projects / deliverable / linux.git / blob
? search:


187be40124749f53ba7b89afa896a957d0f650eb
[deliverable/linux.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 "hash.h"
22 #include "ctree.h"
23 #include "disk-io.h"
24 #include "print-tree.h"
25 #include "transaction.h"
26
27 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
28 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
29 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
30
31 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
32 btrfs_root *extent_root);
33 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
34 btrfs_root *extent_root);
35
36 static int cache_block_group(struct btrfs_root *root,
37 struct btrfs_block_group_cache *block_group)
38 {
39 struct btrfs_path *path;
40 int ret;
41 struct btrfs_key key;
42 struct extent_buffer *leaf;
43 struct extent_map_tree *free_space_cache;
44 int slot;
45 u64 last = 0;
46 u64 hole_size;
47 u64 first_free;
48 int found = 0;
49
50 if (!block_group)
51 return 0;
52
53 root = root->fs_info->extent_root;
54 free_space_cache = &root->fs_info->free_space_cache;
55
56 if (block_group->cached)
57 return 0;
58
59 path = btrfs_alloc_path();
60 if (!path)
61 return -ENOMEM;
62
63 path->reada = 2;
64 first_free = block_group->key.objectid;
65 key.objectid = block_group->key.objectid;
66 key.offset = 0;
67
68 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
69 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
70
71 if (ret < 0)
72 return ret;
73
74 if (ret && path->slots[0] > 0)
75 path->slots[0]--;
76
77 while(1) {
78 leaf = path->nodes[0];
79 slot = path->slots[0];
80 if (slot >= btrfs_header_nritems(leaf)) {
81 ret = btrfs_next_leaf(root, path);
82 if (ret < 0)
83 goto err;
84 if (ret == 0) {
85 continue;
86 } else {
87 break;
88 }
89 }
90
91 btrfs_item_key_to_cpu(leaf, &key, slot);
92 if (key.objectid < block_group->key.objectid) {
93 if (btrfs_key_type(&key) != BTRFS_EXTENT_REF_KEY &&
94 key.objectid + key.offset > first_free)
95 first_free = key.objectid + key.offset;
96 goto next;
97 }
98
99 if (key.objectid >= block_group->key.objectid +
100 block_group->key.offset) {
101 break;
102 }
103
104 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
105 if (!found) {
106 last = first_free;
107 found = 1;
108 }
109 if (key.objectid > last) {
110 hole_size = key.objectid - last;
111 set_extent_dirty(free_space_cache, last,
112 last + hole_size - 1,
113 GFP_NOFS);
114 }
115 last = key.objectid + key.offset;
116 }
117 next:
118 path->slots[0]++;
119 }
120
121 if (!found)
122 last = first_free;
123 if (block_group->key.objectid +
124 block_group->key.offset > last) {
125 hole_size = block_group->key.objectid +
126 block_group->key.offset - last;
127 set_extent_dirty(free_space_cache, last,
128 last + hole_size - 1, GFP_NOFS);
129 }
130 block_group->cached = 1;
131 err:
132 btrfs_free_path(path);
133 return 0;
134 }
135
136 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
137 btrfs_fs_info *info,
138 u64 bytenr)
139 {
140 struct extent_map_tree *block_group_cache;
141 struct btrfs_block_group_cache *block_group = NULL;
142 u64 ptr;
143 u64 start;
144 u64 end;
145 int ret;
146
147 block_group_cache = &info->block_group_cache;
148 ret = find_first_extent_bit(block_group_cache,
149 bytenr, &start, &end,
150 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA);
151 if (ret) {
152 return NULL;
153 }
154 ret = get_state_private(block_group_cache, start, &ptr);
155 if (ret)
156 return NULL;
157
158 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
159 if (block_group->key.objectid <= bytenr && bytenr <
160 block_group->key.objectid + block_group->key.offset)
161 return block_group;
162 return NULL;
163 }
164 static u64 find_search_start(struct btrfs_root *root,
165 struct btrfs_block_group_cache **cache_ret,
166 u64 search_start, int num,
167 int data, int full_scan)
168 {
169 int ret;
170 struct btrfs_block_group_cache *cache = *cache_ret;
171 u64 last;
172 u64 start = 0;
173 u64 end = 0;
174 u64 cache_miss = 0;
175 int wrapped = 0;
176
177 if (!cache) {
178 goto out;
179 }
180 again:
181 ret = cache_block_group(root, cache);
182 if (ret)
183 goto out;
184
185 last = max(search_start, cache->key.objectid);
186
187 while(1) {
188 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
189 last, &start, &end, EXTENT_DIRTY);
190 if (ret) {
191 if (!cache_miss)
192 cache_miss = last;
193 goto new_group;
194 }
195
196 start = max(last, start);
197 last = end + 1;
198 if (last - start < num) {
199 if (last == cache->key.objectid + cache->key.offset)
200 cache_miss = start;
201 continue;
202 }
203 if (data != BTRFS_BLOCK_GROUP_MIXED &&
204 start + num > cache->key.objectid + cache->key.offset)
205 goto new_group;
206 return start;
207 }
208 out:
209 cache = btrfs_lookup_block_group(root->fs_info, search_start);
210 if (!cache) {
211 printk("Unable to find block group for %Lu\n",
212 search_start);
213 WARN_ON(1);
214 return search_start;
215 }
216 return search_start;
217
218 new_group:
219 last = cache->key.objectid + cache->key.offset;
220 wrapped:
221 cache = btrfs_lookup_block_group(root->fs_info, last);
222 if (!cache) {
223 no_cache:
224 if (!wrapped) {
225 wrapped = 1;
226 last = search_start;
227 data = BTRFS_BLOCK_GROUP_MIXED;
228 goto wrapped;
229 }
230 goto out;
231 }
232 if (cache_miss && !cache->cached) {
233 cache_block_group(root, cache);
234 last = cache_miss;
235 cache = btrfs_lookup_block_group(root->fs_info, last);
236 }
237 cache = btrfs_find_block_group(root, cache, last, data, 0);
238 if (!cache)
239 goto no_cache;
240 *cache_ret = cache;
241 cache_miss = 0;
242 goto again;
243 }
244
245 static u64 div_factor(u64 num, int factor)
246 {
247 if (factor == 10)
248 return num;
249 num *= factor;
250 do_div(num, 10);
251 return num;
252 }
253
254 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
255 struct btrfs_block_group_cache
256 *hint, u64 search_start,
257 int data, int owner)
258 {
259 struct btrfs_block_group_cache *cache;
260 struct extent_map_tree *block_group_cache;
261 struct btrfs_block_group_cache *found_group = NULL;
262 struct btrfs_fs_info *info = root->fs_info;
263 u64 used;
264 u64 last = 0;
265 u64 hint_last;
266 u64 start;
267 u64 end;
268 u64 free_check;
269 u64 ptr;
270 int bit;
271 int ret;
272 int full_search = 0;
273 int factor = 8;
274 int data_swap = 0;
275
276 block_group_cache = &info->block_group_cache;
277
278 if (!owner)
279 factor = 8;
280
281 if (data == BTRFS_BLOCK_GROUP_MIXED) {
282 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
283 factor = 10;
284 } else if (data)
285 bit = BLOCK_GROUP_DATA;
286 else
287 bit = BLOCK_GROUP_METADATA;
288
289 if (search_start) {
290 struct btrfs_block_group_cache *shint;
291 shint = btrfs_lookup_block_group(info, search_start);
292 if (shint && (shint->data == data ||
293 shint->data == BTRFS_BLOCK_GROUP_MIXED)) {
294 used = btrfs_block_group_used(&shint->item);
295 if (used + shint->pinned <
296 div_factor(shint->key.offset, factor)) {
297 return shint;
298 }
299 }
300 }
301 if (hint && (hint->data == data ||
302 hint->data == BTRFS_BLOCK_GROUP_MIXED)) {
303 used = btrfs_block_group_used(&hint->item);
304 if (used + hint->pinned <
305 div_factor(hint->key.offset, factor)) {
306 return hint;
307 }
308 last = hint->key.objectid + hint->key.offset;
309 hint_last = last;
310 } else {
311 if (hint)
312 hint_last = max(hint->key.objectid, search_start);
313 else
314 hint_last = search_start;
315
316 last = hint_last;
317 }
318 again:
319 while(1) {
