]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/extent-tree.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:


14b93268920e7fa4746880ef375a7ac0719492e9
[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/module.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "print-tree.h"
23 #include "transaction.h"
24
25 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
26 *orig_root, u64 num_blocks, u64 search_start,
27 u64 search_end, u64 hint_block,
28 struct btrfs_key *ins, u64 exclude_start,
29 u64 exclude_nr, int data);
30 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
31 btrfs_root *extent_root);
32 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
33 btrfs_root *extent_root);
34
35 static void reada_extent_leaves(struct btrfs_root *root,
36 struct btrfs_path *path, u64 limit)
37 {
38 struct btrfs_node *node;
39 int i;
40 int nritems;
41 u64 item_objectid;
42 u64 blocknr;
43 int slot;
44 int ret;
45
46 if (!path->nodes[1])
47 return;
48 node = btrfs_buffer_node(path->nodes[1]);
49 slot = path->slots[1] + 1;
50 nritems = btrfs_header_nritems(&node->header);
51 for (i = slot; i < nritems && i < slot + 8; i++) {
52 item_objectid = btrfs_disk_key_objectid(&node->ptrs[i].key);
53 if (item_objectid > limit)
54 break;
55 blocknr = btrfs_node_blockptr(node, i);
56 ret = readahead_tree_block(root, blocknr);
57 if (ret)
58 break;
59 }
60 }
61
62 static int cache_block_group(struct btrfs_root *root,
63 struct btrfs_block_group_cache *block_group)
64 {
65 struct btrfs_path *path;
66 int ret;
67 struct btrfs_key key;
68 struct btrfs_leaf *leaf;
69 struct radix_tree_root *extent_radix;
70 int slot;
71 u64 i;
72 u64 last = 0;
73 u64 hole_size;
74 u64 limit;
75 int found = 0;
76
77 root = root->fs_info->extent_root;
78 extent_radix = &root->fs_info->extent_map_radix;
79
80 if (block_group->cached)
81 return 0;
82 if (block_group->data)
83 return 0;
84 path = btrfs_alloc_path();
85 if (!path)
86 return -ENOMEM;
87 key.objectid = block_group->key.objectid;
88 key.flags = 0;
89 key.offset = 0;
90 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
91 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
92 if (ret < 0)
93 return ret;
94 if (ret && path->slots[0] > 0)
95 path->slots[0]--;
96 limit = block_group->key.objectid + block_group->key.offset;
97 reada_extent_leaves(root, path, limit);
98 while(1) {
99 leaf = btrfs_buffer_leaf(path->nodes[0]);
100 slot = path->slots[0];
101 if (slot >= btrfs_header_nritems(&leaf->header)) {
102 reada_extent_leaves(root, path, limit);
103 ret = btrfs_next_leaf(root, path);
104 if (ret < 0)
105 goto err;
106 if (ret == 0) {
107 continue;
108 } else {
109 if (found) {
110 hole_size = block_group->key.objectid +
111 block_group->key.offset - last;
112 } else {
113 last = block_group->key.objectid;
114 hole_size = block_group->key.offset;
115 }
116 for (i = 0; i < hole_size; i++) {
117 set_radix_bit(extent_radix,
118 last + i);
119 }
120 break;
121 }
122 }
123 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
124 if (key.objectid >= block_group->key.objectid +
125 block_group->key.offset) {
126 if (found) {
127 hole_size = block_group->key.objectid +
128 block_group->key.offset - last;
129 } else {
130 last = block_group->key.objectid;
131 hole_size = block_group->key.offset;
132 }
133 for (i = 0; i < hole_size; i++) {
134 set_radix_bit(extent_radix, last + i);
135 }
136 break;
137 }
138 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
139 if (!found) {
140 last = key.objectid + key.offset;
141 found = 1;
142 } else {
143 hole_size = key.objectid - last;
144 for (i = 0; i < hole_size; i++) {
145 set_radix_bit(extent_radix, last + i);
146 }
147 last = key.objectid + key.offset;
148 }
149 }
150 path->slots[0]++;
151 }
152
153 block_group->cached = 1;
154 err:
155 btrfs_free_path(path);
156 return 0;
157 }
158
159 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
160 btrfs_fs_info *info,
161 u64 blocknr)
162 {
163 struct btrfs_block_group_cache *block_group;
164 int ret;
165
166 ret = radix_tree_gang_lookup(&info->block_group_radix,
167 (void **)&block_group,
168 blocknr, 1);
169 if (ret) {
170 if (block_group->key.objectid <= blocknr && blocknr <=
171 block_group->key.objectid + block_group->key.offset)
172 return block_group;
173 }
174 ret = radix_tree_gang_lookup(&info->block_group_data_radix,
175 (void **)&block_group,
176 blocknr, 1);
177 if (ret) {
178 if (block_group->key.objectid <= blocknr && blocknr <=
179 block_group->key.objectid + block_group->key.offset)
180 return block_group;
181 }
182 return NULL;
183 }
184
185 static u64 leaf_range(struct btrfs_root *root)
186 {
187 u64 size = BTRFS_LEAF_DATA_SIZE(root);
188 do_div(size, sizeof(struct btrfs_extent_item) +
189 sizeof(struct btrfs_item));
190 return size;
191 }
192
193 static u64 find_search_start(struct btrfs_root *root,
194 struct btrfs_block_group_cache **cache_ret,
195 u64 search_start, int num)
196 {
197 unsigned long gang[8];
198 int ret;
199 struct btrfs_block_group_cache *cache = *cache_ret;
200 u64 last = max(search_start, cache->key.objectid);
201
202 if (cache->data)
203 goto out;
204 if (num > 1) {
205 last = max(last, cache->last_prealloc);
206 }
207 again:
208 ret = cache_block_group(root, cache);
209 if (ret)
210 goto out;
211 while(1) {
212 ret = find_first_radix_bit(&root->fs_info->extent_map_radix,
213 gang, last, ARRAY_SIZE(gang));
214 if (!ret)
215 goto out;
216 last = gang[ret-1] + 1;
217 if (num > 1) {
218 if (ret != ARRAY_SIZE(gang)) {
219 goto new_group;
220 }
221 if (gang[ret-1] - gang[0] > leaf_range(root)) {
222 continue;
223 }
224 }
225 if (gang[0] >= cache->key.objectid + cache->key.offset) {
226 goto new_group;
227 }
228 return gang[0];
229 }
230 out:
231 return max(cache->last_alloc, search_start);
232
233 new_group:
234 cache = btrfs_lookup_block_group(root->fs_info,
235 last + cache->key.offset - 1);
236 if (!cache) {
237 return max((*cache_ret)->last_alloc, search_start);
238 }
239 cache = btrfs_find_block_group(root, cache,
240 last + cache->key.offset - 1, 0, 0);
