projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: Allow find_free_extent callers to pass in an exclusion range
[linux-2.6-block.git] / fs / btrfs / extent-tree.c
... / ...
CommitLineData
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
25static 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);
30static int finish_current_insert(struct btrfs_trans_handle *trans, struct
31 btrfs_root *extent_root);
32static int del_pending_extents(struct btrfs_trans_handle *trans, struct
33 btrfs_root *extent_root);
34
35static 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
62static 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;
154err:
155 btrfs_free_path(path);
156 return 0;
157}
158
159struct 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
185static 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
193static 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 }
207again:
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 }
230out:
231 return max(cache->last_alloc, search_start);
232
233new_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
245static u64 div_factor(u64 num, int factor)
246{
247 num *= factor;
248 do_div(num, 10);
249 return num;
250}
251
252struct 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;
345again:
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 }
391found:
392 return found_group;
393}
394
395int 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
441static 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);
465out:
466 btrfs_free_path(path);
467 return 0;
468}
469
470int 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
476int 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;
525fail:
526 for (i =0; i < faili; i++) {
527 if (leaf) {
528 u64 disk_blocknr;
529 key = &buf_leaf->items[i].key;
530 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
531 continue;
532 fi = btrfs_item_ptr(buf_leaf, i,
533 struct btrfs_file_extent_item);
534 if (btrfs_file_extent_type(fi) ==
535 BTRFS_FILE_EXTENT_INLINE)
536 continue;
537 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
538 if (disk_blocknr == 0)
539 continue;
540 err = btrfs_free_extent(trans, root, disk_blocknr,
541 btrfs_file_extent_disk_num_blocks(fi), 0);
542 BUG_ON(err);
543 } else {
544 blocknr = btrfs_node_blockptr(buf_node, i);
545 err = btrfs_free_extent(trans, root, blocknr, 1, 0);
546 BUG_ON(err);
547 }
548 }
549 return ret;
550}
551
552static int write_one_cache_group(struct btrfs_trans_handle *trans,
553 struct btrfs_root *root,
554 struct btrfs_path *path,
555 struct btrfs_block_group_cache *cache)
556{
557 int ret;
558 int pending_ret;
559 struct btrfs_root *extent_root = root->fs_info->extent_root;
560 struct btrfs_block_group_item *bi;
561 struct btrfs_key ins;
562
563 ret = find_free_extent(trans, extent_root, 0, 0, (u64)-1, 0, &ins,
564 0, 0, 0);
565 /* FIXME, set bit to recalc cache groups on next mount */
566 if (ret)
567 return ret;
568 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
569 if (ret < 0)
570 goto fail;
571 BUG_ON(ret);
572 bi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
573 struct btrfs_block_group_item);
574 memcpy(bi, &cache->item, sizeof(*bi));
575 mark_buffer_dirty(path->nodes[0]);
576 btrfs_release_path(extent_root, path);
577fail:
578 finish_current_insert(trans, extent_root);
579 pending_ret = del_pending_extents(trans, extent_root);
580 if (ret)
581 return ret;
582 if (pending_ret)
583 return pending_ret;
584 if (cache->data)
585 cache->last_alloc = cache->first_free;
586 return 0;
587
588}
589
590static int write_dirty_block_radix(struct btrfs_trans_handle *trans,
591 struct btrfs_root *root,
592 struct radix_tree_root *radix)
593{
594 struct btrfs_block_group_cache *cache[8];
595 int ret;
596 int err = 0;
597 int werr = 0;
598 int i;
599 struct btrfs_path *path;
600 unsigned long off = 0;
601
602 path = btrfs_alloc_path();
603 if (!path)
604 return -ENOMEM;
605
606 while(1) {
607 ret = radix_tree_gang_lookup_tag(radix, (void **)cache,
608 off, ARRAY_SIZE(cache),
609 BTRFS_BLOCK_GROUP_DIRTY);
610 if (!ret)
611 break;