320 ret = find_first_extent_bit(block_group_cache, last,
321 &start, &end, bit);
322 if (ret)
323 break;
324
325 ret = get_state_private(block_group_cache, start, &ptr);
326 if (ret)
327 break;
328
329 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
330 last = cache->key.objectid + cache->key.offset;
331 used = btrfs_block_group_used(&cache->item);
332
333 if (full_search)
334 free_check = cache->key.offset;
335 else
336 free_check = div_factor(cache->key.offset, factor);
337 if (used + cache->pinned < free_check) {
338 found_group = cache;
339 goto found;
340 }
341 cond_resched();
342 }
343 if (!full_search) {
344 last = search_start;
345 full_search = 1;
346 goto again;
347 }
348 if (!data_swap) {
349 data_swap = 1;
350 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
351 last = search_start;
352 goto again;
353 }
354 found:
355 return found_group;
356 }
357
358 static u64 hash_extent_ref(u64 root_objectid, u64 ref_generation,
359 u64 owner, u64 owner_offset)
360 {
361 u32 high_crc = ~(u32)0;
362 u32 low_crc = ~(u32)0;
363 __le64 lenum;
364
365 lenum = cpu_to_le64(root_objectid);
366 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
367 lenum = cpu_to_le64(ref_generation);
368 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
369
370 #if 0
371 lenum = cpu_to_le64(owner);
372 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
373 lenum = cpu_to_le64(owner_offset);
374 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
375 #endif
376 return ((u64)high_crc << 32) | (u64)low_crc;
377 }
378
379 static int match_extent_ref(struct extent_buffer *leaf,
380 struct btrfs_extent_ref *disk_ref,
381 struct btrfs_extent_ref *cpu_ref)
382 {
383 int ret;
384 int len;
385
386 if (cpu_ref->objectid)
387 len = sizeof(*cpu_ref);
388 else
389 len = 2 * sizeof(u64);
390 ret = memcmp_extent_buffer(leaf, cpu_ref, (unsigned long)disk_ref,
391 len);
392 return ret == 0;
393 }
394
395 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
396 struct btrfs_root *root,
397 struct btrfs_path *path, u64 bytenr,
398 u64 root_objectid, u64 ref_generation,
399 u64 owner, u64 owner_offset, int del)
400 {
401 u64 hash;
402 struct btrfs_key key;
403 struct btrfs_key found_key;
404 struct btrfs_extent_ref ref;
405 struct extent_buffer *leaf;
406 struct btrfs_extent_ref *disk_ref;
407 int ret;
408 int ret2;
409
410 btrfs_set_stack_ref_root(&ref, root_objectid);
411 btrfs_set_stack_ref_generation(&ref, ref_generation);
412 btrfs_set_stack_ref_objectid(&ref, owner);
413 btrfs_set_stack_ref_offset(&ref, owner_offset);
414
415 hash = hash_extent_ref(root_objectid, ref_generation, owner,
416 owner_offset);
417 key.offset = hash;
418 key.objectid = bytenr;
419 key.type = BTRFS_EXTENT_REF_KEY;
420
421 while (1) {
422 ret = btrfs_search_slot(trans, root, &key, path,
423 del ? -1 : 0, del);
424 if (ret < 0)
425 goto out;
426 leaf = path->nodes[0];
427 if (ret != 0) {
428 u32 nritems = btrfs_header_nritems(leaf);
429 if (path->slots[0] >= nritems) {
430 ret2 = btrfs_next_leaf(root, path);
431 if (ret2)
432 goto out;
433 leaf = path->nodes[0];
434 }
435 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
436 if (found_key.objectid != bytenr ||
437 found_key.type != BTRFS_EXTENT_REF_KEY)
438 goto out;
439 key.offset = found_key.offset;
440 if (del) {
441 btrfs_release_path(root, path);
442 continue;
443 }
444 }
445 disk_ref = btrfs_item_ptr(path->nodes[0],
446 path->slots[0],
447 struct btrfs_extent_ref);
448 if (match_extent_ref(path->nodes[0], disk_ref, &ref)) {
449 ret = 0;
450 goto out;
451 }
452 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
453 key.offset = found_key.offset + 1;
454 btrfs_release_path(root, path);
455 }
456 out:
457 return ret;
458 }
459
460 /*
461 * Back reference rules. Back refs have three main goals:
462 *
463 * 1) differentiate between all holders of references to an extent so that
464 * when a reference is dropped we can make sure it was a valid reference
465 * before freeing the extent.
466 *
467 * 2) Provide enough information to quickly find the holders of an extent
468 * if we notice a given block is corrupted or bad.
469 *
470 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
471 * maintenance. This is actually the same as #2, but with a slightly
472 * different use case.
473 *
474 * File extents can be referenced by:
475 *
476 * - multiple snapshots, subvolumes, or different generations in one subvol
477 * - different files inside a single subvolume (in theory, not implemented yet)
478 * - different offsets inside a file (bookend extents in file.c)
479 *
480 * The extent ref structure has fields for:
481 *
482 * - Objectid of the subvolume root
483 * - Generation number of the tree holding the reference
484 * - objectid of the file holding the reference
485 * - offset in the file corresponding to the key holding the reference
486 *
487 * When a file extent is allocated the fields are filled in:
488 * (root_key.objectid, trans->transid, inode objectid, offset in file)
489 *
490 * When a leaf is cow'd new references are added for every file extent found
491 * in the leaf. It looks the same as the create case, but trans->transid
492 * will be different when the block is cow'd.
493 *
494 * (root_key.objectid, trans->transid, inode objectid, offset in file)
495 *
496 * When a file extent is removed either during snapshot deletion or file
497 * truncation, the corresponding back reference is found
498 * by searching for:
499 *
500 * (btrfs_header_owner(leaf), btrfs_header_generation(leaf),
501 * inode objectid, offset in file)
502 *
503 * Btree extents can be referenced by:
504 *
505 * - Different subvolumes
506 * - Different generations of the same subvolume
507 *
508 * Storing sufficient information for a full reverse mapping of a btree
509 * block would require storing the lowest key of the block in the backref,
510 * and it would require updating that lowest key either before write out or
511 * every time it changed. Instead, the objectid of the lowest key is stored
512 * along with the level of the tree block. This provides a hint
513 * about where in the btree the block can be found. Searches through the
514 * btree only need to look for a pointer to that block, so they stop one
515 * level higher than the level recorded in the backref.
516 *
517 * Some btrees do not do reference counting on their extents. These
518 * include the extent tree and the tree of tree roots. Backrefs for these
519 * trees always have a generation of zero.
520 *
521 * When a tree block is created, back references are inserted:
522 *
523 * (root->root_key.objectid, trans->transid or zero, lowest_key_objectid, level)
524 *
525 * When a tree block is cow'd in a reference counted root,
526 * new back references are added for all the blocks it points to.
527 * These are of the form (trans->transid will have increased since creation):
528 *
529 * (root->root_key.objectid, trans->transid, lowest_key_objectid, level)
530 *
531 * Because the lowest_key_objectid and the level are just hints
532 * they are not used when backrefs are deleted. When a backref is deleted:
533 *
534 * if backref was for a tree root:
535 * root_objectid = root->root_key.objectid
536 * else
537 * root_objectid = btrfs_header_owner(parent)
538 *
539 * (root_objectid, btrfs_header_generation(parent) or zero, 0, 0)
540 *
541 * Back Reference Key hashing:
542 *
543 * Back references have four fields, each 64 bits long. Unfortunately,
544 * This is hashed into a single 64 bit number and placed into the key offset.
