]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/extent-tree.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: Add support for duplicate blocks on a single spindle
[mirror_ubuntu-artful-kernel.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 <linux/crc32c.h>
21 #include <linux/pagemap.h>
22 #include "hash.h"
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "print-tree.h"
26 #include "transaction.h"
27 #include "volumes.h"
28
29 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
30 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
31 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
32
33 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
34
35 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
36 btrfs_root *extent_root);
37 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
38 btrfs_root *extent_root);
39 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
40 struct btrfs_root *root, u64 bytes_used,
41 u64 type, u64 chunk_tree, u64 chunk_objectid,
42 u64 size);
43
44
45 static int cache_block_group(struct btrfs_root *root,
46 struct btrfs_block_group_cache *block_group)
47 {
48 struct btrfs_path *path;
49 int ret;
50 struct btrfs_key key;
51 struct extent_buffer *leaf;
52 struct extent_io_tree *free_space_cache;
53 int slot;
54 u64 last = 0;
55 u64 hole_size;
56 u64 first_free;
57 int found = 0;
58
59 if (!block_group)
60 return 0;
61
62 root = root->fs_info->extent_root;
63 free_space_cache = &root->fs_info->free_space_cache;
64
65 if (block_group->cached)
66 return 0;
67
68 path = btrfs_alloc_path();
69 if (!path)
70 return -ENOMEM;
71
72 path->reada = 2;
73 first_free = block_group->key.objectid;
74 key.objectid = block_group->key.objectid;
75 key.offset = 0;
76 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
77 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
78 if (ret < 0)
79 return ret;
80 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
81 if (ret < 0)
82 return ret;
83 if (ret == 0) {
84 leaf = path->nodes[0];
85 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
86 if (key.objectid + key.offset > first_free)
87 first_free = key.objectid + key.offset;
88 }
89 while(1) {
90 leaf = path->nodes[0];
91 slot = path->slots[0];
92 if (slot >= btrfs_header_nritems(leaf)) {
93 ret = btrfs_next_leaf(root, path);
94 if (ret < 0)
95 goto err;
96 if (ret == 0) {
97 continue;
98 } else {
99 break;
100 }
101 }
102 btrfs_item_key_to_cpu(leaf, &key, slot);
103 if (key.objectid < block_group->key.objectid) {
104 goto next;
105 }
106 if (key.objectid >= block_group->key.objectid +
107 block_group->key.offset) {
108 break;
109 }
110
111 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
112 if (!found) {
113 last = first_free;
114 found = 1;
115 }
116 if (key.objectid > last) {
117 hole_size = key.objectid - last;
118 set_extent_dirty(free_space_cache, last,
119 last + hole_size - 1,
120 GFP_NOFS);
121 }
122 last = key.objectid + key.offset;
123 }
124 next:
125 path->slots[0]++;
126 }
127
128 if (!found)
129 last = first_free;
130 if (block_group->key.objectid +
131 block_group->key.offset > last) {
132 hole_size = block_group->key.objectid +
133 block_group->key.offset - last;
134 set_extent_dirty(free_space_cache, last,
135 last + hole_size - 1, GFP_NOFS);
136 }
137 block_group->cached = 1;
138 err:
139 btrfs_free_path(path);
140 return 0;
141 }
142
143 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
144 btrfs_fs_info *info,
145 u64 bytenr)
146 {
147 struct extent_io_tree *block_group_cache;
148 struct btrfs_block_group_cache *block_group = NULL;
149 u64 ptr;
150 u64 start;
151 u64 end;
152 int ret;
153
154 block_group_cache = &info->block_group_cache;
155 ret = find_first_extent_bit(block_group_cache,
156 bytenr, &start, &end,
157 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
158 BLOCK_GROUP_SYSTEM);
159 if (ret) {
160 return NULL;
161 }
162 ret = get_state_private(block_group_cache, start, &ptr);
163 if (ret)
164 return NULL;
165
166 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
167 if (block_group->key.objectid <= bytenr && bytenr <
168 block_group->key.objectid + block_group->key.offset)
169 return block_group;
170 return NULL;
171 }
172
173 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
174 {
175 return (cache->flags & bits) == bits;
176 }
177
178 static int noinline find_search_start(struct btrfs_root *root,
179 struct btrfs_block_group_cache **cache_ret,
180 u64 *start_ret, int num, int data)
181 {
182 int ret;
183 struct btrfs_block_group_cache *cache = *cache_ret;
184 struct extent_io_tree *free_space_cache;
185 struct extent_state *state;
186 u64 last;
187 u64 start = 0;
188 u64 cache_miss = 0;
189 u64 total_fs_bytes;
190 u64 search_start = *start_ret;
191 int wrapped = 0;
192
193 if (!cache)
194 goto out;
195 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
196 free_space_cache = &root->fs_info->free_space_cache;
197
198 again:
199 ret = cache_block_group(root, cache);
200 if (ret)
201 goto out;
202
203 last = max(search_start, cache->key.objectid);
204 if (!block_group_bits(cache, data)) {
205 goto new_group;
206 }
207
208 spin_lock_irq(&free_space_cache->lock);
209 state = find_first_extent_bit_state(free_space_cache, last, EXTENT_DIRTY);
210 while(1) {
211 if (!state) {
212 if (!cache_miss)
213 cache_miss = last;
214 spin_unlock_irq(&free_space_cache->lock);
215 goto new_group;
216 }
217
218 start = max(last, state->start);
219 last = state->end + 1;
220 if (last - start < num) {
221 if (last == cache->key.objectid + cache->key.offset)
222 cache_miss = start;
223 do {
224 state = extent_state_next(state);
225 } while(state && !(state->state & EXTENT_DIRTY));
226 continue;
227 }
228 spin_unlock_irq(&free_space_cache->lock);
229 if (start + num > cache->key.objectid + cache->key.offset)
230 goto new_group;
231 if (start + num > total_fs_bytes)
232 goto new_group;
233 if (!block_group_bits(cache, data)) {
234 printk("block group bits don't match %Lu %d\n", cache->flags, data);
235 }
236 *start_ret = start;
237 return 0;
238 }
239 out:
240 cache = btrfs_lookup_block_group(root->fs_info, search_start);
241 if (!cache) {
242 printk("Unable to find block group for %Lu\n", search_start);
243 WARN_ON(1);
244 }
245 return -ENOSPC;
246
247 new_group:
248 last = cache->key.objectid + cache->key.offset;
249 wrapped:
250 cache = btrfs_lookup_block_group(root->fs_info, last);
251 if (!cache || cache->key.objectid >= total_fs_bytes) {
252 no_cache:
253 if (!wrapped) {
254 wrapped = 1;
255 last = search_start;
256 goto wrapped;
257 }
258 goto out;
259 }
260 if (cache_miss && !cache->cached) {
261 cache_block_group(root, cache);
262 last = cache_miss;
263 cache = btrfs_lookup_block_group(root->fs_info, last);
264 }
265 cache = btrfs_find_block_group(root, cache, last, data, 0);
266 if (!cache)
267 goto no_cache;
268 *cache_ret = cache;
269 cache_miss = 0;
270 goto again;
271 }
272
273 static u64 div_factor(u64 num, int factor)
274 {
275 if (factor == 10)
276 return num;
277 num *= factor;
278 do_div(num, 10);
279 return num;
280 }
281
282 static int block_group_state_bits(u64 flags)
283 {
284 int bits = 0;
285 if (flags & BTRFS_BLOCK_GROUP_DATA)
286 bits |= BLOCK_GROUP_DATA;
287 if (flags & BTRFS_BLOCK_GROUP_METADATA)
288 bits |= BLOCK_GROUP_METADATA;
289 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
290 bits |= BLOCK_GROUP_SYSTEM;
291 return bits;
292 }
293
294 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
295 struct btrfs_block_group_cache
296 *hint, u64 search_start,
297 int data, int owner)
298 {
299 struct btrfs_block_group_cache *cache;
300 struct extent_io_tree *block_group_cache;
301 struct btrfs_block_group_cache *found_group = NULL;
302 struct btrfs_fs_info *info = root->fs_info;
303 u64 used;
304 u64 last = 0;
305 u64 hint_last;
306 u64 start;
307 u64 end;
308 u64 free_check;
309 u64 ptr;
310 u64 total_fs_bytes;
311 int bit;
312 int ret;
313 int full_search = 0;
314 int factor = 8;
315
316 block_group_cache = &info->block_group_cache;
317 total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
318
319 if (!owner)
320 factor = 8;
321
322 bit = block_group_state_bits(data);
323
324 if (search_start && search_start < total_fs_bytes) {
325 struct btrfs_block_group_cache *shint;
326 shint = btrfs_lookup_block_group(info, search_start);
327 if (shint && block_group_bits(shint, data)) {
328 used = btrfs_block_group_used(&shint->item);
329 if (used + shint->pinned <
330 div_factor(shint->key.offset, factor)) {
331 return shint;
332 }
333 }
334 }
335 if (hint && block_group_bits(hint, data) &&
336 hint->key.objectid < total_fs_bytes) {
337 used = btrfs_block_group_used(&hint->item);
338 if (used + hint->pinned <
339 div_factor(hint->key.offset, factor)) {
340 return hint;
341 }
342 last = hint->key.objectid + hint->key.offset;
343 hint_last = last;
344 } else {
345 if (hint)
346 hint_last = max(hint->key.objectid, search_start);
347 else
348 hint_last = search_start;
349
350 if (hint_last >= total_fs_bytes)
351 hint_last = search_start;
352 last = hint_last;
353 }
354 again:
355 while(1) {
356 ret = find_first_extent_bit(block_group_cache, last,
357 &start, &end, bit);
358 if (ret)
359 break;
360
361 ret = get_state_private(block_group_cache, start, &ptr);
362 if (ret)
363 break;
364
365 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
366 last = cache->key.objectid + cache->key.offset;
367 used = btrfs_block_group_used(&cache->item);
368
369 if (cache->key.objectid > total_fs_bytes)
370 break;
371
372 if (block_group_bits(cache, data)) {
373 if (full_search)
374 free_check = cache->key.offset;
375 else
376 free_check = div_factor(cache->key.offset,
377 factor);
378
379 if (used + cache->pinned < free_check) {
380 found_group = cache;
381 goto found;
382 }
383 }
384 cond_resched();
385 }
386 if (!full_search) {
387 last = search_start;
388 full_search = 1;
389 goto again;
390 }
391 found:
392 return found_group;
393 }
394
395 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
396 u64 owner, u64 owner_offset)
397 {
398 u32 high_crc = ~(u32)0;
399 u32 low_crc = ~(u32)0;
400 __le64 lenum;
401
402 lenum = cpu_to_le64(root_objectid);
403 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
404 lenum = cpu_to_le64(ref_generation);
405 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
406 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
407 lenum = cpu_to_le64(owner);
408 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
409 lenum = cpu_to_le64(owner_offset);
410 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
411 }
412 return ((u64)high_crc << 32) | (u64)low_crc;
413 }
414
415 static int match_extent_ref(struct extent_buffer *leaf,
416 struct btrfs_extent_ref *disk_ref,
417 struct btrfs_extent_ref *cpu_ref)
418 {
419 int ret;
420 int len;
421
422 if (cpu_ref->objectid)
423 len = sizeof(*cpu_ref);
424 else
425 len = 2 * sizeof(u64);
426 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
427 len);
428 return ret == 0;
429 }
430
431 static int noinline lookup_extent_backref(struct btrfs_trans_handle *trans,
432 struct btrfs_root *root,
433 struct btrfs_path *path, u64 bytenr,
434 u64 root_objectid,
435 u64 ref_generation, u64 owner,
436 u64 owner_offset, int del)
437 {
438 u64 hash;
439 struct btrfs_key key;
440 struct btrfs_key found_key;
441 struct btrfs_extent_ref ref;
442 struct extent_buffer *leaf;
443 struct btrfs_extent_ref *disk_ref;
444 int ret;
445 int ret2;
446
447 btrfs_set_stack_ref_root(&ref, root_objectid);
448 btrfs_set_stack_ref_generation(&ref, ref_generation);
449 btrfs_set_stack_ref_objectid(&ref, owner);
450 btrfs_set_stack_ref_offset(&ref, owner_offset);
451
452 hash = hash_extent_ref(root_objectid, ref_generation, owner,
453 owner_offset);
454 key.offset = hash;
455 key.objectid = bytenr;
456 key.type = BTRFS_EXTENT_REF_KEY;
457
458 while (1) {
459 ret = btrfs_search_slot(trans, root, &key, path,
460 del ? -1 : 0, del);
461 if (ret < 0)
462 goto out;
463 leaf = path->nodes[0];
464 if (ret != 0) {
465 u32 nritems = btrfs_header_nritems(leaf);
466 if (path->slots[0] >= nritems) {
467 ret2 = btrfs_next_leaf(root, path);
468 if (ret2)
469 goto out;
470 leaf = path->nodes[0];
471 }
472 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
473 if (found_key.objectid != bytenr ||
474 found_key.type != BTRFS_EXTENT_REF_KEY)
475 goto out;
476 key.offset = found_key.offset;
477 if (del) {
478 btrfs_release_path(root, path);
479 continue;
480 }
481 }
482 disk_ref = btrfs_item_ptr(path->nodes[0],
483 path->slots[0],
484 struct btrfs_extent_ref);
485 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
486 ret = 0;
487 goto out;
488 }
489 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
490 key.offset = found_key.offset + 1;
491 btrfs_release_path(root, path);
492 }
493 out:
494 return ret;
495 }
496
497 /*
498 * Back reference rules. Back refs have three main goals:
499 *
500 * 1) differentiate between all holders of references to an extent so that
501 * when a reference is dropped we can make sure it was a valid reference
502 * before freeing the extent.