241 *cache_ret = cache;
242 goto again;
243 }
244
245 static u64 div_factor(u64 num, int factor)
246 {
247 num *= factor;
248 do_div(num, 10);
249 return num;
250 }
251
252 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
253 struct btrfs_block_group_cache
254 *hint, u64 search_start,
255 int data, int owner)
256 {
257 struct btrfs_block_group_cache *cache[8];
258 struct btrfs_block_group_cache *found_group = NULL;
259 struct btrfs_fs_info *info = root->fs_info;
260 struct radix_tree_root *radix;
261 struct radix_tree_root *swap_radix;
262 u64 used;
263 u64 last = 0;
264 u64 hint_last;
265 int i;
266 int ret;
267 int full_search = 0;
268 int factor = 8;
269 int data_swap = 0;
270
271 if (!owner)
272 factor = 5;
273
274 if (data) {
275 radix = &info->block_group_data_radix;
276 swap_radix = &info->block_group_radix;
277 } else {
278 radix = &info->block_group_radix;
279 swap_radix = &info->block_group_data_radix;
280 }
281
282 if (search_start) {
283 struct btrfs_block_group_cache *shint;
284 shint = btrfs_lookup_block_group(info, search_start);
285 if (shint->data == data) {
286 used = btrfs_block_group_used(&shint->item);
287 if (used + shint->pinned <
288 div_factor(shint->key.offset, factor)) {
289 return shint;
290 }
291 }
292 }
293 if (hint && hint->data == data) {
294 used = btrfs_block_group_used(&hint->item);
295 if (used + hint->pinned <
296 div_factor(hint->key.offset, factor)) {
297 return hint;
298 }
299 if (used >= div_factor(hint->key.offset, 8)) {
300 radix_tree_tag_clear(radix,
301 hint->key.objectid +
302 hint->key.offset - 1,
303 BTRFS_BLOCK_GROUP_AVAIL);
304 }
305 last = hint->key.offset * 3;
306 if (hint->key.objectid >= last)
307 last = max(search_start + hint->key.offset - 1,
308 hint->key.objectid - last);
309 else
310 last = hint->key.objectid + hint->key.offset;
311 hint_last = last;
312 } else {
313 if (hint)
314 hint_last = max(hint->key.objectid, search_start);
315 else
316 hint_last = search_start;
317
318 last = hint_last;
319 }
320 while(1) {
321 ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
322 last, ARRAY_SIZE(cache),
323 BTRFS_BLOCK_GROUP_AVAIL);
324 if (!ret)
325 break;
326 for (i = 0; i < ret; i++) {
327 last = cache[i]->key.objectid +
328 cache[i]->key.offset;
329 used = btrfs_block_group_used(&cache[i]->item);
330 if (used + cache[i]->pinned <
331 div_factor(cache[i]->key.offset, factor)) {
332 found_group = cache[i];
333 goto found;
334 }
335 if (used >= div_factor(cache[i]->key.offset, 8)) {
336 radix_tree_tag_clear(radix,
337 cache[i]->key.objectid +
338 cache[i]->key.offset - 1,
339 BTRFS_BLOCK_GROUP_AVAIL);
340 }
341 }
342 cond_resched();
343 }
344 last = hint_last;
345 again:
346 while(1) {
347 ret = radix_tree_gang_lookup(radix, (void **)cache,
348 last, ARRAY_SIZE(cache));
349 if (!ret)
350 break;
351 for (i = 0; i < ret; i++) {
352 last = cache[i]->key.objectid +
353 cache[i]->key.offset;
354 used = btrfs_block_group_used(&cache[i]->item);
355 if (used + cache[i]->pinned < cache[i]->key.offset) {
356 found_group = cache[i];
357 goto found;
358 }
359 if (used >= cache[i]->key.offset) {
360 radix_tree_tag_clear(radix,
361 cache[i]->key.objectid +
362 cache[i]->key.offset - 1,
363 BTRFS_BLOCK_GROUP_AVAIL);
364 }
365 }
366 cond_resched();
367 }
368 if (!full_search) {
369 last = search_start;
370 full_search = 1;
371 goto again;
372 }
373 if (!data_swap) {
374 struct radix_tree_root *tmp = radix;
375 data_swap = 1;
376 radix = swap_radix;
377 swap_radix = tmp;
378 last = search_start;
379 goto again;
380 }
381 if (!found_group) {
382 ret = radix_tree_gang_lookup(radix,
383 (void **)&found_group, 0, 1);
384 if (ret == 0) {
385 ret = radix_tree_gang_lookup(swap_radix,
386 (void **)&found_group,
387 0, 1);
388 }
389 BUG_ON(ret != 1);
390 }
391 found:
392 return found_group;
393 }
394
395 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
396 struct btrfs_root *root,
397 u64 blocknr, u64 num_blocks)
398 {
399 struct btrfs_path *path;
400 int ret;
401 struct btrfs_key key;
402 struct btrfs_leaf *l;
403 struct btrfs_extent_item *item;
404 struct btrfs_key ins;
405 u32 refs;
406
407 path = btrfs_alloc_path();
408 if (!path)
409 return -ENOMEM;
410 ret = find_free_extent(trans, root->fs_info->extent_root, 0, 0,
411 (u64)-1, 0, &ins, 0, 0, 0);
412 if (ret) {
413 btrfs_free_path(path);
414 return ret;
415 }
416 key.objectid = blocknr;
417 key.flags = 0;
418 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
419 key.offset = num_blocks;
420 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
421 0, 1);
422 if (ret < 0)
423 return ret;
424 if (ret != 0) {
425 BUG();
426 }
427 BUG_ON(ret != 0);
428 l = btrfs_buffer_leaf(path->nodes[0]);
429 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
430 refs = btrfs_extent_refs(item);
431 btrfs_set_extent_refs(item, refs + 1);
432 btrfs_mark_buffer_dirty(path->nodes[0]);
433
434 btrfs_release_path(root->fs_info->extent_root, path);
435 btrfs_free_path(path);
436 finish_current_insert(trans, root->fs_info->extent_root);
437 del_pending_extents(trans, root->fs_info->extent_root);
438 return 0;
439 }
440
441 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
442 struct btrfs_root *root, u64 blocknr,
443 u64 num_blocks, u32 *refs)
444 {
445 struct btrfs_path *path;
446 int ret;
447 struct btrfs_key key;
448 struct btrfs_leaf *l;
449 struct btrfs_extent_item *item;
450
451 path = btrfs_alloc_path();
452 key.objectid = blocknr;
453 key.offset = num_blocks;
454 key.flags = 0;
455 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
456 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
457 0, 0);
458 if (ret < 0)
459 goto out;
460 if (ret != 0)
461 BUG();
462 l = btrfs_buffer_leaf(path->nodes[0]);
463 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
464 *refs = btrfs_extent_refs(item);
465 out:
466 btrfs_free_path(path);
467 return 0;
468 }
469
470 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
471 struct btrfs_root *root)
472 {
473 return btrfs_inc_extent_ref(trans, root, bh_blocknr(root->node), 1);
474 }
475