545 * The key objectid corresponds to the first byte in the extent, and the
546 * key type is set to BTRFS_EXTENT_REF_KEY
547 */
548 int btrfs_insert_extent_backref(struct btrfs_trans_handle *trans,
549 struct btrfs_root *root,
550 struct btrfs_path *path, u64 bytenr,
551 u64 root_objectid, u64 ref_generation,
552 u64 owner, u64 owner_offset)
553 {
554 u64 hash;
555 struct btrfs_key key;
556 struct btrfs_extent_ref ref;
557 struct btrfs_extent_ref *disk_ref;
558 int ret;
559
560 btrfs_set_stack_ref_root(&ref, root_objectid);
561 btrfs_set_stack_ref_generation(&ref, ref_generation);
562 btrfs_set_stack_ref_objectid(&ref, owner);
563 btrfs_set_stack_ref_offset(&ref, owner_offset);
564
565 hash = hash_extent_ref(root_objectid, ref_generation, owner,
566 owner_offset);
567 key.offset = hash;
568 key.objectid = bytenr;
569 key.type = BTRFS_EXTENT_REF_KEY;
570
571 ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(ref));
572 while (ret == -EEXIST) {
573 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
574 struct btrfs_extent_ref);
575 if (match_extent_ref(path->nodes[0], disk_ref, &ref))
576 goto out;
577 key.offset++;
578 btrfs_release_path(root, path);
579 ret = btrfs_insert_empty_item(trans, root, path, &key,
580 sizeof(ref));
581 }
582 if (ret)
583 goto out;
584 disk_ref = btrfs_item_ptr(path->nodes[0], path->slots[0],
585 struct btrfs_extent_ref);
586 write_extent_buffer(path->nodes[0], &ref, (unsigned long)disk_ref,
587 sizeof(ref));
588 btrfs_mark_buffer_dirty(path->nodes[0]);
589 out:
590 btrfs_release_path(root, path);
591 return ret;
592 }
593
594 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
595 struct btrfs_root *root,
596 u64 bytenr, u64 num_bytes,
597 u64 root_objectid, u64 ref_generation,
598 u64 owner, u64 owner_offset)
599 {
600 struct btrfs_path *path;
601 int ret;
602 struct btrfs_key key;
603 struct extent_buffer *l;
604 struct btrfs_extent_item *item;
605 u32 refs;
606
607 WARN_ON(num_bytes < root->sectorsize);
608 path = btrfs_alloc_path();
609 if (!path)
610 return -ENOMEM;
611
612 key.objectid = bytenr;
613 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
614 key.offset = num_bytes;
615 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
616 0, 1);
617 if (ret < 0)
618 return ret;
619 if (ret != 0) {
620 BUG();
621 }
622 BUG_ON(ret != 0);
623 l = path->nodes[0];
624 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
625 refs = btrfs_extent_refs(l, item);
626 btrfs_set_extent_refs(l, item, refs + 1);
627 btrfs_mark_buffer_dirty(path->nodes[0]);
628
629 btrfs_release_path(root->fs_info->extent_root, path);
630
631 ret = btrfs_insert_extent_backref(trans, root->fs_info->extent_root,
632 path, bytenr, root_objectid,
633 ref_generation, owner, owner_offset);
634 BUG_ON(ret);
635 finish_current_insert(trans, root->fs_info->extent_root);
636 del_pending_extents(trans, root->fs_info->extent_root);
637
638 btrfs_free_path(path);
639 return 0;
640 }
641
642 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
643 struct btrfs_root *root)
644 {
645 finish_current_insert(trans, root->fs_info->extent_root);
646 del_pending_extents(trans, root->fs_info->extent_root);
647 return 0;
648 }
649
650 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
651 struct btrfs_root *root, u64 bytenr,
652 u64 num_bytes, u32 *refs)
653 {
654 struct btrfs_path *path;
655 int ret;
656 struct btrfs_key key;
657 struct extent_buffer *l;
658 struct btrfs_extent_item *item;
659
660 WARN_ON(num_bytes < root->sectorsize);
661 path = btrfs_alloc_path();
662 key.objectid = bytenr;
663 key.offset = num_bytes;
664 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
665 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
666 0, 0);
667 if (ret < 0)
668 goto out;
669 if (ret != 0) {
670 btrfs_print_leaf(root, path->nodes[0]);
671 printk("failed to find block number %Lu\n", bytenr);
672 BUG();
673 }
674 l = path->nodes[0];
675 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
676 *refs = btrfs_extent_refs(l, item);
677 out:
678 btrfs_free_path(path);
679 return 0;
680 }
681
682 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
683 struct btrfs_root *root, u64 owner_objectid)
684 {
685 u64 generation;
686 u64 key_objectid;
687 u64 level;
688 u32 nritems;
689 struct btrfs_disk_key disk_key;
690
691 level = btrfs_header_level(root->node);
692 generation = trans->transid;
693 nritems = btrfs_header_nritems(root->node);
694 if (nritems > 0) {
695 if (level == 0)
696 btrfs_item_key(root->node, &disk_key, 0);
697 else
698 btrfs_node_key(root->node, &disk_key, 0);
699 key_objectid = btrfs_disk_key_objectid(&disk_key);
700 } else {
701 key_objectid = 0;
702 }
703 return btrfs_inc_extent_ref(trans, root, root->node->start,
704 root->node->len, owner_objectid,
705 generation, 0, 0);
706 }
707
708 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
709 struct extent_buffer *buf)
710 {
711 u64 bytenr;
712 u32 nritems;
713 struct btrfs_key key;
714 struct btrfs_file_extent_item *fi;
715 int i;
716 int level;
717 int ret;
718 int faili;
719
720 if (!root->ref_cows)
721 return 0;
722
723 level = btrfs_header_level(buf);
724 nritems = btrfs_header_nritems(buf);
725 for (i = 0; i < nritems; i++) {
726 if (level == 0) {
727 u64 disk_bytenr;
728 btrfs_item_key_to_cpu(buf, &key, i);
729 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
730 continue;
731 fi = btrfs_item_ptr(buf, i,
732 struct btrfs_file_extent_item);
733 if (btrfs_file_extent_type(buf, fi) ==
734 BTRFS_FILE_EXTENT_INLINE)
735 continue;
736 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
737 if (disk_bytenr == 0)
738 continue;
739 ret = btrfs_inc_extent_ref(trans, root, disk_bytenr,
740 btrfs_file_extent_disk_num_bytes(buf, fi),
741 root->root_key.objectid, trans->transid,
742 key.objectid, key.offset);
743 if (ret) {
744 faili = i;
745 goto fail;
746 }
747 } else {
748 bytenr = btrfs_node_blockptr(buf, i);
749 ret = btrfs_inc_extent_ref(trans, root, bytenr,
750 btrfs_level_size(root, level - 1),
751 root->root_key.objectid,
752 trans->transid, 0, 0);
753 if (ret) {
754 faili = i;
755 goto fail;
756 }
757 }
758 }
759 return 0;
760 fail:
761 WARN_ON(1);
762 #if 0
763 for (i =0; i < faili; i++) {
764 if (level == 0) {
765 u64 disk_bytenr;
766 btrfs_item_key_to_cpu(buf, &key, i);
767 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
768 continue;
769 fi = btrfs_item_ptr(buf, i,
770 struct btrfs_file_extent_item);
771 if (btrfs_file_extent_type(buf, fi) ==
772 BTRFS_FILE_EXTENT_INLINE)
773 continue;
774 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
775 if (disk_bytenr == 0)
776 continue;
777 err = btrfs_free_extent(trans, root, disk_bytenr,
778 btrfs_file_extent_disk_num_bytes(buf,
779 fi), 0);
780 BUG_ON(err);
781 } else {