612 for (i = 0; i < ret; i++) {
613 err = write_one_cache_group(trans, root,
614 path, cache[i]);
615 /*
616 * if we fail to write the cache group, we want
617 * to keep it marked dirty in hopes that a later
618 * write will work
619 */
620 if (err) {
621 werr = err;
622 off = cache[i]->key.objectid +
623 cache[i]->key.offset;
624 continue;
625 }
626
627 radix_tree_tag_clear(radix, cache[i]->key.objectid +
628 cache[i]->key.offset - 1,
629 BTRFS_BLOCK_GROUP_DIRTY);
630 }
631 }
632 btrfs_free_path(path);
633 return werr;
634}
635
636int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
637 struct btrfs_root *root)
638{
639 int ret;
640 int ret2;
641 ret = write_dirty_block_radix(trans, root,
642 &root->fs_info->block_group_radix);
643 ret2 = write_dirty_block_radix(trans, root,
644 &root->fs_info->block_group_data_radix);
645 if (ret)
646 return ret;
647 if (ret2)
648 return ret2;
649 return 0;
650}
651
652static int update_block_group(struct btrfs_trans_handle *trans,
653 struct btrfs_root *root,
654 u64 blocknr, u64 num, int alloc, int mark_free,
655 int data)
656{
657 struct btrfs_block_group_cache *cache;
658 struct btrfs_fs_info *info = root->fs_info;
659 u64 total = num;
660 u64 old_val;
661 u64 block_in_group;
662 u64 i;
663 int ret;
664
665 while(total) {
666 cache = btrfs_lookup_block_group(info, blocknr);
667 if (!cache) {
668 return -1;
669 }
670 block_in_group = blocknr - cache->key.objectid;
671 WARN_ON(block_in_group > cache->key.offset);
672 radix_tree_tag_set(cache->radix, cache->key.objectid +
673 cache->key.offset - 1,
674 BTRFS_BLOCK_GROUP_DIRTY);
675
676 old_val = btrfs_block_group_used(&cache->item);
677 num = min(total, cache->key.offset - block_in_group);
678 if (alloc) {
679 if (blocknr > cache->last_alloc)
680 cache->last_alloc = blocknr;
681 if (!cache->data) {
682 for (i = 0; i < num; i++) {
683 clear_radix_bit(&info->extent_map_radix,
684 blocknr + i);
685 }
686 }
687 if (cache->data != data &&
688 old_val < (cache->key.offset >> 1)) {
689 cache->data = data;
690 radix_tree_delete(cache->radix,
691 cache->key.objectid +
692 cache->key.offset - 1);
693
694 if (data) {
695 cache->radix =
696 &info->block_group_data_radix;
697 cache->item.flags |=
698 BTRFS_BLOCK_GROUP_DATA;
699 } else {
700 cache->radix = &info->block_group_radix;
701 cache->item.flags &=
702 ~BTRFS_BLOCK_GROUP_DATA;
703 }
704 ret = radix_tree_insert(cache->radix,
705 cache->key.objectid +
706 cache->key.offset - 1,
707 (void *)cache);
708 }
709 old_val += num;
710 } else {
711 old_val -= num;
712 if (blocknr < cache->first_free)
713 cache->first_free = blocknr;
714 if (!cache->data && mark_free) {
715 for (i = 0; i < num; i++) {
716 set_radix_bit(&info->extent_map_radix,
717 blocknr + i);
718 }
719 }
720 if (old_val < (cache->key.offset >> 1) &&
721 old_val + num >= (cache->key.offset >> 1)) {
722 radix_tree_tag_set(cache->radix,
723 cache->key.objectid +
724 cache->key.offset - 1,
725 BTRFS_BLOCK_GROUP_AVAIL);
726 }
727 }
728 btrfs_set_block_group_used(&cache->item, old_val);
729 total -= num;
730 blocknr += num;
731 }
732 return 0;
733}
734
735static int try_remove_page(struct address_space *mapping, unsigned long index)
736{
737 int ret;
738 ret = invalidate_mapping_pages(mapping, index, index);
739 return ret;
740}
741
742int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct
743 btrfs_root *root)
744{
745 unsigned long gang[8];
746 struct inode *btree_inode = root->fs_info->btree_inode;
747 struct btrfs_block_group_cache *block_group;
748 u64 first = 0;
749 int ret;
750 int i;
751 struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
752 struct radix_tree_root *extent_radix = &root->fs_info->extent_map_radix;
753
754 while(1) {
755 ret = find_first_radix_bit(pinned_radix, gang, 0,
756 ARRAY_SIZE(gang));
757 if (!ret)
758 break;
759 if (!first)
760 first = gang[0];
761 for (i = 0; i < ret; i++) {
762 clear_radix_bit(pinned_radix, gang[i]);
763 block_group = btrfs_lookup_block_group(root->fs_info,
764 gang[i]);
765 if (block_group) {
766 WARN_ON(block_group->pinned == 0);
767 block_group->pinned--;