503 *
504 * 2) Provide enough information to quickly find the holders of an extent
505 * if we notice a given block is corrupted or bad.
506 *
507 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
508 * maintenance. This is actually the same as #2, but with a slightly
509 * different use case.
510 *
511 * File extents can be referenced by:
512 *
513 * - multiple snapshots, subvolumes, or different generations in one subvol
514 * - different files inside a single subvolume (in theory, not implemented yet)
515 * - different offsets inside a file (bookend extents in file.c)
516 *
517 * The extent ref structure has fields for:
518 *
519 * - Objectid of the subvolume root
520 * - Generation number of the tree holding the reference
521 * - objectid of the file holding the reference
522 * - offset in the file corresponding to the key holding the reference
523 *
524 * When a file extent is allocated the fields are filled in:
525 * (root_key.objectid, trans->transid, inode objectid, offset in file)
526 *
527 * When a leaf is cow'd new references are added for every file extent found
528 * in the leaf. It looks the same as the create case, but trans->transid
529 * will be different when the block is cow'd.
530 *
531 * (root_key.objectid, trans->transid, inode objectid, offset in file)
532 *
533 * When a file extent is removed either during snapshot deletion or file
534 * truncation, the corresponding back reference is found
535 * by searching for:
536 *
537 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
538 * inode objectid, offset in file)
539 *
540 * Btree extents can be referenced by:
541 *
542 * - Different subvolumes
543 * - Different generations of the same subvolume
544 *
545 * Storing sufficient information for a full reverse mapping of a btree
546 * block would require storing the lowest key of the block in the backref,
547 * and it would require updating that lowest key either before write out or
548 * every time it changed. Instead, the objectid of the lowest key is stored
549 * along with the level of the tree block. This provides a hint
550 * about where in the btree the block can be found. Searches through the
551 * btree only need to look for a pointer to that block, so they stop one
552 * level higher than the level recorded in the backref.
553 *
554 * Some btrees do not do reference counting on their extents. These
555 * include the extent tree and the tree of tree roots. Backrefs for these
556 * trees always have a generation of zero.
557 *
558 * When a tree block is created, back references are inserted:
559 *
560 * (root->root_key.objectid, trans->transid or zero, level, lowest_key_objectid)
561 *
562 * When a tree block is cow'd in a reference counted root,
563 * new back references are added for all the blocks it points to.
564 * These are of the form (trans->transid will have increased since creation):
565 *
566 * (root->root_key.objectid, trans->transid, level, lowest_key_objectid)
567 *
568 * Because the lowest_key_objectid and the level are just hints
569 * they are not used when backrefs are deleted. When a backref is deleted:
570 *
571 * if backref was for a tree root:
572 * root_objectid = root->root_key.objectid
573 * else
574 * root_objectid = btrfs_header_owner(parent)
575 *
576 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
577 *
578 * Back Reference Key hashing:
579 *
580 * Back references have four fields, each 64 bits long. Unfortunately,
581 * This is hashed into a single 64 bit number and placed into the key offset.
582 * The key objectid corresponds to the first byte in the extent, and the
583 * key type is set to BTRFS_EXTENT_REF_KEY
584 */
585 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
586 struct btrfs_root *root,
587 struct btrfs_path *path, u64 bytenr,
588 u64 root_objectid, u64 ref_generation,
589 u64 owner, u64 owner_offset)
590 {
591 u64 hash;
592 struct btrfs_key key;
593 struct btrfs_extent_ref ref;
594 struct btrfs_extent_ref *disk_ref;
595 int ret;
596
597 btrfs_set_stack_ref_root(&ref, root_objectid);
598 btrfs_set_stack_ref_generation(&ref, ref_generation);
599 btrfs_set_stack_ref_objectid(&ref, owner);
600 btrfs_set_stack_ref_offset(&ref, owner_offset);
601
602 hash = hash_extent_ref(root_objectid, ref_generation, owner,
603 owner_offset);
604 key.offset = hash;
605 key.objectid = bytenr;
606 key.type = BTRFS_EXTENT_REF_KEY;
607
608 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
609 while (ret == -EEXIST) {
610 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
611 struct btrfs_extent_ref);
612 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
613 goto out;
614 key.offset++;
615 btrfs_release_path(root, path);
616 ret = btrfs_insert_empty_item(trans, root, path, &key,
617 sizeof(ref));
618 }
619 if (ret)
620 goto out;
621 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
622 struct btrfs_extent_ref);
623 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
624 sizeof(ref));
625 btrfs_mark_buffer_dirty(path->nodes[0]);
626 out:
627 btrfs_release_path(root, path);
628 return ret;
629 }
630
631 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
632 struct btrfs_root *root,
633 u64 bytenr, u64 num_bytes,
634 u64 root_objectid, u64 ref_generation,
635 u64 owner, u64 owner_offset)
636 {
637 struct btrfs_path *path;
638 int ret;
639 struct btrfs_key key;
640 struct extent_buffer *l;
641 struct btrfs_extent_item *item;
642 u32 refs;
643
644 WARN_ON(num_bytes < root->sectorsize);
645 path = btrfs_alloc_path();
646 if (!path)
647 return -ENOMEM;
648
649 path->reada = 0;
650 key.objectid = bytenr;
651 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
652 key.offset = num_bytes;
653 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
654 0, 1);
655 if (ret < 0)
656 return ret;
657 if (ret != 0) {
658 BUG();
659 }
660 BUG_ON(ret != 0);
661 l = path->nodes[0];
662 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
663 refs = btrfs_extent_refs(l, item);
664 btrfs_set_extent_refs(l, item, refs + 1);
665 btrfs_mark_buffer_dirty(path->nodes[0]);
666
667 btrfs_release_path(root->fs_info->extent_root, path);
668
669 path->reada = 0;
670 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
671 path, bytenr, root_objectid,
672 ref_generation, owner, owner_offset);
673 BUG_ON(ret);
674 finish_current_insert(trans, root->fs_info->extent_root);
675 del_pending_extents(trans, root->fs_info->extent_root);
676
677 btrfs_free_path(path);
678 return 0;
679 }
680
681 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
682 struct btrfs_root *root)
683 {
684 finish_current_insert(trans, root->fs_info->extent_root);
685 del_pending_extents(trans, root->fs_info->extent_root);
686 return 0;
687 }
688
689 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
690 struct btrfs_root *root, u64 bytenr,
691 u64 num_bytes, u32 *refs)
692 {
693 struct btrfs_path *path;
694 int ret;
695 struct btrfs_key key;
696 struct extent_buffer *l;
697 struct btrfs_extent_item *item;
698
699 WARN_ON(num_bytes < root->sectorsize);
700 path = btrfs_alloc_path();
701 path->reada = 0;
702 key.objectid = bytenr;
703 key.offset = num_bytes;
704 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
705 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
706 0, 0);
707 if (ret < 0)
708 goto out;
709 if (ret != 0) {
710 btrfs_print_leaf(root, path->nodes[0]);
711 printk("failed to find block number %Lu\n", bytenr);
712 BUG();
713 }
714 l = path->nodes[0];
715 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
716 *refs = btrfs_extent_refs(l, item);
717 out:
718 btrfs_free_path(path);
719 return 0;
720 }
721
722 u32 btrfs_count_snapshots_in_path(struct btrfs_root *root,
723 struct btrfs_path *count_path,
724 u64 first_extent)
725 {
726 struct btrfs_root *extent_root = root->fs_info->extent_root;
727 struct btrfs_path *path;
728 u64 bytenr;
729 u64 found_objectid;
730 u64 root_objectid = root->root_key.objectid;
731 u32 total_count = 0;
732 u32 cur_count;
733 u32 nritems;
734 int ret;
735 struct btrfs_key key;
736 struct btrfs_key found_key;
737 struct extent_buffer *l;
738 struct btrfs_extent_item *item;
739 struct btrfs_extent_ref *ref_item;
740 int level = -1;
741
742 path = btrfs_alloc_path();
743 again:
744 if (level == -1)
745 bytenr = first_extent;
746 else
747 bytenr = count_path->nodes[level]->start;
748
749 cur_count = 0;
750 key.objectid = bytenr;
751 key.offset = 0;
752
753 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
754 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
755 if (ret < 0)
756 goto out;
757 BUG_ON(ret == 0);
758
759 l = path->nodes[0];
760 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
761
762 if (found_key.objectid != bytenr ||
763 found_key.type != BTRFS_EXTENT_ITEM_KEY) {
764 goto out;
765 }
766
767 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
768 while (1) {
769 l = path->nodes[0];
770 nritems = btrfs_header_nritems(l);
771 if (path->slots[0] >= nritems) {
772 ret = btrfs_next_leaf(extent_root, path);
773 if (ret == 0)
774 continue;
775 break;
776 }
777 btrfs_item_key_to_cpu(l, &found_key, path->slots[0]);
778 if (found_key.objectid != bytenr)
779 break;
780
781 if (found_key.type != BTRFS_EXTENT_REF_KEY) {
782 path->slots[0]++;
783 continue;
784 }
785
786 cur_count++;
787 ref_item = btrfs_item_ptr(l, path->slots[0],
788 struct btrfs_extent_ref);
789 found_objectid = btrfs_ref_root(l, ref_item);
790
791 if (found_objectid != root_objectid) {
792 total_count = 2;
793 goto out;
794 }
795 total_count = 1;
796 path->slots[0]++;
797 }
798 if (cur_count == 0) {
799 total_count = 0;
800 goto out;
801 }
802 if (level >= 0 && root->node == count_path->nodes[level])
803 goto out;
804 level++;
805 btrfs_release_path(root, path);