476 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
477 struct buffer_head *buf)
478 {
479 u64 blocknr;
480 struct btrfs_node *buf_node;
481 struct btrfs_leaf *buf_leaf;
482 struct btrfs_disk_key *key;
483 struct btrfs_file_extent_item *fi;
484 int i;
485 int leaf;
486 int ret;
487 int faili;
488 int err;
489
490 if (!root->ref_cows)
491 return 0;
492 buf_node = btrfs_buffer_node(buf);
493 leaf = btrfs_is_leaf(buf_node);
494 buf_leaf = btrfs_buffer_leaf(buf);
495 for (i = 0; i < btrfs_header_nritems(&buf_node->header); i++) {
496 if (leaf) {
497 u64 disk_blocknr;
498 key = &buf_leaf->items[i].key;
499 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
500 continue;
501 fi = btrfs_item_ptr(buf_leaf, i,
502 struct btrfs_file_extent_item);
503 if (btrfs_file_extent_type(fi) ==
504 BTRFS_FILE_EXTENT_INLINE)
505 continue;
506 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
507 if (disk_blocknr == 0)
508 continue;
509 ret = btrfs_inc_extent_ref(trans, root, disk_blocknr,
510 btrfs_file_extent_disk_num_blocks(fi));
511 if (ret) {
512 faili = i;
513 goto fail;
514 }
515 } else {
516 blocknr = btrfs_node_blockptr(buf_node, i);
517 ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
518 if (ret) {
519 faili = i;
520 goto fail;
521 }
522 }
523 }
524 return 0;
525 fail:
526 WARN_ON(1);
527 for (i =0; i < faili; i++) {
528 if (leaf) {
529 u64 disk_blocknr;
530 key = &buf_leaf->items[i].key;
531 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
532 continue;
533 fi = btrfs_item_ptr(buf_leaf, i,
534 struct btrfs_file_extent_item);
535 if (btrfs_file_extent_type(fi) ==
536 BTRFS_FILE_EXTENT_INLINE)
537 continue;
538 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
539 if (disk_blocknr == 0)
540 continue;
541 err = btrfs_free_extent(trans, root, disk_blocknr,
542 btrfs_file_extent_disk_num_blocks(fi), 0);
543 BUG_ON(err);
544 } else {
545 blocknr = btrfs_node_blockptr(buf_node, i);
546 err = btrfs_free_extent(trans, root, blocknr, 1, 0);
547 BUG_ON(err);
548 }
549 }
550 return ret;
551 }
552
553 static int write_one_cache_group(struct btrfs_trans_handle *trans,
554 struct btrfs_root *root,
555 struct btrfs_path *path,
556 struct btrfs_block_group_cache *cache)
557 {
558 int ret;
559 int pending_ret;
560 struct btrfs_root *extent_root = root->fs_info->extent_root;
561 struct btrfs_block_group_item *bi;
562 struct btrfs_key ins;
563
564 ret = find_free_extent(trans, extent_root, 0, 0, (u64)-1, 0, &ins,
565 0, 0, 0);
566 /* FIXME, set bit to recalc cache groups on next mount */
567 if (ret)
568 return ret;
569 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
570 if (ret < 0)
571 goto fail;
572 BUG_ON(ret);
573 bi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
574 struct btrfs_block_group_item);
575 memcpy(bi, &cache->item, sizeof(*bi));
576 btrfs_mark_buffer_dirty(path->nodes[0]);
577 btrfs_release_path(extent_root, path);
578 fail:
579 finish_current_insert(trans, extent_root);
580 pending_ret = del_pending_extents(trans, extent_root);
581 if (ret)
582 return ret;
583 if (pending_ret)
584 return pending_ret;
585 if (cache->data)
586 cache->last_alloc = cache->first_free;
587 return 0;
588
589 }
590
591 static int write_dirty_block_radix(struct btrfs_trans_handle *trans,
592 struct btrfs_root *root,
593 struct radix_tree_root *radix)
594 {
595 struct btrfs_block_group_cache *cache[8];
596 int ret;
597 int err = 0;
598 int werr = 0;
599 int i;
600 struct btrfs_path *path;
601 unsigned long off = 0;
602
603 path = btrfs_alloc_path();
604 if (!path)
605 return -ENOMEM;
606
607 while(1) {
608 ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
609 off, ARRAY_SIZE(cache),
610 BTRFS_BLOCK_GROUP_DIRTY);
611 if (!ret)
612 break;
613 for (i = 0; i < ret; i++) {
614 err = write_one_cache_group(trans, root,
615 path, cache[i]);
616 /*
617 * if we fail to write the cache group, we want
618 * to keep it marked dirty in hopes that a later
619 * write will work
620 */
621 if (err) {
622 werr = err;
623 off = cache[i]->key.objectid +
624 cache[i]->key.offset;
625 continue;
626 }
627
628 radix_tree_tag_clear(radix, cache[i]->key.objectid +
629 cache[i]->key.offset - 1,
630 BTRFS_BLOCK_GROUP_DIRTY);
631 }
632 }
633 btrfs_free_path(path);
634 return werr;
635 }
636
637 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
638 struct btrfs_root *root)
639 {
640 int ret;
641 int ret2;
642 ret = write_dirty_block_radix(trans, root,
643 &root->fs_info->block_group_radix);
644 ret2 = write_dirty_block_radix(trans, root,
645 &root->fs_info->block_group_data_radix);
646 if (ret)
647 return ret;
648 if (ret2)
649 return ret2;
650 return 0;
651 }
652
653 static int update_block_group(struct btrfs_trans_handle *trans,
654 struct btrfs_root *root,
655 u64 blocknr, u64 num, int alloc, int mark_free,
656 int data)
657 {
658 struct btrfs_block_group_cache *cache;
659 struct btrfs_fs_info *info = root->fs_info;
660 u64 total = num;
661 u64 old_val;
662 u64 block_in_group;
663 u64 i;
664 int ret;
665
666 while(total) {
667 cache = btrfs_lookup_block_group(info, blocknr);
668 if (!cache) {
669 return -1;
670 }
671 block_in_group = blocknr - cache->key.objectid;
672 WARN_ON(block_in_group > cache->key.offset);
673 radix_tree_tag_set(cache->radix, cache->key.objectid +
674 cache->key.offset - 1,
675 BTRFS_BLOCK_GROUP_DIRTY);
676
677 old_val = btrfs_block_group_used(&cache->item);
678 num = min(total, cache->key.offset - block_in_group);
679 if (alloc) {
680 if (blocknr > cache->last_alloc)
681 cache->last_alloc = blocknr;
682 if (!cache->data) {
683 for (i = 0; i < num; i++) {
684 clear_radix_bit(&info->extent_map_radix,
685 blocknr + i);
686 }
687 }
688 if (cache->data != data &&
689 old_val < (cache->key.offset >> 1)) {
690 cache->data = data;
691 radix_tree_delete(cache->radix,
692 cache->key.objectid +
693 cache->key.offset - 1);
694
695 if (data) {
696 cache->radix =
697 &info->block_group_data_radix;
698 cache->item.flags |=
699 BTRFS_BLOCK_GROUP_DATA;
700 } else {
701 cache->radix = &info->block_group_radix;
702 cache->item.flags &=
703 ~BTRFS_BLOCK_GROUP_DATA;
