Breakout BTRFS_SETGET_FUNCS into a separate C file, the inlines were too big.
[linux-2.6-block.git] / fs / btrfs / 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 <linux/sched.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "print-tree.h"
23 #include "transaction.h"
24
25 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
26 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
27 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
28
29 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
30                                  btrfs_root *extent_root);
31 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
32                                btrfs_root *extent_root);
33
34 static int cache_block_group(struct btrfs_root *root,
35                              struct btrfs_block_group_cache *block_group)
36 {
37         struct btrfs_path *path;
38         int ret;
39         struct btrfs_key key;
40         struct extent_buffer *leaf;
41         struct extent_map_tree *free_space_cache;
42         int slot;
43         u64 last = 0;
44         u64 hole_size;
45         u64 first_free;
46         int found = 0;
47
48         root = root->fs_info->extent_root;
49         free_space_cache = &root->fs_info->free_space_cache;
50
51         if (block_group->cached)
52                 return 0;
53
54         path = btrfs_alloc_path();
55         if (!path)
56                 return -ENOMEM;
57
58         path->reada = 2;
59         first_free = block_group->key.objectid;
60         key.objectid = block_group->key.objectid;
61         key.offset = 0;
62
63         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
64         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
65
66         if (ret < 0)
67                 return ret;
68
69         if (ret && path->slots[0] > 0)
70                 path->slots[0]--;
71
72         while(1) {
73                 leaf = path->nodes[0];
74                 slot = path->slots[0];
75                 if (slot >= btrfs_header_nritems(leaf)) {
76                         ret = btrfs_next_leaf(root, path);
77                         if (ret < 0)
78                                 goto err;
79                         if (ret == 0) {
80                                 continue;
81                         } else {
82                                 break;
83                         }
84                 }
85
86                 btrfs_item_key_to_cpu(leaf, &key, slot);
87                 if (key.objectid < block_group->key.objectid) {
88                         if (key.objectid + key.offset > first_free)
89                                 first_free = key.objectid + key.offset;
90                         goto next;
91                 }
92
93                 if (key.objectid >= block_group->key.objectid +
94                     block_group->key.offset) {
95                         break;
96                 }
97
98                 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
99                         if (!found) {
100                                 last = first_free;
101                                 found = 1;
102                         }
103                         if (key.objectid > last) {
104                                 hole_size = key.objectid - last;
105                                 set_extent_dirty(free_space_cache, last,
106                                                  last + hole_size - 1,
107                                                  GFP_NOFS);
108                         }
109                         last = key.objectid + key.offset;
110                 }
111 next:
112                 path->slots[0]++;
113         }
114
115         if (!found)
116                 last = first_free;
117         if (block_group->key.objectid +
118             block_group->key.offset > last) {
119                 hole_size = block_group->key.objectid +
120                         block_group->key.offset - last;
121                 set_extent_dirty(free_space_cache, last,
122                                  last + hole_size - 1, GFP_NOFS);
123         }
124         block_group->cached = 1;
125 err:
126         btrfs_free_path(path);
127         return 0;
128 }
129
130 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
131                                                          btrfs_fs_info *info,
132                                                          u64 bytenr)
133 {
134         struct extent_map_tree *block_group_cache;
135         struct btrfs_block_group_cache *block_group = NULL;
136         u64 ptr;
137         u64 start;
138         u64 end;
139         int ret;
140
141         block_group_cache = &info->block_group_cache;
142         ret = find_first_extent_bit(block_group_cache,
143                                     bytenr, &start, &end,
144                                     BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA);
145         if (ret) {
146                 return NULL;
147         }
148         ret = get_state_private(block_group_cache, start, &ptr);
149         if (ret)
150                 return NULL;
151
152         block_group = (struct btrfs_block_group_cache *)ptr;
153
154
155         if (block_group->key.objectid <= bytenr && bytenr <=
156             block_group->key.objectid + block_group->key.offset)
157                 return block_group;
158
159         return NULL;
160 }
161
162 static u64 find_search_start(struct btrfs_root *root,
163                              struct btrfs_block_group_cache **cache_ret,
164                              u64 search_start, int num, int data)
165 {
166         int ret;
167         struct btrfs_block_group_cache *cache = *cache_ret;
168         u64 last;
169         u64 start = 0;
170         u64 end = 0;
171
172 again:
173         ret = cache_block_group(root, cache);
174         if (ret)
175                 goto out;
176         last = max(search_start, cache->key.objectid);
177
178         while(1) {
179                 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
180                                             last, &start, &end, EXTENT_DIRTY);
181                 if (ret) {
182                         goto new_group;
183                 }
184
185                 start = max(last, start);
186                 last = end + 1;
187                 if (end + 1 - start < num)
188                         continue;
189                 if (start + num >= cache->key.objectid + cache->key.offset)
190                         goto new_group;
191                 return start;
192         }
193 out:
194         return search_start;
195
196 new_group:
197         last = cache->key.objectid + cache->key.offset;
198         cache = btrfs_lookup_block_group(root->fs_info, last);
199         if (!cache) {
200                 return search_start;
201         }
202         cache = btrfs_find_block_group(root, cache, last, data, 0);
203         *cache_ret = cache;
204         goto again;
205 }
206
207 static u64 div_factor(u64 num, int factor)
208 {
209         num *= factor;
210         do_div(num, 10);
211         return num;
212 }
213
214 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
215                                                  struct btrfs_block_group_cache
216                                                  *hint, u64 search_start,
217                                                  int data, int owner)
218 {
219         struct btrfs_block_group_cache *cache;
220         struct extent_map_tree *block_group_cache;
221         struct btrfs_block_group_cache *found_group = NULL;
222         struct btrfs_fs_info *info = root->fs_info;
223         u64 used;
224         u64 last = 0;
225         u64 hint_last;
226         u64 start;
227         u64 end;
228         u64 free_check;
229         u64 ptr;
230         int bit;
231         int ret;
232         int full_search = 0;
233         int factor = 8;
234         int data_swap = 0;
235
236         block_group_cache = &info->block_group_cache;
237
238         if (!owner)
239                 factor = 5;
240
241         if (data)
242                 bit = BLOCK_GROUP_DATA;
243         else
244                 bit = BLOCK_GROUP_METADATA;
