932         u64 start;
1933         u64 end;
1934
1935         /* block accounting for super block */
1936         old_val = btrfs_super_bytes_used(info->super_copy);
1937         if (alloc)
1938                 old_val += num_bytes;
1939         else
1940                 old_val -= num_bytes;
1941         btrfs_set_super_bytes_used(info->super_copy, old_val);
1942
1943         /* block accounting for root item */
1944         old_val = btrfs_root_used(&root->root_item);
1945         if (alloc)
1946                 old_val += num_bytes;
1947         else
1948                 old_val -= num_bytes;
1949         btrfs_set_root_used(&root->root_item, old_val);
1950
1951         while(total) {
1952                 cache = btrfs_lookup_block_group(info, bytenr);
1953                 if (!cache) {
1954                         return -1;
1955                 }
1956                 byte_in_group = bytenr - cache->key.objectid;
1957                 WARN_ON(byte_in_group > cache->key.offset);
1958                 start = cache->key.objectid;
1959                 end = start + cache->key.offset - 1;
1960                 set_extent_bits(&info->block_group_cache, start, end,
1961                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
1962
1963                 old_val = btrfs_block_group_used(&cache->item);
1964                 num_bytes = min(total, cache->key.offset - byte_in_group);
1965
1966                 if (alloc) {
1967                         old_val += num_bytes;
1968                         cache->space_info->bytes_used += num_bytes;
1969                 } else {
1970                         old_val -= num_bytes;
1971                         cache->space_info->bytes_used -= num_bytes;
1972                         if (mark_free) {
1973                                 set_extent_dirty(&info->free_space_cache,
1974                                                  bytenr, bytenr + num_bytes - 1,
1975                                                  GFP_NOFS);
1976                         }
1977                 }
1978                 btrfs_set_block_group_used(&cache->item, old_val);
1979                 total -= num_bytes;
1980                 bytenr += num_bytes;
1981         }
1982         return 0;
1983 }
1984
1985 static int update_pinned_extents(struct btrfs_root *root,
1986                                 u64 bytenr, u64 num, int pin)
1987 {
1988         u64 len;
1989         struct btrfs_block_group_cache *cache;
1990         struct btrfs_fs_info *fs_info = root->fs_info;
1991
1992         if (pin) {
1993                 set_extent_dirty(&fs_info->pinned_extents,
1994                                 bytenr, bytenr + num - 1, GFP_NOFS);
1995         } else {
1996                 clear_extent_dirty(&fs_info->pinned_extents,
1997                                 bytenr, bytenr + num - 1, GFP_NOFS);
1998         }
1999         while (num > 0) {
2000                 cache = btrfs_lookup_block_group(fs_info, bytenr);
2001                 if (!cache) {
2002                         len = min((u64)root->sectorsize, num);
2003                         goto next;
2004                 }
2005                 WARN_ON(!cache);
2006                 len = min(num, cache->key.offset -
2007                           (bytenr - cache->key.objectid));
2008                 if (pin) {
2009                         cache->pinned += len;
2010                         cache->space_info->bytes_pinned += len;
2011                         fs_info->total_pinned += len;
2012                 } else {
2013                         cache->pinned -= len;
2014                         cache->space_info->bytes_pinned -= len;
2015                         fs_info->total_pinned -= len;
2016                 }
2017 next:
2018                 bytenr += len;
2019                 num -= len;
2020         }
2021         return 0;
2022 }
2023
2024 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2025                                struct btrfs_root *root,
2026                                struct extent_io_tree *unpin)
2027 {
2028         u64 start;
2029         u64 end;
2030         int ret;
2031         struct extent_io_tree *free_space_cache;
2032         free_space_cache = &root->fs_info->free_space_cache;
2033
2034         while(1) {
2035                 ret = find_first_extent_bit(unpin, 0, &start, &end,
2036                                             EXTENT_DIRTY);
2037                 if (ret)
2038                         break;
2039                 update_pinned_extents(root, start, end + 1 - start, 0);
2040                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
2041                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
2042         }
2043         return 0;
2044 }
2045
2046 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2047 {
2048         u64 start;
2049         u64 end;
2050         int ret;
2051
2052         ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2053                                     &end, EXTENT_LOCKED);
2054         if (!ret) {
2055                 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2056                                             EXTENT_LOCKED);
2057         }
2058         return ret == 0;
2059
2060 }
2061 static int finish_current_insert(struct btrfs_trans_handle *trans,
2062                                  struct btrfs_root *extent_root)
2063 {
2064         u64 start;
2065         u64 end;
2066         u64 priv;
2067         struct btrfs_fs_info *info = extent_root->fs_info;
2068         struct pending_extent_op *extent_op;
2069         struct btrfs_key key;
2070         int ret;
2071         int skinny_metadata =
2072                 btrfs_fs_incompat(extent_root->fs_info,
2073                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2074
2075         while(1) {
2076                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2077                                             &end, EXTENT_LOCKED);
2078                 if (ret)
2079                         break;
2080
2081                 ret = get_state_private(&info->extent_ins, start, &priv);
2082                 BUG_ON(ret);
2083                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2084
2085                 if (extent_op->type == PENDING_EXTENT_INSERT) {
2086                         key.objectid = start;
2087                         if (skinny_metadata) {
2088                                 key.offset = extent_op->level;
2089                                 key.type = BTRFS_METADATA_ITEM_KEY;
2090                         } else {
2091                                 key.offset = extent_op->num_bytes;
2092                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2093                         }
2094                         ret = alloc_reserved_tree_block(trans, extent_root,
2095                                                 extent_root->root_key.objectid,
2096                                                 trans->transid,
2097                                                 extent_op->flags,
2098                                                 &extent_op->key,
2099                                                 extent_op->level, &key);
2100                         BUG_ON(ret);
2101                 } else {
2102                         BUG_ON(1);
2103                 }
2104
2105                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2106                                   GFP_NOFS);
2107                 kfree(extent_op);
2108         }
2109         return 0;
2110 }
2111
2112 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2113                           struct btrfs_root *root,
2114                           u64 bytenr, u64 num_bytes, int is_data)
2115 {
2116         int err = 0;
2117         struct extent_buffer *buf;
2118
2119         if (is_data)
2120                 goto pinit;
2121
2122         buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2123         if (!buf)
2124                 goto pinit;
2125
2126         /* we can reuse a block if it hasn't been written
2127          * and it is from this transaction.  We can't
2128          * reuse anything from the tree log root because
2129          * it has tiny sub-transactions.
2130          */
2131         if (btrfs_buffer_uptodate(buf, 0)) {
2132                 u64 header_owner = btrfs_header_owner(buf);
2133                 u64 header_transid = btrfs_header_generation(buf);
2134                 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2135                     header_transid == trans->transid &&
2136                     !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2137                         clean_tree_block(NULL, root, buf);
2138                         free_extent_buffer(buf);
2139                         return 1;
2140                 }
2141         }
2142         free_extent_buffer(buf);
2143 pinit:
2144         update_pinned_extents(root, bytenr, num_bytes, 1);
2145
2146         BUG_ON(err < 0);
2147         return 0;
2148 }
2149
2150 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2151                        u64 bytenr, u64 num_bytes)
2152 {
2153         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2154 }
2155
2156 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2157                         u64 bytenr, u64 num_bytes)
2158 {
2159         update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2160 }
2161
2162 /*
2163  * remove an extent from the root, returns 0 on success
2164  */
2165 static int __free_extent(struct btrfs_trans_handle *trans,
2166                          struct btrfs_root *root,
2167                          u64 bytenr, u64 num_bytes, u64 parent,
2168                          u64 root_objectid, u64 owner_objectid,
2169                          u64 owner_offset, int refs_to_drop)
2170 {
2171
2172         struct btrfs_key key;
2173         struct btrfs_path *path;
2174         struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2175         struct btrfs_root *extent_root = root->fs_info->extent_root;
2176         struct extent_buffer *leaf;
2177         struct btrfs_extent_item *ei;
2178         struct btrfs_extent_inline_ref *iref;
2179         int ret;
2180         int is_data;
2181         int extent_slot = 0;
2182         int found_extent = 0;
2183         int num_to_del = 1;
2184         u32 item_size;
2185         u64 refs;
2186         int skinny_metadata =
2187                 btrfs_fs_incompat(extent_root->fs_info,
2188                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2189
2190         if (root->fs_info->free_extent_hook) {
2191                 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2192                                                 parent, root_objectid, owner_objectid,
2193                                                 owner_offset, refs_to_drop);
2194
2195         }
2196         path = btrfs_alloc_path();
2197         if (!path)
2198                 return -ENOMEM;
2199
2200         path->reada = 1;
2201
2202         is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2203         if (is_data)
2204                 skinny_metadata = 0;
2205         BUG_ON(!is_data && refs_to_drop != 1);
2206
2207         ret = lookup_extent_backref(trans, extent_root, path, &iref,
2208                                     bytenr, num_bytes, parent,
2209                                     root_objectid, owner_objectid,
2210                                     owner_offset);
2211         if (ret == 0) {
2212                 extent_slot = path->slots[0];
2213                 while (extent_slot >= 0) {
2214                         btrfs_item_key_to_cpu(path->nodes[0], &key,
2215                                               extent_slot);
2216                         if (key.objectid != bytenr)
2217                                 break;
2218                         if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2219                             key.offset == num_bytes) {
2220                                 found_extent = 1;
2221                                 break;
2222                         }
2223                         if (key.type == BTRFS_METADATA_ITEM_KEY &&
2224                             key.offset == owner_objectid) {
2225                                 found_extent = 1;
2226                                 break;
2227                         }
2228                         if (path->slots[0] - extent_slot > 5)
2229                                 break;
2230                         extent_slot--;
2231                 }
2232 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2233                 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2234                 if (found_extent && item_size < sizeof(*ei))