806 goto again;
807
808 out:
809 btrfs_free_path(path);
810 return total_count;
811 }
812 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
813 struct btrfs_root *root, u64 owner_objectid)
814 {
815 u64 generation;
816 u64 key_objectid;
817 u64 level;
818 u32 nritems;
819 struct btrfs_disk_key disk_key;
820
821 level = btrfs_header_level(root->node);
822 generation = trans->transid;
823 nritems = btrfs_header_nritems(root->node);
824 if (nritems > 0) {
825 if (level == 0)
826 btrfs_item_key(root->node, &disk_key, 0);
827 else
828 btrfs_node_key(root->node, &disk_key, 0);
829 key_objectid = btrfs_disk_key_objectid(&disk_key);
830 } else {
831 key_objectid = 0;
832 }
833 return btrfs_inc_extent_ref(trans, root, root->node->start,
834 root->node->len, owner_objectid,
835 generation, level, key_objectid);
836 }
837
838 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
839 struct extent_buffer *buf)
840 {
841 u64 bytenr;
842 u32 nritems;
843 struct btrfs_key key;
844 struct btrfs_file_extent_item *fi;
845 int i;
846 int level;
847 int ret;
848 int faili;
849
850 if (!root->ref_cows)
851 return 0;
852
853 level = btrfs_header_level(buf);
854 nritems = btrfs_header_nritems(buf);
855 for (i = 0; i < nritems; i++) {
856 if (level == 0) {
857 u64 disk_bytenr;
858 btrfs_item_key_to_cpu(buf, &key, i);
859 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
860 continue;
861 fi = btrfs_item_ptr(buf, i,
862 struct btrfs_file_extent_item);
863 if (btrfs_file_extent_type(buf, fi) ==
864 BTRFS_FILE_EXTENT_INLINE)
865 continue;
866 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
867 if (disk_bytenr == 0)
868 continue;
869 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
870 btrfs_file_extent_disk_num_bytes(buf, fi),
871 root->root_key.objectid, trans->transid,
872 key.objectid, key.offset);
873 if (ret) {
874 faili = i;
875 goto fail;
876 }
877 } else {
878 bytenr = btrfs_node_blockptr(buf, i);
879 btrfs_node_key_to_cpu(buf, &key, i);
880 ret = btrfs_inc_extent_ref(trans, root, bytenr,
881 btrfs_level_size(root, level - 1),
882 root->root_key.objectid,
883 trans->transid,
884 level - 1, key.objectid);
885 if (ret) {
886 faili = i;
887 goto fail;
888 }
889 }
890 }
891 return 0;
892 fail:
893 WARN_ON(1);
894 #if 0
895 for (i =0; i < faili; i++) {
896 if (level == 0) {
897 u64 disk_bytenr;
898 btrfs_item_key_to_cpu(buf, &key, i);
899 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
900 continue;
901 fi = btrfs_item_ptr(buf, i,
902 struct btrfs_file_extent_item);
903 if (btrfs_file_extent_type(buf, fi) ==
904 BTRFS_FILE_EXTENT_INLINE)
905 continue;
906 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
907 if (disk_bytenr == 0)
908 continue;
909 err = btrfs_free_extent(trans, root, disk_bytenr,
910 btrfs_file_extent_disk_num_bytes(buf,
911 fi), 0);
912 BUG_ON(err);
913 } else {
914 bytenr = btrfs_node_blockptr(buf, i);
915 err = btrfs_free_extent(trans, root, bytenr,
916 btrfs_level_size(root, level - 1), 0);
917 BUG_ON(err);
918 }
919 }
920 #endif
921 return ret;
922 }
923
924 static int write_one_cache_group(struct btrfs_trans_handle *trans,
925 struct btrfs_root *root,
926 struct btrfs_path *path,
927 struct btrfs_block_group_cache *cache)
928 {
929 int ret;
930 int pending_ret;
931 struct btrfs_root *extent_root = root->fs_info->extent_root;
932 unsigned long bi;
933 struct extent_buffer *leaf;
934
935 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
936 if (ret < 0)
937 goto fail;
938 BUG_ON(ret);
939
940 leaf = path->nodes[0];
941 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
942 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
943 btrfs_mark_buffer_dirty(leaf);
944 btrfs_release_path(extent_root, path);
945 fail:
946 finish_current_insert(trans, extent_root);
947 pending_ret = del_pending_extents(trans, extent_root);
948 if (ret)
949 return ret;
950 if (pending_ret)
951 return pending_ret;
952 return 0;
953
954 }
955
956 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
957 struct btrfs_root *root)
958 {
959 struct extent_io_tree *block_group_cache;
960 struct btrfs_block_group_cache *cache;
961 int ret;
962 int err = 0;
963 int werr = 0;
964 struct btrfs_path *path;
965 u64 last = 0;
966 u64 start;
967 u64 end;
968 u64 ptr;
969
970 block_group_cache = &root->fs_info->block_group_cache;
971 path = btrfs_alloc_path();
972 if (!path)
973 return -ENOMEM;
974
975 while(1) {
976 ret = find_first_extent_bit(block_group_cache, last,
977 &start, &end, BLOCK_GROUP_DIRTY);
978 if (ret)
979 break;
980
981 last = end + 1;
982 ret = get_state_private(block_group_cache, start, &ptr);
983 if (ret)
984 break;
985
986 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
987 err = write_one_cache_group(trans, root,
988 path, cache);
989 /*
990 * if we fail to write the cache group, we want
991 * to keep it marked dirty in hopes that a later
992 * write will work
993 */
994 if (err) {
995 werr = err;
996 continue;
997 }
998 clear_extent_bits(block_group_cache, start, end,
999 BLOCK_GROUP_DIRTY, GFP_NOFS);
1000 }
1001 btrfs_free_path(path);
1002 return werr;
1003 }
1004
1005 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1006 u64 flags)
1007 {
1008 struct list_head *head = &info->space_info;
1009 struct list_head *cur;
1010 struct btrfs_space_info *found;
1011 list_for_each(cur, head) {
1012 found = list_entry(cur, struct btrfs_space_info, list);
1013 if (found->flags == flags)
1014 return found;
1015 }
1016 return NULL;
1017
1018 }
1019
1020 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1021 u64 total_bytes, u64 bytes_used,
1022 struct btrfs_space_info **space_info)
1023 {
1024 struct btrfs_space_info *found;
1025
1026 found = __find_space_info(info, flags);
1027 if (found) {
1028 found->total_bytes += total_bytes;
1029 found->bytes_used += bytes_used;
1030 WARN_ON(found->total_bytes < found->bytes_used);
1031 *space_info = found;
1032 return 0;
1033 }
1034 found = kmalloc(sizeof(*found), GFP_NOFS);
1035 if (!found)
1036 return -ENOMEM;
1037
1038 list_add(&found->list, &info->space_info);
1039 found->flags = flags;
1040 found->total_bytes = total_bytes;
1041 found->bytes_used = bytes_used;
1042 found->bytes_pinned = 0;
1043 found->full = 0;
1044 *space_info = found;
1045 return 0;
1046 }
1047
1048 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1049 {
1050 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1051 BTRFS_BLOCK_GROUP_RAID1 |
1052 BTRFS_BLOCK_GROUP_DUP);
1053 if (extra_flags) {
1054 if (flags & BTRFS_BLOCK_GROUP_DATA)
1055 fs_info->avail_data_alloc_bits |= extra_flags;
1056 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1057 fs_info->avail_metadata_alloc_bits |= extra_flags;
1058 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1059 fs_info->avail_system_alloc_bits |= extra_flags;
1060 }
1061 }
1062
1063 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1064 struct btrfs_root *extent_root, u64 alloc_bytes,
1065 u64 flags)
1066 {
1067 struct btrfs_space_info *space_info;
1068 u64 thresh;
1069 u64 start;
1070 u64 num_bytes;
1071 int ret;
1072
1073 space_info = __find_space_info(extent_root->fs_info, flags);
1074 if (!space_info) {
1075 ret = update_space_info(extent_root->fs_info, flags,
1076 0, 0, &space_info);
1077 BUG_ON(ret);
1078 }
1079 BUG_ON(!space_info);
1080
1081 if (space_info->full)
1082 return 0;
1083
1084 thresh = div_factor(space_info->total_bytes, 6);
1085 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1086 thresh)
1087 return 0;
1088
1089 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1090 if (ret == -ENOSPC) {
1091 printk("space info full %Lu\n", flags);
1092 space_info->full = 1;
1093 return 0;
1094 }
1095
1096 BUG_ON(ret);
1097
1098 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1099 extent_root->fs_info->chunk_root->root_key.objectid,
1100 start, num_bytes);
1101 BUG_ON(ret);
1102
1103 set_avail_alloc_bits(extent_root->fs_info, flags);
1104 return 0;
1105 }
1106
1107 static int update_block_group(struct btrfs_trans_handle *trans,
1108 struct btrfs_root *root,
1109 u64 bytenr, u64 num_bytes, int alloc,
1110 int mark_free)
1111 {
1112 struct btrfs_block_group_cache *cache;
1113 struct btrfs_fs_info *info = root->fs_info;
1114 u64 total = num_bytes;
1115 u64 old_val;
1116 u64 byte_in_group;
1117 u64 start;
1118 u64 end;
1119
1120 while(total) {
1121 cache = btrfs_lookup_block_group(info, bytenr);
1122 if (!cache) {
1123 return -1;
1124 }
1125 byte_in_group = bytenr - cache->key.objectid;
1126 WARN_ON(byte_in_group > cache->key.offset);
1127 start = cache->key.objectid;
1128 end = start + cache->key.offset - 1;
1129 set_extent_bits(&info->block_group_cache, start, end,
1130 BLOCK_GROUP_DIRTY, GFP_NOFS);
1131
1132 old_val = btrfs_block_group_used(&cache->item);
1133 num_bytes = min(total, cache->key.offset - byte_in_group);
1134 if (alloc) {
1135 old_val += num_bytes;
1136 cache->space_info->bytes_used += num_bytes;
1137 } else {
1138 old_val -= num_bytes;
1139 cache->space_info->bytes_used -= num_bytes;
1140 if (mark_free) {
1141 set_extent_dirty(&info->free_space_cache,
1142 bytenr, bytenr + num_bytes - 1,
1143 GFP_NOFS);
1144 }
1145 }
1146 btrfs_set_block_group_used(&cache->item, old_val);
1147 total -= num_bytes;
1148 bytenr += num_bytes;
1149 }
1150 return 0;
1151 }
1152
1153 static int update_pinned_extents(struct btrfs_root *root,
1154 u64 bytenr, u64 num, int pin)
1155 {
1156 u64 len;
1157 struct btrfs_block_group_cache *cache;
1158 struct btrfs_fs_info *fs_info = root->fs_info;
1159
1160 if (pin) {
1161 set_extent_dirty(&fs_info->pinned_extents,
1162 bytenr, bytenr + num - 1, GFP_NOFS);
1163 } else {
1164 clear_extent_dirty(&fs_info->pinned_extents,
1165 bytenr, bytenr + num - 1, GFP_NOFS);