768 if (gang[i] < block_group->last_alloc)
769 block_group->last_alloc = gang[i];
770 if (gang[i] < block_group->last_prealloc)
771 block_group->last_prealloc = gang[i];
772 if (!block_group->data)
773 set_radix_bit(extent_radix, gang[i]);
774 }
775 try_remove_page(btree_inode->i_mapping,
776 gang[i] << (PAGE_CACHE_SHIFT -
777 btree_inode->i_blkbits));
778 }
779 }
780 return 0;
781}
782
783static int finish_current_insert(struct btrfs_trans_handle *trans, struct
784 btrfs_root *extent_root)
785{
786 struct btrfs_key ins;
787 struct btrfs_extent_item extent_item;
788 int i;
789 int ret;
790 u64 super_blocks_used;
791 struct btrfs_fs_info *info = extent_root->fs_info;
792
793 btrfs_set_extent_refs(&extent_item, 1);
794 ins.offset = 1;
795 ins.flags = 0;
796 btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
797 btrfs_set_extent_owner(&extent_item, extent_root->root_key.objectid);
798
799 for (i = 0; i < extent_root->fs_info->extent_tree_insert_nr; i++) {
800 ins.objectid = extent_root->fs_info->extent_tree_insert[i];
801 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
802 btrfs_set_super_blocks_used(&info->super_copy,
803 super_blocks_used + 1);
804 ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item,
805 sizeof(extent_item));
806 BUG_ON(ret);
807 }
808 extent_root->fs_info->extent_tree_insert_nr = 0;
809 return 0;
810}
811
812static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
813{
814 int err;
815 struct btrfs_header *header;
816 struct buffer_head *bh;
817
818 if (!pending) {
819 bh = btrfs_find_tree_block(root, blocknr);
820 if (bh) {
821 if (buffer_uptodate(bh)) {
822 u64 transid =
823 root->fs_info->running_transaction->transid;
824 header = btrfs_buffer_header(bh);
825 if (btrfs_header_generation(header) ==
826 transid) {
827 btrfs_block_release(root, bh);
828 return 0;
829 }
830 }
831 btrfs_block_release(root, bh);
832 }
833 err = set_radix_bit(&root->fs_info->pinned_radix, blocknr);
834 if (!err) {
835 struct btrfs_block_group_cache *cache;
836 cache = btrfs_lookup_block_group(root->fs_info,
837 blocknr);
838 if (cache)
839 cache->pinned++;
840 }
841 } else {
842 err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr);
843 }
844 BUG_ON(err < 0);
845 return 0;
846}
847
848/*
849 * remove an extent from the root, returns 0 on success
850 */
851static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
852 *root, u64 blocknr, u64 num_blocks, int pin,
853 int mark_free)
854{
855 struct btrfs_path *path;
856 struct btrfs_key key;
857 struct btrfs_fs_info *info = root->fs_info;
858 struct btrfs_root *extent_root = info->extent_root;
859 int ret;
860 struct btrfs_extent_item *ei;
861 struct btrfs_key ins;
862 u32 refs;
863
864 key.objectid = blocknr;
865 key.flags = 0;
866 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
867 key.offset = num_blocks;
868
869 path = btrfs_alloc_path();
870 if (!path)
871 return -ENOMEM;
872
873 ret = find_free_extent(trans, root, 0, 0, (u64)-1, 0, &ins, 0, 0, 0);
874 if (ret) {
875 btrfs_free_path(path);
876 return ret;
877 }
878
879 ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
880 if (ret < 0)
881 return ret;
882 BUG_ON(ret);
883 ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
884 struct btrfs_extent_item);
885 BUG_ON(ei->refs == 0);
886 refs = btrfs_extent_refs(ei) - 1;
887 btrfs_set_extent_refs(ei, refs);
888 btrfs_mark_buffer_dirty(path->nodes[0]);
889 if (refs == 0) {
890 u64 super_blocks_used;
891
892 if (pin) {
893 ret = pin_down_block(root, blocknr, 0);
894 BUG_ON(ret);
895 }
896
897 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
898 btrfs_set_super_blocks_used(&info->super_copy,
899 super_blocks_used - num_blocks);
900 ret = btrfs_del_item(trans, extent_root, path);
901 if (ret) {
902 return ret;
903 }
904 ret = update_block_group(trans, root, blocknr, num_blocks, 0,
905 mark_free, 0);
906 BUG_ON(ret);
907 }
908 btrfs_free_path(path);
909 finish_current_insert(trans, extent_root);
910 return ret;
911}
912
913/*
914 * find all the blocks marked as pending in the radix tree and remove
915 * them from the extent map
916 */