2235                         found_extent = 0;
2236 #endif
2237                 if (!found_extent) {
2238                         BUG_ON(iref);
2239                         ret = remove_extent_backref(trans, extent_root, path,
2240                                                     NULL, refs_to_drop,
2241                                                     is_data);
2242                         BUG_ON(ret);
2243                         btrfs_release_path(path);
2244
2245                         key.objectid = bytenr;
2246
2247                         if (skinny_metadata) {
2248                                 key.type = BTRFS_METADATA_ITEM_KEY;
2249                                 key.offset = owner_objectid;
2250                         } else {
2251                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2252                                 key.offset = num_bytes;
2253                         }
2254
2255                         ret = btrfs_search_slot(trans, extent_root,
2256                                                 &key, path, -1, 1);
2257                         if (ret > 0 && skinny_metadata && path->slots[0]) {
2258                                 path->slots[0]--;
2259                                 btrfs_item_key_to_cpu(path->nodes[0],
2260                                                       &key,
2261                                                       path->slots[0]);
2262                                 if (key.objectid == bytenr &&
2263                                     key.type == BTRFS_EXTENT_ITEM_KEY &&
2264                                     key.offset == num_bytes)
2265                                         ret = 0;
2266                         }
2267
2268                         if (ret > 0 && skinny_metadata) {
2269                                 skinny_metadata = 0;
2270                                 btrfs_release_path(path);
2271                                 key.type = BTRFS_EXTENT_ITEM_KEY;
2272                                 key.offset = num_bytes;
2273                                 ret = btrfs_search_slot(trans, extent_root,
2274                                                         &key, path, -1, 1);
2275                         }
2276
2277                         if (ret) {
2278                                 printk(KERN_ERR "umm, got %d back from search"
2279                                        ", was looking for %llu\n", ret,
2280                                        (unsigned long long)bytenr);
2281                                 btrfs_print_leaf(extent_root, path->nodes[0]);
2282                         }
2283                         BUG_ON(ret);
2284                         extent_slot = path->slots[0];
2285                 }
2286         } else {
2287                 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2288                        "parent %llu root %llu  owner %llu offset %llu\n",
2289                        (unsigned long long)bytenr,
2290                        (unsigned long long)parent,
2291                        (unsigned long long)root_objectid,
2292                        (unsigned long long)owner_objectid,
2293                        (unsigned long long)owner_offset);
2294                 ret = -EIO;
2295                 goto fail;
2296         }
2297
2298         leaf = path->nodes[0];
2299         item_size = btrfs_item_size_nr(leaf, extent_slot);
2300 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2301         if (item_size < sizeof(*ei)) {
2302                 BUG_ON(found_extent || extent_slot != path->slots[0]);
2303                 ret = convert_extent_item_v0(trans, extent_root, path,
2304                                              owner_objectid, 0);
2305                 BUG_ON(ret < 0);
2306
2307                 btrfs_release_path(path);
2308
2309                 key.objectid = bytenr;
2310                 key.type = BTRFS_EXTENT_ITEM_KEY;
2311                 key.offset = num_bytes;
2312
2313                 ret = btrfs_search_slot(trans, extent_root, &key, path,
2314                                         -1, 1);
2315                 if (ret) {
2316                         printk(KERN_ERR "umm, got %d back from search"
2317                                ", was looking for %llu\n", ret,
2318                                (unsigned long long)bytenr);
2319                         btrfs_print_leaf(extent_root, path->nodes[0]);
2320                 }
2321                 BUG_ON(ret);
2322                 extent_slot = path->slots[0];
2323                 leaf = path->nodes[0];
2324                 item_size = btrfs_item_size_nr(leaf, extent_slot);
2325         }
2326 #endif
2327         BUG_ON(item_size < sizeof(*ei));
2328         ei = btrfs_item_ptr(leaf, extent_slot,
2329                             struct btrfs_extent_item);
2330         if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2331             key.type == BTRFS_EXTENT_ITEM_KEY) {
2332                 struct btrfs_tree_block_info *bi;
2333                 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2334                 bi = (struct btrfs_tree_block_info *)(ei + 1);
2335                 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2336         }
2337
2338         refs = btrfs_extent_refs(leaf, ei);
2339         BUG_ON(refs < refs_to_drop);
2340         refs -= refs_to_drop;
2341
2342         if (refs > 0) {
2343                 /*
2344                  * In the case of inline back ref, reference count will
2345                  * be updated by remove_extent_backref
2346                  */
2347                 if (iref) {
2348                         BUG_ON(!found_extent);
2349                 } else {
2350                         btrfs_set_extent_refs(leaf, ei, refs);
2351                         btrfs_mark_buffer_dirty(leaf);
2352                 }
2353                 if (found_extent) {
2354                         ret = remove_extent_backref(trans, extent_root, path,
2355                                                     iref, refs_to_drop,
2356                                                     is_data);
2357                         BUG_ON(ret);
2358                 }
2359         } else {
2360                 int mark_free = 0;
2361                 int pin = 1;
2362
2363                 if (found_extent) {
2364                         BUG_ON(is_data && refs_to_drop !=
2365                                extent_data_ref_count(root, path, iref));
2366                         if (iref) {
2367                                 BUG_ON(path->slots[0] != extent_slot);
2368                         } else {
2369                                 BUG_ON(path->slots[0] != extent_slot + 1);
2370                                 path->slots[0] = extent_slot;
2371                                 num_to_del = 2;
2372                         }
2373                 }
2374
2375                 if (ops && ops->free_extent) {
2376                         ret = ops->free_extent(root, bytenr, num_bytes);
2377                         if (ret > 0) {
2378                                 pin = 0;
2379                                 mark_free = 0;
2380                         }
2381                 }
2382
2383                 if (pin) {
2384                         ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2385                                              is_data);
2386                         if (ret > 0)
2387                                 mark_free = 1;
2388                         BUG_ON(ret < 0);
2389                 }
2390
2391                 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2392                                       num_to_del);
2393                 BUG_ON(ret);
2394                 btrfs_release_path(path);
2395
2396                 if (is_data) {
2397                         ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2398                         BUG_ON(ret);
2399                 }
2400
2401                 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2402         }
2403 fail:
2404         btrfs_free_path(path);
2405         finish_current_insert(trans, extent_root);
2406         return ret;
2407 }
2408
2409 /*
2410  * find all the blocks marked as pending in the radix tree and remove
2411  * them from the extent map
2412  */
2413 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2414                                btrfs_root *extent_root)
2415 {
2416         int ret;
2417         int err = 0;
2418         u64 start;
2419         u64 end;
2420         u64 priv;
2421         struct extent_io_tree *pending_del;
2422         struct extent_io_tree *extent_ins;
2423         struct pending_extent_op *extent_op;
2424
2425         extent_ins = &extent_root->fs_info->extent_ins;
2426         pending_del = &extent_root->fs_info->pending_del;
2427
2428         while(1) {
2429                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2430                                             EXTENT_LOCKED);
2431                 if (ret)
2432                         break;
2433
2434                 ret = get_state_private(pending_del, start, &priv);
2435                 BUG_ON(ret);
2436                 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2437
2438                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2439                                   GFP_NOFS);
2440
2441                 if (!test_range_bit(extent_ins, start, end,
2442                                     EXTENT_LOCKED, 0)) {
2443                         ret = __free_extent(trans, extent_root,
2444                                             start, end + 1 - start, 0,
2445                                             extent_root->root_key.objectid,
2446                                             extent_op->level, 0, 1);
2447                         kfree(extent_op);
2448                 } else {
2449                         kfree(extent_op);
2450                         ret = get_state_private(extent_ins, start, &priv);
2451                         BUG_ON(ret);
2452                         extent_op = (struct pending_extent_op *)
2453                                                         (unsigned long)priv;
2454
2455                         clear_extent_bits(extent_ins, start, end,
2456                                           EXTENT_LOCKED, GFP_NOFS);
2457
2458                         if (extent_op->type == PENDING_BACKREF_UPDATE)
2459                                 BUG_ON(1);
2460
2461                         kfree(extent_op);
2462                 }
2463                 if (ret)
2464                         err = ret;
2465         }
2466         return err;
2467 }
2468
2469 /*
2470  * remove an extent from the root, returns 0 on success
2471  */
2472
2473 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2474                       struct btrfs_root *root,
2475                       u64 bytenr, u64 num_bytes, u64 parent,
2476                       u64 root_objectid, u64 owner, u64 offset)
2477 {
2478         struct btrfs_root *extent_root = root->fs_info->extent_root;
2479         int pending_ret;
2480         int ret;
2481
2482         WARN_ON(num_bytes < root->sectorsize);
2483         if (root == extent_root) {
2484                 struct pending_extent_op *extent_op;
2485
2486                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2487                 BUG_ON(!extent_op);
2488
2489                 extent_op->type = PENDING_EXTENT_DELETE;
2490                 extent_op->bytenr = bytenr;
2491                 extent_op->num_bytes = num_bytes;
2492                 extent_op->level = (int)owner;
2493
2494                 set_extent_bits(&root->fs_info->pending_del,
2495                                 bytenr, bytenr + num_bytes - 1,
2496                                 EXTENT_LOCKED, GFP_NOFS);
2497                 set_state_private(&root->fs_info->pending_del,
2498                                   bytenr, (unsigned long)extent_op);
2499                 return 0;
2500         }
2501         ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2502                             root_objectid, owner, offset, 1);
2503         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2504         return ret ? ret : pending_ret;
2505 }
2506
2507 static u64 stripe_align(struct btrfs_root *root, u64 val)
2508 {
2509         u64 mask = ((u64)root->stripesize - 1);
2510         u64 ret = (val + mask) & ~mask;
2511         return ret;
2512 }
2513
2514 /*
2515  * walks the btree of allocated extents and find a hole of a given size.
2516  * The key ins is changed to record the hole:
2517  * ins->objectid == block start
2518  * ins->flags = BTRFS_EXTENT_ITEM_KEY
2519  * ins->offset == number of blocks
2520  * Any available blocks before search_start are skipped.