245
246         if (search_start) {
247                 struct btrfs_block_group_cache *shint;
248                 shint = btrfs_lookup_block_group(info, search_start);
249                 if (shint && shint->data == data) {
250                         used = btrfs_block_group_used(&shint->item);
251                         if (used < div_factor(shint->key.offset, factor)) {
252                                 return shint;
253                         }
254                 }
255         }
256         if (hint && hint->data == data) {
257                 used = btrfs_block_group_used(&hint->item);
258                 if (used < div_factor(hint->key.offset, factor)) {
259                         return hint;
260                 }
261                 last = hint->key.objectid + hint->key.offset;
262                 hint_last = last;
263         } else {
264                 if (hint)
265                         hint_last = max(hint->key.objectid, search_start);
266                 else
267                         hint_last = search_start;
268
269                 last = hint_last;
270         }
271 again:
272         while(1) {
273                 ret = find_first_extent_bit(block_group_cache, last,
274                                             &start, &end, bit);
275                 if (ret)
276                         break;
277
278                 ret = get_state_private(block_group_cache, start, &ptr);
279                 if (ret)
280                         break;
281
282                 cache = (struct btrfs_block_group_cache *)ptr;
283                 last = cache->key.objectid + cache->key.offset;
284                 used = btrfs_block_group_used(&cache->item);
285
286                 if (full_search)
287                         free_check = cache->key.offset;
288                 else
289                         free_check = div_factor(cache->key.offset, factor);
290
291                 if (used < free_check) {
292                         found_group = cache;
293                         goto found;
294                 }
295                 cond_resched();
296         }
297         if (!full_search) {
298                 last = search_start;
299                 full_search = 1;
300                 goto again;
301         }
302         if (!data_swap) {
303                 data_swap = 1;
304                 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
305                 last = search_start;
306                 goto again;
307         }
308 found:
309         return found_group;
310 }
311
312 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
313                                 struct btrfs_root *root,
314                                 u64 bytenr, u64 num_bytes)
315 {
316         struct btrfs_path *path;
317         int ret;
318         struct btrfs_key key;
319         struct extent_buffer *l;
320         struct btrfs_extent_item *item;
321         u32 refs;
322
323         WARN_ON(num_bytes < root->sectorsize);
324         path = btrfs_alloc_path();
325         if (!path)
326                 return -ENOMEM;
327
328         key.objectid = bytenr;
329         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
330         key.offset = num_bytes;
331         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
332                                 0, 1);
333         if (ret < 0)
334                 return ret;
335         if (ret != 0) {
336                 BUG();
337         }
338         BUG_ON(ret != 0);
339         l = path->nodes[0];
340         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
341         refs = btrfs_extent_refs(l, item);
342         btrfs_set_extent_refs(l, item, refs + 1);
343         btrfs_mark_buffer_dirty(path->nodes[0]);
344
345         btrfs_release_path(root->fs_info->extent_root, path);
346         btrfs_free_path(path);
347         finish_current_insert(trans, root->fs_info->extent_root);
348         del_pending_extents(trans, root->fs_info->extent_root);
349         return 0;
350 }
351
352 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
353                          struct btrfs_root *root)
354 {
355         finish_current_insert(trans, root->fs_info->extent_root);
356         del_pending_extents(trans, root->fs_info->extent_root);
357         return 0;
358 }
359
360 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
361                              struct btrfs_root *root, u64 bytenr,
362                              u64 num_bytes, u32 *refs)
363 {
364         struct btrfs_path *path;
365         int ret;
366         struct btrfs_key key;
367         struct extent_buffer *l;
368         struct btrfs_extent_item *item;
369
370         WARN_ON(num_bytes < root->sectorsize);
371         path = btrfs_alloc_path();
372         key.objectid = bytenr;
373         key.offset = num_bytes;
374         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
375         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
376                                 0, 0);
377         if (ret < 0)
378                 goto out;
379         if (ret != 0) {
380                 btrfs_print_leaf(root, path->nodes[0]);
381                 printk("failed to find block number %Lu\n", bytenr);
382                 BUG();
383         }
384         l = path->nodes[0];
385         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
386         *refs = btrfs_extent_refs(l, item);
387 out:
388         btrfs_free_path(path);
389         return 0;
390 }
391
392 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
393                        struct btrfs_root *root)
394 {
395         return btrfs_inc_extent_ref(trans, root, root->node->start,
396                                     root->node->len);
397 }
398
399 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
400                   struct extent_buffer *buf)
401 {
402         u64 bytenr;
403         u32 nritems;
404         struct btrfs_key key;
405         struct btrfs_file_extent_item *fi;
406         int i;
407         int level;
408         int ret;
409         int faili;
410         int err;
411
412         if (!root->ref_cows)
413                 return 0;
414
415         level = btrfs_header_level(buf);
416         nritems = btrfs_header_nritems(buf);
417         for (i = 0; i < nritems; i++) {
418                 if (level == 0) {
419                         u64 disk_bytenr;
420                         btrfs_item_key_to_cpu(buf, &key, i);
421                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
422                                 continue;
423                         fi = btrfs_item_ptr(buf, i,
424                                             struct btrfs_file_extent_item);
425                         if (btrfs_file_extent_type(buf, fi) ==
426                             BTRFS_FILE_EXTENT_INLINE)
427                                 continue;
428                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
429                         if (disk_bytenr == 0)
430                                 continue;
431                         ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
432                                     btrfs_file_extent_disk_num_bytes(buf, fi));
433                         if (ret) {
434                                 faili = i;
435                                 goto fail;
436                         }
437                 } else {
438                         bytenr = btrfs_node_blockptr(buf, i);
439                         ret = btrfs_inc_extent_ref(trans, root, bytenr,
440                                            btrfs_level_size(root, level - 1));
441                         if (ret) {
442                                 faili = i;
443                                 goto fail;
444                         }
445                 }
446         }
447         return 0;
448 fail:
449         WARN_ON(1);
450         for (i =0; i < faili; i++) {
451                 if (level == 0) {
452                         u64 disk_bytenr;
453                         btrfs_item_key_to_cpu(buf, &key, i);
454                         if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