1166 }
1167 while (num > 0) {
1168 cache = btrfs_lookup_block_group(fs_info, bytenr);
1169 WARN_ON(!cache);
1170 len = min(num, cache->key.offset -
1171 (bytenr - cache->key.objectid));
1172 if (pin) {
1173 cache->pinned += len;
1174 cache->space_info->bytes_pinned += len;
1175 fs_info->total_pinned += len;
1176 } else {
1177 cache->pinned -= len;
1178 cache->space_info->bytes_pinned -= len;
1179 fs_info->total_pinned -= len;
1180 }
1181 bytenr += len;
1182 num -= len;
1183 }
1184 return 0;
1185 }
1186
1187 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1188 {
1189 u64 last = 0;
1190 u64 start;
1191 u64 end;
1192 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1193 int ret;
1194
1195 while(1) {
1196 ret = find_first_extent_bit(pinned_extents, last,
1197 &start, &end, EXTENT_DIRTY);
1198 if (ret)
1199 break;
1200 set_extent_dirty(copy, start, end, GFP_NOFS);
1201 last = end + 1;
1202 }
1203 return 0;
1204 }
1205
1206 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1207 struct btrfs_root *root,
1208 struct extent_io_tree *unpin)
1209 {
1210 u64 start;
1211 u64 end;
1212 int ret;
1213 struct extent_io_tree *free_space_cache;
1214 free_space_cache = &root->fs_info->free_space_cache;
1215
1216 while(1) {
1217 ret = find_first_extent_bit(unpin, 0, &start, &end,
1218 EXTENT_DIRTY);
1219 if (ret)
1220 break;
1221 update_pinned_extents(root, start, end + 1 - start, 0);
1222 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1223 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1224 }
1225 return 0;
1226 }
1227
1228 static int finish_current_insert(struct btrfs_trans_handle *trans,
1229 struct btrfs_root *extent_root)
1230 {
1231 u64 start;
1232 u64 end;
1233 struct btrfs_fs_info *info = extent_root->fs_info;
1234 struct extent_buffer *eb;
1235 struct btrfs_path *path;
1236 struct btrfs_key ins;
1237 struct btrfs_disk_key first;
1238 struct btrfs_extent_item extent_item;
1239 int ret;
1240 int level;
1241 int err = 0;
1242
1243 btrfs_set_stack_extent_refs(&extent_item, 1);
1244 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1245 path = btrfs_alloc_path();
1246
1247 while(1) {
1248 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1249 &end, EXTENT_LOCKED);
1250 if (ret)
1251 break;
1252
1253 ins.objectid = start;
1254 ins.offset = end + 1 - start;
1255 err = btrfs_insert_item(trans, extent_root, &ins,
1256 &extent_item, sizeof(extent_item));
1257 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1258 GFP_NOFS);
1259 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1260 level = btrfs_header_level(eb);
1261 if (level == 0) {
1262 btrfs_item_key(eb, &first, 0);
1263 } else {
1264 btrfs_node_key(eb, &first, 0);
1265 }
1266 err = btrfs_insert_extent_backref(trans, extent_root, path,
1267 start, extent_root->root_key.objectid,
1268 0, level,
1269 btrfs_disk_key_objectid(&first));
1270 BUG_ON(err);
1271 free_extent_buffer(eb);
1272 }
1273 btrfs_free_path(path);
1274 return 0;
1275 }
1276
1277 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1278 int pending)
1279 {
1280 int err = 0;
1281 struct extent_buffer *buf;
1282
1283 if (!pending) {
1284 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1285 if (buf) {
1286 if (btrfs_buffer_uptodate(buf)) {
1287 u64 transid =
1288 root->fs_info->running_transaction->transid;
1289 u64 header_transid =
1290 btrfs_header_generation(buf);
1291 if (header_transid == transid) {
1292 clean_tree_block(NULL, root, buf);
1293 free_extent_buffer(buf);
1294 return 1;
1295 }
1296 }
1297 free_extent_buffer(buf);
1298 }
1299 update_pinned_extents(root, bytenr, num_bytes, 1);
1300 } else {
1301 set_extent_bits(&root->fs_info->pending_del,
1302 bytenr, bytenr + num_bytes - 1,
1303 EXTENT_LOCKED, GFP_NOFS);
1304 }
1305 BUG_ON(err < 0);
1306 return 0;
1307 }
1308
1309 /*
1310 * remove an extent from the root, returns 0 on success
1311 */
1312 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1313 *root, u64 bytenr, u64 num_bytes,
1314 u64 root_objectid, u64 ref_generation,
1315 u64 owner_objectid, u64 owner_offset, int pin,
1316 int mark_free)
1317 {
1318 struct btrfs_path *path;
1319 struct btrfs_key key;
1320 struct btrfs_fs_info *info = root->fs_info;
1321 struct btrfs_root *extent_root = info->extent_root;
1322 struct extent_buffer *leaf;
1323 int ret;
1324 int extent_slot = 0;
1325 int found_extent = 0;
1326 int num_to_del = 1;
1327 struct btrfs_extent_item *ei;
1328 u32 refs;
1329
1330 key.objectid = bytenr;
1331 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1332 key.offset = num_bytes;
1333 path = btrfs_alloc_path();
1334 if (!path)
1335 return -ENOMEM;
1336
1337 path->reada = 0;
1338 ret = lookup_extent_backref(trans, extent_root, path,
1339 bytenr, root_objectid,
1340 ref_generation,
1341 owner_objectid, owner_offset, 1);
1342 if (ret == 0) {
1343 struct btrfs_key found_key;
1344 extent_slot = path->slots[0];
1345 while(extent_slot > 0) {
1346 extent_slot--;
1347 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1348 extent_slot);
1349 if (found_key.objectid != bytenr)
1350 break;
1351 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1352 found_key.offset == num_bytes) {
1353 found_extent = 1;
1354 break;
1355 }
1356 if (path->slots[0] - extent_slot > 5)
1357 break;
1358 }
1359 if (!found_extent)
1360 ret = btrfs_del_item(trans, extent_root, path);
1361 } else {
1362 btrfs_print_leaf(extent_root, path->nodes[0]);
1363 WARN_ON(1);
1364 printk("Unable to find ref byte nr %Lu root %Lu "
1365 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1366 root_objectid, ref_generation, owner_objectid,
1367 owner_offset);
1368 }
1369 if (!found_extent) {
1370 btrfs_release_path(extent_root, path);
1371 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1372 if (ret < 0)
1373 return ret;
1374 BUG_ON(ret);
1375 extent_slot = path->slots[0];
1376 }
1377
1378 leaf = path->nodes[0];
1379 ei = btrfs_item_ptr(leaf, extent_slot,
1380 struct btrfs_extent_item);
1381 refs = btrfs_extent_refs(leaf, ei);
1382 BUG_ON(refs == 0);
1383 refs -= 1;
1384 btrfs_set_extent_refs(leaf, ei, refs);
1385
1386 btrfs_mark_buffer_dirty(leaf);
1387
1388 if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) {
1389 /* if the back ref and the extent are next to each other
1390 * they get deleted below in one shot
1391 */
1392 path->slots[0] = extent_slot;
1393 num_to_del = 2;
1394 } else if (found_extent) {
1395 /* otherwise delete the extent back ref */
1396 ret = btrfs_del_item(trans, extent_root, path);
1397 BUG_ON(ret);
1398 /* if refs are 0, we need to setup the path for deletion */
1399 if (refs == 0) {
1400 btrfs_release_path(extent_root, path);
1401 ret = btrfs_search_slot(trans, extent_root, &key, path,
1402 -1, 1);
1403 if (ret < 0)
1404 return ret;
1405 BUG_ON(ret);
1406 }
1407 }
1408
1409 if (refs == 0) {
1410 u64 super_used;
1411 u64 root_used;
1412
1413 if (pin) {
1414 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1415 if (ret > 0)
1416 mark_free = 1;
1417 BUG_ON(ret < 0);
1418 }
1419
1420 /* block accounting for super block */
1421 super_used = btrfs_super_bytes_used(&info->super_copy);
1422 btrfs_set_super_bytes_used(&info->super_copy,
1423 super_used - num_bytes);
1424
1425 /* block accounting for root item */
1426 root_used = btrfs_root_used(&root->root_item);
1427 btrfs_set_root_used(&root->root_item,
1428 root_used - num_bytes);
1429 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
1430 num_to_del);
1431 if (ret) {
1432 return ret;
1433 }
1434 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1435 mark_free);
1436 BUG_ON(ret);
1437 }
1438 btrfs_free_path(path);
1439 finish_current_insert(trans, extent_root);
1440 return ret;
1441 }
1442
1443 /*
1444 * find all the blocks marked as pending in the radix tree and remove
1445 * them from the extent map
1446 */
1447 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1448 btrfs_root *extent_root)
1449 {
1450 int ret;
1451 int err = 0;
1452 u64 start;
1453 u64 end;
1454 struct extent_io_tree *pending_del;
1455 struct extent_io_tree *pinned_extents;
1456
1457 pending_del = &extent_root->fs_info->pending_del;
1458 pinned_extents = &extent_root->fs_info->pinned_extents;
1459
1460 while(1) {
1461 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1462 EXTENT_LOCKED);
1463 if (ret)
1464 break;
1465 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1466 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1467 GFP_NOFS);
1468 ret = __free_extent(trans, extent_root,
1469 start, end + 1 - start,
1470 extent_root->root_key.objectid,
1471 0, 0, 0, 0, 0);
1472 if (ret)
1473 err = ret;
1474 }
1475 return err;
1476 }
1477
1478 /*
1479 * remove an extent from the root, returns 0 on success
1480 */
1481 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1482 *root, u64 bytenr, u64 num_bytes,
1483 u64 root_objectid, u64 ref_generation,
1484 u64 owner_objectid, u64 owner_offset, int pin)
1485 {
1486 struct btrfs_root *extent_root = root->fs_info->extent_root;
1487 int pending_ret;
1488 int ret;
1489
1490 WARN_ON(num_bytes < root->sectorsize);
1491 if (!root->ref_cows)
1492 ref_generation = 0;
1493
1494 if (root == extent_root) {
1495 pin_down_bytes(root, bytenr, num_bytes, 1);
1496 return 0;
1497 }
1498 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1499 ref_generation, owner_objectid, owner_offset,
1500 pin, pin == 0);
1501 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1502 return ret ? ret : pending_ret;
1503 }
1504
1505 static u64 stripe_align(struct btrfs_root *root, u64 val)
1506 {
1507 u64 mask = ((u64)root->stripesize - 1);
1508 u64 ret = (val + mask) & ~mask;
1509 return ret;
1510 }
1511
1512 /*
1513 * walks the btree of allocated extents and find a hole of a given size.