2521  */
2522 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2523                                      struct btrfs_root *orig_root,
2524                                      u64 num_bytes, u64 empty_size,
2525                                      u64 search_start, u64 search_end,
2526                                      u64 hint_byte, struct btrfs_key *ins,
2527                                      u64 exclude_start, u64 exclude_nr,
2528                                      int data)
2529 {
2530         int ret;
2531         u64 orig_search_start = search_start;
2532         struct btrfs_root * root = orig_root->fs_info->extent_root;
2533         struct btrfs_fs_info *info = root->fs_info;
2534         u64 total_needed = num_bytes;
2535         struct btrfs_block_group_cache *block_group;
2536         int full_scan = 0;
2537         int wrapped = 0;
2538
2539         WARN_ON(num_bytes < root->sectorsize);
2540         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2541
2542         search_start = stripe_align(root, search_start);
2543
2544         if (hint_byte) {
2545                 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2546                 if (!block_group)
2547                         hint_byte = search_start;
2548                 block_group = btrfs_find_block_group(root, block_group,
2549                                                      hint_byte, data, 1);
2550         } else {
2551                 block_group = btrfs_find_block_group(root,
2552                                                      trans->block_group,
2553                                                      search_start, data, 1);
2554         }
2555
2556         total_needed += empty_size;
2557
2558 check_failed:
2559         search_start = stripe_align(root, search_start);
2560         if (!block_group) {
2561                 block_group = btrfs_lookup_first_block_group(info,
2562                                                              search_start);
2563                 if (!block_group)
2564                         block_group = btrfs_lookup_first_block_group(info,
2565                                                        orig_search_start);
2566         }
2567         ret = find_search_start(root, &block_group, &search_start,
2568                                 total_needed, data);
2569         if (ret)
2570                 goto new_group;
2571
2572         ins->objectid = search_start;
2573         ins->offset = num_bytes;
2574
2575         if (ins->objectid + num_bytes >
2576             block_group->key.objectid + block_group->key.offset) {
2577                 search_start = block_group->key.objectid +
2578                         block_group->key.offset;
2579                 goto new_group;
2580         }
2581
2582         if (test_range_bit(&info->extent_ins, ins->objectid,
2583                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2584                 search_start = ins->objectid + num_bytes;
2585                 goto new_group;
2586         }
2587
2588         if (test_range_bit(&info->pinned_extents, ins->objectid,
2589                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2590                 search_start = ins->objectid + num_bytes;
2591                 goto new_group;
2592         }
2593
2594         if (info->excluded_extents &&
2595             test_range_bit(info->excluded_extents, ins->objectid,
2596                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2597                 search_start = ins->objectid + num_bytes;
2598                 goto new_group;
2599         }
2600
2601         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2602             ins->objectid < exclude_start + exclude_nr)) {
2603                 search_start = exclude_start + exclude_nr;
2604                 goto new_group;
2605         }
2606
2607         if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2608                 if (check_crossing_stripes(ins->objectid, num_bytes)) {
2609                         search_start = round_down(ins->objectid + num_bytes,
2610                                                   BTRFS_STRIPE_LEN);
2611                         goto new_group;
2612                 }
2613                 block_group = btrfs_lookup_block_group(info, ins->objectid);
2614                 if (block_group)
2615                         trans->block_group = block_group;
2616         }
2617         ins->offset = num_bytes;
2618         return 0;
2619
2620 new_group:
2621         block_group = btrfs_lookup_first_block_group(info, search_start);
2622         if (!block_group) {
2623                 search_start = orig_search_start;
2624                 if (full_scan) {
2625                         ret = -ENOSPC;
2626                         goto error;
2627                 }
2628                 if (wrapped) {
2629                         if (!full_scan)
2630                                 total_needed -= empty_size;
2631                         full_scan = 1;
2632                 } else
2633                         wrapped = 1;
2634         }
2635         cond_resched();
2636         block_group = btrfs_find_block_group(root, block_group,
2637                                              search_start, data, 0);
2638         goto check_failed;
2639
2640 error:
2641         return ret;
2642 }
2643
2644 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2645                          struct btrfs_root *root,
2646                          u64 num_bytes, u64 empty_size,
2647                          u64 hint_byte, u64 search_end,
2648                          struct btrfs_key *ins, int data)
2649 {
2650         int ret;
2651         u64 search_start = 0;
2652         u64 alloc_profile;
2653         struct btrfs_fs_info *info = root->fs_info;
2654
2655         if (info->extent_ops) {
2656                 struct btrfs_extent_ops *ops = info->extent_ops;
2657                 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins, !data);
2658                 BUG_ON(ret);
2659                 goto found;
2660         }
2661
2662         if (data) {
2663                 alloc_profile = info->avail_data_alloc_bits &
2664                                 info->data_alloc_profile;
2665                 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2666         } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2667                    info->system_allocs >= 0) {
2668                 alloc_profile = info->avail_system_alloc_bits &
2669                                 info->system_alloc_profile;
2670                 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2671         } else {
2672                 alloc_profile = info->avail_metadata_alloc_bits &
2673                                 info->metadata_alloc_profile;
2674                 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2675         }
2676
2677         if (root->ref_cows) {
2678                 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2679                         ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2680                                              num_bytes,
2681                                              BTRFS_BLOCK_GROUP_METADATA);
2682                         BUG_ON(ret);
2683                 }
2684                 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2685                                      num_bytes + 2 * 1024 * 1024, data);
2686                 BUG_ON(ret);
2687         }
2688
2689         WARN_ON(num_bytes < root->sectorsize);
2690         ret = find_free_extent(trans, root, num_bytes, empty_size,
2691                                search_start, search_end, hint_byte, ins,
2692                                trans->alloc_exclude_start,
2693                                trans->alloc_exclude_nr, data);
2694         BUG_ON(ret);
2695 found:
2696         clear_extent_dirty(&root->fs_info->free_space_cache,
2697                            ins->objectid, ins->objectid + ins->offset - 1,
2698                            GFP_NOFS);
2699         return ret;
2700 }
2701
2702 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2703                                      struct btrfs_root *root,
2704                                      u64 root_objectid, u64 generation,
2705                                      u64 flags, struct btrfs_disk_key *key,
2706                                      int level, struct btrfs_key *ins)
2707 {
2708         int ret;
2709         struct btrfs_fs_info *fs_info = root->fs_info;
2710         struct btrfs_extent_item *extent_item;
2711         struct btrfs_tree_block_info *block_info;
2712         struct btrfs_extent_inline_ref *iref;
2713         struct btrfs_path *path;
2714         struct extent_buffer *leaf;
2715         u32 size = sizeof(*extent_item) + sizeof(*iref);
2716         int skinny_metadata =
2717                 btrfs_fs_incompat(fs_info,
2718                                   BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA);
2719
2720         if (!skinny_metadata)
2721                 size += sizeof(*block_info);
2722
2723         path = btrfs_alloc_path();
2724         BUG_ON(!path);
2725
2726         ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2727                                       ins, size);
2728         BUG_ON(ret);
2729
2730         leaf = path->nodes[0];
2731         extent_item = btrfs_item_ptr(leaf, path->slots[0],
2732                                      struct btrfs_extent_item);
2733         btrfs_set_extent_refs(leaf, extent_item, 1);
2734         btrfs_set_extent_generation(leaf, extent_item, generation);
2735         btrfs_set_extent_flags(leaf, extent_item,
2736                                flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2737
2738         if (skinny_metadata) {
2739                 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2740         } else {
2741                 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2742                 btrfs_set_tree_block_key(leaf, block_info, key);
2743                 btrfs_set_tree_block_level(leaf, block_info, level);
2744                 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2745         }
2746
2747         btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2748         btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2749
2750         btrfs_mark_buffer_dirty(leaf);
2751         btrfs_free_path(path);
2752
2753         ret = update_block_group(trans, root, ins->objectid, root->nodesize,
2754                                  1, 0);
2755         return ret;
2756 }
2757
2758 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2759                             struct btrfs_root *root, u64 num_bytes,
2760                             u64 root_objectid, u64 generation,
2761                             u64 flags, struct btrfs_disk_key *key,
2762                             int level, u64 empty_size, u64 hint_byte,
2763                             u64 search_end, struct btrfs_key *ins)
2764 {
2765         int ret;
2766         ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2767                                    hint_byte, search_end, ins, 0);
2768         BUG_ON(ret);
2769
2770         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2771                 struct pending_extent_op *extent_op;
2772
2773                 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2774                 BUG_ON(!extent_op);
2775
2776                 extent_op->type = PENDING_EXTENT_INSERT;
2777                 extent_op->bytenr = ins->objectid;
2778                 extent_op->num_bytes = ins->offset;
2779                 extent_op->level = level;
2780                 extent_op->flags = flags;
2781                 memcpy(&extent_op->key, key, sizeof(*key));
2782
2783                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2784                                 ins->objectid + ins->offset - 1,
2785                                 EXTENT_LOCKED, GFP_NOFS);
2786                 set_state_private(&root->fs_info->extent_ins,
2787                                   ins->objectid, (unsigned long)extent_op);
2788         } else {
2789                 if (btrfs_fs_incompat(root->fs_info,
2790                                 BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)) {
2791                         ins->offset = level;
2792                         ins->type = BTRFS_METADATA_ITEM_KEY;
2793                 }
2794                 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2795                                                 generation, flags,
2796                                                 key, level, ins);
2797                 finish_current_insert(trans, root->fs_info->extent_root);
2798                 del_pending_extents(trans, root->fs_info->extent_root);
2799         }
2800         return ret;
2801 }
2802
2803 /*
2804  * helper function to allocate a block for a given tree
2805  * returns the tree buffer or NULL.