782 bytenr = btrfs_node_blockptr(buf, i);
783 err = btrfs_free_extent(trans, root, bytenr,
784 btrfs_level_size(root, level - 1), 0);
785 BUG_ON(err);
786 }
787 }
788 #endif
789 return ret;
790 }
791
792 static int write_one_cache_group(struct btrfs_trans_handle *trans,
793 struct btrfs_root *root,
794 struct btrfs_path *path,
795 struct btrfs_block_group_cache *cache)
796 {
797 int ret;
798 int pending_ret;
799 struct btrfs_root *extent_root = root->fs_info->extent_root;
800 unsigned long bi;
801 struct extent_buffer *leaf;
802
803 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
804 if (ret < 0)
805 goto fail;
806 BUG_ON(ret);
807
808 leaf = path->nodes[0];
809 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
810 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
811 btrfs_mark_buffer_dirty(leaf);
812 btrfs_release_path(extent_root, path);
813 fail:
814 finish_current_insert(trans, extent_root);
815 pending_ret = del_pending_extents(trans, extent_root);
816 if (ret)
817 return ret;
818 if (pending_ret)
819 return pending_ret;
820 return 0;
821
822 }
823
824 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
825 struct btrfs_root *root)
826 {
827 struct extent_map_tree *block_group_cache;
828 struct btrfs_block_group_cache *cache;
829 int ret;
830 int err = 0;
831 int werr = 0;
832 struct btrfs_path *path;
833 u64 last = 0;
834 u64 start;
835 u64 end;
836 u64 ptr;
837
838 block_group_cache = &root->fs_info->block_group_cache;
839 path = btrfs_alloc_path();
840 if (!path)
841 return -ENOMEM;
842
843 while(1) {
844 ret = find_first_extent_bit(block_group_cache, last,
845 &start, &end, BLOCK_GROUP_DIRTY);
846 if (ret)
847 break;
848
849 last = end + 1;
850 ret = get_state_private(block_group_cache, start, &ptr);
851 if (ret)
852 break;
853
854 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
855 err = write_one_cache_group(trans, root,
856 path, cache);
857 /*
858 * if we fail to write the cache group, we want
859 * to keep it marked dirty in hopes that a later
860 * write will work
861 */
862 if (err) {
863 werr = err;
864 continue;
865 }
866 clear_extent_bits(block_group_cache, start, end,
867 BLOCK_GROUP_DIRTY, GFP_NOFS);
868 }
869 btrfs_free_path(path);
870 return werr;
871 }
872
873 static int update_block_group(struct btrfs_trans_handle *trans,
874 struct btrfs_root *root,
875 u64 bytenr, u64 num_bytes, int alloc,
876 int mark_free, int data)
877 {
878 struct btrfs_block_group_cache *cache;
879 struct btrfs_fs_info *info = root->fs_info;
880 u64 total = num_bytes;
881 u64 old_val;
882 u64 byte_in_group;
883 u64 start;
884 u64 end;
885
886 while(total) {
887 cache = btrfs_lookup_block_group(info, bytenr);
888 if (!cache) {
889 return -1;
890 }
891 byte_in_group = bytenr - cache->key.objectid;
892 WARN_ON(byte_in_group > cache->key.offset);
893 start = cache->key.objectid;
894 end = start + cache->key.offset - 1;
895 set_extent_bits(&info->block_group_cache, start, end,
896 BLOCK_GROUP_DIRTY, GFP_NOFS);
897
898 old_val = btrfs_block_group_used(&cache->item);
899 num_bytes = min(total, cache->key.offset - byte_in_group);
900 if (alloc) {
901 if (cache->data != data &&
902 old_val < (cache->key.offset >> 1)) {
903 int bit_to_clear;
904 int bit_to_set;
905 cache->data = data;
906 if (data) {
907 bit_to_clear = BLOCK_GROUP_METADATA;
908 bit_to_set = BLOCK_GROUP_DATA;
909 cache->item.flags &=
910 ~BTRFS_BLOCK_GROUP_MIXED;
911 cache->item.flags |=
912 BTRFS_BLOCK_GROUP_DATA;
913 } else {
914 bit_to_clear = BLOCK_GROUP_DATA;
915 bit_to_set = BLOCK_GROUP_METADATA;
916 cache->item.flags &=
917 ~BTRFS_BLOCK_GROUP_MIXED;
918 cache->item.flags &=
919 ~BTRFS_BLOCK_GROUP_DATA;
920 }
921 clear_extent_bits(&info->block_group_cache,
922 start, end, bit_to_clear,
923 GFP_NOFS);
924 set_extent_bits(&info->block_group_cache,
925 start, end, bit_to_set,
926 GFP_NOFS);
927 } else if (cache->data != data &&
928 cache->data != BTRFS_BLOCK_GROUP_MIXED) {
929 cache->data = BTRFS_BLOCK_GROUP_MIXED;
930 set_extent_bits(&info->block_group_cache,
931 start, end,
932 BLOCK_GROUP_DATA |
933 BLOCK_GROUP_METADATA,
934 GFP_NOFS);
935 }
936 old_val += num_bytes;
937 } else {
938 old_val -= num_bytes;
939 if (mark_free) {
940 set_extent_dirty(&info->free_space_cache,
941 bytenr, bytenr + num_bytes - 1,
942 GFP_NOFS);
943 }
944 }
945 btrfs_set_block_group_used(&cache->item, old_val);
946 total -= num_bytes;
947 bytenr += num_bytes;
948 }
949 return 0;
950 }
951 static int update_pinned_extents(struct btrfs_root *root,
952 u64 bytenr, u64 num, int pin)
953 {
954 u64 len;
955 struct btrfs_block_group_cache *cache;
956 struct btrfs_fs_info *fs_info = root->fs_info;
957
958 if (pin) {
959 set_extent_dirty(&fs_info->pinned_extents,
960 bytenr, bytenr + num - 1, GFP_NOFS);
961 } else {
962 clear_extent_dirty(&fs_info->pinned_extents,
963 bytenr, bytenr + num - 1, GFP_NOFS);
964 }
965 while (num > 0) {
966 cache = btrfs_lookup_block_group(fs_info, bytenr);
967 WARN_ON(!cache);
968 len = min(num, cache->key.offset -
969 (bytenr - cache->key.objectid));
970 if (pin) {
971 cache->pinned += len;
972 fs_info->total_pinned += len;
973 } else {
974 cache->pinned -= len;
975 fs_info->total_pinned -= len;
976 }
977 bytenr += len;
978 num -= len;
979 }
980 return 0;
981 }
982
983 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy)
984 {
985 u64 last = 0;
986 u64 start;
987 u64 end;
988 struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents;
989 int ret;
990
991 while(1) {
992 ret = find_first_extent_bit(pinned_extents, last,
993 &start, &end, EXTENT_DIRTY);
994 if (ret)
995 break;
996 set_extent_dirty(copy, start, end, GFP_NOFS);
997 last = end + 1;
998 }
999 return 0;
1000 }
1001
1002 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1003 struct btrfs_root *root,
1004 struct extent_map_tree *unpin)
1005 {
1006 u64 start;
1007 u64 end;
1008 int ret;
1009 struct extent_map_tree *free_space_cache;
1010 free_space_cache = &root->fs_info->free_space_cache;
1011
1012 while(1) {
1013 ret = find_first_extent_bit(unpin, 0, &start, &end,
1014 EXTENT_DIRTY);
1015 if (ret)
1016 break;
1017 update_pinned_extents(root, start, end + 1 - start, 0);
1018 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1019 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1020 }
1021 return 0;
1022 }
1023
1024 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
1025 btrfs_root *extent_root)
1026 {
1027 u64 start;
1028 u64 end;
1029 struct btrfs_fs_info *info = extent_root->fs_info;
1030 struct extent_buffer *eb;
1031 struct btrfs_path *path;
1032 struct btrfs_key ins;
1033 struct btrfs_disk_key first;
1034 struct btrfs_extent_item extent_item;
1035 int ret;
1036 int level;
1037 int err = 0;
1038
1039 btrfs_set_stack_extent_refs(&extent_item, 1);