704 }
705 ret = radix_tree_insert(cache->radix,
706 cache->key.objectid +
707 cache->key.offset - 1,
708 (void *)cache);
709 }
710 old_val += num;
711 } else {
712 old_val -= num;
713 if (blocknr < cache->first_free)
714 cache->first_free = blocknr;
715 if (!cache->data && mark_free) {
716 for (i = 0; i < num; i++) {
717 set_radix_bit(&info->extent_map_radix,
718 blocknr + i);
719 }
720 }
721 if (old_val < (cache->key.offset >> 1) &&
722 old_val + num >= (cache->key.offset >> 1)) {
723 radix_tree_tag_set(cache->radix,
724 cache->key.objectid +
725 cache->key.offset - 1,
726 BTRFS_BLOCK_GROUP_AVAIL);
727 }
728 }
729 btrfs_set_block_group_used(&cache->item, old_val);
730 total -= num;
731 blocknr += num;
732 }
733 return 0;
734 }
735
736 static int try_remove_page(struct address_space *mapping, unsigned long index)
737 {
738 int ret;
739 ret = invalidate_mapping_pages(mapping, index, index);
740 return ret;
741 }
742
743 int btrfs_copy_pinned(struct btrfs_root *root, struct radix_tree_root *copy)
744 {
745 unsigned long gang[8];
746 u64 last = 0;
747 struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
748 int ret;
749 int i;
750
751 while(1) {
752 ret = find_first_radix_bit(pinned_radix, gang, last,
753 ARRAY_SIZE(gang));
754 if (!ret)
755 break;
756 for (i = 0 ; i < ret; i++) {
757 set_radix_bit(copy, gang[i]);
758 last = gang[i] + 1;
759 }
760 }
761 return 0;
762 }
763
764 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
765 struct btrfs_root *root,
766 struct radix_tree_root *unpin_radix)
767 {
768 unsigned long gang[8];
769 struct inode *btree_inode = root->fs_info->btree_inode;
770 struct btrfs_block_group_cache *block_group;
771 u64 first = 0;
772 int ret;
773 int i;
774 struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
775 struct radix_tree_root *extent_radix = &root->fs_info->extent_map_radix;
776
777 while(1) {
778 ret = find_first_radix_bit(unpin_radix, gang, 0,
779 ARRAY_SIZE(gang));
780 if (!ret)
781 break;
782 if (!first)
783 first = gang[0];
784 for (i = 0; i < ret; i++) {
785 clear_radix_bit(pinned_radix, gang[i]);
786 clear_radix_bit(unpin_radix, gang[i]);
787 block_group = btrfs_lookup_block_group(root->fs_info,
788 gang[i]);
789 if (block_group) {
790 WARN_ON(block_group->pinned == 0);
791 block_group->pinned--;
792 if (gang[i] < block_group->last_alloc)
793 block_group->last_alloc = gang[i];
794 if (gang[i] < block_group->last_prealloc)
795 block_group->last_prealloc = gang[i];
796 if (!block_group->data)
797 set_radix_bit(extent_radix, gang[i]);
798 }
799 try_remove_page(btree_inode->i_mapping,
800 gang[i] << (PAGE_CACHE_SHIFT -
801 btree_inode->i_blkbits));
802 }
803 }
804 return 0;
805 }
806
807 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
808 btrfs_root *extent_root)
809 {
810 struct btrfs_key ins;
811 struct btrfs_extent_item extent_item;
812 int i;
813 int ret;
814 u64 super_blocks_used;
815 struct btrfs_fs_info *info = extent_root->fs_info;
816
817 btrfs_set_extent_refs(&extent_item, 1);
818 ins.offset = 1;
819 ins.flags = 0;
820 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
821 btrfs_set_extent_owner(&extent_item, extent_root->root_key.objectid);
822
823 for (i = 0; i < extent_root->fs_info->extent_tree_insert_nr; i++) {
824 ins.objectid = extent_root->fs_info->extent_tree_insert[i];
825 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
826 btrfs_set_super_blocks_used(&info->super_copy,
827 super_blocks_used + 1);
828 ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item,
829 sizeof(extent_item));
830 BUG_ON(ret);
831 }
832 extent_root->fs_info->extent_tree_insert_nr = 0;
833 return 0;
834 }
835
836 static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
837 {
838 int err;
839 struct btrfs_header *header;
840 struct buffer_head *bh;
841
842 if (!pending) {
843 bh = btrfs_find_tree_block(root, blocknr);
844 if (bh) {
845 if (buffer_uptodate(bh)) {
846 u64 transid =
847 root->fs_info->running_transaction->transid;
848 header = btrfs_buffer_header(bh);
849 if (btrfs_header_generation(header) ==
850 transid) {
851 btrfs_block_release(root, bh);
852 return 0;
853 }
854 }
855 btrfs_block_release(root, bh);
856 }
857 err = set_radix_bit(&root->fs_info->pinned_radix, blocknr);
858 if (!err) {
859 struct btrfs_block_group_cache *cache;
860 cache = btrfs_lookup_block_group(root->fs_info,
861 blocknr);
862 if (cache)
863 cache->pinned++;
864 }
865 } else {
866 err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr);
867 }
868 BUG_ON(err < 0);
869 return 0;
870 }
871
872 /*
873 * remove an extent from the root, returns 0 on success
874 */
875 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
876 *root, u64 blocknr, u64 num_blocks, int pin,
877 int mark_free)
878 {
879 struct btrfs_path *path;
880 struct btrfs_key key;
881 struct btrfs_fs_info *info = root->fs_info;
882 struct btrfs_root *extent_root = info->extent_root;
883 int ret;
884 struct btrfs_extent_item *ei;
885 struct btrfs_key ins;
886 u32 refs;
887
888 key.objectid = blocknr;
889 key.flags = 0;
890 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
891 key.offset = num_blocks;
892
893 path = btrfs_alloc_path();
894 if (!path)
895 return -ENOMEM;
896
897 ret = find_free_extent(trans, root, 0, 0, (u64)-1, 0, &ins, 0, 0, 0);
898 if (ret) {
899 btrfs_free_path(path);
900 return ret;
901 }
902
903 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
904 if (ret < 0)
905 return ret;
906 BUG_ON(ret);
907 ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
908 struct btrfs_extent_item);
909 BUG_ON(ei->refs == 0);
910 refs = btrfs_extent_refs(ei) - 1;
911 btrfs_set_extent_refs(ei, refs);
912 btrfs_mark_buffer_dirty(path->nodes[0]);
913 if (refs == 0) {
914 u64 super_blocks_used;
915
916 if (pin) {
917 ret = pin_down_block(root, blocknr, 0);
918 BUG_ON(ret);
919 }
920
921 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
922 btrfs_set_super_blocks_used(&info->super_copy,
923 super_blocks_used - num_blocks);
924 ret = btrfs_del_item(trans, extent_root, path);
925 if (ret) {
926 return ret;