2806  */
2807 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2808                                         struct btrfs_root *root,
2809                                         u32 blocksize, u64 root_objectid,
2810                                         struct btrfs_disk_key *key, int level,
2811                                         u64 hint, u64 empty_size)
2812 {
2813         struct btrfs_key ins;
2814         int ret;
2815         struct extent_buffer *buf;
2816
2817         ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2818                                trans->transid, 0, key, level,
2819                                empty_size, hint, (u64)-1, &ins);
2820         if (ret) {
2821                 BUG_ON(ret > 0);
2822                 return ERR_PTR(ret);
2823         }
2824
2825         buf = btrfs_find_create_tree_block(root->fs_info, ins.objectid,
2826                                            blocksize);
2827         if (!buf) {
2828                 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2829                                   0, root->root_key.objectid, level, 0);
2830                 BUG_ON(1);
2831                 return ERR_PTR(-ENOMEM);
2832         }
2833         btrfs_set_buffer_uptodate(buf);
2834         trans->blocks_used++;
2835
2836         return buf;
2837 }
2838
2839 #if 0
2840
2841 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2842                                   struct btrfs_root *root,
2843                                   struct extent_buffer *leaf)
2844 {
2845         u64 leaf_owner;
2846         u64 leaf_generation;
2847         struct btrfs_key key;
2848         struct btrfs_file_extent_item *fi;
2849         int i;
2850         int nritems;
2851         int ret;
2852
2853         BUG_ON(!btrfs_is_leaf(leaf));
2854         nritems = btrfs_header_nritems(leaf);
2855         leaf_owner = btrfs_header_owner(leaf);
2856         leaf_generation = btrfs_header_generation(leaf);
2857
2858         for (i = 0; i < nritems; i++) {
2859                 u64 disk_bytenr;
2860
2861                 btrfs_item_key_to_cpu(leaf, &key, i);
2862                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2863                         continue;
2864                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2865                 if (btrfs_file_extent_type(leaf, fi) ==
2866                     BTRFS_FILE_EXTENT_INLINE)
2867                         continue;
2868                 /*
2869                  * FIXME make sure to insert a trans record that
2870                  * repeats the snapshot del on crash
2871                  */
2872                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2873                 if (disk_bytenr == 0)
2874                         continue;
2875                 ret = btrfs_free_extent(trans, root, disk_bytenr,
2876                                 btrfs_file_extent_disk_num_bytes(leaf, fi),
2877                                 leaf->start, leaf_owner, leaf_generation,
2878                                 key.objectid, 0);
2879                 BUG_ON(ret);
2880         }
2881         return 0;
2882 }
2883
2884 static void noinline reada_walk_down(struct btrfs_root *root,
2885                                      struct extent_buffer *node,
2886                                      int slot)
2887 {
2888         u64 bytenr;
2889         u64 last = 0;
2890         u32 nritems;
2891         u32 refs;
2892         u32 blocksize;
2893         int ret;
2894         int i;
2895         int level;
2896         int skipped = 0;
2897
2898         nritems = btrfs_header_nritems(node);
2899         level = btrfs_header_level(node);
2900         if (level)
2901                 return;
2902
2903         for (i = slot; i < nritems && skipped < 32; i++) {
2904                 bytenr = btrfs_node_blockptr(node, i);
2905                 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2906                              (last > bytenr && last - bytenr > 32 * 1024))) {
2907                         skipped++;
2908                         continue;
2909                 }
2910                 blocksize = btrfs_level_size(root, level - 1);
2911                 if (i != slot) {
2912                         ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2913                                                       blocksize, &refs);
2914                         BUG_ON(ret);
2915                         if (refs != 1) {
2916                                 skipped++;
2917                                 continue;
2918                         }
2919                 }
2920                 mutex_unlock(&root->fs_info->fs_mutex);
2921                 ret = readahead_tree_block(root, bytenr, blocksize,
2922                                            btrfs_node_ptr_generation(node, i));
2923                 last = bytenr + blocksize;
2924                 cond_resched();
2925                 mutex_lock(&root->fs_info->fs_mutex);
2926                 if (ret)
2927                         break;
2928         }
2929 }
2930
2931 /*
2932  * helper function for drop_snapshot, this walks down the tree dropping ref
2933  * counts as it goes.
2934  */
2935 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2936                                    struct btrfs_root *root,
2937                                    struct btrfs_path *path, int *level)
2938 {
2939         u64 root_owner;
2940         u64 root_gen;
2941         u64 bytenr;
2942         u64 ptr_gen;
2943         struct extent_buffer *next;
2944         struct extent_buffer *cur;
2945         struct extent_buffer *parent;
2946         u32 blocksize;
2947         int ret;
2948         u32 refs;
2949
2950         WARN_ON(*level < 0);
2951         WARN_ON(*level >= BTRFS_MAX_LEVEL);
2952         ret = btrfs_lookup_extent_ref(trans, root,
2953                                       path->nodes[*level]->start,
2954                                       path->nodes[*level]->len, &refs);
2955         BUG_ON(ret);
2956         if (refs > 1)
2957                 goto out;
2958
2959         /*
2960          * walk down to the last node level and free all the leaves
2961          */
2962         while(*level >= 0) {
2963                 WARN_ON(*level < 0);
2964                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2965                 cur = path->nodes[*level];
2966
2967                 if (btrfs_header_level(cur) != *level)
2968                         WARN_ON(1);
2969
2970                 if (path->slots[*level] >=
2971                     btrfs_header_nritems(cur))
2972                         break;
2973                 if (*level == 0) {
2974                         ret = drop_leaf_ref(trans, root, cur);
2975                         BUG_ON(ret);
2976                         break;
2977                 }
2978                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2979                 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2980                 blocksize = btrfs_level_size(root, *level - 1);
2981                 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2982                                               &refs);
2983                 BUG_ON(ret);
2984                 if (refs != 1) {
2985                         parent = path->nodes[*level];
2986                         root_owner = btrfs_header_owner(parent);
2987                         root_gen = btrfs_header_generation(parent);
2988                         path->slots[*level]++;
2989                         ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2990                                                 parent->start, root_owner,
2991                                                 root_gen, *level - 1, 1);
2992                         BUG_ON(ret);
2993                         continue;
2994                 }
2995                 next = btrfs_find_tree_block(root, bytenr, blocksize);
2996                 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2997                         free_extent_buffer(next);
2998                         reada_walk_down(root, cur, path->slots[*level]);
2999                         mutex_unlock(&root->fs_info->fs_mutex);
3000                         next = read_tree_block(root, bytenr, blocksize,
3001                                                ptr_gen);
3002                         mutex_lock(&root->fs_info->fs_mutex);
3003                         if (!extent_buffer_uptodate(next)) {
3004                                 if (IS_ERR(next))
3005                                         ret = PTR_ERR(next);
3006                                 else
3007                                         ret = -EIO;
3008                                 break;
3009                         }
3010                 }
3011                 WARN_ON(*level <= 0);
3012                 if (path->nodes[*level-1])
3013                         free_extent_buffer(path->nodes[*level-1]);
3014                 path->nodes[*level-1] = next;
3015                 *level = btrfs_header_level(next);
3016                 path->slots[*level] = 0;
3017         }
3018 out:
3019         WARN_ON(*level < 0);
3020         WARN_ON(*level >= BTRFS_MAX_LEVEL);
3021
3022         if (path->nodes[*level] == root->node) {
3023                 root_owner = root->root_key.objectid;
3024                 parent = path->nodes[*level];
3025         } else {
3026                 parent = path->nodes[*level + 1];
3027                 root_owner = btrfs_header_owner(parent);
3028         }
3029
3030         root_gen = btrfs_header_generation(parent);
3031         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
3032                                 path->nodes[*level]->len, parent->start,
3033                                 root_owner, root_gen, *level, 1);
3034         free_extent_buffer(path->nodes[*level]);
3035         path->nodes[*level] = NULL;
3036         *level += 1;
3037         BUG_ON(ret);
3038         return 0;
3039 }
3040
3041 /*
3042  * helper for dropping snapshots.  This walks back up the tree in the path
3043  * to find the first node higher up where we haven't yet gone through
3044  * all the slots
3045  */
3046 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3047                                  struct btrfs_root *root,
3048                                  struct btrfs_path *path, int *level)
3049 {
3050         u64 root_owner;
3051         u64 root_gen;
3052         struct btrfs_root_item *root_item = &root->root_item;
3053         int i;
3054         int slot;
3055         int ret;
3056
3057         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3058                 slot = path->slots[i];
3059                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3060                         struct extent_buffer *node;
3061                         struct btrfs_disk_key disk_key;
3062                         node = path->nodes[i];
3063                         path->slots[i]++;
3064                         *level = i;
3065                         WARN_ON(*level == 0);
3066                         btrfs_node_key(node, &disk_key, path->slots[i]);
3067                         memcpy(&root_item->drop_progress,
3068                                &disk_key, sizeof(disk_key));
3069                         root_item->drop_level = i;
3070                         return 0;
3071                 } else {
3072                         struct extent_buffer *parent;
3073                         if (path->nodes[*level] == root->node)
3074                                 parent = path->nodes[*level];
3075                         else
3076                                 parent = path->nodes[*level + 1];
3077
3078                         root_owner = btrfs_header_owner(parent);
3079                         root_gen = btrfs_header_generation(parent);
3080                         ret = btrfs_free_extent(trans, root,
3081                                                 path->nodes[*level]->start,