1514 * The key ins is changed to record the hole:
1515 * ins->objectid == block start
1516 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1517 * ins->offset == number of blocks
1518 * Any available blocks before search_start are skipped.
1519 */
1520 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
1521 struct btrfs_root *orig_root,
1522 u64 num_bytes, u64 empty_size,
1523 u64 search_start, u64 search_end,
1524 u64 hint_byte, struct btrfs_key *ins,
1525 u64 exclude_start, u64 exclude_nr,
1526 int data)
1527 {
1528 int ret;
1529 u64 orig_search_start = search_start;
1530 struct btrfs_root * root = orig_root->fs_info->extent_root;
1531 struct btrfs_fs_info *info = root->fs_info;
1532 u64 total_needed = num_bytes;
1533 u64 *last_ptr = NULL;
1534 struct btrfs_block_group_cache *block_group;
1535 int full_scan = 0;
1536 int wrapped = 0;
1537 int empty_cluster = 2 * 1024 * 1024;
1538
1539 WARN_ON(num_bytes < root->sectorsize);
1540 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1541
1542 if (data & BTRFS_BLOCK_GROUP_METADATA) {
1543 last_ptr = &root->fs_info->last_alloc;
1544 empty_cluster = 256 * 1024;
1545 }
1546
1547 if ((data & BTRFS_BLOCK_GROUP_DATA) && btrfs_test_opt(root, SSD)) {
1548 last_ptr = &root->fs_info->last_data_alloc;
1549 }
1550
1551 if (last_ptr) {
1552 if (*last_ptr)
1553 hint_byte = *last_ptr;
1554 else {
1555 empty_size += empty_cluster;
1556 }
1557 }
1558
1559 if (search_end == (u64)-1)
1560 search_end = btrfs_super_total_bytes(&info->super_copy);
1561
1562 if (hint_byte) {
1563 block_group = btrfs_lookup_block_group(info, hint_byte);
1564 if (!block_group)
1565 hint_byte = search_start;
1566 block_group = btrfs_find_block_group(root, block_group,
1567 hint_byte, data, 1);
1568 if (last_ptr && *last_ptr == 0 && block_group)
1569 hint_byte = block_group->key.objectid;
1570 } else {
1571 block_group = btrfs_find_block_group(root,
1572 trans->block_group,
1573 search_start, data, 1);
1574 }
1575 search_start = max(search_start, hint_byte);
1576
1577 total_needed += empty_size;
1578
1579 check_failed:
1580 if (!block_group) {
1581 block_group = btrfs_lookup_block_group(info, search_start);
1582 if (!block_group)
1583 block_group = btrfs_lookup_block_group(info,
1584 orig_search_start);
1585 }
1586 ret = find_search_start(root, &block_group, &search_start,
1587 total_needed, data);
1588 if (ret == -ENOSPC && last_ptr && *last_ptr) {
1589 *last_ptr = 0;
1590 block_group = btrfs_lookup_block_group(info,
1591 orig_search_start);
1592 search_start = orig_search_start;
1593 ret = find_search_start(root, &block_group, &search_start,
1594 total_needed, data);
1595 }
1596 if (ret == -ENOSPC)
1597 goto enospc;
1598 if (ret)
1599 goto error;
1600
1601 if (last_ptr && *last_ptr && search_start != *last_ptr) {
1602 *last_ptr = 0;
1603 if (!empty_size) {
1604 empty_size += empty_cluster;
1605 total_needed += empty_size;
1606 }
1607 block_group = btrfs_lookup_block_group(info,
1608 orig_search_start);
1609 search_start = orig_search_start;
1610 ret = find_search_start(root, &block_group,
1611 &search_start, total_needed, data);
1612 if (ret == -ENOSPC)
1613 goto enospc;
1614 if (ret)
1615 goto error;
1616 }
1617
1618 search_start = stripe_align(root, search_start);
1619 ins->objectid = search_start;
1620 ins->offset = num_bytes;
1621
1622 if (ins->objectid + num_bytes >= search_end)
1623 goto enospc;
1624
1625 if (ins->objectid + num_bytes >
1626 block_group->key.objectid + block_group->key.offset) {
1627 search_start = block_group->key.objectid +
1628 block_group->key.offset;
1629 goto new_group;
1630 }
1631
1632 if (test_range_bit(&info->extent_ins, ins->objectid,
1633 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1634 search_start = ins->objectid + num_bytes;
1635 goto new_group;
1636 }
1637
1638 if (test_range_bit(&info->pinned_extents, ins->objectid,
1639 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1640 search_start = ins->objectid + num_bytes;
1641 goto new_group;
1642 }
1643
1644 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1645 ins->objectid < exclude_start + exclude_nr)) {
1646 search_start = exclude_start + exclude_nr;
1647 goto new_group;
1648 }
1649
1650 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
1651 block_group = btrfs_lookup_block_group(info, ins->objectid);
1652 if (block_group)
1653 trans->block_group = block_group;
1654 }
1655 ins->offset = num_bytes;
1656 if (last_ptr) {
1657 *last_ptr = ins->objectid + ins->offset;
1658 if (*last_ptr ==
1659 btrfs_super_total_bytes(&root->fs_info->super_copy)) {
1660 *last_ptr = 0;
1661 }
1662 }
1663 return 0;
1664
1665 new_group:
1666 if (search_start + num_bytes >= search_end) {
1667 enospc:
1668 search_start = orig_search_start;
1669 if (full_scan) {
1670 ret = -ENOSPC;
1671 goto error;
1672 }
1673 if (wrapped) {
1674 if (!full_scan)
1675 total_needed -= empty_size;
1676 full_scan = 1;
1677 } else
1678 wrapped = 1;
1679 }
1680 block_group = btrfs_lookup_block_group(info, search_start);
1681 cond_resched();
1682 block_group = btrfs_find_block_group(root, block_group,
1683 search_start, data, 0);
1684 goto check_failed;
1685
1686 error:
1687 return ret;
1688 }
1689 /*
1690 * finds a free extent and does all the dirty work required for allocation
1691 * returns the key for the extent through ins, and a tree buffer for
1692 * the first block of the extent through buf.
1693 *
1694 * returns 0 if everything worked, non-zero otherwise.
1695 */
1696 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1697 struct btrfs_root *root,
1698 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1699 u64 owner, u64 owner_offset,
1700 u64 empty_size, u64 hint_byte,
1701 u64 search_end, struct btrfs_key *ins, int data)
1702 {
1703 int ret;
1704 int pending_ret;
1705 u64 super_used;
1706 u64 root_used;
1707 u64 search_start = 0;
1708 u64 new_hint;
1709 u64 alloc_profile;
1710 u32 sizes[2];
1711 struct btrfs_fs_info *info = root->fs_info;
1712 struct btrfs_root *extent_root = info->extent_root;
1713 struct btrfs_extent_item *extent_item;
1714 struct btrfs_extent_ref *ref;
1715 struct btrfs_path *path;
1716 struct btrfs_key keys[2];
1717
1718 if (data) {
1719 alloc_profile = info->avail_data_alloc_bits &
1720 info->data_alloc_profile;
1721 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
1722 } else if (root == root->fs_info->chunk_root) {
1723 alloc_profile = info->avail_system_alloc_bits &
1724 info->system_alloc_profile;
1725 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
1726 } else {
1727 alloc_profile = info->avail_metadata_alloc_bits &
1728 info->metadata_alloc_profile;
1729 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
1730 }
1731
1732 if (root->ref_cows) {
1733 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
1734 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1735 2 * 1024 * 1024,
1736 BTRFS_BLOCK_GROUP_METADATA |
1737 (info->metadata_alloc_profile &
1738 info->avail_metadata_alloc_bits));
1739 BUG_ON(ret);
1740 }
1741 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
1742 num_bytes + 2 * 1024 * 1024, data);
1743 BUG_ON(ret);
1744 }
1745
1746 new_hint = max(hint_byte, root->fs_info->alloc_start);
1747 if (new_hint < btrfs_super_total_bytes(&info->super_copy))
1748 hint_byte = new_hint;
1749
1750 WARN_ON(num_bytes < root->sectorsize);
1751 ret = find_free_extent(trans, root, num_bytes, empty_size,
1752 search_start, search_end, hint_byte, ins,
1753 trans->alloc_exclude_start,
1754 trans->alloc_exclude_nr, data);
1755 BUG_ON(ret);
1756 if (ret)
1757 return ret;
1758
1759 /* block accounting for super block */
1760 super_used = btrfs_super_bytes_used(&info->super_copy);
1761 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1762
1763 /* block accounting for root item */
1764 root_used = btrfs_root_used(&root->root_item);
1765 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1766
1767 clear_extent_dirty(&root->fs_info->free_space_cache,
1768 ins->objectid, ins->objectid + ins->offset - 1,
1769 GFP_NOFS);
1770
1771 if (root == extent_root) {
1772 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1773 ins->objectid + ins->offset - 1,
1774 EXTENT_LOCKED, GFP_NOFS);
1775 goto update_block;
1776 }
1777
1778 WARN_ON(trans->alloc_exclude_nr);
1779 trans->alloc_exclude_start = ins->objectid;
1780 trans->alloc_exclude_nr = ins->offset;
1781
1782 memcpy(&keys[0], ins, sizeof(*ins));
1783 keys[1].offset = hash_extent_ref(root_objectid, ref_generation,
1784 owner, owner_offset);
1785 keys[1].objectid = ins->objectid;
1786 keys[1].type = BTRFS_EXTENT_REF_KEY;
1787 sizes[0] = sizeof(*extent_item);
1788 sizes[1] = sizeof(*ref);
1789
1790 path = btrfs_alloc_path();
1791 BUG_ON(!path);
1792
1793 ret = btrfs_insert_empty_items(trans, extent_root, path, keys,
1794 sizes, 2);
1795
1796 BUG_ON(ret);
1797 extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1798 struct btrfs_extent_item);
1799 btrfs_set_extent_refs(path->nodes[0], extent_item, 1);
1800 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1801 struct btrfs_extent_ref);
1802
1803 btrfs_set_ref_root(path->nodes[0], ref, root_objectid);
1804 btrfs_set_ref_generation(path->nodes[0], ref, ref_generation);
1805 btrfs_set_ref_objectid(path->nodes[0], ref, owner);
1806 btrfs_set_ref_offset(path->nodes[0], ref, owner_offset);
1807
1808 btrfs_mark_buffer_dirty(path->nodes[0]);
1809
1810 trans->alloc_exclude_start = 0;
1811 trans->alloc_exclude_nr = 0;
1812 btrfs_free_path(path);
1813 finish_current_insert(trans, extent_root);
1814 pending_ret = del_pending_extents(trans, extent_root);
1815
1816 if (ret) {
1817 return ret;
1818 }
1819 if (pending_ret) {
1820 return pending_ret;
1821 }
1822
1823 update_block:
1824 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0);
1825 if (ret) {
1826 printk("update block group failed for %Lu %Lu\n",
1827 ins->objectid, ins->offset);
1828 BUG();
1829 }
1830 return 0;
1831 }
1832
1833 /*
1834 * helper function to allocate a block for a given tree
1835 * returns the tree buffer or NULL.