927 }
928 ret = update_block_group(trans, root, blocknr, num_blocks, 0,
929 mark_free, 0);
930 BUG_ON(ret);
931 }
932 btrfs_free_path(path);
933 finish_current_insert(trans, extent_root);
934 return ret;
935 }
936
937 /*
938 * find all the blocks marked as pending in the radix tree and remove
939 * them from the extent map
940 */
941 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
942 btrfs_root *extent_root)
943 {
944 int ret;
945 int wret;
946 int err = 0;
947 unsigned long gang[4];
948 int i;
949 struct radix_tree_root *pending_radix;
950 struct radix_tree_root *pinned_radix;
951 struct btrfs_block_group_cache *cache;
952
953 pending_radix = &extent_root->fs_info->pending_del_radix;
954 pinned_radix = &extent_root->fs_info->pinned_radix;
955
956 while(1) {
957 ret = find_first_radix_bit(pending_radix, gang, 0,
958 ARRAY_SIZE(gang));
959 if (!ret)
960 break;
961 for (i = 0; i < ret; i++) {
962 wret = set_radix_bit(pinned_radix, gang[i]);
963 if (wret == 0) {
964 cache =
965 btrfs_lookup_block_group(extent_root->fs_info,
966 gang[i]);
967 if (cache)
968 cache->pinned++;
969 }
970 if (wret < 0) {
971 printk(KERN_CRIT "set_radix_bit, err %d\n",
972 wret);
973 BUG_ON(wret < 0);
974 }
975 wret = clear_radix_bit(pending_radix, gang[i]);
976 BUG_ON(wret);
977 wret = __free_extent(trans, extent_root,
978 gang[i], 1, 0, 0);
979 if (wret)
980 err = wret;
981 }
982 }
983 return err;
984 }
985
986 /*
987 * remove an extent from the root, returns 0 on success
988 */
989 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
990 *root, u64 blocknr, u64 num_blocks, int pin)
991 {
992 struct btrfs_root *extent_root = root->fs_info->extent_root;
993 int pending_ret;
994 int ret;
995
996 if (root == extent_root) {
997 pin_down_block(root, blocknr, 1);
998 return 0;
999 }
1000 ret = __free_extent(trans, root, blocknr, num_blocks, pin, pin == 0);
1001 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
1002 return ret ? ret : pending_ret;
1003 }
1004
1005 /*
1006 * walks the btree of allocated extents and find a hole of a given size.
1007 * The key ins is changed to record the hole:
1008 * ins->objectid == block start
1009 * ins->flags = BTRFS_EXTENT_ITEM_KEY
1010 * ins->offset == number of blocks
1011 * Any available blocks before search_start are skipped.
1012 */
1013 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
1014 *orig_root, u64 num_blocks, u64 search_start, u64
1015 search_end, u64 hint_block,
1016 struct btrfs_key *ins, u64 exclude_start,
1017 u64 exclude_nr, int data)
1018 {
1019 struct btrfs_path *path;
1020 struct btrfs_key key;
1021 int ret;
1022 u64 hole_size = 0;
1023 int slot = 0;
1024 u64 last_block = 0;
1025 u64 test_block;
1026 u64 orig_search_start = search_start;
1027 int start_found;
1028 struct btrfs_leaf *l;
1029 struct btrfs_root * root = orig_root->fs_info->extent_root;
1030 struct btrfs_fs_info *info = root->fs_info;
1031 int total_needed = num_blocks;
1032 int total_found = 0;
1033 int fill_prealloc = 0;
1034 int level;
1035 struct btrfs_block_group_cache *block_group;
1036 int full_scan = 0;
1037 int wrapped = 0;
1038 u64 limit;
1039
1040 ins->flags = 0;
1041 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1042
1043 level = btrfs_header_level(btrfs_buffer_header(root->node));
1044 if (num_blocks == 0) {
1045 fill_prealloc = 1;
1046 num_blocks = 1;
1047 total_needed = (min(level + 1, BTRFS_MAX_LEVEL) + 2) * 3;
1048 }
1049 if (fill_prealloc) {
1050 u64 first;
1051 int nr = info->extent_tree_prealloc_nr;
1052 first = info->extent_tree_prealloc[nr - 1];
1053 if (info->extent_tree_prealloc_nr >= total_needed &&
1054 first >= search_start) {
1055 ins->objectid = info->extent_tree_prealloc[0];
1056 ins->offset = 1;
1057 return 0;
1058 }
1059 info->extent_tree_prealloc_nr = 0;
1060 }
1061 if (search_end == (u64)-1)
1062 search_end = btrfs_super_total_blocks(&info->super_copy);
1063 if (hint_block) {
1064 block_group = btrfs_lookup_block_group(info, hint_block);
1065 block_group = btrfs_find_block_group(root, block_group,
1066 hint_block, data, 1);
1067 } else {
1068 block_group = btrfs_find_block_group(root,
1069 trans->block_group, 0,
1070 data, 1);
1071 }
1072
1073 path = btrfs_alloc_path();
1074
1075 check_failed:
1076 if (!block_group->data)
1077 search_start = find_search_start(root, &block_group,
1078 search_start, total_needed);
1079 else if (!full_scan)
1080 search_start = max(block_group->last_alloc, search_start);
1081
1082 btrfs_init_path(path);
1083 ins->objectid = search_start;
1084 ins->offset = 0;
1085 start_found = 0;
1086
1087 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1088 if (ret < 0)
1089 goto error;
1090
1091 if (path->slots[0] > 0) {
1092 path->slots[0]--;
1093 }
1094
1095 l = btrfs_buffer_leaf(path->nodes[0]);
1096 btrfs_disk_key_to_cpu(&key, &l->items[path->slots[0]].key);
1097 /*
1098 * a rare case, go back one key if we hit a block group item
1099 * instead of an extent item
1100 */
1101 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
1102 key.objectid + key.offset >= search_start) {
1103 ins->objectid = key.objectid;
1104 ins->offset = key.offset - 1;
1105 btrfs_release_path(root, path);
1106 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1107 if (ret < 0)
1108 goto error;
1109
1110 if (path->slots[0] > 0) {
1111 path->slots[0]--;
1112 }
1113 }
1114
1115 while (1) {
1116 l = btrfs_buffer_leaf(path->nodes[0]);
1117 slot = path->slots[0];
1118 if (slot >= btrfs_header_nritems(&l->header)) {
1119 if (fill_prealloc) {
1120 info->extent_tree_prealloc_nr = 0;
1121 total_found = 0;
1122 }
1123 if (start_found)
1124 limit = last_block +
1125 (block_group->key.offset >> 1);
1126 else
1127 limit = search_start +
1128 (block_group->key.offset >> 1);
1129 ret = btrfs_next_leaf(root, path);
1130 if (ret == 0)
1131 continue;
1132 if (ret < 0)
1133 goto error;
1134 if (!start_found) {
1135 ins->objectid = search_start;
1136 ins->offset = search_end - search_start;
1137 start_found = 1;
1138 goto check_pending;
1139 }
1140 ins->objectid = last_block > search_start ?