3082                                                 path->nodes[*level]->len,
3083                                                 parent->start, root_owner,
3084                                                 root_gen, *level, 1);
3085                         BUG_ON(ret);
3086                         free_extent_buffer(path->nodes[*level]);
3087                         path->nodes[*level] = NULL;
3088                         *level = i + 1;
3089                 }
3090         }
3091         return 1;
3092 }
3093
3094 #endif
3095
3096 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3097 {
3098         struct btrfs_space_info *sinfo;
3099         struct btrfs_block_group_cache *cache;
3100         u64 start;
3101         u64 end;
3102         u64 ptr;
3103         int ret;
3104
3105         while(1) {
3106                 ret = find_first_extent_bit(&info->block_group_cache, 0,
3107                                             &start, &end, (unsigned int)-1);
3108                 if (ret)
3109                         break;
3110                 ret = get_state_private(&info->block_group_cache, start, &ptr);
3111                 if (!ret) {
3112                         cache = u64_to_ptr(ptr);
3113                         if (cache->free_space_ctl) {
3114                                 btrfs_remove_free_space_cache(cache);
3115                                 kfree(cache->free_space_ctl);
3116                         }
3117                         kfree(cache);
3118                 }
3119                 clear_extent_bits(&info->block_group_cache, start,
3120                                   end, (unsigned int)-1, GFP_NOFS);
3121         }
3122         while(1) {
3123                 ret = find_first_extent_bit(&info->free_space_cache, 0,
3124                                             &start, &end, EXTENT_DIRTY);
3125                 if (ret)
3126                         break;
3127                 clear_extent_dirty(&info->free_space_cache, start,
3128                                    end, GFP_NOFS);
3129         }
3130
3131         while (!list_empty(&info->space_info)) {
3132                 sinfo = list_entry(info->space_info.next,
3133                                    struct btrfs_space_info, list);
3134                 list_del_init(&sinfo->list);
3135                 kfree(sinfo);
3136         }
3137         return 0;
3138 }
3139
3140 static int find_first_block_group(struct btrfs_root *root,
3141                 struct btrfs_path *path, struct btrfs_key *key)
3142 {
3143         int ret;
3144         struct btrfs_key found_key;
3145         struct extent_buffer *leaf;
3146         int slot;
3147
3148         ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3149         if (ret < 0)
3150                 return ret;
3151         while(1) {
3152                 slot = path->slots[0];
3153                 leaf = path->nodes[0];
3154                 if (slot >= btrfs_header_nritems(leaf)) {
3155                         ret = btrfs_next_leaf(root, path);
3156                         if (ret == 0)
3157                                 continue;
3158                         if (ret < 0)
3159                                 goto error;
3160                         break;
3161                 }
3162                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3163
3164                 if (found_key.objectid >= key->objectid &&
3165                     found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3166                         return 0;
3167                 path->slots[0]++;
3168         }
3169         ret = -ENOENT;
3170 error:
3171         return ret;
3172 }
3173
3174 static void account_super_bytes(struct btrfs_fs_info *fs_info,
3175                                 struct btrfs_block_group_cache *cache)
3176 {
3177         u64 bytenr;
3178         u64 *logical;
3179         int stripe_len;
3180         int i, nr, ret;
3181
3182         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
3183                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
3184                 cache->bytes_super += stripe_len;
3185         }
3186
3187         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3188                 bytenr = btrfs_sb_offset(i);
3189                 ret = btrfs_rmap_block(&fs_info->mapping_tree,
3190                                        cache->key.objectid, bytenr,
3191                                        0, &logical, &nr, &stripe_len);
3192                 if (ret)
3193                         return;
3194
3195                 while (nr--) {
3196                         u64 start, len;
3197
3198                         if (logical[nr] > cache->key.objectid +
3199                             cache->key.offset)
3200                                 continue;
3201
3202                         if (logical[nr] + stripe_len <= cache->key.objectid)
3203                                 continue;
3204
3205                         start = logical[nr];
3206                         if (start < cache->key.objectid) {
3207                                 start = cache->key.objectid;
3208                                 len = (logical[nr] + stripe_len) - start;
3209                         } else {
3210                                 len = min_t(u64, stripe_len,
3211                                             cache->key.objectid +
3212                                             cache->key.offset - start);
3213                         }
3214
3215                         cache->bytes_super += len;
3216                 }
3217
3218                 kfree(logical);
3219         }
3220 }
3221
3222 int btrfs_read_block_groups(struct btrfs_root *root)
3223 {
3224         struct btrfs_path *path;
3225         int ret;
3226         int bit;
3227         struct btrfs_block_group_cache *cache;
3228         struct btrfs_fs_info *info = root->fs_info;
3229         struct btrfs_space_info *space_info;
3230         struct extent_io_tree *block_group_cache;
3231         struct btrfs_key key;
3232         struct btrfs_key found_key;
3233         struct extent_buffer *leaf;
3234
3235         block_group_cache = &info->block_group_cache;
3236
3237         root = info->extent_root;
3238         key.objectid = 0;
3239         key.offset = 0;
3240         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3241         path = btrfs_alloc_path();
3242         if (!path)
3243                 return -ENOMEM;
3244
3245         while(1) {
3246                 ret = find_first_block_group(root, path, &key);
3247                 if (ret > 0) {
3248                         ret = 0;
3249                         goto error;
3250                 }
3251                 if (ret != 0) {
3252                         goto error;
3253                 }
3254                 leaf = path->nodes[0];
3255                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3256                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3257                 if (!cache) {
3258                         ret = -ENOMEM;
3259                         goto error;
3260                 }
3261
3262                 read_extent_buffer(leaf, &cache->item,
3263                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
3264                                    sizeof(cache->item));
3265                 memcpy(&cache->key, &found_key, sizeof(found_key));
3266                 cache->cached = 0;
3267                 cache->pinned = 0;
3268                 key.objectid = found_key.objectid + found_key.offset;
3269                 btrfs_release_path(path);
3270                 cache->flags = btrfs_block_group_flags(&cache->item);
3271                 bit = 0;
3272                 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3273                         bit = BLOCK_GROUP_DATA;
3274                 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3275                         bit = BLOCK_GROUP_SYSTEM;
3276                 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3277                         bit = BLOCK_GROUP_METADATA;
3278                 }
3279                 set_avail_alloc_bits(info, cache->flags);
3280                 if (btrfs_chunk_readonly(root, cache->key.objectid))
3281                         cache->ro = 1;
3282
3283                 account_super_bytes(info, cache);
3284
3285                 ret = update_space_info(info, cache->flags, found_key.offset,
3286                                         btrfs_block_group_used(&cache->item),
3287                                         &space_info);
3288                 BUG_ON(ret);
3289                 cache->space_info = space_info;
3290
3291                 /* use EXTENT_LOCKED to prevent merging */
3292                 set_extent_bits(block_group_cache, found_key.objectid,
3293                                 found_key.objectid + found_key.offset - 1,
3294                                 bit | EXTENT_LOCKED, GFP_NOFS);
3295                 set_state_private(block_group_cache, found_key.objectid,
3296                                   (unsigned long)cache);
3297         }
3298         ret = 0;
3299 error:
3300         btrfs_free_path(path);
3301         return ret;
3302 }
3303
3304 struct btrfs_block_group_cache *
3305 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3306                       u64 chunk_objectid, u64 chunk_offset, u64 size)
3307 {
3308         int ret;
3309         int bit = 0;
3310         struct btrfs_block_group_cache *cache;
3311         struct extent_io_tree *block_group_cache;
3312
3313         block_group_cache = &fs_info->block_group_cache;
3314
3315         cache = kzalloc(sizeof(*cache), GFP_NOFS);
3316         BUG_ON(!cache);
3317         cache->key.objectid = chunk_offset;
3318         cache->key.offset = size;
3319
3320         btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3321         btrfs_set_block_group_used(&cache->item, bytes_used);
3322         btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3323         cache->flags = type;
3324         btrfs_set_block_group_flags(&cache->item, type);
3325
3326         account_super_bytes(fs_info, cache);
3327         ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3328                                 &cache->space_info);
3329         BUG_ON(ret);
3330
3331         bit = block_group_state_bits(type);
3332         ret = set_extent_bits(block_group_cache, chunk_offset,
3333                               chunk_offset + size - 1,
3334                               bit | EXTENT_LOCKED, GFP_NOFS);
3335         BUG_ON(ret);
3336
3337         ret = set_state_private(block_group_cache, chunk_offset,
3338                                 (unsigned long)cache);
3339         BUG_ON(ret);
3340         set_avail_alloc_bits(fs_info, type);
3341
3342         return cache;
3343 }
3344
3345 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3346                            struct btrfs_root *root, u64 bytes_used,
3347                            u64 type, u64 chunk_objectid, u64 chunk_offset,
3348                            u64 size)
3349 {
3350         int ret;
3351         struct btrfs_root *extent_root;
3352         struct btrfs_block_group_cache *cache;
3353
3354         cache = btrfs_add_block_group(root->fs_info, bytes_used, type,
3355                                       chunk_objectid, chunk_offset, size);
3356         extent_root = root->fs_info->extent_root;
3357         ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3358                                 sizeof(cache->item));
3359         BUG_ON(ret);
3360
3361         ret = finish_current_insert(trans, extent_root);
3362         BUG_ON(ret);
3363         ret = del_pending_extents(trans, extent_root);
3364         BUG_ON(ret);
3365
3366         return 0;
3367 }
3368
3369 /*
3370  * This is for converter use only.
3371  *
3372  * In that case, we don't know where are free blocks located.
3373  * Therefore all block group cache entries must be setup properly
3374  * before doing any block allocation.