1836 */
1837 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1838 struct btrfs_root *root,
1839 u32 blocksize,
1840 u64 root_objectid, u64 hint,
1841 u64 empty_size)
1842 {
1843 u64 ref_generation;
1844
1845 if (root->ref_cows)
1846 ref_generation = trans->transid;
1847 else
1848 ref_generation = 0;
1849
1850
1851 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1852 ref_generation, 0, 0, hint, empty_size);
1853 }
1854
1855 /*
1856 * helper function to allocate a block for a given tree
1857 * returns the tree buffer or NULL.
1858 */
1859 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *root,
1861 u32 blocksize,
1862 u64 root_objectid,
1863 u64 ref_generation,
1864 u64 first_objectid,
1865 int level,
1866 u64 hint,
1867 u64 empty_size)
1868 {
1869 struct btrfs_key ins;
1870 int ret;
1871 struct extent_buffer *buf;
1872
1873 ret = btrfs_alloc_extent(trans, root, blocksize,
1874 root_objectid, ref_generation,
1875 level, first_objectid, empty_size, hint,
1876 (u64)-1, &ins, 0);
1877 if (ret) {
1878 BUG_ON(ret > 0);
1879 return ERR_PTR(ret);
1880 }
1881 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1882 if (!buf) {
1883 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1884 root->root_key.objectid, ref_generation,
1885 0, 0, 0);
1886 return ERR_PTR(-ENOMEM);
1887 }
1888 btrfs_set_header_generation(buf, trans->transid);
1889 clean_tree_block(trans, root, buf);
1890 wait_on_tree_block_writeback(root, buf);
1891 btrfs_set_buffer_uptodate(buf);
1892
1893 if (PageDirty(buf->first_page)) {
1894 printk("page %lu dirty\n", buf->first_page->index);
1895 WARN_ON(1);
1896 }
1897
1898 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1899 buf->start + buf->len - 1, GFP_NOFS);
1900 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->io_tree,
1901 buf->start, buf->start + buf->len - 1,
1902 EXTENT_CSUM, GFP_NOFS);
1903 buf->flags |= EXTENT_CSUM;
1904 if (!btrfs_test_opt(root, SSD))
1905 btrfs_set_buffer_defrag(buf);
1906 trans->blocks_used++;
1907 return buf;
1908 }
1909
1910 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
1911 struct btrfs_root *root,
1912 struct extent_buffer *leaf)
1913 {
1914 u64 leaf_owner;
1915 u64 leaf_generation;
1916 struct btrfs_key key;
1917 struct btrfs_file_extent_item *fi;
1918 int i;
1919 int nritems;
1920 int ret;
1921
1922 BUG_ON(!btrfs_is_leaf(leaf));
1923 nritems = btrfs_header_nritems(leaf);
1924 leaf_owner = btrfs_header_owner(leaf);
1925 leaf_generation = btrfs_header_generation(leaf);
1926
1927 for (i = 0; i < nritems; i++) {
1928 u64 disk_bytenr;
1929
1930 btrfs_item_key_to_cpu(leaf, &key, i);
1931 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1932 continue;
1933 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1934 if (btrfs_file_extent_type(leaf, fi) ==
1935 BTRFS_FILE_EXTENT_INLINE)
1936 continue;
1937 /*
1938 * FIXME make sure to insert a trans record that
1939 * repeats the snapshot del on crash
1940 */
1941 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1942 if (disk_bytenr == 0)
1943 continue;
1944 ret = btrfs_free_extent(trans, root, disk_bytenr,
1945 btrfs_file_extent_disk_num_bytes(leaf, fi),
1946 leaf_owner, leaf_generation,
1947 key.objectid, key.offset, 0);
1948 BUG_ON(ret);
1949 }
1950 return 0;
1951 }
1952
1953 static void noinline reada_walk_down(struct btrfs_root *root,
1954 struct extent_buffer *node,
1955 int slot)
1956 {
1957 u64 bytenr;
1958 u64 last = 0;
1959 u32 nritems;
1960 u32 refs;
1961 u32 blocksize;
1962 int ret;
1963 int i;
1964 int level;
1965 int skipped = 0;
1966
1967 nritems = btrfs_header_nritems(node);
1968 level = btrfs_header_level(node);
1969 if (level)
1970 return;
1971
1972 for (i = slot; i < nritems && skipped < 32; i++) {
1973 bytenr = btrfs_node_blockptr(node, i);
1974 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
1975 (last > bytenr && last - bytenr > 32 * 1024))) {
1976 skipped++;
1977 continue;
1978 }
1979 blocksize = btrfs_level_size(root, level - 1);
1980 if (i != slot) {
1981 ret = lookup_extent_ref(NULL, root, bytenr,
1982 blocksize, &refs);
1983 BUG_ON(ret);
1984 if (refs != 1) {
1985 skipped++;
1986 continue;
1987 }
1988 }
1989 mutex_unlock(&root->fs_info->fs_mutex);
1990 ret = readahead_tree_block(root, bytenr, blocksize);
1991 last = bytenr + blocksize;
1992 cond_resched();
1993 mutex_lock(&root->fs_info->fs_mutex);
1994 if (ret)
1995 break;
1996 }
1997 }
1998
1999 /*
2000 * helper function for drop_snapshot, this walks down the tree dropping ref
2001 * counts as it goes.
2002 */
2003 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2004 struct btrfs_root *root,
2005 struct btrfs_path *path, int *level)
2006 {
2007 u64 root_owner;
2008 u64 root_gen;
2009 u64 bytenr;
2010 struct extent_buffer *next;
2011 struct extent_buffer *cur;
2012 struct extent_buffer *parent;
2013 u32 blocksize;
2014 int ret;
2015 u32 refs;
2016
2017 WARN_ON(*level < 0);
2018 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2019 ret = lookup_extent_ref(trans, root,
2020 path->nodes[*level]->start,
2021 path->nodes[*level]->len, &refs);
2022 BUG_ON(ret);
2023 if (refs > 1)
2024 goto out;
2025
2026 /*
2027 * walk down to the last node level and free all the leaves
2028 */
2029 while(*level >= 0) {
2030 WARN_ON(*level < 0);
2031 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2032 cur = path->nodes[*level];
2033
2034 if (btrfs_header_level(cur) != *level)
2035 WARN_ON(1);
2036
2037 if (path->slots[*level] >=
2038 btrfs_header_nritems(cur))
2039 break;
2040 if (*level == 0) {
2041 ret = drop_leaf_ref(trans, root, cur);
2042 BUG_ON(ret);
2043 break;
2044 }
2045 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2046 blocksize = btrfs_level_size(root, *level - 1);
2047 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
2048 BUG_ON(ret);
2049 if (refs != 1) {
2050 parent = path->nodes[*level];
2051 root_owner = btrfs_header_owner(parent);
2052 root_gen = btrfs_header_generation(parent);
2053 path->slots[*level]++;
2054 ret = btrfs_free_extent(trans, root, bytenr,
2055 blocksize, root_owner,
2056 root_gen, 0, 0, 1);
2057 BUG_ON(ret);
2058 continue;
2059 }
2060 next = btrfs_find_tree_block(root, bytenr, blocksize);
2061 if (!next || !btrfs_buffer_uptodate(next)) {
2062 free_extent_buffer(next);
2063 reada_walk_down(root, cur, path->slots[*level]);
2064
2065 mutex_unlock(&root->fs_info->fs_mutex);
2066 next = read_tree_block(root, bytenr, blocksize);
2067 mutex_lock(&root->fs_info->fs_mutex);
2068
2069 /* we've dropped the lock, double check */
2070 ret = lookup_extent_ref(trans, root, bytenr,
2071 blocksize, &refs);
2072 BUG_ON(ret);
2073 if (refs != 1) {
2074 parent = path->nodes[*level];
2075 root_owner = btrfs_header_owner(parent);
2076 root_gen = btrfs_header_generation(parent);
2077
2078 path->slots[*level]++;
2079 free_extent_buffer(next);
2080 ret = btrfs_free_extent(trans, root, bytenr,
2081 blocksize,
2082 root_owner,
2083 root_gen, 0, 0, 1);
2084 BUG_ON(ret);
2085 continue;
2086 }
2087 } else if (next) {
2088 btrfs_verify_block_csum(root, next);
2089 }
2090 WARN_ON(*level <= 0);
2091 if (path->nodes[*level-1])
2092 free_extent_buffer(path->nodes[*level-1]);
2093 path->nodes[*level-1] = next;
2094 *level = btrfs_header_level(next);
2095 path->slots[*level] = 0;
2096 }
2097 out:
2098 WARN_ON(*level < 0);
2099 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2100
2101 if (path->nodes[*level] == root->node) {
2102 root_owner = root->root_key.objectid;
2103 parent = path->nodes[*level];
2104 } else {
2105 parent = path->nodes[*level + 1];
2106 root_owner = btrfs_header_owner(parent);
2107 }
2108
2109 root_gen = btrfs_header_generation(parent);
2110 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2111 path->nodes[*level]->len,
2112 root_owner, root_gen, 0, 0, 1);
2113 free_extent_buffer(path->nodes[*level]);
2114 path->nodes[*level] = NULL;
2115 *level += 1;
2116 BUG_ON(ret);
2117 return 0;
2118 }
2119
2120 /*
2121 * helper for dropping snapshots. This walks back up the tree in the path
2122 * to find the first node higher up where we haven't yet gone through
2123 * all the slots
2124 */
2125 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2126 struct btrfs_root *root,
2127 struct btrfs_path *path, int *level)
2128 {
2129 u64 root_owner;
2130 u64 root_gen;
2131 struct btrfs_root_item *root_item = &root->root_item;
2132 int i;
2133 int slot;
2134 int ret;
2135
2136 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2137 slot = path->slots[i];
2138 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2139 struct extent_buffer *node;
2140 struct btrfs_disk_key disk_key;
2141 node = path->nodes[i];
2142 path->slots[i]++;
2143 *level = i;
2144 WARN_ON(*level == 0);
2145 btrfs_node_key(node, &disk_key, path->slots[i]);
2146 memcpy(&root_item->drop_progress,
2147 &disk_key, sizeof(disk_key));
2148 root_item->drop_level = i;
2149 return 0;
2150 } else {
2151 if (path->nodes[*level] == root->node) {
2152 root_owner = root->root_key.objectid;
2153 root_gen =
2154 btrfs_header_generation(path->nodes[*level]);
2155 } else {
2156 struct extent_buffer *node;
2157 node = path->nodes[*level + 1];
2158 root_owner = btrfs_header_owner(node);
2159 root_gen = btrfs_header_generation(node);
2160 }
2161 ret = btrfs_free_extent(trans, root,
2162 path->nodes[*level]->start,
2163 path->nodes[*level]->len,
2164 root_owner, root_gen, 0, 0, 1);
2165 BUG_ON(ret);
2166 free_extent_buffer(path->nodes[*level]);
2167 path->nodes[*level] = NULL;
2168 *level = i + 1;
2169 }
2170 }
2171 return 1;
2172 }
2173
2174 /*
2175 * drop the reference count on the tree rooted at 'snap'. This traverses
2176 * the tree freeing any blocks that have a ref count of zero after being
2177 * decremented.