1141 last_block : search_start;
1142 ins->offset = search_end - ins->objectid;
1143 goto check_pending;
1144 }
1145
1146 btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
1147 if (key.objectid >= search_start && key.objectid > last_block &&
1148 start_found) {
1149 if (last_block < search_start)
1150 last_block = search_start;
1151 hole_size = key.objectid - last_block;
1152 if (hole_size >= num_blocks) {
1153 ins->objectid = last_block;
1154 ins->offset = hole_size;
1155 goto check_pending;
1156 }
1157 }
1158
1159 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
1160 goto next;
1161
1162 start_found = 1;
1163 last_block = key.objectid + key.offset;
1164 if (!full_scan && last_block >= block_group->key.objectid +
1165 block_group->key.offset) {
1166 btrfs_release_path(root, path);
1167 search_start = block_group->key.objectid +
1168 block_group->key.offset * 2;
1169 goto new_group;
1170 }
1171 next:
1172 path->slots[0]++;
1173 cond_resched();
1174 }
1175 check_pending:
1176 /* we have to make sure we didn't find an extent that has already
1177 * been allocated by the map tree or the original allocation
1178 */
1179 btrfs_release_path(root, path);
1180 BUG_ON(ins->objectid < search_start);
1181
1182 if (ins->objectid + num_blocks >= search_end) {
1183 if (full_scan) {
1184 ret = -ENOSPC;
1185 goto error;
1186 }
1187 search_start = orig_search_start;
1188 if (wrapped)
1189 full_scan = 1;
1190 else
1191 wrapped = 1;
1192 goto new_group;
1193 }
1194 for (test_block = ins->objectid;
1195 test_block < ins->objectid + num_blocks; test_block++) {
1196 if (test_radix_bit(&info->pinned_radix, test_block)) {
1197 search_start = test_block + 1;
1198 goto new_group;
1199 }
1200 }
1201 if (!fill_prealloc && info->extent_tree_insert_nr) {
1202 u64 last =
1203 info->extent_tree_insert[info->extent_tree_insert_nr - 1];
1204 if (ins->objectid + num_blocks >
1205 info->extent_tree_insert[0] &&
1206 ins->objectid <= last) {
1207 search_start = last + 1;
1208 WARN_ON(!full_scan);
1209 goto new_group;
1210 }
1211 }
1212 if (!fill_prealloc && info->extent_tree_prealloc_nr) {
1213 u64 first =
1214 info->extent_tree_prealloc[info->extent_tree_prealloc_nr - 1];
1215 if (ins->objectid + num_blocks > first &&
1216 ins->objectid <= info->extent_tree_prealloc[0]) {
1217 search_start = info->extent_tree_prealloc[0] + 1;
1218 goto new_group;
1219 }
1220 }
1221 if (exclude_nr > 0 && (ins->objectid + num_blocks > exclude_start &&
1222 ins->objectid < exclude_start + exclude_nr)) {
1223 search_start = exclude_start + exclude_nr;
1224 goto new_group;
1225 }
1226 if (fill_prealloc) {
1227 int nr;
1228 test_block = ins->objectid;
1229 if (test_block - info->extent_tree_prealloc[total_needed - 1] >=
1230 leaf_range(root)) {
1231 total_found = 0;
1232 info->extent_tree_prealloc_nr = total_found;
1233 }
1234 while(test_block < ins->objectid + ins->offset &&
1235 total_found < total_needed) {
1236 nr = total_needed - total_found - 1;
1237 BUG_ON(nr < 0);
1238 info->extent_tree_prealloc[nr] = test_block;
1239 total_found++;
1240 test_block++;
1241 }
1242 if (total_found < total_needed) {
1243 search_start = test_block;
1244 goto new_group;
1245 }
1246 info->extent_tree_prealloc_nr = total_found;
1247 }
1248 if (!data) {
1249 block_group = btrfs_lookup_block_group(info, ins->objectid);
1250 if (block_group) {
1251 if (fill_prealloc)
1252 block_group->last_prealloc =
1253 info->extent_tree_prealloc[total_needed-1];
1254 else
1255 trans->block_group = block_group;
1256 }
1257 }
1258 ins->offset = num_blocks;
1259 btrfs_free_path(path);
1260 return 0;
1261
1262 new_group:
1263 if (search_start + num_blocks >= search_end) {
1264 search_start = orig_search_start;
1265 if (full_scan) {
1266 ret = -ENOSPC;
1267 goto error;
1268 }
1269 if (wrapped)
1270 full_scan = 1;
1271 else
1272 wrapped = 1;
1273 }
1274 block_group = btrfs_lookup_block_group(info, search_start);
1275 cond_resched();
1276 if (!full_scan)
1277 block_group = btrfs_find_block_group(root, block_group,
1278 search_start, data, 0);
1279 goto check_failed;
1280
1281 error:
1282 btrfs_release_path(root, path);
1283 btrfs_free_path(path);
1284 return ret;
1285 }
1286 /*
1287 * finds a free extent and does all the dirty work required for allocation
1288 * returns the key for the extent through ins, and a tree buffer for
1289 * the first block of the extent through buf.
1290 *
1291 * returns 0 if everything worked, non-zero otherwise.
1292 */
1293 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1294 struct btrfs_root *root, u64 owner,
1295 u64 num_blocks, u64 hint_block,
1296 u64 search_end, struct btrfs_key *ins, int data)
1297 {
1298 int ret;
1299 int pending_ret;
1300 u64 super_blocks_used;
1301 u64 search_start = 0;
1302 u64 exclude_start = 0;
1303 u64 exclude_nr = 0;
1304 struct btrfs_fs_info *info = root->fs_info;
1305 struct btrfs_root *extent_root = info->extent_root;
1306 struct btrfs_extent_item extent_item;
1307 struct btrfs_key prealloc_key;
1308
1309 btrfs_set_extent_refs(&extent_item, 1);
1310 btrfs_set_extent_owner(&extent_item, owner);
1311
1312 if (root == extent_root) {
1313 int nr;
1314 BUG_ON(info->extent_tree_prealloc_nr == 0);
1315 BUG_ON(num_blocks != 1);
1316 ins->offset = 1;
1317 info->extent_tree_prealloc_nr--;
1318 nr = info->extent_tree_prealloc_nr;
1319 ins->objectid = info->extent_tree_prealloc[nr];
1320 info->extent_tree_insert[info->extent_tree_insert_nr++] =
1321 ins->objectid;
1322 ret = update_block_group(trans, root,
1323 ins->objectid, ins->offset, 1, 0, 0);
1324 BUG_ON(ret);
1325 return 0;
1326 }
1327
1328 /*
1329 * if we're doing a data allocation, preallocate room in the
1330 * extent tree first. This way the extent tree blocks end up
1331 * in the correct block group.