3375  */
3376 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3377                             struct btrfs_root *root)
3378 {
3379         u64 total_bytes;
3380         u64 cur_start;
3381         u64 group_type;
3382         u64 group_size;
3383         u64 group_align;
3384         u64 total_data = 0;
3385         u64 total_metadata = 0;
3386         u64 chunk_objectid;
3387         int ret;
3388         int bit;
3389         struct btrfs_root *extent_root;
3390         struct btrfs_block_group_cache *cache;
3391         struct extent_io_tree *block_group_cache;
3392
3393         extent_root = root->fs_info->extent_root;
3394         block_group_cache = &root->fs_info->block_group_cache;
3395         chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3396         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
3397         group_align = 64 * root->sectorsize;
3398
3399         cur_start = 0;
3400         while (cur_start < total_bytes) {
3401                 group_size = total_bytes / 12;
3402                 group_size = min_t(u64, group_size, total_bytes - cur_start);
3403                 if (cur_start == 0) {
3404                         bit = BLOCK_GROUP_SYSTEM;
3405                         group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3406                         group_size /= 4;
3407                         group_size &= ~(group_align - 1);
3408                         group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3409                         group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3410                 } else {
3411                         group_size &= ~(group_align - 1);
3412                         if (total_data >= total_metadata * 2) {
3413                                 group_type = BTRFS_BLOCK_GROUP_METADATA;
3414                                 group_size = min_t(u64, group_size,
3415                                                    1ULL * 1024 * 1024 * 1024);
3416                                 total_metadata += group_size;
3417                         } else {
3418                                 group_type = BTRFS_BLOCK_GROUP_DATA;
3419                                 group_size = min_t(u64, group_size,
3420                                                    5ULL * 1024 * 1024 * 1024);
3421                                 total_data += group_size;
3422                         }
3423                         if ((total_bytes - cur_start) * 4 < group_size * 5)
3424                                 group_size = total_bytes - cur_start;
3425                 }
3426
3427                 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3428                 BUG_ON(!cache);
3429
3430                 cache->key.objectid = cur_start;
3431                 cache->key.offset = group_size;
3432                 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3433
3434                 btrfs_set_block_group_used(&cache->item, 0);
3435                 btrfs_set_block_group_chunk_objectid(&cache->item,
3436                                                      chunk_objectid);
3437                 btrfs_set_block_group_flags(&cache->item, group_type);
3438
3439                 cache->flags = group_type;
3440
3441                 ret = update_space_info(root->fs_info, group_type, group_size,
3442                                         0, &cache->space_info);
3443                 BUG_ON(ret);
3444                 set_avail_alloc_bits(extent_root->fs_info, group_type);
3445
3446                 set_extent_bits(block_group_cache, cur_start,
3447                                 cur_start + group_size - 1,
3448                                 bit | EXTENT_LOCKED, GFP_NOFS);
3449                 set_state_private(block_group_cache, cur_start,
3450                                   (unsigned long)cache);
3451                 cur_start += group_size;
3452         }
3453         /* then insert all the items */
3454         cur_start = 0;
3455         while(cur_start < total_bytes) {
3456                 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3457                 BUG_ON(!cache);
3458
3459                 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3460                                         sizeof(cache->item));
3461                 BUG_ON(ret);
3462
3463                 finish_current_insert(trans, extent_root);
3464                 ret = del_pending_extents(trans, extent_root);
3465                 BUG_ON(ret);
3466
3467                 cur_start = cache->key.objectid + cache->key.offset;
3468         }
3469         return 0;
3470 }
3471
3472 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3473                              struct btrfs_root *root,
3474                              u64 bytenr, u64 num_bytes, int alloc,
3475                              int mark_free)
3476 {
3477         return update_block_group(trans, root, bytenr, num_bytes,
3478                                   alloc, mark_free);
3479 }
3480
3481 /*
3482  * Just remove a block group item in extent tree
3483  * Caller should ensure the block group is empty and all space is pinned.
3484  * Or new tree block/data may be allocated into it.
3485  */
3486 static int free_block_group_item(struct btrfs_trans_handle *trans,
3487                                  struct btrfs_fs_info *fs_info,
3488                                  u64 bytenr, u64 len)
3489 {
3490         struct btrfs_path *path;
3491         struct btrfs_key key;
3492         struct btrfs_root *root = fs_info->extent_root;
3493         int ret = 0;
3494
3495         key.objectid = bytenr;
3496         key.offset = len;
3497         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3498
3499         path = btrfs_alloc_path();
3500         if (!path)
3501                 return -ENOMEM;
3502
3503         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3504         if (ret > 0) {
3505                 ret = -ENOENT;
3506                 goto out;
3507         }
3508         if (ret < 0)
3509                 goto out;
3510
3511         ret = btrfs_del_item(trans, root, path);
3512 out:
3513         btrfs_free_path(path);
3514         return ret;
3515 }
3516
3517 static int free_dev_extent_item(struct btrfs_trans_handle *trans,
3518                                 struct btrfs_fs_info *fs_info,
3519                                 u64 devid, u64 dev_offset)
3520 {
3521         struct btrfs_root *root = fs_info->dev_root;
3522         struct btrfs_path *path;
3523         struct btrfs_key key;
3524         int ret;
3525
3526         path = btrfs_alloc_path();
3527         if (!path)
3528                 return -ENOMEM;
3529
3530         key.objectid = devid;
3531         key.type = BTRFS_DEV_EXTENT_KEY;
3532         key.offset = dev_offset;
3533
3534         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3535         if (ret < 0)
3536                 goto out;
3537         if (ret > 0) {
3538                 ret = -ENOENT;
3539                 goto out;
3540         }
3541
3542         ret = btrfs_del_item(trans, root, path);
3543 out:
3544         btrfs_free_path(path);
3545         return ret;
3546 }
3547
3548 static int free_chunk_dev_extent_items(struct btrfs_trans_handle *trans,
3549                                        struct btrfs_fs_info *fs_info,
3550                                        u64 chunk_offset)
3551 {
3552         struct btrfs_chunk *chunk = NULL;
3553         struct btrfs_root *root= fs_info->chunk_root;
3554         struct btrfs_path *path;
3555         struct btrfs_key key;
3556         u16 num_stripes;
3557         int i;
3558         int ret;
3559
3560         path = btrfs_alloc_path();
3561         if (!path)
3562                 return -ENOMEM;
3563
3564         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3565         key.type = BTRFS_CHUNK_ITEM_KEY;
3566         key.offset = chunk_offset;
3567
3568         ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3569         if (ret < 0)
3570                 goto out;
3571         if (ret > 0) {
3572                 ret = -ENOENT;
3573                 goto out;
3574         }
3575         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3576                                struct btrfs_chunk);
3577         num_stripes = btrfs_chunk_num_stripes(path->nodes[0], chunk);
3578         for (i = 0; i < num_stripes; i++) {
3579                 ret = free_dev_extent_item(trans, fs_info,
3580                         btrfs_stripe_devid_nr(path->nodes[0], chunk, i),
3581                         btrfs_stripe_offset_nr(path->nodes[0], chunk, i));
3582                 if (ret < 0)
3583                         goto out;
3584         }
3585 out:
3586         btrfs_free_path(path);
3587         return ret;
3588 }
3589
3590 static int free_system_chunk_item(struct btrfs_super_block *super,
3591                                   struct btrfs_key *key)
3592 {
3593         struct btrfs_disk_key *disk_key;
3594         struct btrfs_key cpu_key;
3595         u32 array_size = btrfs_super_sys_array_size(super);
3596         char *ptr = (char *)super->sys_chunk_array;
3597         int cur = 0;
3598         int ret = -ENOENT;
3599
3600         while (cur < btrfs_super_sys_array_size(super)) {
3601                 struct btrfs_chunk *chunk;
3602                 u32 num_stripes;
3603                 u32 chunk_len;
3604
3605                 disk_key = (struct btrfs_disk_key *)(ptr + cur);
3606                 btrfs_disk_key_to_cpu(&cpu_key, disk_key);
3607                 if (cpu_key.type != BTRFS_CHUNK_ITEM_KEY) {
3608                         /* just in case */
3609                         ret = -EIO;
3610                         goto out;
3611                 }
3612
3613                 chunk = (struct btrfs_chunk *)(ptr + cur + sizeof(*disk_key));
3614                 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3615                 chunk_len = btrfs_chunk_item_size(num_stripes) +
3616                             sizeof(*disk_key);
3617
3618                 if (key->objectid == cpu_key.objectid &&
3619                     key->offset == cpu_key.offset &&
3620                     key->type == cpu_key.type) {
3621                         memmove(ptr + cur, ptr + cur + chunk_len,
3622                                 array_size - cur - chunk_len);
3623                         array_size -= chunk_len;
3624                         btrfs_set_super_sys_array_size(super, array_size);
3625                         ret = 0;
3626                         goto out;
3627                 }
3628
3629                 cur += chunk_len;
3630         }
3631 out:
3632         return ret;
3633 }
3634
3635 static int free_chunk_item(struct btrfs_trans_handle *trans,
3636                            struct btrfs_fs_info *fs_info,
3637                            u64 bytenr, u64 len)
3638 {
3639         struct btrfs_path *path;
3640         struct btrfs_key key;
3641         struct btrfs_root *root = fs_info->chunk_root;
3642         struct btrfs_chunk *chunk;
3643         u64 chunk_type;
3644         int ret;
3645
3646         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3647         key.offset = bytenr;
3648         key.type = BTRFS_CHUNK_ITEM_KEY;
3649
3650         path = btrfs_alloc_path();
3651         if (!path)
3652                 return -ENOMEM;
3653
3654         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3655         if (ret > 0) {
3656                 ret = -ENOENT;
3657                 goto out;
3658         }
3659         if (ret < 0)
3660                 goto out;
3661         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3662                                struct btrfs_chunk);
3663         chunk_type = btrfs_chunk_type(path->nodes[0], chunk);
3664
3665         ret = btrfs_del_item(trans, root, path);
3666         if (ret < 0)
3667                 goto out;
3668
3669         if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3670                 ret = free_system_chunk_item(fs_info->super_copy, &key);
3671 out:
3672         btrfs_free_path(path);
3673         return ret;
3674 }
3675
3676 static u64 get_dev_extent_len(struct map_lookup *map)
3677 {
3678         int div;
3679
3680         switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3681         case 0: /* Single */
3682         case BTRFS_BLOCK_GROUP_DUP:
3683         case BTRFS_BLOCK_GROUP_RAID1:
3684                 div = 1;
3685                 break;