2178 */
2179 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2180 *root)
2181 {
2182 int ret = 0;
2183 int wret;
2184 int level;
2185 struct btrfs_path *path;
2186 int i;
2187 int orig_level;
2188 struct btrfs_root_item *root_item = &root->root_item;
2189
2190 path = btrfs_alloc_path();
2191 BUG_ON(!path);
2192
2193 level = btrfs_header_level(root->node);
2194 orig_level = level;
2195 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2196 path->nodes[level] = root->node;
2197 extent_buffer_get(root->node);
2198 path->slots[level] = 0;
2199 } else {
2200 struct btrfs_key key;
2201 struct btrfs_disk_key found_key;
2202 struct extent_buffer *node;
2203
2204 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2205 level = root_item->drop_level;
2206 path->lowest_level = level;
2207 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2208 if (wret < 0) {
2209 ret = wret;
2210 goto out;
2211 }
2212 node = path->nodes[level];
2213 btrfs_node_key(node, &found_key, path->slots[level]);
2214 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2215 sizeof(found_key)));
2216 }
2217 while(1) {
2218 wret = walk_down_tree(trans, root, path, &level);
2219 if (wret > 0)
2220 break;
2221 if (wret < 0)
2222 ret = wret;
2223
2224 wret = walk_up_tree(trans, root, path, &level);
2225 if (wret > 0)
2226 break;
2227 if (wret < 0)
2228 ret = wret;
2229 ret = -EAGAIN;
2230 break;
2231 }
2232 for (i = 0; i <= orig_level; i++) {
2233 if (path->nodes[i]) {
2234 free_extent_buffer(path->nodes[i]);
2235 path->nodes[i] = NULL;
2236 }
2237 }
2238 out:
2239 btrfs_free_path(path);
2240 return ret;
2241 }
2242
2243 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2244 {
2245 u64 start;
2246 u64 end;
2247 u64 ptr;
2248 int ret;
2249 while(1) {
2250 ret = find_first_extent_bit(&info->block_group_cache, 0,
2251 &start, &end, (unsigned int)-1);
2252 if (ret)
2253 break;
2254 ret = get_state_private(&info->block_group_cache, start, &ptr);
2255 if (!ret)
2256 kfree((void *)(unsigned long)ptr);
2257 clear_extent_bits(&info->block_group_cache, start,
2258 end, (unsigned int)-1, GFP_NOFS);
2259 }
2260 while(1) {
2261 ret = find_first_extent_bit(&info->free_space_cache, 0,
2262 &start, &end, EXTENT_DIRTY);
2263 if (ret)
2264 break;
2265 clear_extent_dirty(&info->free_space_cache, start,
2266 end, GFP_NOFS);
2267 }
2268 return 0;
2269 }
2270
2271 static int noinline relocate_inode_pages(struct inode *inode, u64 start,
2272 u64 len)
2273 {
2274 u64 page_start;
2275 u64 page_end;
2276 u64 delalloc_start;
2277 u64 existing_delalloc;
2278 unsigned long last_index;
2279 unsigned long i;
2280 struct page *page;
2281 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2282 struct file_ra_state *ra;
2283
2284 ra = kzalloc(sizeof(*ra), GFP_NOFS);
2285
2286 mutex_lock(&inode->i_mutex);
2287 i = start >> PAGE_CACHE_SHIFT;
2288 last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
2289
2290 file_ra_state_init(ra, inode->i_mapping);
2291 btrfs_force_ra(inode->i_mapping, ra, NULL, i, last_index);
2292 kfree(ra);
2293
2294 for (; i <= last_index; i++) {
2295 page = grab_cache_page(inode->i_mapping, i);
2296 if (!page)
2297 goto out_unlock;
2298 if (!PageUptodate(page)) {
2299 btrfs_readpage(NULL, page);
2300 lock_page(page);
2301 if (!PageUptodate(page)) {
2302 unlock_page(page);
2303 page_cache_release(page);
2304 goto out_unlock;
2305 }
2306 }
2307 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2308 page_end = page_start + PAGE_CACHE_SIZE - 1;
2309
2310 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2311
2312 delalloc_start = page_start;
2313 existing_delalloc = count_range_bits(io_tree,
2314 &delalloc_start, page_end,
2315 PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2316
2317 set_extent_delalloc(io_tree, page_start,
2318 page_end, GFP_NOFS);
2319
2320 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2321 set_page_dirty(page);
2322 unlock_page(page);
2323 page_cache_release(page);
2324 }
2325
2326 out_unlock:
2327 mutex_unlock(&inode->i_mutex);
2328 return 0;
2329 }
2330
2331 /*
2332 * note, this releases the path
2333 */
2334 static int noinline relocate_one_reference(struct btrfs_root *extent_root,
2335 struct btrfs_path *path,
2336 struct btrfs_key *extent_key)
2337 {
2338 struct inode *inode;
2339 struct btrfs_root *found_root;
2340 struct btrfs_key *root_location;
2341 struct btrfs_extent_ref *ref;
2342 u64 ref_root;
2343 u64 ref_gen;
2344 u64 ref_objectid;
2345 u64 ref_offset;
2346 int ret;
2347
2348 ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
2349 struct btrfs_extent_ref);
2350 ref_root = btrfs_ref_root(path->nodes[0], ref);
2351 ref_gen = btrfs_ref_generation(path->nodes[0], ref);
2352 ref_objectid = btrfs_ref_objectid(path->nodes[0], ref);
2353 ref_offset = btrfs_ref_offset(path->nodes[0], ref);
2354 btrfs_release_path(extent_root, path);
2355
2356 root_location = kmalloc(sizeof(*root_location), GFP_NOFS);
2357 root_location->objectid = ref_root;
2358 if (ref_gen == 0)
2359 root_location->offset = 0;
2360 else
2361 root_location->offset = (u64)-1;
2362 root_location->type = BTRFS_ROOT_ITEM_KEY;
2363
2364 found_root = btrfs_read_fs_root_no_name(extent_root->fs_info,
2365 root_location);
2366 BUG_ON(!found_root);
2367 kfree(root_location);
2368
2369 if (ref_objectid >= BTRFS_FIRST_FREE_OBJECTID) {
2370 mutex_unlock(&extent_root->fs_info->fs_mutex);
2371 inode = btrfs_iget_locked(extent_root->fs_info->sb,
2372 ref_objectid, found_root);
2373 if (inode->i_state & I_NEW) {
2374 /* the inode and parent dir are two different roots */
2375 BTRFS_I(inode)->root = found_root;
2376 BTRFS_I(inode)->location.objectid = ref_objectid;
2377 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
2378 BTRFS_I(inode)->location.offset = 0;
2379 btrfs_read_locked_inode(inode);
2380 unlock_new_inode(inode);
2381
2382 }
2383 /* this can happen if the reference is not against
2384 * the latest version of the tree root
2385 */
2386 if (is_bad_inode(inode)) {
2387 mutex_lock(&extent_root->fs_info->fs_mutex);
2388 goto out;
2389 }
2390 relocate_inode_pages(inode, ref_offset, extent_key->offset);
2391 /* FIXME, data=ordered will help get rid of this */
2392 filemap_fdatawrite(inode->i_mapping);
2393 iput(inode);
2394 mutex_lock(&extent_root->fs_info->fs_mutex);
2395 } else {
2396 struct btrfs_trans_handle *trans;
2397 struct btrfs_key found_key;
2398 struct extent_buffer *eb;
2399 int level;
2400 int i;
2401
2402 trans = btrfs_start_transaction(found_root, 1);
2403 eb = read_tree_block(found_root, extent_key->objectid,
2404 extent_key->offset);
2405 level = btrfs_header_level(eb);
2406
2407 if (level == 0)
2408 btrfs_item_key_to_cpu(eb, &found_key, 0);
2409 else
2410 btrfs_node_key_to_cpu(eb, &found_key, 0);
2411
2412 free_extent_buffer(eb);
2413
2414 path->lowest_level = level;
2415 path->reada = 2;
2416 ret = btrfs_search_slot(trans, found_root, &found_key, path,
2417 0, 1);
2418 path->lowest_level = 0;
2419 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2420 if (!path->nodes[i])
2421 break;
2422 free_extent_buffer(path->nodes[i]);
2423 path->nodes[i] = NULL;
2424 }
2425 btrfs_release_path(found_root, path);
2426 btrfs_end_transaction(trans, found_root);
2427 }
2428
2429 out:
2430 return 0;
2431 }
2432
2433 static int noinline relocate_one_extent(struct btrfs_root *extent_root,
2434 struct btrfs_path *path,
2435 struct btrfs_key *extent_key)
2436 {
2437 struct btrfs_key key;
2438 struct btrfs_key found_key;
2439 struct extent_buffer *leaf;
2440 u32 nritems;
2441 u32 item_size;
2442 int ret = 0;
2443
2444 key.objectid = extent_key->objectid;
2445 key.type = BTRFS_EXTENT_REF_KEY;
2446 key.offset = 0;
2447
2448 while(1) {
2449 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2450
2451 if (ret < 0)
2452 goto out;
2453
2454 ret = 0;
2455 leaf = path->nodes[0];
2456 nritems = btrfs_header_nritems(leaf);
2457 if (path->slots[0] == nritems)
2458 goto out;
2459
2460 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2461 if (found_key.objectid != extent_key->objectid)
2462 break;
2463
2464 if (found_key.type != BTRFS_EXTENT_REF_KEY)
2465 break;
2466
2467 key.offset = found_key.offset + 1;
2468 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2469
2470 ret = relocate_one_reference(extent_root, path, extent_key);
2471 if (ret)
2472 goto out;
2473 }
2474 ret = 0;
2475 out:
2476 btrfs_release_path(extent_root, path);
2477 return ret;
2478 }
2479
2480 int btrfs_shrink_extent_tree(struct btrfs_root *root, u64 new_size)
2481 {
2482 struct btrfs_trans_handle *trans;
2483 struct btrfs_root *tree_root = root->fs_info->tree_root;
2484 struct btrfs_path *path;
2485 u64 cur_byte;
2486 u64 total_found;
2487 struct btrfs_fs_info *info = root->fs_info;
2488 struct extent_io_tree *block_group_cache;
2489 struct btrfs_key key;
2490 struct btrfs_key found_key;
2491 struct extent_buffer *leaf;
2492 u32 nritems;
2493 int ret;
2494 int progress = 0;
2495
2496 btrfs_set_super_total_bytes(&info->super_copy, new_size);
2497 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2498 GFP_NOFS);
2499 block_group_cache = &info->block_group_cache;
2500 path = btrfs_alloc_path();
2501 root = root->fs_info->extent_root;
2502 path->reada = 2;
2503
2504 again:
2505 total_found = 0;
2506 key.objectid = new_size;
2507 key.offset = 0;
2508 key.type = 0;
2509 cur_byte = key.objectid;
2510