1332 */
1333 if (data) {
1334 ret = find_free_extent(trans, root, 0, 0,
1335 search_end, 0, &prealloc_key, 0, 0, 0);
1336 BUG_ON(ret);
1337 if (ret)
1338 return ret;
1339 exclude_nr = info->extent_tree_prealloc_nr;
1340 exclude_start = info->extent_tree_prealloc[exclude_nr - 1];
1341 }
1342
1343 /* do the real allocation */
1344 ret = find_free_extent(trans, root, num_blocks, search_start,
1345 search_end, hint_block, ins,
1346 exclude_start, exclude_nr, data);
1347 BUG_ON(ret);
1348 if (ret)
1349 return ret;
1350
1351 /*
1352 * if we're doing a metadata allocation, preallocate space in the
1353 * extent tree second. This way, we don't create a tiny hole
1354 * in the allocation map between any unused preallocation blocks
1355 * and the metadata block we're actually allocating. On disk,
1356 * it'll go:
1357 * [block we've allocated], [used prealloc 1], [ unused prealloc ]
1358 * The unused prealloc will get reused the next time around.
1359 */
1360 if (!data) {
1361 exclude_start = ins->objectid;
1362 exclude_nr = ins->offset;
1363 hint_block = exclude_start + exclude_nr;
1364 ret = find_free_extent(trans, root, 0, search_start,
1365 search_end, hint_block,
1366 &prealloc_key, exclude_start,
1367 exclude_nr, 0);
1368 BUG_ON(ret);
1369 if (ret)
1370 return ret;
1371 }
1372
1373 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
1374 btrfs_set_super_blocks_used(&info->super_copy, super_blocks_used +
1375 num_blocks);
1376 ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1377 sizeof(extent_item));
1378
1379 BUG_ON(ret);
1380 finish_current_insert(trans, extent_root);
1381 pending_ret = del_pending_extents(trans, extent_root);
1382 if (ret) {
1383 return ret;
1384 }
1385 if (pending_ret) {
1386 return pending_ret;
1387 }
1388 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1389 data);
1390 BUG_ON(ret);
1391 return 0;
1392 }
1393
1394 /*
1395 * helper function to allocate a block for a given tree
1396 * returns the tree buffer or NULL.
1397 */
1398 struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1399 struct btrfs_root *root, u64 hint)
1400 {
1401 struct btrfs_key ins;
1402 int ret;
1403 struct buffer_head *buf;
1404
1405 ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1406 1, hint, (unsigned long)-1, &ins, 0);
1407 if (ret) {
1408 BUG_ON(ret > 0);
1409 return ERR_PTR(ret);
1410 }
1411 buf = btrfs_find_create_tree_block(root, ins.objectid);
1412 if (!buf) {
1413 btrfs_free_extent(trans, root, ins.objectid, 1, 0);
1414 return ERR_PTR(-ENOMEM);
1415 }
1416 set_buffer_uptodate(buf);
1417 set_buffer_checked(buf);
1418 set_radix_bit(&trans->transaction->dirty_pages, buf->b_page->index);
1419 return buf;
1420 }
1421
1422 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1423 struct btrfs_root *root, struct buffer_head *cur)
1424 {
1425 struct btrfs_disk_key *key;
1426 struct btrfs_leaf *leaf;
1427 struct btrfs_file_extent_item *fi;
1428 int i;
1429 int nritems;
1430 int ret;
1431
1432 BUG_ON(!btrfs_is_leaf(btrfs_buffer_node(cur)));
1433 leaf = btrfs_buffer_leaf(cur);
1434 nritems = btrfs_header_nritems(&leaf->header);
1435 for (i = 0; i < nritems; i++) {
1436 u64 disk_blocknr;
1437 key = &leaf->items[i].key;
1438 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
1439 continue;
1440 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1441 if (btrfs_file_extent_type(fi) == BTRFS_FILE_EXTENT_INLINE)
1442 continue;
1443 /*
1444 * FIXME make sure to insert a trans record that
1445 * repeats the snapshot del on crash
1446 */
1447 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
1448 if (disk_blocknr == 0)
1449 continue;
1450 ret = btrfs_free_extent(trans, root, disk_blocknr,
1451 btrfs_file_extent_disk_num_blocks(fi),
1452 0);
1453 BUG_ON(ret);
1454 }
1455 return 0;
1456 }
1457
1458 static void reada_walk_down(struct btrfs_root *root,
1459 struct btrfs_node *node)
1460 {
1461 int i;
1462 u32 nritems;
1463 u64 blocknr;
1464 int ret;
1465 u32 refs;
1466
1467 nritems = btrfs_header_nritems(&node->header);
1468 for (i = 0; i < nritems; i++) {
1469 blocknr = btrfs_node_blockptr(node, i);
1470 ret = lookup_extent_ref(NULL, root, blocknr, 1, &refs);
1471 BUG_ON(ret);
1472 if (refs != 1)
1473 continue;
1474 ret = readahead_tree_block(root, blocknr);
1475 if (ret)
1476 break;
1477 }
1478 }
1479
1480 /*
1481 * helper function for drop_snapshot, this walks down the tree dropping ref
1482 * counts as it goes.
1483 */
1484 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1485 *root, struct btrfs_path *path, int *level)
1486 {
1487 struct buffer_head *next;
1488 struct buffer_head *cur;
1489 u64 blocknr;
1490 int ret;
1491 u32 refs;
1492
1493 WARN_ON(*level < 0);
1494 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1495 ret = lookup_extent_ref(trans, root, bh_blocknr(path->nodes[*level]),
1496 1, &refs);
1497 BUG_ON(ret);
1498 if (refs > 1)
1499 goto out;
1500
1501 /*
1502 * walk down to the last node level and free all the leaves
1503 */
1504 while(*level >= 0) {
1505 WARN_ON(*level < 0);
1506 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1507 cur = path->nodes[*level];
1508
1509 if (*level > 0 && path->slots[*level] == 0)
1510 reada_walk_down(root, btrfs_buffer_node(cur));
1511
1512 if (btrfs_header_level(btrfs_buffer_header(cur)) != *level)
1513 WARN_ON(1);
1514
1515 if (path->slots[*level] >=
1516 btrfs_header_nritems(btrfs_buffer_header(cur)))
1517 break;
1518 if (*level == 0) {
1519 ret = drop_leaf_ref(trans, root, cur);
1520 BUG_ON(ret);
1521 break;
1522 }
1523 blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur),
1524 path->slots[*level]);
1525 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1526 BUG_ON(ret);
1527 if (refs != 1) {
1528 path->slots[*level]++;
1529 ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
1530 BUG_ON(ret);
1531 continue;
1532 }
1533 next = read_tree_block(root, blocknr);
1534 WARN_ON(*level <= 0);
1535 if (path->nodes[*level-1])
1536 btrfs_block_release(root, path->nodes[*level-1]);
1537 path->nodes[*level-1] = next;
1538 *level = btrfs_header_level(btrfs_buffer_header(next));
1539 path->slots[*level] = 0;
1540 }
1541 out:
1542 WARN_ON(*level < 0);
1543 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1544 ret = btrfs_free_extent(trans, root,
1545 bh_blocknr(path->nodes[*level]), 1, 1);
1546 btrfs_block_release(root, path->nodes[*level]);
1547 path->nodes[*level] = NULL;
1548 *level += 1;
1549 BUG_ON(ret);
1550 return 0;
1551 }
1552
1553 /*
1554 * helper for dropping snapshots. This walks back up the tree in the path
1555 * to find the first node higher up where we haven't yet gone through
1556 * all the slots
1557 */
1558 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1559 *root, struct btrfs_path *path, int *level)
1560 {
1561 int i;
1562 int slot;
1563 int ret;
1564 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1565 slot = path->slots[i];
1566 if (slot < btrfs_header_nritems(
1567 btrfs_buffer_header(path->nodes[i])) - 1) {
1568 path->slots[i]++;
1569 *level = i;
1570 return 0;
1571 } else {
1572 ret = btrfs_free_extent(trans, root,
1573 bh_blocknr(path->nodes[*level]),
1574 1, 1);
1575 BUG_ON(ret);
1576 btrfs_block_release(root, path->nodes[*level]);
1577 path->nodes[*level] = NULL;
1578 *level = i + 1;
1579 }
1580 }
1581 return 1;
1582 }
1583
1584 /*
1585 * drop the reference count on the tree rooted at 'snap'. This traverses
1586 * the tree freeing any blocks that have a ref count of zero after being
1587 * decremented.