1040 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
1041 path = btrfs_alloc_path();
1042
1043 while(1) {
1044 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1045 &end, EXTENT_LOCKED);
1046 if (ret)
1047 break;
1048
1049 ins.objectid = start;
1050 ins.offset = end + 1 - start;
1051 err = btrfs_insert_item(trans, extent_root, &ins,
1052 &extent_item, sizeof(extent_item));
1053 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
1054 GFP_NOFS);
1055 eb = read_tree_block(extent_root, ins.objectid, ins.offset);
1056 level = btrfs_header_level(eb);
1057 if (level == 0) {
1058 btrfs_item_key(eb, &first, 0);
1059 } else {
1060 btrfs_node_key(eb, &first, 0);
1061 }
1062 err = btrfs_insert_extent_backref(trans, extent_root, path,
1063 start, extent_root->root_key.objectid,
1064 0, btrfs_disk_key_objectid(&first),
1065 level);
1066 BUG_ON(err);
1067 free_extent_buffer(eb);
1068 }
1069 btrfs_free_path(path);
1070 return 0;
1071 }
1072
1073 static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes,
1074 int pending)
1075 {
1076 int err = 0;
1077 struct extent_buffer *buf;
1078
1079 if (!pending) {
1080 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
1081 if (buf) {
1082 if (btrfs_buffer_uptodate(buf)) {
1083 u64 transid =
1084 root->fs_info->running_transaction->transid;
1085 if (btrfs_header_generation(buf) == transid) {
1086 free_extent_buffer(buf);
1087 return 1;
1088 }
1089 }
1090 free_extent_buffer(buf);
1091 }
1092 update_pinned_extents(root, bytenr, num_bytes, 1);
1093 } else {
1094 set_extent_bits(&root->fs_info->pending_del,
1095 bytenr, bytenr + num_bytes - 1,
1096 EXTENT_LOCKED, GFP_NOFS);
1097 }
1098 BUG_ON(err < 0);
1099 return 0;
1100 }
1101
1102 /*
1103 * remove an extent from the root, returns 0 on success
1104 */
1105 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1106 *root, u64 bytenr, u64 num_bytes,
1107 u64 root_objectid, u64 ref_generation,
1108 u64 owner_objectid, u64 owner_offset, int pin,
1109 int mark_free)
1110 {
1111 struct btrfs_path *path;
1112 struct btrfs_key key;
1113 struct btrfs_fs_info *info = root->fs_info;
1114 struct btrfs_root *extent_root = info->extent_root;
1115 struct extent_buffer *leaf;
1116 int ret;
1117 struct btrfs_extent_item *ei;
1118 u32 refs;
1119
1120 key.objectid = bytenr;
1121 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1122 key.offset = num_bytes;
1123
1124 path = btrfs_alloc_path();
1125 if (!path)
1126 return -ENOMEM;
1127
1128 if (ref_generation && owner_objectid == 0 && root_objectid == 3) {
1129 //printk("drop backref root %Lu gen %Lu byte %Lu\n", root_objectid, ref_generation, bytenr );
1130 }
1131 ret = lookup_extent_backref(trans, extent_root, path,
1132 bytenr, root_objectid,
1133 ref_generation,
1134 owner_objectid, owner_offset, 1);
1135 if (ret == 0) {
1136 ret = btrfs_del_item(trans, extent_root, path);
1137 } else {
1138 btrfs_print_leaf(extent_root, path->nodes[0]);
1139 WARN_ON(1);
1140 printk("Unable to find ref byte nr %Lu root %Lu "
1141 " gen %Lu owner %Lu offset %Lu\n", bytenr,
1142 root_objectid, ref_generation, owner_objectid,
1143 owner_offset);
1144 }
1145 btrfs_release_path(extent_root, path);
1146 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
1147 if (ret < 0)
1148 return ret;
1149 BUG_ON(ret);
1150
1151 leaf = path->nodes[0];
1152 ei = btrfs_item_ptr(leaf, path->slots[0],
1153 struct btrfs_extent_item);
1154 refs = btrfs_extent_refs(leaf, ei);
1155 BUG_ON(refs == 0);
1156 refs -= 1;
1157 btrfs_set_extent_refs(leaf, ei, refs);
1158 btrfs_mark_buffer_dirty(leaf);
1159
1160 if (refs == 0) {
1161 u64 super_used;
1162 u64 root_used;
1163
1164 if (pin) {
1165 ret = pin_down_bytes(root, bytenr, num_bytes, 0);
1166 if (ret > 0)
1167 mark_free = 1;
1168 BUG_ON(ret < 0);
1169 }
1170
1171 /* block accounting for super block */
1172 super_used = btrfs_super_bytes_used(&info->super_copy);
1173 btrfs_set_super_bytes_used(&info->super_copy,
1174 super_used - num_bytes);
1175
1176 /* block accounting for root item */
1177 root_used = btrfs_root_used(&root->root_item);
1178 btrfs_set_root_used(&root->root_item,
1179 root_used - num_bytes);
1180
1181 ret = btrfs_del_item(trans, extent_root, path);
1182 if (ret) {
1183 return ret;
1184 }
1185 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
1186 mark_free, 0);
1187 BUG_ON(ret);
1188 }
1189 btrfs_free_path(path);
1190 finish_current_insert(trans, extent_root);
1191 return ret;
1192 }
1193
1194 /*
1195 * find all the blocks marked as pending in the radix tree and remove
1196 * them from the extent map
1197 */
1198 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
1199 btrfs_root *extent_root)
1200 {
1201 int ret;
1202 int err = 0;
1203 u64 start;
1204 u64 end;
1205 struct extent_map_tree *pending_del;
1206 struct extent_map_tree *pinned_extents;
1207
1208 pending_del = &extent_root->fs_info->pending_del;
1209 pinned_extents = &extent_root->fs_info->pinned_extents;
1210
1211 while(1) {
1212 ret = find_first_extent_bit(pending_del, 0, &start, &end,
1213 EXTENT_LOCKED);
1214 if (ret)
1215 break;
1216 update_pinned_extents(extent_root, start, end + 1 - start, 1);
1217 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
1218 GFP_NOFS);
1219 ret = __free_extent(trans, extent_root,
1220 start, end + 1 - start,
1221 extent_root->root_key.objectid,
1222 0, 0, 0, 0, 0);
1223 if (ret)
1224 err = ret;
1225 }
1226 return err;
1227 }
1228
1229 /*
1230 * remove an extent from the root, returns 0 on success
1231 */
1232 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1233 *root, u64 bytenr, u64 num_bytes,
1234 u64 root_objectid, u64 ref_generation,
1235 u64 owner_objectid, u64 owner_offset, int pin)
1236 {
1237 struct btrfs_root *extent_root = root->fs_info->extent_root;
1238 int pending_ret;
1239 int ret;
1240
1241 WARN_ON(num_bytes < root->sectorsize);
1242 if (!root->ref_cows)
1243 ref_generation = 0;
1244
1245 if (root == extent_root) {
1246 pin_down_bytes(root, bytenr, num_bytes, 1);
1247 return 0;
1248 }
1249 ret = __free_extent(trans, root, bytenr, num_bytes, root_objectid,
1250 ref_generation, owner_objectid, owner_offset,
1251 pin, pin == 0);
1252 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1253 return ret ? ret : pending_ret;
1254 }
1255
1256 static u64 stripe_align(struct btrfs_root *root, u64 val)
1257 {
1258 u64 mask = ((u64)root->stripesize - 1);
1259 u64 ret = (val + mask) & ~mask;
1260 return ret;
1261 }
1262
1263 /*
1264 * walks the btree of allocated extents and find a hole of a given size.
1265 * The key ins is changed to record the hole:
1266 * ins->objectid == block start
1267 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1268 * ins->offset == number of blocks
1269 * Any available blocks before search_start are skipped.