455                                 continue;
456                         fi = btrfs_item_ptr(buf, i,
457                                             struct btrfs_file_extent_item);
458                         if (btrfs_file_extent_type(buf, fi) ==
459                             BTRFS_FILE_EXTENT_INLINE)
460                                 continue;
461                         disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
462                         if (disk_bytenr == 0)
463                                 continue;
464                         err = btrfs_free_extent(trans, root, disk_bytenr,
465                                     btrfs_file_extent_disk_num_bytes(buf,
466                                                                       fi), 0);
467                         BUG_ON(err);
468                 } else {
469                         bytenr = btrfs_node_blockptr(buf, i);
470                         err = btrfs_free_extent(trans, root, bytenr,
471                                         btrfs_level_size(root, level - 1), 0);
472                         BUG_ON(err);
473                 }
474         }
475         return ret;
476 }
477
478 static int write_one_cache_group(struct btrfs_trans_handle *trans,
479                                  struct btrfs_root *root,
480                                  struct btrfs_path *path,
481                                  struct btrfs_block_group_cache *cache)
482 {
483         int ret;
484         int pending_ret;
485         struct btrfs_root *extent_root = root->fs_info->extent_root;
486         unsigned long bi;
487         struct extent_buffer *leaf;
488
489         ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
490         if (ret < 0)
491                 goto fail;
492         BUG_ON(ret);
493
494         leaf = path->nodes[0];
495         bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
496         write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
497         btrfs_mark_buffer_dirty(leaf);
498         btrfs_release_path(extent_root, path);
499 fail:
500         finish_current_insert(trans, extent_root);
501         pending_ret = del_pending_extents(trans, extent_root);
502         if (ret)
503                 return ret;
504         if (pending_ret)
505                 return pending_ret;
506         return 0;
507
508 }
509
510 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
511                                    struct btrfs_root *root)
512 {
513         struct extent_map_tree *block_group_cache;
514         struct btrfs_block_group_cache *cache;
515         int ret;
516         int err = 0;
517         int werr = 0;
518         struct btrfs_path *path;
519         u64 last = 0;
520         u64 start;
521         u64 end;
522         u64 ptr;
523
524         block_group_cache = &root->fs_info->block_group_cache;
525         path = btrfs_alloc_path();
526         if (!path)
527                 return -ENOMEM;
528
529         while(1) {
530                 ret = find_first_extent_bit(block_group_cache, last,
531                                             &start, &end, BLOCK_GROUP_DIRTY);
532                 if (ret)
533                         break;
534
535                 last = end + 1;
536                 ret = get_state_private(block_group_cache, start, &ptr);
537                 if (ret)
538                         break;
539
540                 cache = (struct btrfs_block_group_cache *)ptr;
541                 err = write_one_cache_group(trans, root,
542                                             path, cache);
543                 /*
544                  * if we fail to write the cache group, we want
545                  * to keep it marked dirty in hopes that a later
546                  * write will work
547                  */
548                 if (err) {
549                         werr = err;
550                         continue;
551                 }
552                 clear_extent_bits(block_group_cache, start, end,
553                                   BLOCK_GROUP_DIRTY, GFP_NOFS);
554         }
555         btrfs_free_path(path);
556         return werr;
557 }
558
559 static int update_block_group(struct btrfs_trans_handle *trans,
560                               struct btrfs_root *root,
561                               u64 bytenr, u64 num_bytes, int alloc,
562                               int mark_free, int data)
563 {
564         struct btrfs_block_group_cache *cache;
565         struct btrfs_fs_info *info = root->fs_info;
566         u64 total = num_bytes;
567         u64 old_val;
568         u64 byte_in_group;
569         u64 start;
570         u64 end;
571
572         while(total) {
573                 cache = btrfs_lookup_block_group(info, bytenr);
574                 if (!cache) {
575                         return -1;
576                 }
577                 byte_in_group = bytenr - cache->key.objectid;
578                 WARN_ON(byte_in_group > cache->key.offset);
579                 start = cache->key.objectid;
580                 end = start + cache->key.offset - 1;
581                 set_extent_bits(&info->block_group_cache, start, end,
582                                 BLOCK_GROUP_DIRTY, GFP_NOFS);
583
584                 old_val = btrfs_block_group_used(&cache->item);
585                 num_bytes = min(total, cache->key.offset - byte_in_group);
586                 if (alloc) {
587                         if (cache->data != data &&
588                             old_val < (cache->key.offset >> 1)) {
589                                 int bit_to_clear;
590                                 int bit_to_set;
591
592                                 cache->data = data;
593                                 if (data) {
594                                         bit_to_clear = BLOCK_GROUP_DATA;
595                                         bit_to_set = BLOCK_GROUP_METADATA;
596                                         cache->item.flags |=
597                                                 BTRFS_BLOCK_GROUP_DATA;
598                                 } else {
599                                         bit_to_clear = BLOCK_GROUP_METADATA;
600                                         bit_to_set = BLOCK_GROUP_DATA;
601                                         cache->item.flags &=
602                                                 ~BTRFS_BLOCK_GROUP_DATA;
603                                 }
604                                 clear_extent_bits(&info->block_group_cache,
605                                                   start, end, bit_to_clear,
606                                                   GFP_NOFS);
607                                 set_extent_bits(&info->block_group_cache,
608                                                 start, end, bit_to_set,
609                                                 GFP_NOFS);
610                         }
611                         old_val += num_bytes;
612                 } else {
613                         old_val -= num_bytes;
614                         if (mark_free) {
615                                 set_extent_dirty(&info->free_space_cache,
616                                                  bytenr, bytenr + num_bytes - 1,
617                                                  GFP_NOFS);
618                         }
619                 }
620                 btrfs_set_block_group_used(&cache->item, old_val);
621                 total -= num_bytes;
622                 bytenr += num_bytes;
623         }
624         return 0;
625 }
626
627 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy)
628 {
629         u64 last = 0;
630         u64 start;
631         u64 end;
632         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
633         int ret;
634
635         while(1) {
636                 ret = find_first_extent_bit(pinned_extents, last,
637                                             &start, &end, EXTENT_DIRTY);
638                 if (ret)
639                         break;
640                 set_extent_dirty(copy, start, end, GFP_NOFS);
641                 last = end + 1;
642         }
643         return 0;
644 }
645
646 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