2511 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2512 if (ret < 0)
2513 goto out;
2514
2515 ret = btrfs_previous_item(root, path, 0, BTRFS_EXTENT_ITEM_KEY);
2516 if (ret < 0)
2517 goto out;
2518 if (ret == 0) {
2519 leaf = path->nodes[0];
2520 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2521 if (found_key.objectid + found_key.offset > new_size) {
2522 cur_byte = found_key.objectid;
2523 key.objectid = cur_byte;
2524 }
2525 }
2526 btrfs_release_path(root, path);
2527
2528 while(1) {
2529 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2530 if (ret < 0)
2531 goto out;
2532
2533 leaf = path->nodes[0];
2534 nritems = btrfs_header_nritems(leaf);
2535 next:
2536 if (path->slots[0] >= nritems) {
2537 ret = btrfs_next_leaf(root, path);
2538 if (ret < 0)
2539 goto out;
2540 if (ret == 1) {
2541 ret = 0;
2542 break;
2543 }
2544 leaf = path->nodes[0];
2545 nritems = btrfs_header_nritems(leaf);
2546 }
2547
2548 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2549
2550 if (progress && need_resched()) {
2551 memcpy(&key, &found_key, sizeof(key));
2552 mutex_unlock(&root->fs_info->fs_mutex);
2553 cond_resched();
2554 mutex_lock(&root->fs_info->fs_mutex);
2555 btrfs_release_path(root, path);
2556 btrfs_search_slot(NULL, root, &key, path, 0, 0);
2557 progress = 0;
2558 goto next;
2559 }
2560 progress = 1;
2561
2562 if (btrfs_key_type(&found_key) != BTRFS_EXTENT_ITEM_KEY ||
2563 found_key.objectid + found_key.offset <= cur_byte) {
2564 path->slots[0]++;
2565 goto next;
2566 }
2567
2568 total_found++;
2569 cur_byte = found_key.objectid + found_key.offset;
2570 key.objectid = cur_byte;
2571 btrfs_release_path(root, path);
2572 ret = relocate_one_extent(root, path, &found_key);
2573 }
2574
2575 btrfs_release_path(root, path);
2576
2577 if (total_found > 0) {
2578 trans = btrfs_start_transaction(tree_root, 1);
2579 btrfs_commit_transaction(trans, tree_root);
2580
2581 mutex_unlock(&root->fs_info->fs_mutex);
2582 btrfs_clean_old_snapshots(tree_root);
2583 mutex_lock(&root->fs_info->fs_mutex);
2584
2585 trans = btrfs_start_transaction(tree_root, 1);
2586 btrfs_commit_transaction(trans, tree_root);
2587 goto again;
2588 }
2589
2590 trans = btrfs_start_transaction(root, 1);
2591 key.objectid = new_size;
2592 key.offset = 0;
2593 key.type = 0;
2594 while(1) {
2595 u64 ptr;
2596
2597 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2598 if (ret < 0)
2599 goto out;
2600
2601 leaf = path->nodes[0];
2602 nritems = btrfs_header_nritems(leaf);
2603 bg_next:
2604 if (path->slots[0] >= nritems) {
2605 ret = btrfs_next_leaf(root, path);
2606 if (ret < 0)
2607 break;
2608 if (ret == 1) {
2609 ret = 0;
2610 break;
2611 }
2612 leaf = path->nodes[0];
2613 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2614
2615 /*
2616 * btrfs_next_leaf doesn't cow buffers, we have to
2617 * do the search again
2618 */
2619 memcpy(&key, &found_key, sizeof(key));
2620 btrfs_release_path(root, path);
2621 goto resched_check;
2622 }
2623
2624 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2625 if (btrfs_key_type(&found_key) != BTRFS_BLOCK_GROUP_ITEM_KEY) {
2626 printk("shrinker found key %Lu %u %Lu\n",
2627 found_key.objectid, found_key.type,
2628 found_key.offset);
2629 path->slots[0]++;
2630 goto bg_next;
2631 }
2632 ret = get_state_private(&info->block_group_cache,
2633 found_key.objectid, &ptr);
2634 if (!ret)
2635 kfree((void *)(unsigned long)ptr);
2636
2637 clear_extent_bits(&info->block_group_cache, found_key.objectid,
2638 found_key.objectid + found_key.offset - 1,
2639 (unsigned int)-1, GFP_NOFS);
2640
2641 key.objectid = found_key.objectid + 1;
2642 btrfs_del_item(trans, root, path);
2643 btrfs_release_path(root, path);
2644 resched_check:
2645 if (need_resched()) {
2646 mutex_unlock(&root->fs_info->fs_mutex);
2647 cond_resched();
2648 mutex_lock(&root->fs_info->fs_mutex);
2649 }
2650 }
2651 clear_extent_dirty(&info->free_space_cache, new_size, (u64)-1,
2652 GFP_NOFS);
2653 btrfs_commit_transaction(trans, root);
2654 out:
2655 btrfs_free_path(path);
2656 return ret;
2657 }
2658
2659 int btrfs_grow_extent_tree(struct btrfs_trans_handle *trans,
2660 struct btrfs_root *root, u64 new_size)
2661 {
2662 btrfs_set_super_total_bytes(&root->fs_info->super_copy, new_size);
2663 return 0;
2664 }
2665
2666 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
2667 struct btrfs_key *key)
2668 {
2669 int ret;
2670 struct btrfs_key found_key;
2671 struct extent_buffer *leaf;
2672 int slot;
2673
2674 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
2675 if (ret < 0)
2676 return ret;
2677 while(1) {
2678 slot = path->slots[0];
2679 leaf = path->nodes[0];
2680 if (slot >= btrfs_header_nritems(leaf)) {
2681 ret = btrfs_next_leaf(root, path);
2682 if (ret == 0)
2683 continue;
2684 if (ret < 0)
2685 goto error;
2686 break;
2687 }
2688 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2689
2690 if (found_key.objectid >= key->objectid &&
2691 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
2692 return 0;
2693 path->slots[0]++;
2694 }
2695 ret = -ENOENT;
2696 error:
2697 return ret;
2698 }
2699
2700 int btrfs_read_block_groups(struct btrfs_root *root)
2701 {
2702 struct btrfs_path *path;
2703 int ret;
2704 int bit;
2705 struct btrfs_block_group_cache *cache;
2706 struct btrfs_fs_info *info = root->fs_info;
2707 struct btrfs_space_info *space_info;
2708 struct extent_io_tree *block_group_cache;
2709 struct btrfs_key key;
2710 struct btrfs_key found_key;
2711 struct extent_buffer *leaf;
2712
2713 block_group_cache = &info->block_group_cache;
2714 root = info->extent_root;
2715 key.objectid = 0;
2716 key.offset = 0;
2717 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2718 path = btrfs_alloc_path();
2719 if (!path)
2720 return -ENOMEM;
2721
2722 while(1) {
2723 ret = find_first_block_group(root, path, &key);
2724 if (ret > 0) {
2725 ret = 0;
2726 goto error;
2727 }
2728 if (ret != 0)
2729 goto error;
2730
2731 leaf = path->nodes[0];
2732 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2733 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2734 if (!cache) {
2735 ret = -ENOMEM;
2736 break;
2737 }
2738
2739 read_extent_buffer(leaf, &cache->item,
2740 btrfs_item_ptr_offset(leaf, path->slots[0]),
2741 sizeof(cache->item));
2742 memcpy(&cache->key, &found_key, sizeof(found_key));
2743 cache->cached = 0;
2744 cache->pinned = 0;
2745
2746 key.objectid = found_key.objectid + found_key.offset;
2747 btrfs_release_path(root, path);
2748 cache->flags = btrfs_block_group_flags(&cache->item);
2749 bit = 0;
2750 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
2751 bit = BLOCK_GROUP_DATA;
2752 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
2753 bit = BLOCK_GROUP_SYSTEM;
2754 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
2755 bit = BLOCK_GROUP_METADATA;
2756 }
2757 set_avail_alloc_bits(info, cache->flags);
2758
2759 ret = update_space_info(info, cache->flags, found_key.offset,
2760 btrfs_block_group_used(&cache->item),
2761 &space_info);
2762 BUG_ON(ret);
2763 cache->space_info = space_info;
2764
2765 /* use EXTENT_LOCKED to prevent merging */
2766 set_extent_bits(block_group_cache, found_key.objectid,
2767 found_key.objectid + found_key.offset - 1,
2768 bit | EXTENT_LOCKED, GFP_NOFS);
2769 set_state_private(block_group_cache, found_key.objectid,
2770 (unsigned long)cache);
2771
2772 if (key.objectid >=
2773 btrfs_super_total_bytes(&info->super_copy))
2774 break;
2775 }
2776 ret = 0;
2777 error:
2778 btrfs_free_path(path);
2779 return ret;
2780 }
2781
2782 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
2783 struct btrfs_root *root, u64 bytes_used,
2784 u64 type, u64 chunk_tree, u64 chunk_objectid,
2785 u64 size)
2786 {
2787 int ret;
2788 int bit = 0;
2789 struct btrfs_root *extent_root;
2790 struct btrfs_block_group_cache *cache;
2791 struct extent_io_tree *block_group_cache;
2792
2793 extent_root = root->fs_info->extent_root;
2794 block_group_cache = &root->fs_info->block_group_cache;
2795
2796 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2797 BUG_ON(!cache);
2798 cache->key.objectid = chunk_objectid;
2799 cache->key.offset = size;
2800 cache->cached = 0;
2801 cache->pinned = 0;
2802 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2803 memset(&cache->item, 0, sizeof(cache->item));
2804 btrfs_set_block_group_used(&cache->item, bytes_used);
2805 btrfs_set_block_group_chunk_tree(&cache->item, chunk_tree);
2806 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
2807 cache->flags = type;
2808 btrfs_set_block_group_flags(&cache->item, type);
2809
2810 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
2811 &cache->space_info);
2812 BUG_ON(ret);
2813
2814 if (type & BTRFS_BLOCK_GROUP_DATA) {
2815 bit = BLOCK_GROUP_DATA;
2816 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2817 bit = BLOCK_GROUP_SYSTEM;
2818 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2819 bit = BLOCK_GROUP_METADATA;
2820 }
2821 set_extent_bits(block_group_cache, chunk_objectid,
2822 chunk_objectid + size - 1,
2823 bit | EXTENT_LOCKED, GFP_NOFS);
2824 set_state_private(block_group_cache, chunk_objectid,
2825 (unsigned long)cache);
2826
2827 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
2828 sizeof(cache->item));
2829 BUG_ON(ret);
2830
2831 finish_current_insert(trans, extent_root);
2832 ret = del_pending_extents(trans, extent_root);
2833 BUG_ON(ret);
2834 return 0;
2835 }
Ubuntu Artful kernel mirror