1270 */
1271 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1272 *orig_root, u64 num_bytes, u64 empty_size,
1273 u64 search_start, u64 search_end, u64 hint_byte,
1274 struct btrfs_key *ins, u64 exclude_start,
1275 u64 exclude_nr, int data)
1276 {
1277 struct btrfs_path *path;
1278 struct btrfs_key key;
1279 u64 hole_size = 0;
1280 u64 aligned;
1281 int ret;
1282 int slot = 0;
1283 u64 last_byte = 0;
1284 u64 orig_search_start = search_start;
1285 int start_found;
1286 struct extent_buffer *l;
1287 struct btrfs_root * root = orig_root->fs_info->extent_root;
1288 struct btrfs_fs_info *info = root->fs_info;
1289 u64 total_needed = num_bytes;
1290 int level;
1291 struct btrfs_block_group_cache *block_group;
1292 int full_scan = 0;
1293 int wrapped = 0;
1294 u64 cached_start;
1295
1296 WARN_ON(num_bytes < root->sectorsize);
1297 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1298
1299 level = btrfs_header_level(root->node);
1300
1301 if (num_bytes >= 32 * 1024 * 1024 && hint_byte) {
1302 data = BTRFS_BLOCK_GROUP_MIXED;
1303 }
1304
1305 if (search_end == (u64)-1)
1306 search_end = btrfs_super_total_bytes(&info->super_copy);
1307 if (hint_byte) {
1308 block_group = btrfs_lookup_block_group(info, hint_byte);
1309 if (!block_group)
1310 hint_byte = search_start;
1311 block_group = btrfs_find_block_group(root, block_group,
1312 hint_byte, data, 1);
1313 } else {
1314 block_group = btrfs_find_block_group(root,
1315 trans->block_group,
1316 search_start, data, 1);
1317 }
1318
1319 total_needed += empty_size;
1320 path = btrfs_alloc_path();
1321 check_failed:
1322 search_start = find_search_start(root, &block_group, search_start,
1323 total_needed, data, full_scan);
1324 search_start = stripe_align(root, search_start);
1325 cached_start = search_start;
1326 btrfs_init_path(path);
1327 ins->objectid = search_start;
1328 ins->offset = 0;
1329 start_found = 0;
1330 path->reada = 2;
1331
1332 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1333 if (ret < 0)
1334 goto error;
1335
1336 if (path->slots[0] > 0) {
1337 path->slots[0]--;
1338 }
1339
1340 l = path->nodes[0];
1341 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1342
1343 /*
1344 * walk backwards to find the first extent item key
1345 */
1346 while(btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1347 if (path->slots[0] == 0) {
1348 ret = btrfs_prev_leaf(root, path);
1349 if (ret != 0) {
1350 ret = btrfs_search_slot(trans, root, ins,
1351 path, 0, 0);
1352 if (ret < 0)
1353 goto error;
1354 if (path->slots[0] > 0)
1355 path->slots[0]--;
1356 break;
1357 }
1358 } else {
1359 path->slots[0]--;
1360 }
1361 l = path->nodes[0];
1362 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1363 }
1364 while (1) {
1365 l = path->nodes[0];
1366 slot = path->slots[0];
1367 if (slot >= btrfs_header_nritems(l)) {
1368 ret = btrfs_next_leaf(root, path);
1369 if (ret == 0)
1370 continue;
1371 if (ret < 0)
1372 goto error;
1373
1374 search_start = max(search_start,
1375 block_group->key.objectid);
1376 if (!start_found) {
1377 aligned = stripe_align(root, search_start);
1378 ins->objectid = aligned;
1379 if (aligned >= search_end) {
1380 ret = -ENOSPC;
1381 goto error;
1382 }
1383 ins->offset = search_end - aligned;
1384 start_found = 1;
1385 goto check_pending;
1386 }
1387 ins->objectid = stripe_align(root,
1388 last_byte > search_start ?
1389 last_byte : search_start);
1390 if (search_end <= ins->objectid) {
1391 ret = -ENOSPC;
1392 goto error;
1393 }
1394 ins->offset = search_end - ins->objectid;
1395 BUG_ON(ins->objectid >= search_end);
1396 goto check_pending;
1397 }
1398 btrfs_item_key_to_cpu(l, &key, slot);
1399
1400 if (key.objectid >= search_start && key.objectid > last_byte &&
1401 start_found) {
1402 if (last_byte < search_start)
1403 last_byte = search_start;
1404 aligned = stripe_align(root, last_byte);
1405 hole_size = key.objectid - aligned;
1406 if (key.objectid > aligned && hole_size >= num_bytes) {
1407 ins->objectid = aligned;
1408 ins->offset = hole_size;
1409 goto check_pending;
1410 }
1411 }
1412 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) {
1413 if (!start_found && btrfs_key_type(&key) ==
1414 BTRFS_BLOCK_GROUP_ITEM_KEY) {
1415 last_byte = key.objectid;
1416 start_found = 1;
1417 }
1418 goto next;
1419 }
1420
1421
1422 start_found = 1;
1423 last_byte = key.objectid + key.offset;
1424
1425 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1426 last_byte >= block_group->key.objectid +
1427 block_group->key.offset) {
1428 btrfs_release_path(root, path);
1429 search_start = block_group->key.objectid +
1430 block_group->key.offset;
1431 goto new_group;
1432 }
1433 next:
1434 path->slots[0]++;
1435 cond_resched();
1436 }
1437 check_pending:
1438 /* we have to make sure we didn't find an extent that has already
1439 * been allocated by the map tree or the original allocation
1440 */
1441 btrfs_release_path(root, path);
1442 BUG_ON(ins->objectid < search_start);
1443
1444 if (ins->objectid + num_bytes >= search_end)
1445 goto enospc;
1446 if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED &&
1447 ins->objectid + num_bytes > block_group->
1448 key.objectid + block_group->key.offset) {
1449 search_start = block_group->key.objectid +
1450 block_group->key.offset;
1451 goto new_group;
1452 }
1453 if (test_range_bit(&info->extent_ins, ins->objectid,
1454 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
1455 search_start = ins->objectid + num_bytes;
1456 goto new_group;
1457 }
1458 if (test_range_bit(&info->pinned_extents, ins->objectid,
1459 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
1460 search_start = ins->objectid + num_bytes;
1461 goto new_group;
1462 }
1463 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
1464 ins->objectid < exclude_start + exclude_nr)) {
1465 search_start = exclude_start + exclude_nr;
1466 goto new_group;
1467 }
1468 if (!data) {
1469 block_group = btrfs_lookup_block_group(info, ins->objectid);
1470 if (block_group)
1471 trans->block_group = block_group;
1472 }
1473 ins->offset = num_bytes;
1474 btrfs_free_path(path);
1475 return 0;
1476
1477 new_group:
1478 if (search_start + num_bytes >= search_end) {
1479 enospc:
1480 search_start = orig_search_start;
1481 if (full_scan) {
1482 ret = -ENOSPC;
1483 goto error;
1484 }
1485 if (wrapped) {
1486 if (!full_scan)
1487 total_needed -= empty_size;
1488 full_scan = 1;
1489 data = BTRFS_BLOCK_GROUP_MIXED;
1490 } else
1491 wrapped = 1;
1492 }
1493 block_group = btrfs_lookup_block_group(info, search_start);
1494 cond_resched();
1495 block_group = btrfs_find_block_group(root, block_group,
1496 search_start, data, 0);
1497 goto check_failed;
1498
1499 error:
1500 btrfs_release_path(root, path);
1501 btrfs_free_path(path);
1502 return ret;
1503 }
1504 /*
1505 * finds a free extent and does all the dirty work required for allocation
1506 * returns the key for the extent through ins, and a tree buffer for
1507 * the first block of the extent through buf.
1508 *
1509 * returns 0 if everything worked, non-zero otherwise.
1510 */
1511 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1512 struct btrfs_root *root,
1513 u64 num_bytes, u64 root_objectid, u64 ref_generation,
1514 u64 owner, u64 owner_offset,
1515 u64 empty_size, u64 hint_byte,
1516 u64 search_end, struct btrfs_key *ins, int data)
1517 {
1518 int ret;
1519 int pending_ret;
1520 u64 super_used, root_used;
1521 u64 search_start = 0;
1522 struct btrfs_fs_info *info = root->fs_info;
1523 struct btrfs_root *extent_root = info->extent_root;
1524 struct btrfs_extent_item extent_item;
1525 struct btrfs_path *path;
1526
1527 btrfs_set_stack_extent_refs(&extent_item, 1);
1528
1529 WARN_ON(num_bytes < root->sectorsize);
1530 ret = find_free_extent(trans, root, num_bytes, empty_size,
1531 search_start, search_end, hint_byte, ins,
1532 trans->alloc_exclude_start,
1533 trans->alloc_exclude_nr, data);
1534 BUG_ON(ret);
1535 if (ret)
1536 return ret;
1537
1538 /* block accounting for super block */
1539 super_used = btrfs_super_bytes_used(&info->super_copy);
1540 btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes);
1541
1542 /* block accounting for root item */
1543 root_used = btrfs_root_used(&root->root_item);
1544 btrfs_set_root_used(&root->root_item, root_used + num_bytes);
1545
1546 clear_extent_dirty(&root->fs_info->free_space_cache,
1547 ins->objectid, ins->objectid + ins->offset - 1,
1548 GFP_NOFS);
1549
1550 if (root == extent_root) {
1551 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
1552 ins->objectid + ins->offset - 1,
1553 EXTENT_LOCKED, GFP_NOFS);
1554 WARN_ON(data == 1);
1555 goto update_block;
1556 }
1557
1558 WARN_ON(trans->alloc_exclude_nr);
1559 trans->alloc_exclude_start = ins->objectid;
1560 trans->alloc_exclude_nr = ins->offset;
1561 ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1562 sizeof(extent_item));
1563
1564 trans->alloc_exclude_start = 0;
1565 trans->alloc_exclude_nr = 0;
1566 BUG_ON(ret);
1567
1568 path = btrfs_alloc_path();
1569 BUG_ON(!path);
1570 ret = btrfs_insert_extent_backref(trans, extent_root, path,
1571 ins->objectid, root_objectid,
1572 ref_generation, owner, owner_offset);
1573
1574 BUG_ON(ret);
1575 btrfs_free_path(path);
1576 finish_current_insert(trans, extent_root);
1577 pending_ret = del_pending_extents(trans, extent_root);
1578
1579 if (ret) {
1580 return ret;
1581 }
1582 if (pending_ret) {
1583 return pending_ret;
1584 }
1585
1586 update_block:
1587 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1588 data);
1589 BUG_ON(ret);
1590 return 0;
1591 }
1592
1593 /*
1594 * helper function to allocate a block for a given tree
1595 * returns the tree buffer or NULL.