647                                struct btrfs_root *root,
648                                struct extent_map_tree *unpin)
649 {
650         u64 start;
651         u64 end;
652         int ret;
653         struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
654         struct extent_map_tree *free_space_cache;
655
656         free_space_cache = &root->fs_info->free_space_cache;
657
658         while(1) {
659                 ret = find_first_extent_bit(unpin, 0, &start, &end,
660                                             EXTENT_DIRTY);
661                 if (ret)
662                         break;
663
664                 clear_extent_dirty(pinned_extents, start, end,
665                                    GFP_NOFS);
666                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
667                 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
668         }
669         return 0;
670 }
671
672 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
673                                  btrfs_root *extent_root)
674 {
675         struct btrfs_key ins;
676         struct btrfs_extent_item extent_item;
677         int ret;
678         int err = 0;
679         u64 start;
680         u64 end;
681         struct btrfs_fs_info *info = extent_root->fs_info;
682
683         btrfs_set_stack_extent_refs(&extent_item, 1);
684         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
685         btrfs_set_stack_extent_owner(&extent_item,
686                                      extent_root->root_key.objectid);
687
688         while(1) {
689                 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
690                                             &end, EXTENT_LOCKED);
691                 if (ret)
692                         break;
693
694                 ins.objectid = start;
695                 ins.offset = end + 1 - start;
696                 err = btrfs_insert_item(trans, extent_root, &ins,
697                                         &extent_item, sizeof(extent_item));
698                 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
699                                   GFP_NOFS);
700         }
701         return 0;
702 }
703
704 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
705                           int pending)
706 {
707         int err = 0;
708         struct extent_buffer *buf;
709
710         if (!pending) {
711                 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
712                 if (buf) {
713                         if (btrfs_buffer_uptodate(buf)) {
714                                 u64 transid =
715                                     root->fs_info->running_transaction->transid;
716                                 if (btrfs_header_generation(buf) == transid) {
717                                         free_extent_buffer(buf);
718                                         return 0;
719                                 }
720                         }
721                         free_extent_buffer(buf);
722                 }
723                 set_extent_dirty(&root->fs_info->pinned_extents,
724                                  bytenr, bytenr + num_bytes - 1, GFP_NOFS);
725         } else {
726                 set_extent_bits(&root->fs_info->pending_del,
727                                 bytenr, bytenr + num_bytes - 1,
728                                 EXTENT_LOCKED, GFP_NOFS);
729         }
730         BUG_ON(err < 0);
731         return 0;
732 }
733
734 /*
735  * remove an extent from the root, returns 0 on success
736  */
737 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
738                          *root, u64 bytenr, u64 num_bytes, int pin,
739                          int mark_free)
740 {
741         struct btrfs_path *path;
742         struct btrfs_key key;
743         struct btrfs_fs_info *info = root->fs_info;
744         struct btrfs_root *extent_root = info->extent_root;
745         struct extent_buffer *leaf;
746         int ret;
747         struct btrfs_extent_item *ei;
748         u32 refs;
749
750         key.objectid = bytenr;
751         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
752         key.offset = num_bytes;
753
754         path = btrfs_alloc_path();
755         if (!path)
756                 return -ENOMEM;
757
758         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
759         if (ret < 0)
760                 return ret;
761         BUG_ON(ret);
762
763         leaf = path->nodes[0];
764         ei = btrfs_item_ptr(leaf, path->slots[0],
765                             struct btrfs_extent_item);
766         refs = btrfs_extent_refs(leaf, ei);
767         BUG_ON(refs == 0);
768         refs -= 1;
769         btrfs_set_extent_refs(leaf, ei, refs);
770         btrfs_mark_buffer_dirty(leaf);
771
772         if (refs == 0) {
773                 u64 super_used;
774                 u64 root_used;
775
776                 if (pin) {
777                         ret = pin_down_bytes(root, bytenr, num_bytes, 0);
778                         BUG_ON(ret);
779                 }
780
781                 /* block accounting for super block */
782                 super_used = btrfs_super_bytes_used(&info->super_copy);
783                 btrfs_set_super_bytes_used(&info->super_copy,
784                                            super_used - num_bytes);
785
786                 /* block accounting for root item */
787                 root_used = btrfs_root_used(&root->root_item);
788                 btrfs_set_root_used(&root->root_item,
789                                            root_used - num_bytes);
790
791                 ret = btrfs_del_item(trans, extent_root, path);
792                 if (ret) {
793                         return ret;
794                 }
795                 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
796                                          mark_free, 0);
797                 BUG_ON(ret);
798         }
799         btrfs_free_path(path);
800         finish_current_insert(trans, extent_root);
801         return ret;
802 }
803
804 /*
805  * find all the blocks marked as pending in the radix tree and remove
806  * them from the extent map
807  */
808 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
809                                btrfs_root *extent_root)
810 {
811         int ret;
812         int err = 0;
813         u64 start;
814         u64 end;
815         struct extent_map_tree *pending_del;
816         struct extent_map_tree *pinned_extents;
817
818         pending_del = &extent_root->fs_info->pending_del;
819         pinned_extents = &extent_root->fs_info->pinned_extents;
820
821         while(1) {
822                 ret = find_first_extent_bit(pending_del, 0, &start, &end,
823                                             EXTENT_LOCKED);
824                 if (ret)
825                         break;
826
827                 set_extent_dirty(pinned_extents, start, end, GFP_NOFS);
828                 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
829                                   GFP_NOFS);
830                 ret = __free_extent(trans, extent_root,
831                                      start, end + 1 - start, 0, 0);
832                 if (ret)
833                         err = ret;
834         }
835         return err;
836 }
837
838 /*
839  * remove an extent from the root, returns 0 on success
840  */
841 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
842                       *root, u64 bytenr, u64 num_bytes, int pin)
843 {
844         struct btrfs_root *extent_root = root->fs_info->extent_root;
845         int pending_ret;
846         int ret;
847
848         WARN_ON(num_bytes < root->sectorsize);
849         if (root == extent_root) {
850                 pin_down_bytes(root, bytenr, num_bytes, 1);
851                 return 0;
852         }
853         ret = __free_extent(trans, root, bytenr, num_bytes, pin, pin == 0);
854         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
855         return ret ? ret : pending_ret;
856 }
857
858 /*
859  * walks the btree of allocated extents and find a hole of a given size.