3686         case BTRFS_BLOCK_GROUP_RAID5:
3687                 div = (map->num_stripes - 1);
3688                 break;
3689         case BTRFS_BLOCK_GROUP_RAID6:
3690                 div = (map->num_stripes - 2);
3691                 break;
3692         case BTRFS_BLOCK_GROUP_RAID10:
3693                 div = (map->num_stripes / map->sub_stripes);
3694                 break;
3695         default:
3696                 /* normally, read chunk security hook should handled it */
3697                 BUG_ON(1);
3698         }
3699         return map->ce.size / div;
3700 }
3701
3702 /* free block group/chunk related caches */
3703 static int free_block_group_cache(struct btrfs_trans_handle *trans,
3704                                   struct btrfs_fs_info *fs_info,
3705                                   u64 bytenr, u64 len)
3706 {
3707         struct btrfs_block_group_cache *cache;
3708         struct cache_extent *ce;
3709         struct map_lookup *map;
3710         int ret;
3711         int i;
3712         u64 flags;
3713
3714         /* Free block group cache first */
3715         cache = btrfs_lookup_block_group(fs_info, bytenr);
3716         if (!cache)
3717                 return -ENOENT;
3718         flags = cache->flags;
3719         if (cache->free_space_ctl) {
3720                 btrfs_remove_free_space_cache(cache);
3721                 kfree(cache->free_space_ctl);
3722         }
3723         clear_extent_bits(&fs_info->block_group_cache, bytenr, bytenr + len,
3724                           (unsigned int)-1, GFP_NOFS);
3725         ret = free_space_info(fs_info, flags, len, 0, NULL);
3726         if (ret < 0)
3727                 goto out;
3728         kfree(cache);
3729
3730         /* Then free mapping info and dev usage info */
3731         ce = search_cache_extent(&fs_info->mapping_tree.cache_tree, bytenr);
3732         if (!ce || ce->start != bytenr) {
3733                 ret = -ENOENT;
3734                 goto out;
3735         }
3736         map = container_of(ce, struct map_lookup, ce);
3737         for (i = 0; i < map->num_stripes; i++) {
3738                 struct btrfs_device *device;
3739
3740                 device = map->stripes[i].dev;
3741                 device->bytes_used -= get_dev_extent_len(map);
3742                 ret = btrfs_update_device(trans, device);
3743                 if (ret < 0)
3744                         goto out;
3745         }
3746         remove_cache_extent(&fs_info->mapping_tree.cache_tree, ce);
3747         free(map);
3748 out:
3749         return ret;
3750 }
3751
3752 int btrfs_free_block_group(struct btrfs_trans_handle *trans,
3753                            struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
3754 {
3755         struct btrfs_root *extent_root = fs_info->extent_root;
3756         struct btrfs_path *path;
3757         struct btrfs_block_group_item *bgi;
3758         struct btrfs_key key;
3759         int ret = 0;
3760
3761         path = btrfs_alloc_path();
3762         if (!path)
3763                 return -ENOMEM;
3764
3765         key.objectid = bytenr;
3766         key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3767         key.offset = len;
3768
3769         /* Double check the block group to ensure it's empty */
3770         ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3771         if (ret > 0) {
3772                 ret = -ENONET;
3773                 goto out;
3774         }
3775         if (ret < 0)
3776                 goto out;
3777
3778         bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3779                              struct btrfs_block_group_item);
3780         if (btrfs_disk_block_group_used(path->nodes[0], bgi)) {
3781                 fprintf(stderr,
3782                         "WARNING: block group [%llu,%llu) is not empty\n",
3783                         bytenr, bytenr + len);
3784                 ret = -EINVAL;
3785                 goto out;
3786         }
3787         btrfs_release_path(path);
3788
3789         /*
3790          * Now pin all space in the block group, to prevent further transaction
3791          * allocate space from it.
3792          * Every operation needs a transaction must be in the range.
3793          */
3794         btrfs_pin_extent(fs_info, bytenr, len);
3795
3796         /* delete block group item and chunk item */
3797         ret = free_block_group_item(trans, fs_info, bytenr, len);
3798         if (ret < 0) {
3799                 fprintf(stderr,
3800                         "failed to free block group item for [%llu,%llu)\n",
3801                         bytenr, bytenr + len);
3802                 btrfs_unpin_extent(fs_info, bytenr, len);
3803                 goto out;
3804         }
3805
3806         ret = free_chunk_dev_extent_items(trans, fs_info, bytenr);
3807         if (ret < 0) {
3808                 fprintf(stderr,
3809                         "failed to dev extents belongs to [%llu,%llu)\n",
3810                         bytenr, bytenr + len);
3811                 btrfs_unpin_extent(fs_info, bytenr, len);
3812                 goto out;
3813         }
3814         ret = free_chunk_item(trans, fs_info, bytenr, len);
3815         if (ret < 0) {
3816                 fprintf(stderr,
3817                         "failed to free chunk for [%llu,%llu)\n",
3818                         bytenr, bytenr + len);
3819                 btrfs_unpin_extent(fs_info, bytenr, len);
3820                 goto out;
3821         }
3822
3823         /* Now release the block_group_cache */
3824         ret = free_block_group_cache(trans, fs_info, bytenr, len);
3825         btrfs_unpin_extent(fs_info, bytenr, len);
3826
3827 out:
3828         btrfs_free_path(path);
3829         return ret;
3830 }
3831
3832 /*
3833  * Fixup block accounting. The initial block accounting created by
3834  * make_block_groups isn't accuracy in this case.
3835  */
3836 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3837                                struct btrfs_root *root)
3838 {
3839         int ret;
3840         int slot;
3841         u64 start = 0;
3842         u64 bytes_used = 0;
3843         struct btrfs_path path;
3844         struct btrfs_key key;
3845         struct extent_buffer *leaf;
3846         struct btrfs_block_group_cache *cache;
3847         struct btrfs_fs_info *fs_info = root->fs_info;
3848
3849         root = root->fs_info->extent_root;
3850
3851         while(extent_root_pending_ops(fs_info)) {
3852                 ret = finish_current_insert(trans, root);
3853                 if (ret)
3854                         return ret;
3855                 ret = del_pending_extents(trans, root);
3856                 if (ret)
3857                         return ret;
3858         }
3859
3860         while(1) {
3861                 cache = btrfs_lookup_first_block_group(fs_info, start);
3862                 if (!cache)
3863                         break;
3864                 start = cache->key.objectid + cache->key.offset;
3865                 btrfs_set_block_group_used(&cache->item, 0);
3866                 cache->space_info->bytes_used = 0;
3867                 set_extent_bits(&root->fs_info->block_group_cache,
3868                                 cache->key.objectid,
3869                                 cache->key.objectid + cache->key.offset -1,
3870                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
3871         }
3872
3873         btrfs_init_path(&path);
3874         key.offset = 0;
3875         key.objectid = 0;
3876         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
3877         ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3878                                 &key, &path, 0, 0);
3879         if (ret < 0)
3880                 return ret;
3881         while(1) {
3882                 leaf = path.nodes[0];
3883                 slot = path.slots[0];
3884                 if (slot >= btrfs_header_nritems(leaf)) {
3885                         ret = btrfs_next_leaf(root, &path);
3886                         if (ret < 0)
3887                                 return ret;
3888                         if (ret > 0)
3889                                 break;
3890                         leaf = path.nodes[0];
3891                         slot = path.slots[0];
3892                 }
3893                 btrfs_item_key_to_cpu(leaf, &key, slot);
3894                 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3895                         bytes_used += key.offset;
3896                         ret = btrfs_update_block_group(trans, root,
3897                                   key.objectid, key.offset, 1, 0);
3898                         BUG_ON(ret);
3899                 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3900                         bytes_used += root->nodesize;
3901                         ret = btrfs_update_block_group(trans, root,
3902                                   key.objectid, root->nodesize, 1, 0);
3903                         BUG_ON(ret);
3904                 }
3905                 path.slots[0]++;
3906         }
3907         btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3908         btrfs_release_path(&path);
3909         return 0;
3910 }
3911
3912 static void __get_extent_size(struct btrfs_root *root, struct btrfs_path *path,
3913                               u64 *start, u64 *len)
3914 {
3915         struct btrfs_key key;
3916
3917         btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3918         BUG_ON(!(key.type == BTRFS_EXTENT_ITEM_KEY ||
3919                  key.type == BTRFS_METADATA_ITEM_KEY));
3920         *start = key.objectid;
3921         if (key.type == BTRFS_EXTENT_ITEM_KEY)
3922                 *len = key.offset;
3923         else
3924                 *len = root->nodesize;
3925 }
3926
3927 /*
3928  * Find first overlap extent for range [bytenr, bytenr + len)
3929  * Return 0 for found and point path to it.
3930  * Return >0 for not found.
3931  * Return <0 for err
3932  */
3933 int btrfs_search_overlap_extent(struct btrfs_root *root,
3934                                 struct btrfs_path *path, u64 bytenr, u64 len)
3935 {
3936         struct btrfs_key key;
3937         u64 cur_start;
3938         u64 cur_len;
3939         int ret;
3940
3941         key.objectid = bytenr;
3942         key.type = BTRFS_EXTENT_DATA_KEY;
3943         key.offset = (u64)-1;
3944
3945         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3946         if (ret < 0)
3947                 return ret;
3948         BUG_ON(ret == 0);
3949
3950         ret = btrfs_previous_extent_item(root, path, 0);
3951         if (ret < 0)
3952                 return ret;
3953         /* no previous, check next extent */
3954         if (ret > 0)
3955                 goto next;
3956         __get_extent_size(root, path, &cur_start, &cur_len);
3957         /* Tail overlap */
3958         if (cur_start + cur_len > bytenr)
3959                 return 1;
3960
3961 next:
3962         ret = btrfs_next_extent_item(root, path, bytenr + len);
3963         if (ret < 0)
3964                 return ret;
3965         /* No next, prev already checked, no overlap */
3966         if (ret > 0)
3967                 return 0;
3968         __get_extent_size(root, path, &cur_start, &cur_len);
3969         /* head overlap*/
3970         if (cur_start < bytenr + len)
3971                 return 1;
3972         return 0;
3973 }
3974
3975 static int __btrfs_record_file_extent(struct btrfs_trans_handle *trans,
3976                                       struct btrfs_root *root, u64 objectid,
3977                                       struct btrfs_inode_item *inode,
3978                                       u64 file_pos, u64 disk_bytenr,
3979                                       u64 *ret_num_bytes)
3980 {
3981         int ret;
3982         struct btrfs_fs_info *info = root->fs_info;
3983         struct btrfs_root *extent_root = info->extent_root;
3984         struct extent_buffer *leaf;
3985         struct btrfs_file_extent_item *fi;
3986         struct btrfs_key ins_key;
3987         struct btrfs_path *path;
3988         struct btrfs_extent_item *ei;
3989         u64 nbytes;
3990         u64 extent_num_bytes;
3991         u64 extent_bytenr;
3992         u64 extent_offset;
3993         u64 num_bytes = *ret_num_bytes;
3994
3995         num_bytes = min_t(u64, num_bytes, BTRFS_MAX_EXTENT_SIZE);
3996         /*
3997          * All supported file system should not use its 0 extent.