1596 */
1597 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1598 struct btrfs_root *root,
1599 u32 blocksize,
1600 u64 root_objectid, u64 hint,
1601 u64 empty_size)
1602 {
1603 u64 ref_generation;
1604
1605 if (root->ref_cows)
1606 ref_generation = trans->transid;
1607 else
1608 ref_generation = 0;
1609
1610
1611 return __btrfs_alloc_free_block(trans, root, blocksize, root_objectid,
1612 ref_generation, 0, 0, hint, empty_size);
1613 }
1614
1615 /*
1616 * helper function to allocate a block for a given tree
1617 * returns the tree buffer or NULL.
1618 */
1619 struct extent_buffer *__btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1620 struct btrfs_root *root,
1621 u32 blocksize,
1622 u64 root_objectid,
1623 u64 ref_generation,
1624 u64 first_objectid,
1625 int level,
1626 u64 hint,
1627 u64 empty_size)
1628 {
1629 struct btrfs_key ins;
1630 int ret;
1631 struct extent_buffer *buf;
1632
1633 ret = btrfs_alloc_extent(trans, root, blocksize,
1634 root_objectid, ref_generation,
1635 first_objectid, level, empty_size, hint,
1636 (u64)-1, &ins, 0);
1637 if (ret) {
1638 BUG_ON(ret > 0);
1639 return ERR_PTR(ret);
1640 }
1641 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
1642 if (!buf) {
1643 btrfs_free_extent(trans, root, ins.objectid, blocksize,
1644 root->root_key.objectid, ref_generation,
1645 0, 0, 0);
1646 return ERR_PTR(-ENOMEM);
1647 }
1648 btrfs_set_buffer_uptodate(buf);
1649 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
1650 buf->start + buf->len - 1, GFP_NOFS);
1651 set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->extent_tree,
1652 buf->start, buf->start + buf->len - 1,
1653 EXTENT_CSUM, GFP_NOFS);
1654 buf->flags |= EXTENT_CSUM;
1655 btrfs_set_buffer_defrag(buf);
1656 trans->blocks_used++;
1657 return buf;
1658 }
1659
1660 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1661 struct btrfs_root *root, struct extent_buffer *leaf)
1662 {
1663 u64 leaf_owner;
1664 u64 leaf_generation;
1665 struct btrfs_key key;
1666 struct btrfs_file_extent_item *fi;
1667 int i;
1668 int nritems;
1669 int ret;
1670
1671 BUG_ON(!btrfs_is_leaf(leaf));
1672 nritems = btrfs_header_nritems(leaf);
1673 leaf_owner = btrfs_header_owner(leaf);
1674 leaf_generation = btrfs_header_generation(leaf);
1675
1676 for (i = 0; i < nritems; i++) {
1677 u64 disk_bytenr;
1678
1679 btrfs_item_key_to_cpu(leaf, &key, i);
1680 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1681 continue;
1682 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1683 if (btrfs_file_extent_type(leaf, fi) ==
1684 BTRFS_FILE_EXTENT_INLINE)
1685 continue;
1686 /*
1687 * FIXME make sure to insert a trans record that
1688 * repeats the snapshot del on crash
1689 */
1690 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1691 if (disk_bytenr == 0)
1692 continue;
1693 ret = btrfs_free_extent(trans, root, disk_bytenr,
1694 btrfs_file_extent_disk_num_bytes(leaf, fi),
1695 leaf_owner, leaf_generation,
1696 key.objectid, key.offset, 0);
1697 BUG_ON(ret);
1698 }
1699 return 0;
1700 }
1701
1702 static void reada_walk_down(struct btrfs_root *root,
1703 struct extent_buffer *node)
1704 {
1705 int i;
1706 u32 nritems;
1707 u64 bytenr;
1708 int ret;
1709 u32 refs;
1710 int level;
1711 u32 blocksize;
1712
1713 nritems = btrfs_header_nritems(node);
1714 level = btrfs_header_level(node);
1715 for (i = 0; i < nritems; i++) {
1716 bytenr = btrfs_node_blockptr(node, i);
1717 blocksize = btrfs_level_size(root, level - 1);
1718 ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs);
1719 BUG_ON(ret);
1720 if (refs != 1)
1721 continue;
1722 mutex_unlock(&root->fs_info->fs_mutex);
1723 ret = readahead_tree_block(root, bytenr, blocksize);
1724 cond_resched();
1725 mutex_lock(&root->fs_info->fs_mutex);
1726 if (ret)
1727 break;
1728 }
1729 }
1730
1731 /*
1732 * helper function for drop_snapshot, this walks down the tree dropping ref
1733 * counts as it goes.
1734 */
1735 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1736 *root, struct btrfs_path *path, int *level)
1737 {
1738 u64 root_owner;
1739 u64 root_gen;
1740 u64 bytenr;
1741 struct extent_buffer *next;
1742 struct extent_buffer *cur;
1743 struct extent_buffer *parent;
1744 u32 blocksize;
1745 int ret;
1746 u32 refs;
1747
1748 WARN_ON(*level < 0);
1749 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1750 ret = lookup_extent_ref(trans, root,
1751 path->nodes[*level]->start,
1752 path->nodes[*level]->len, &refs);
1753 BUG_ON(ret);
1754 if (refs > 1)
1755 goto out;
1756
1757 /*
1758 * walk down to the last node level and free all the leaves
1759 */
1760 while(*level >= 0) {
1761 WARN_ON(*level < 0);
1762 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1763 cur = path->nodes[*level];
1764
1765 if (*level > 0 && path->slots[*level] == 0)
1766 reada_walk_down(root, cur);
1767
1768 if (btrfs_header_level(cur) != *level)
1769 WARN_ON(1);
1770
1771 if (path->slots[*level] >=
1772 btrfs_header_nritems(cur))
1773 break;
1774 if (*level == 0) {
1775 ret = drop_leaf_ref(trans, root, cur);
1776 BUG_ON(ret);
1777 break;
1778 }
1779 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
1780 blocksize = btrfs_level_size(root, *level - 1);
1781 ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs);
1782 BUG_ON(ret);
1783 if (refs != 1) {
1784 parent = path->nodes[*level];
1785 root_owner = btrfs_header_owner(parent);
1786 root_gen = btrfs_header_generation(parent);
1787 path->slots[*level]++;
1788 ret = btrfs_free_extent(trans, root, bytenr,
1789 blocksize, root_owner,
1790 root_gen, 0, 0, 1);
1791 BUG_ON(ret);
1792 continue;
1793 }
1794 next = btrfs_find_tree_block(root, bytenr, blocksize);
1795 if (!next || !btrfs_buffer_uptodate(next)) {
1796 free_extent_buffer(next);
1797 mutex_unlock(&root->fs_info->fs_mutex);
1798 next = read_tree_block(root, bytenr, blocksize);
1799 mutex_lock(&root->fs_info->fs_mutex);
1800
1801 /* we dropped the lock, check one more time */
1802 ret = lookup_extent_ref(trans, root, bytenr,
1803 blocksize, &refs);
1804 BUG_ON(ret);
1805 if (refs != 1) {
1806 parent = path->nodes[*level];
1807 root_owner = btrfs_header_owner(parent);
1808 root_gen = btrfs_header_generation(parent);
1809
1810 path->slots[*level]++;
1811 free_extent_buffer(next);
1812 ret = btrfs_free_extent(trans, root, bytenr,
1813 blocksize,
1814 root_owner,
1815 root_gen, 0, 0, 1);
1816 BUG_ON(ret);
1817 continue;
1818 }
1819 }
1820 WARN_ON(*level <= 0);
1821 if (path->nodes[*level-1])
1822 free_extent_buffer(path->nodes[*level-1]);
1823 path->nodes[*level-1] = next;
1824 *level = btrfs_header_level(next);
1825 path->slots[*level] = 0;
1826 }
1827 out:
1828 WARN_ON(*level < 0);
1829 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1830
1831 if (path->nodes[*level] == root->node) {
1832 root_owner = root->root_key.objectid;
1833 parent = path->nodes[*level];