860  * The key ins is changed to record the hole:
861  * ins->objectid == block start
862  * ins->flags = BTRFS_EXTENT_ITEM_KEY
863  * ins->offset == number of blocks
864  * Any available blocks before search_start are skipped.
865  */
866 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
867                             *orig_root, u64 num_bytes, u64 empty_size,
868                             u64 search_start, u64 search_end, u64 hint_byte,
869                             struct btrfs_key *ins, u64 exclude_start,
870                             u64 exclude_nr, int data)
871 {
872         struct btrfs_path *path;
873         struct btrfs_key key;
874         int ret;
875         u64 hole_size = 0;
876         int slot = 0;
877         u64 last_byte = 0;
878         u64 orig_search_start = search_start;
879         int start_found;
880         struct extent_buffer *l;
881         struct btrfs_root * root = orig_root->fs_info->extent_root;
882         struct btrfs_fs_info *info = root->fs_info;
883         u64 total_needed = num_bytes;
884         int level;
885         struct btrfs_block_group_cache *block_group;
886         int full_scan = 0;
887         int wrapped = 0;
888
889         WARN_ON(num_bytes < root->sectorsize);
890         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
891
892         level = btrfs_header_level(root->node);
893
894         if (search_end == (u64)-1)
895                 search_end = btrfs_super_total_bytes(&info->super_copy);
896         if (hint_byte) {
897                 block_group = btrfs_lookup_block_group(info, hint_byte);
898                 block_group = btrfs_find_block_group(root, block_group,
899                                                      hint_byte, data, 1);
900         } else {
901                 block_group = btrfs_find_block_group(root,
902                                                      trans->block_group, 0,
903                                                      data, 1);
904         }
905
906         total_needed += empty_size;
907         path = btrfs_alloc_path();
908
909 check_failed:
910         search_start = find_search_start(root, &block_group,
911                                          search_start, total_needed, data);
912         btrfs_init_path(path);
913         ins->objectid = search_start;
914         ins->offset = 0;
915         start_found = 0;
916         path->reada = 2;
917
918         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
919         if (ret < 0)
920                 goto error;
921
922         if (path->slots[0] > 0) {
923                 path->slots[0]--;
924         }
925
926         l = path->nodes[0];
927         btrfs_item_key_to_cpu(l, &key, path->slots[0]);
928
929         /*
930          * a rare case, go back one key if we hit a block group item
931          * instead of an extent item
932          */
933         if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
934             key.objectid + key.offset >= search_start) {
935                 ins->objectid = key.objectid;
936                 ins->offset = key.offset - 1;
937                 btrfs_release_path(root, path);
938                 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
939                 if (ret < 0)
940                         goto error;
941
942                 if (path->slots[0] > 0) {
943                         path->slots[0]--;
944                 }
945         }
946
947         while (1) {
948                 l = path->nodes[0];
949                 slot = path->slots[0];
950                 if (slot >= btrfs_header_nritems(l)) {
951                         ret = btrfs_next_leaf(root, path);
952                         if (ret == 0)
953                                 continue;
954                         if (ret < 0)
955                                 goto error;
956
957                         search_start = max(search_start,
958                                            block_group->key.objectid);
959                         if (!start_found) {
960                                 ins->objectid = search_start;
961                                 ins->offset = search_end - search_start;
962                                 start_found = 1;
963                                 goto check_pending;
964                         }
965                         ins->objectid = last_byte > search_start ?