3998          * As it's for hole
3999          */
4000         if (disk_bytenr == 0) {
4001                 ret = btrfs_insert_file_extent(trans, root, objectid,
4002                                                 file_pos, disk_bytenr,
4003                                                 num_bytes, num_bytes);
4004                 return ret;
4005         }
4006
4007         path = btrfs_alloc_path();
4008         if (!path)
4009                 return -ENOMEM;
4010
4011         /* First to check extent overlap */
4012         ret = btrfs_search_overlap_extent(extent_root, path, disk_bytenr,
4013                                           num_bytes);
4014         if (ret < 0)
4015                 goto fail;
4016         if (ret > 0) {
4017                 /* Found overlap */
4018                 u64 cur_start;
4019                 u64 cur_len;
4020
4021                 __get_extent_size(extent_root, path, &cur_start, &cur_len);
4022                 /*
4023                  * For convert case, this extent should be a subset of
4024                  * existing one.
4025                  */
4026                 BUG_ON(disk_bytenr < cur_start);
4027
4028                 extent_bytenr = cur_start;
4029                 extent_num_bytes = cur_len;
4030                 extent_offset = disk_bytenr - extent_bytenr;
4031         } else {
4032                 /* No overlap, create new extent */
4033                 btrfs_release_path(path);
4034                 ins_key.objectid = disk_bytenr;
4035                 ins_key.offset = num_bytes;
4036                 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
4037
4038                 ret = btrfs_insert_empty_item(trans, extent_root, path,
4039                                               &ins_key, sizeof(*ei));
4040                 if (ret == 0) {
4041                         leaf = path->nodes[0];
4042                         ei = btrfs_item_ptr(leaf, path->slots[0],
4043                                             struct btrfs_extent_item);
4044
4045                         btrfs_set_extent_refs(leaf, ei, 0);
4046                         btrfs_set_extent_generation(leaf, ei, 0);
4047                         btrfs_set_extent_flags(leaf, ei,
4048                                                BTRFS_EXTENT_FLAG_DATA);
4049                         btrfs_mark_buffer_dirty(leaf);
4050
4051                         ret = btrfs_update_block_group(trans, root, disk_bytenr,
4052                                                        num_bytes, 1, 0);
4053                         if (ret)
4054                                 goto fail;
4055                 } else if (ret != -EEXIST) {
4056                         goto fail;
4057                 }
4058                 btrfs_extent_post_op(trans, extent_root);
4059                 extent_bytenr = disk_bytenr;
4060                 extent_num_bytes = num_bytes;
4061                 extent_offset = 0;
4062         }
4063         btrfs_release_path(path);
4064         ins_key.objectid = objectid;
4065         ins_key.offset = file_pos;
4066         btrfs_set_key_type(&ins_key, BTRFS_EXTENT_DATA_KEY);
4067         ret = btrfs_insert_empty_item(trans, root, path, &ins_key,
4068                                       sizeof(*fi));
4069         if (ret)
4070                 goto fail;
4071         leaf = path->nodes[0];
4072         fi = btrfs_item_ptr(leaf, path->slots[0],
4073                             struct btrfs_file_extent_item);
4074         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4075         btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
4076         btrfs_set_file_extent_disk_bytenr(leaf, fi, extent_bytenr);
4077         btrfs_set_file_extent_disk_num_bytes(leaf, fi, extent_num_bytes);
4078         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
4079         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
4080         btrfs_set_file_extent_ram_bytes(leaf, fi, extent_num_bytes);
4081         btrfs_set_file_extent_compression(leaf, fi, 0);
4082         btrfs_set_file_extent_encryption(leaf, fi, 0);
4083         btrfs_set_file_extent_other_encoding(leaf, fi, 0);
4084         btrfs_mark_buffer_dirty(leaf);
4085
4086         nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
4087         btrfs_set_stack_inode_nbytes(inode, nbytes);
4088         btrfs_release_path(path);
4089
4090         ret = btrfs_inc_extent_ref(trans, root, extent_bytenr, extent_num_bytes,
4091                                    0, root->root_key.objectid, objectid,
4092                                    file_pos - extent_offset);
4093         if (ret)
4094                 goto fail;
4095         ret = 0;
4096         *ret_num_bytes = min(extent_num_bytes - extent_offset, num_bytes);
4097 fail:
4098         btrfs_free_path(path);
4099         return ret;
4100 }
4101
4102 /*
4103  * Record a file extent. Do all the required works, such as inserting
4104  * file extent item, inserting extent item and backref item into extent
4105  * tree and updating block accounting.
4106  */
4107 int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
4108                               struct btrfs_root *root, u64 objectid,
4109                               struct btrfs_inode_item *inode,
4110                               u64 file_pos, u64 disk_bytenr,
4111                               u64 num_bytes)
4112 {
4113         u64 cur_disk_bytenr = disk_bytenr;
4114         u64 cur_file_pos = file_pos;
4115         u64 cur_num_bytes = num_bytes;
4116         int ret = 0;
4117
4118         while (num_bytes > 0) {
4119                 ret = __btrfs_record_file_extent(trans, root, objectid,
4120                                                  inode, cur_file_pos,
4121                                                  cur_disk_bytenr,
4122                                                  &cur_num_bytes);
4123                 if (ret < 0)
4124                         break;
4125                 cur_disk_bytenr += cur_num_bytes;
4126                 cur_file_pos += cur_num_bytes;
4127                 num_bytes -= cur_num_bytes;
4128         }
4129         return ret;
4130 }
4131
4132
4133 static int add_excluded_extent(struct btrfs_root *root,
4134                                u64 start, u64 num_bytes)
4135 {
4136         u64 end = start + num_bytes - 1;
4137         set_extent_bits(&root->fs_info->pinned_extents,
4138                         start, end, EXTENT_UPTODATE, GFP_NOFS);
4139         return 0;
4140 }
4141
4142 void free_excluded_extents(struct btrfs_root *root,
4143                            struct btrfs_block_group_cache *cache)
4144 {
4145         u64 start, end;
4146
4147         start = cache->key.objectid;
4148         end = start + cache->key.offset - 1;
4149
4150         clear_extent_bits(&root->fs_info->pinned_extents,
4151                           start, end, EXTENT_UPTODATE, GFP_NOFS);
4152 }
4153
4154 int exclude_super_stripes(struct btrfs_root *root,
4155                           struct btrfs_block_group_cache *cache)
4156 {
4157         u64 bytenr;
4158         u64 *logical;
4159         int stripe_len;
4160         int i, nr, ret;
4161
4162         if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
4163                 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
4164                 cache->bytes_super += stripe_len;
4165                 ret = add_excluded_extent(root, cache->key.objectid,
4166                                           stripe_len);
4167                 if (ret)
4168                         return ret;
4169         }
4170
4171         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4172                 bytenr = btrfs_sb_offset(i);
4173                 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
4174                                        cache->key.objectid, bytenr,
4175                                        0, &logical, &nr, &stripe_len);
4176                 if (ret)
4177                         return ret;
4178
4179                 while (nr--) {
4180                         u64 start, len;
4181
4182                         if (logical[nr] > cache->key.objectid +
4183                             cache->key.offset)
4184                                 continue;
4185
4186                         if (logical[nr] + stripe_len <= cache->key.objectid)
4187                                 continue;
4188
4189                         start = logical[nr];
4190                         if (start < cache->key.objectid) {
4191                                 start = cache->key.objectid;
4192                                 len = (logical[nr] + stripe_len) - start;
4193                         } else {
4194                                 len = min_t(u64, stripe_len,
4195                                             cache->key.objectid +
4196                                             cache->key.offset - start);
4197                         }
4198
4199                         cache->bytes_super += len;
4200                         ret = add_excluded_extent(root, start, len);
4201                         if (ret) {
4202                                 kfree(logical);
4203                                 return ret;
4204                         }
4205                 }
4206
4207                 kfree(logical);
4208         }
4209         return 0;
4210 }
4211
4212 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
4213                        struct btrfs_fs_info *info, u64 start, u64 end)
4214 {
4215         u64 extent_start, extent_end, size, total_added = 0;
4216         int ret;
4217
4218         while (start < end) {
4219                 ret = find_first_extent_bit(&info->pinned_extents, start,
4220                                             &extent_start, &extent_end,
4221                                             EXTENT_DIRTY | EXTENT_UPTODATE);
4222                 if (ret)
4223                         break;
4224
4225                 if (extent_start <= start) {
4226                         start = extent_end + 1;
4227                 } else if (extent_start > start && extent_start < end) {
4228                         size = extent_start - start;
4229                         total_added += size;
4230                         ret = btrfs_add_free_space(block_group->free_space_ctl,
4231                                                    start, size);
4232                         BUG_ON(ret); /* -ENOMEM or logic error */
4233                         start = extent_end + 1;
4234                 } else {
4235                         break;
4236                 }
4237         }
4238
4239         if (start < end) {
4240                 size = end - start;
4241                 total_added += size;
4242                 ret = btrfs_add_free_space(block_group->free_space_ctl, start,
4243                                            size);
4244                 BUG_ON(ret); /* -ENOMEM or logic error */
4245         }
4246
4247         return total_added;
4248 }