1834 } else {
1835 parent = path->nodes[*level + 1];
1836 root_owner = btrfs_header_owner(parent);
1837 }
1838
1839 root_gen = btrfs_header_generation(parent);
1840 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
1841 path->nodes[*level]->len,
1842 root_owner, root_gen, 0, 0, 1);
1843 free_extent_buffer(path->nodes[*level]);
1844 path->nodes[*level] = NULL;
1845 *level += 1;
1846 BUG_ON(ret);
1847 return 0;
1848 }
1849
1850 /*
1851 * helper for dropping snapshots. This walks back up the tree in the path
1852 * to find the first node higher up where we haven't yet gone through
1853 * all the slots
1854 */
1855 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1856 *root, struct btrfs_path *path, int *level)
1857 {
1858 u64 root_owner;
1859 u64 root_gen;
1860 struct btrfs_root_item *root_item = &root->root_item;
1861 int i;
1862 int slot;
1863 int ret;
1864
1865 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1866 slot = path->slots[i];
1867 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
1868 struct extent_buffer *node;
1869 struct btrfs_disk_key disk_key;
1870 node = path->nodes[i];
1871 path->slots[i]++;
1872 *level = i;
1873 WARN_ON(*level == 0);
1874 btrfs_node_key(node, &disk_key, path->slots[i]);
1875 memcpy(&root_item->drop_progress,
1876 &disk_key, sizeof(disk_key));
1877 root_item->drop_level = i;
1878 return 0;
1879 } else {
1880 if (path->nodes[*level] == root->node) {
1881 root_owner = root->root_key.objectid;
1882 root_gen =
1883 btrfs_header_generation(path->nodes[*level]);
1884 } else {
1885 struct extent_buffer *node;
1886 node = path->nodes[*level + 1];
1887 root_owner = btrfs_header_owner(node);
1888 root_gen = btrfs_header_generation(node);
1889 }
1890 ret = btrfs_free_extent(trans, root,
1891 path->nodes[*level]->start,
1892 path->nodes[*level]->len,
1893 root_owner, root_gen, 0, 0, 1);
1894 BUG_ON(ret);
1895 free_extent_buffer(path->nodes[*level]);
1896 path->nodes[*level] = NULL;
1897 *level = i + 1;
1898 }
1899 }
1900 return 1;
1901 }
1902
1903 /*
1904 * drop the reference count on the tree rooted at 'snap'. This traverses
1905 * the tree freeing any blocks that have a ref count of zero after being
1906 * decremented.
1907 */
1908 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1909 *root)
1910 {
1911 int ret = 0;
1912 int wret;
1913 int level;
1914 struct btrfs_path *path;
1915 int i;
1916 int orig_level;
1917 struct btrfs_root_item *root_item = &root->root_item;
1918
1919 path = btrfs_alloc_path();
1920 BUG_ON(!path);
1921
1922 level = btrfs_header_level(root->node);
1923 orig_level = level;
1924 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1925 path->nodes[level] = root->node;
1926 extent_buffer_get(root->node);
1927 path->slots[level] = 0;
1928 } else {
1929 struct btrfs_key key;
1930 struct btrfs_disk_key found_key;
1931 struct extent_buffer *node;
1932
1933 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1934 level = root_item->drop_level;
1935 path->lowest_level = level;
1936 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1937 if (wret < 0) {
1938 ret = wret;
1939 goto out;
1940 }
1941 node = path->nodes[level];
1942 btrfs_node_key(node, &found_key, path->slots[level]);
1943 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
1944 sizeof(found_key)));
1945 }
1946 while(1) {
1947 wret = walk_down_tree(trans, root, path, &level);
1948 if (wret > 0)
1949 break;
1950 if (wret < 0)
1951 ret = wret;
1952
1953 wret = walk_up_tree(trans, root, path, &level);
1954 if (wret > 0)
1955 break;
1956 if (wret < 0)
1957 ret = wret;
1958 ret = -EAGAIN;
1959 break;
1960 }
1961 for (i = 0; i <= orig_level; i++) {
1962 if (path->nodes[i]) {
1963 free_extent_buffer(path->nodes[i]);
1964 path->nodes[i] = NULL;
1965 }
1966 }
1967 out:
1968 btrfs_free_path(path);
1969 return ret;
1970 }
1971
1972 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1973 {
1974 u64 start;
1975 u64 end;
1976 u64 ptr;
1977 int ret;
1978 while(1) {
1979 ret = find_first_extent_bit(&info->block_group_cache, 0,
1980 &start, &end, (unsigned int)-1);
1981 if (ret)
1982 break;
1983 ret = get_state_private(&info->block_group_cache, start, &ptr);
1984 if (!ret)
1985 kfree((void *)(unsigned long)ptr);
1986 clear_extent_bits(&info->block_group_cache, start,
1987 end, (unsigned int)-1, GFP_NOFS);
1988 }
1989 while(1) {
1990 ret = find_first_extent_bit(&info->free_space_cache, 0,
1991 &start, &end, EXTENT_DIRTY);
1992 if (ret)
1993 break;
1994 clear_extent_dirty(&info->free_space_cache, start,
1995 end, GFP_NOFS);
1996 }
1997 return 0;
1998 }
1999
2000 int btrfs_read_block_groups(struct btrfs_root *root)
2001 {
2002 struct btrfs_path *path;
2003 int ret;
2004 int err = 0;
2005 int bit;
2006 struct btrfs_block_group_cache *cache;
2007 struct btrfs_fs_info *info = root->fs_info;
2008 struct extent_map_tree *block_group_cache;
2009 struct btrfs_key key;
2010 struct btrfs_key found_key;
2011 struct extent_buffer *leaf;
2012
2013 block_group_cache = &info->block_group_cache;
2014
2015 root = info->extent_root;
2016 key.objectid = 0;
2017 key.offset = BTRFS_BLOCK_GROUP_SIZE;
2018 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
2019
2020 path = btrfs_alloc_path();
2021 if (!path)
2022 return -ENOMEM;
2023
2024 while(1) {
2025 ret = btrfs_search_slot(NULL, info->extent_root,
2026 &key, path, 0, 0);
2027 if (ret != 0) {
2028 err = ret;
2029 break;
2030 }
2031 leaf = path->nodes[0];
2032 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2033 cache = kmalloc(sizeof(*cache), GFP_NOFS);
2034 if (!cache) {
2035 err = -1;
2036 break;
2037 }
2038
2039 read_extent_buffer(leaf, &cache->item,
2040 btrfs_item_ptr_offset(leaf, path->slots[0]),
2041 sizeof(cache->item));
2042 memcpy(&cache->key, &found_key, sizeof(found_key));
2043 cache->cached = 0;
2044 cache->pinned = 0;
2045 key.objectid = found_key.objectid + found_key.offset;
2046 btrfs_release_path(root, path);
2047
2048 if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) {
2049 bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA;
2050 cache->data = BTRFS_BLOCK_GROUP_MIXED;
2051 } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) {
2052 bit = BLOCK_GROUP_DATA;
2053 cache->data = BTRFS_BLOCK_GROUP_DATA;
2054 } else {
2055 bit = BLOCK_GROUP_METADATA;
2056 cache->data = 0;
2057 }
2058
2059 /* use EXTENT_LOCKED to prevent merging */
2060 set_extent_bits(block_group_cache, found_key.objectid,
2061 found_key.objectid + found_key.offset - 1,
2062 bit | EXTENT_LOCKED, GFP_NOFS);
2063 set_state_private(block_group_cache, found_key.objectid,
2064 (unsigned long)cache);
2065
2066 if (key.objectid >=
2067 btrfs_super_total_bytes(&info->super_copy))
2068 break;
2069 }
2070
2071 btrfs_free_path(path);
2072 return 0;
2073 }
Linux kernel - Deliverables
This page took 0.082403 seconds and 5 git commands to generate.