966                                         last_byte : search_start;
967                         ins->offset = search_end - ins->objectid;
968                         BUG_ON(ins->objectid >= search_end);
969                         goto check_pending;
970                 }
971                 btrfs_item_key_to_cpu(l, &key, slot);
972
973                 if (key.objectid >= search_start && key.objectid > last_byte &&
974                     start_found) {
975                         if (last_byte < search_start)
976                                 last_byte = search_start;
977                         hole_size = key.objectid - last_byte;
978                         if (hole_size >= num_bytes) {
979                                 ins->objectid = last_byte;
980                                 ins->offset = hole_size;
981                                 goto check_pending;
982                         }
983                 }
984                 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
985                         if (!start_found) {
986                                 last_byte = key.objectid;
987                                 start_found = 1;
988                         }
989                         goto next;
990                 }
991
992
993                 start_found = 1;
994                 last_byte = key.objectid + key.offset;
995
996                 if (!full_scan && last_byte >= block_group->key.objectid +
997                     block_group->key.offset) {
998                         btrfs_release_path(root, path);
999                         search_start = block_group->key.objectid +
1000                                 block_group->key.offset;
1001                         goto new_group;
1002                 }
1003 next:
1004                 path->slots[0]++;
1005                 cond_resched();
1006         }
1007 check_pending:
1008         /* we have to make sure we didn't find an extent that has already
1009          * been allocated by the map tree or the original allocation
1010          */
1011         btrfs_release_path(root, path);
1012         BUG_ON(ins->objectid < search_start);
1013
1014         if (ins->objectid + num_bytes >= search_end)
1015                 goto enospc;
1016
1017         if (!full_scan && ins->objectid + num_bytes >= block_group->
1018             key.objectid + block_group->key.offset) {
1019                 search_start = block_group->key.objectid +
1020                         block_group->key.offset;
1021                 goto new_group;
1022         }
1023         if (test_range_bit(&info->extent_ins, ins->objectid,
1024                            ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1025                 search_start = ins->objectid + num_bytes;
1026                 goto new_group;
1027         }
1028         if (test_range_bit(&info->pinned_extents, ins->objectid,
1029                            ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1030                 search_start = ins->objectid + num_bytes;
1031                 goto new_group;
1032         }
1033         if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1034             ins->objectid < exclude_start + exclude_nr)) {
1035                 search_start = exclude_start + exclude_nr;
1036                 goto new_group;
1037         }
1038         if (!data) {
1039                 block_group = btrfs_lookup_block_group(info, ins->objectid);
1040                 if (block_group)
1041                         trans->block_group = block_group;
1042         }
1043         ins->offset = num_bytes;
1044         btrfs_free_path(path);
1045         return 0;
1046
1047 new_group:
1048         if (search_start + num_bytes >= search_end) {
1049 enospc:
1050                 search_start = orig_search_start;
1051                 if (full_scan) {
1052                         ret = -ENOSPC;
1053                         goto error;
1054                 }
1055                 if (wrapped) {
1056                         if (!full_scan)
1057                                 total_needed -= empty_size;
1058                         full_scan = 1;
1059                 } else
1060                         wrapped = 1;
1061         }
1062         block_group = btrfs_lookup_block_group(info, search_start);
1063         cond_resched();
1064         if (!full_scan)
1065                 block_group = btrfs_find_block_group(root, block_group,
1066                                                      search_start, data, 0);
1067         goto check_failed;
1068
1069 error:
1070         btrfs_release_path(root, path);
1071         btrfs_free_path(path);
1072         return ret;
1073 }
1074 /*
1075  * finds a free extent and does all the dirty work required for allocation
1076  * returns the key for the extent through ins, and a tree buffer for
1077  * the first block of the extent through buf.
1078  *
1079  * returns 0 if everything worked, non-zero otherwise.
1080  */
1081 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1082                        struct btrfs_root *root, u64 owner,
1083                        u64 num_bytes, u64 empty_size, u64 hint_byte,
1084                        u64 search_end, struct btrfs_key *ins, int data)
1085 {
1086         int ret;
1087         int pending_ret;
1088         u64 super_used, root_used;
1089         u64 search_start = 0;
1090         struct btrfs_fs_info *info = root->fs_info;
1091         struct btrfs_root *extent_root = info->extent_root;
1092         struct btrfs_extent_item extent_item;
1093
1094         btrfs_set_stack_extent_refs(&extent_item, 1);
1095         btrfs_set_stack_extent_owner(&extent_item, owner);
1096
1097         WARN_ON(num_bytes < root->sectorsize);
1098         ret = find_free_extent(trans, root, num_bytes, empty_size,
1099                                search_start, search_end, hint_byte, ins,
1100                                trans->alloc_exclude_start,
1101                                trans->alloc_exclude_nr, data);
1102         BUG_ON(ret);
1103         if (ret)
1104                 return ret;
1105
1106         /* block accounting for super block */
1107         super_used = btrfs_super_bytes_used(&info->super_copy);
1108         btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1109
1110         /* block accounting for root item */
1111         root_used = btrfs_root_used(&root->root_item);
1112         btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1113
1114         clear_extent_dirty(&root->fs_info->free_space_cache,
1115                            ins->objectid, ins->objectid + ins->offset - 1,
1116                            GFP_NOFS);
1117
1118         if (root == extent_root) {
1119                 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1120                                 ins->objectid + ins->offset - 1,
1121                                 EXTENT_LOCKED, GFP_NOFS);
1122                 WARN_ON(data == 1);
1123                 goto update_block;
1124         }
1125
1126         WARN_ON(trans->alloc_exclude_nr);
1127         trans->alloc_exclude_start = ins->objectid;
1128         trans->alloc_exclude_nr = ins->offset;
1129         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1130                                 sizeof(extent_item));
1131
1132         trans->alloc_exclude_start = 0;
1133         trans->alloc_exclude_nr = 0;
1134
1135         BUG_ON(ret);
1136         finish_current_insert(trans, extent_root);
1137         pending_ret = del_pending_extents(trans, extent_root);
1138
1139         if (ret) {
1140                 return ret;
1141         }
1142         if (pending_ret) {
1143                 return pending_ret;
1144         }
1145
1146 update_block:
1147         ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1148                                  data);
1149         BUG_ON(ret);
1150         return 0;
1151 }
1152
1153 /*
1154  * helper function to allocate a block for a given tree
1155  * returns the tree buffer or NULL.