1588 */
1589 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1590 *root, struct buffer_head *snap)
1591 {
1592 int ret = 0;
1593 int wret;
1594 int level;
1595 struct btrfs_path *path;
1596 int i;
1597 int orig_level;
1598
1599 path = btrfs_alloc_path();
1600 BUG_ON(!path);
1601
1602 level = btrfs_header_level(btrfs_buffer_header(snap));
1603 orig_level = level;
1604 path->nodes[level] = snap;
1605 path->slots[level] = 0;
1606 while(1) {
1607 wret = walk_down_tree(trans, root, path, &level);
1608 if (wret > 0)
1609 break;
1610 if (wret < 0)
1611 ret = wret;
1612
1613 wret = walk_up_tree(trans, root, path, &level);
1614 if (wret > 0)
1615 break;
1616 if (wret < 0)
1617 ret = wret;
1618 }
1619 for (i = 0; i <= orig_level; i++) {
1620 if (path->nodes[i]) {
1621 btrfs_block_release(root, path->nodes[i]);
1622 }
1623 }
1624 btrfs_free_path(path);
1625 return ret;
1626 }
1627
1628 static int free_block_group_radix(struct radix_tree_root *radix)
1629 {
1630 int ret;
1631 struct btrfs_block_group_cache *cache[8];
1632 int i;
1633
1634 while(1) {
1635 ret = radix_tree_gang_lookup(radix, (void **)cache, 0,
1636 ARRAY_SIZE(cache));
1637 if (!ret)
1638 break;
1639 for (i = 0; i < ret; i++) {
1640 radix_tree_delete(radix, cache[i]->key.objectid +
1641 cache[i]->key.offset - 1);
1642 kfree(cache[i]);
1643 }
1644 }
1645 return 0;
1646 }
1647
1648 int btrfs_free_block_groups(struct btrfs_fs_info *info)
1649 {
1650 int ret;
1651 int ret2;
1652 unsigned long gang[16];
1653 int i;
1654
1655 ret = free_block_group_radix(&info->block_group_radix);
1656 ret2 = free_block_group_radix(&info->block_group_data_radix);
1657 if (ret)
1658 return ret;
1659 if (ret2)
1660 return ret2;
1661
1662 while(1) {
1663 ret = find_first_radix_bit(&info->extent_map_radix,
1664 gang, 0, ARRAY_SIZE(gang));
1665 if (!ret)
1666 break;
1667 for (i = 0; i < ret; i++) {
1668 clear_radix_bit(&info->extent_map_radix, gang[i]);
1669 }
1670 }
1671 return 0;
1672 }
1673
1674 int btrfs_read_block_groups(struct btrfs_root *root)
1675 {
1676 struct btrfs_path *path;
1677 int ret;
1678 int err = 0;
1679 struct btrfs_block_group_item *bi;
1680 struct btrfs_block_group_cache *cache;
1681 struct btrfs_fs_info *info = root->fs_info;
1682 struct radix_tree_root *radix;
1683 struct btrfs_key key;
1684 struct btrfs_key found_key;
1685 struct btrfs_leaf *leaf;
1686 u64 group_size_blocks;
1687 u64 used;
1688
1689 group_size_blocks = BTRFS_BLOCK_GROUP_SIZE >>
1690 root->fs_info->sb->s_blocksize_bits;
1691 root = info->extent_root;
1692 key.objectid = 0;
1693 key.offset = group_size_blocks;
1694 key.flags = 0;
1695 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1696
1697 path = btrfs_alloc_path();
1698 if (!path)
1699 return -ENOMEM;
1700
1701 while(1) {
1702 ret = btrfs_search_slot(NULL, info->extent_root,
1703 &key, path, 0, 0);
1704 if (ret != 0) {
1705 err = ret;
1706 break;
1707 }
1708 leaf = btrfs_buffer_leaf(path->nodes[0]);
1709 btrfs_disk_key_to_cpu(&found_key,
1710 &leaf->items[path->slots[0]].key);
1711 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1712 if (!cache) {
1713 err = -1;
1714 break;
1715 }
1716
1717 bi = btrfs_item_ptr(leaf, path->slots[0],
1718 struct btrfs_block_group_item);
1719 if (bi->flags & BTRFS_BLOCK_GROUP_DATA) {
1720 radix = &info->block_group_data_radix;
1721 cache->data = 1;
1722 } else {
1723 radix = &info->block_group_radix;
1724 cache->data = 0;
1725 }
1726
1727 memcpy(&cache->item, bi, sizeof(*bi));
1728 memcpy(&cache->key, &found_key, sizeof(found_key));
1729 cache->last_alloc = cache->key.objectid;
1730 cache->first_free = cache->key.objectid;
1731 cache->last_prealloc = cache->key.objectid;
1732 cache->pinned = 0;
1733 cache->cached = 0;
1734
1735 cache->radix = radix;
1736
1737 key.objectid = found_key.objectid + found_key.offset;
1738 btrfs_release_path(root, path);
1739 ret = radix_tree_insert(radix, found_key.objectid +
1740 found_key.offset - 1,
1741 (void *)cache);
1742 BUG_ON(ret);
1743 used = btrfs_block_group_used(bi);
1744 if (used < div_factor(key.offset, 8)) {
1745 radix_tree_tag_set(radix, found_key.objectid +
1746 found_key.offset - 1,
1747 BTRFS_BLOCK_GROUP_AVAIL);
1748 }
1749 if (key.objectid >=
1750 btrfs_super_total_blocks(&info->super_copy))
1751 break;
1752 }
1753
1754 btrfs_free_path(path);
1755 return 0;
1756 }
Ubuntu Artful kernel mirror