1156  */
1157 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1158                                              struct btrfs_root *root,
1159                                              u32 blocksize, u64 hint,
1160                                              u64 empty_size)
1161 {
1162         struct btrfs_key ins;
1163         int ret;
1164         struct extent_buffer *buf;
1165
1166         ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1167                                  blocksize, empty_size, hint,
1168                                  (u64)-1, &ins, 0);
1169         if (ret) {
1170                 BUG_ON(ret > 0);
1171                 return ERR_PTR(ret);
1172         }
1173         buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1174         if (!buf) {
1175                 btrfs_free_extent(trans, root, ins.objectid, blocksize, 0);
1176                 return ERR_PTR(-ENOMEM);
1177         }
1178         btrfs_set_buffer_uptodate(buf);
1179         set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1180                          buf->start + buf->len - 1, GFP_NOFS);
1181         btrfs_set_buffer_defrag(buf);
1182         trans->blocks_used++;
1183         return buf;
1184 }
1185
1186 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1187                          struct btrfs_root *root, struct extent_buffer *leaf)
1188 {
1189         struct btrfs_key key;
1190         struct btrfs_file_extent_item *fi;
1191         int i;
1192         int nritems;
1193         int ret;
1194
1195         BUG_ON(!btrfs_is_leaf(leaf));
1196         nritems = btrfs_header_nritems(leaf);
1197         for (i = 0; i < nritems; i++) {
1198                 u64 disk_bytenr;
1199
1200                 btrfs_item_key_to_cpu(leaf, &key, i);
1201                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1202                         continue;
1203                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1204                 if (btrfs_file_extent_type(leaf, fi) ==
1205                     BTRFS_FILE_EXTENT_INLINE)
1206                         continue;
1207                 /*
1208                  * FIXME make sure to insert a trans record that
1209                  * repeats the snapshot del on crash
1210                  */
1211                 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1212                 if (disk_bytenr == 0)
1213                         continue;
1214                 ret = btrfs_free_extent(trans, root, disk_bytenr,
1215                                 btrfs_file_extent_disk_num_bytes(leaf, fi), 0);
1216                 BUG_ON(ret);
1217         }
1218         return 0;
1219 }
1220
1221 static void reada_walk_down(struct btrfs_root *root,
1222                             struct extent_buffer *node)
1223 {
1224         int i;
1225         u32 nritems;
1226         u64 bytenr;
1227         int ret;
1228         u32 refs;
1229         int level;
1230         u32 blocksize;
1231
1232         nritems = btrfs_header_nritems(node);
1233         level = btrfs_header_level(node);
1234         for (i = 0; i < nritems; i++) {
1235                 bytenr = btrfs_node_blockptr(node, i);
1236                 blocksize = btrfs_level_size(root, level - 1);
1237                 ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs);
1238                 BUG_ON(ret);
1239                 if (refs != 1)
1240                         continue;
1241                 mutex_unlock(&root->fs_info->fs_mutex);
1242                 ret = readahead_tree_block(root, bytenr, blocksize);
1243                 cond_resched();
1244                 mutex_lock(&root->fs_info->fs_mutex);
1245                 if (ret)
1246                         break;
1247         }
1248 }
1249
1250 /*
1251  * helper function for drop_snapshot, this walks down the tree dropping ref
1252  * counts as it goes.
1253  */
1254 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1255                           *root, struct btrfs_path *path, int *level)
1256 {
1257         struct extent_buffer *next;
1258         struct extent_buffer *cur;
1259         u64 bytenr;
1260         u32 blocksize;
1261         int ret;
1262         u32 refs;
1263
1264         WARN_ON(*level < 0);
1265         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1266         ret = lookup_extent_ref(trans, root,
1267                                 path->nodes[*level]->start,
1268                                 path->nodes[*level]->len, &refs);
1269         BUG_ON(ret);
1270         if (refs > 1)
1271                 goto out;
1272
1273         /*
1274          * walk down to the last node level and free all the leaves
1275          */
1276         while(*level >= 0) {
1277                 WARN_ON(*level < 0);
1278                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1279                 cur = path->nodes[*level];
1280
1281                 if (*level > 0 && path->slots[*level] == 0)
1282                         reada_walk_down(root, cur);
1283
1284                 if (btrfs_header_level(cur) != *level)
1285                         WARN_ON(1);
1286
1287                 if (path->slots[*level] >=
1288                     btrfs_header_nritems(cur))
1289                         break;
1290                 if (*level == 0) {
1291                         ret = drop_leaf_ref(trans, root, cur);
1292                         BUG_ON(ret);
1293                         break;
1294                 }
1295                 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1296                 blocksize = btrfs_level_size(root, *level - 1);
1297                 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1298                 BUG_ON(ret);
1299                 if (refs != 1) {
1300                         path->slots[*level]++;
1301                         ret = btrfs_free_extent(trans, root, bytenr,
1302                                                 blocksize, 1);
1303                         BUG_ON(ret);
1304                         continue;
1305                 }
1306                 next = btrfs_find_tree_block(root, bytenr, blocksize);
1307                 if (!next || !btrfs_buffer_uptodate(next)) {
1308                         free_extent_buffer(next);
1309                         mutex_unlock(&root->fs_info->fs_mutex);
1310                         next = read_tree_block(root, bytenr, blocksize);
1311                         mutex_lock(&root->fs_info->fs_mutex);
1312
1313                         /* we dropped the lock, check one more time */
1314                         ret = lookup_extent_ref(trans, root, bytenr,
1315                                                 blocksize, &refs);
1316                         BUG_ON(ret);
1317                         if (refs != 1) {
1318                                 path->slots[*level]++;
1319                                 free_extent_buffer(next);
1320                                 ret = btrfs_free_extent(trans, root,
1321                                                         bytenr, blocksize, 1);
1322                                 BUG_ON(ret);
1323                                 continue;
1324                         }
1325                 }
1326                 WARN_ON(*level <= 0);
1327                 if (path->nodes[*level-1])
1328                         free_extent_buffer(path->nodes[*level-1]);
1329                 path->nodes[*level-1] = next;
1330                 *level = btrfs_header_level(next);
1331                 path->slots[*level] = 0;
1332         }
1333 out:
1334         WARN_ON(*level < 0);
1335         WARN_ON(*level >= BTRFS_MAX_LEVEL);
1336         ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1337                                 path->nodes[*level]->len, 1);
1338         free_extent_buffer(path->nodes[*level]);
1339         path->nodes[*level] = NULL;
1340         *level += 1;
1341         BUG_ON(ret);
1342         return 0;
1343 }
1344
1345 /*
1346  * helper for dropping snapshots.  This walks back up the tree in the path
1347  * to find the first node higher up where we haven't yet gone through
1348  * all the slots
1349  */
1350 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1351                         *root, struct btrfs_path *path, int *level)
1352 {
1353         int i;
1354         int slot;
1355         int ret;
1356         struct btrfs_root_item *root_item = &root->root_item;
1357
1358         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1359                 slot = path->slots[i];
1360                 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1361                         struct extent_buffer *node;
1362                         struct btrfs_disk_key disk_key;
1363                         node = path->nodes[i];
1364                         path->slots[i]++;
1365                         *level = i;
1366                         WARN_ON(*level == 0);
1367                         btrfs_node_key(node, &disk_key, path->slots[i]);
1368                         memcpy(&root_item->drop_progress,
1369                                &disk_key, sizeof(disk_key));
1370                         root_item->drop_level = i;
1371                         return 0;
1372                 } else {
1373                         ret = btrfs_free_extent(trans, root,
1374                                                 path->nodes[*level]->start,
1375                                                 path->nodes[*level]->len, 1);
1376                         BUG_ON(ret);
1377                         free_extent_buffer(path->nodes[*level]);
1378                         path->nodes[*level] = NULL;
1379                         *level = i + 1;
1380                 }
1381         }
1382         return 1;
1383 }
1384
1385 /*
1386  * drop the reference count on the tree rooted at 'snap'.  This traverses
1387  * the tree freeing any blocks that have a ref count of zero after being
1388  * decremented.
1389  */
1390 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1391                         *root)
1392 {
1393         int ret = 0;
1394         int wret;
1395         int level;
1396         struct btrfs_path *path;
1397         int i;
1398         int orig_level;
1399         struct btrfs_root_item *root_item = &root->root_item;
1400
1401         path = btrfs_alloc_path();
1402         BUG_ON(!path);
1403
1404         level = btrfs_header_level(root->node);
1405         orig_level = level;
1406         if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1407                 path->nodes[level] = root->node;
1408                 extent_buffer_get(root->node);
1409                 path->slots[level] = 0;
1410         } else {
1411                 struct btrfs_key key;
1412                 struct btrfs_disk_key found_key;
1413                 struct extent_buffer *node;
1414
1415                 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1416                 level = root_item->drop_level;
1417                 path->lowest_level = level;
1418                 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1419                 if (wret < 0) {
1420                         ret = wret;
1421                         goto out;
1422                 }
1423                 node = path->nodes[level];
1424                 btrfs_node_key(node, &found_key, path->slots[level]);
1425                 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1426                                sizeof(found_key)));
1427         }
1428         while(1) {
1429                 wret = walk_down_tree(trans, root, path, &level);
1430                 if (wret > 0)
1431                         break;
1432                 if (wret < 0)
1433                         ret = wret;
1434
1435                 wret = walk_up_tree(trans, root, path, &level);
1436                 if (wret > 0)
1437                         break;
1438                 if (wret < 0)
1439                         ret = wret;
1440                 ret = -EAGAIN;
1441                 break;
1442         }
1443         for (i = 0; i <= orig_level; i++) {
1444                 if (path->nodes[i]) {
1445                         free_extent_buffer(path->nodes[i]);
1446                         path->nodes[i] = NULL;
1447                 }
1448         }
1449 out:
1450         btrfs_free_path(path);
1451         return ret;
1452 }
1453
1454 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1455 {
1456         u64 start;
1457         u64 end;
1458         int ret;
1459
1460         while(1) {
1461                 ret = find_first_extent_bit(&info->block_group_cache, 0,
1462                                             &start, &end, (unsigned int)-1);
1463                 if (ret)
1464                         break;
1465                 clear_extent_bits(&info->block_group_cache, start,
1466                                   end, (unsigned int)-1, GFP_NOFS);
1467         }
1468         while(1) {
1469                 ret = find_first_extent_bit(&info->free_space_cache, 0,
1470                                             &start, &end, EXTENT_DIRTY);
1471                 if (ret)
1472                         break;
1473                 clear_extent_dirty(&info->free_space_cache, start,
1474                                    end, GFP_NOFS);
1475         }
1476         return 0;
1477 }
1478
1479 int btrfs_read_block_groups(struct btrfs_root *root)
1480 {
1481         struct btrfs_path *path;
1482         int ret;
1483         int err = 0;
1484         int bit;
1485         struct btrfs_block_group_cache *cache;
1486         struct btrfs_fs_info *info = root->fs_info;
1487         struct extent_map_tree *block_group_cache;
1488         struct btrfs_key key;
1489         struct btrfs_key found_key;
1490         struct extent_buffer *leaf;
1491
1492         block_group_cache = &info->block_group_cache;
1493
1494         root = info->extent_root;
1495         key.objectid = 0;
1496         key.offset = BTRFS_BLOCK_GROUP_SIZE;
1497         btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1498
1499         path = btrfs_alloc_path();
1500         if (!path)
1501                 return -ENOMEM;
1502
1503         while(1) {
1504                 ret = btrfs_search_slot(NULL, info->extent_root,
1505                                         &key, path, 0, 0);
1506                 if (ret != 0) {
1507                         err = ret;
1508                         break;
1509                 }
1510                 leaf = path->nodes[0];
1511                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1512                 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1513                 if (!cache) {
1514                         err = -1;
1515                         break;
1516                 }
1517
1518                 read_extent_buffer(leaf, &cache->item,
1519                                    btrfs_item_ptr_offset(leaf, path->slots[0]),
1520                                    sizeof(cache->item));
1521                 memcpy(&cache->key, &found_key, sizeof(found_key));
1522                 cache->cached = 0;
1523
1524                 key.objectid = found_key.objectid + found_key.offset;
1525                 btrfs_release_path(root, path);
1526
1527                 if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
1528                         bit = BLOCK_GROUP_DATA;
1529                         cache->data = 1;
1530                 } else {
1531                         bit = BLOCK_GROUP_METADATA;
1532                         cache->data = 0;
1533                 }
1534
1535                 /* use EXTENT_LOCKED to prevent merging */
1536                 set_extent_bits(block_group_cache, found_key.objectid,
1537                                 found_key.objectid + found_key.offset - 1,
1538                                 bit | EXTENT_LOCKED, GFP_NOFS);
1539                 set_state_private(block_group_cache, found_key.objectid,
1540                                   (u64)cache);
1541
1542                 if (key.objectid >=
1543                     btrfs_super_total_bytes(&info->super_copy))
1544                         break;
1545         }
1546
1547         btrfs_free_path(path);
1548         return 0;
1549 }