917static int del_pending_extents(struct btrfs_trans_handle *trans, struct
918 btrfs_root *extent_root)
919{
920 int ret;
921 int wret;
922 int err = 0;
923 unsigned long gang[4];
924 int i;
925 struct radix_tree_root *pending_radix;
926 struct radix_tree_root *pinned_radix;
927 struct btrfs_block_group_cache *cache;
928
929 pending_radix = &extent_root->fs_info->pending_del_radix;
930 pinned_radix = &extent_root->fs_info->pinned_radix;
931
932 while(1) {
933 ret = find_first_radix_bit(pending_radix, gang, 0,
934 ARRAY_SIZE(gang));
935 if (!ret)
936 break;
937 for (i = 0; i < ret; i++) {
938 wret = set_radix_bit(pinned_radix, gang[i]);
939 if (wret == 0) {
940 cache =
941 btrfs_lookup_block_group(extent_root->fs_info,
942 gang[i]);
943 if (cache)
944 cache->pinned++;
945 }
946 if (wret < 0) {
947 printk(KERN_CRIT "set_radix_bit, err %d\n",
948 wret);
949 BUG_ON(wret < 0);
950 }
951 wret = clear_radix_bit(pending_radix, gang[i]);
952 BUG_ON(wret);
953 wret = __free_extent(trans, extent_root,
954 gang[i], 1, 0, 0);
955 if (wret)
956 err = wret;
957 }
958 }
959 return err;
960}
961
962/*
963 * remove an extent from the root, returns 0 on success
964 */
965int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
966 *root, u64 blocknr, u64 num_blocks, int pin)
967{
968 struct btrfs_root *extent_root = root->fs_info->extent_root;
969 int pending_ret;
970 int ret;
971
972 if (root == extent_root) {
973 pin_down_block(root, blocknr, 1);
974 return 0;
975 }
976 ret = __free_extent(trans, root, blocknr, num_blocks, pin, pin == 0);
977 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
978 return ret ? ret : pending_ret;
979}
980
981/*
982 * walks the btree of allocated extents and find a hole of a given size.
983 * The key ins is changed to record the hole:
984 * ins->objectid == block start
985 * ins->flags = BTRFS_EXTENT_ITEM_KEY
986 * ins->offset == number of blocks
987 * Any available blocks before search_start are skipped.
988 */
989static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
990 *orig_root, u64 num_blocks, u64 search_start, u64
991 search_end, u64 hint_block,
992 struct btrfs_key *ins, u64 exclude_start,
993 u64 exclude_nr, int data)
994{
995 struct btrfs_path *path;
996 struct btrfs_key key;
997 int ret;
998 u64 hole_size = 0;
999 int slot = 0;
1000 u64 last_block = 0;
1001 u64 test_block;
1002 u64 orig_search_start = search_start;
1003 int start_found;
1004 struct btrfs_leaf *l;
1005 struct btrfs_root * root = orig_root->fs_info->extent_root;
1006 struct btrfs_fs_info *info = root->fs_info;
1007 int total_needed = num_blocks;
1008 int total_found = 0;
1009 int fill_prealloc = 0;
1010 int level;
1011 struct btrfs_block_group_cache *block_group;
1012 int full_scan = 0;
1013 int wrapped = 0;
1014 u64 limit;
1015
1016 ins->flags = 0;
1017 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
1018
1019 level = btrfs_header_level(btrfs_buffer_header(root->node));
1020 if (num_blocks == 0) {
1021 fill_prealloc = 1;
1022 num_blocks = 1;
1023 total_needed = (min(level + 1, BTRFS_MAX_LEVEL) + 2) * 3;
1024 }
1025 if (fill_prealloc) {
1026 u64 first;
1027 int nr = info->extent_tree_prealloc_nr;
1028 first = info->extent_tree_prealloc[nr - 1];
1029 if (info->extent_tree_prealloc_nr >= total_needed &&
1030 first >= search_start) {
1031 ins->objectid = info->extent_tree_prealloc[0];
1032 ins->offset = 1;
1033 return 0;
1034 }
1035 info->extent_tree_prealloc_nr = 0;
1036 }
1037 if (search_end == (u64)-1)
1038 search_end = btrfs_super_total_blocks(&info->super_copy);
1039 if (hint_block) {
1040 block_group = btrfs_lookup_block_group(info, hint_block);
1041 block_group = btrfs_find_block_group(root, block_group,
1042 hint_block, data, 1);
1043 } else {
1044 block_group = btrfs_find_block_group(root,
1045 trans->block_group, 0,
1046 data, 1);
1047 }
1048
1049 path = btrfs_alloc_path();
1050
1051check_failed:
1052 if (!block_group->data)
1053 search_start = find_search_start(root, &block_group,
1054 search_start, total_needed);
1055 else if (!full_scan)
1056 search_start = max(block_group->last_alloc, search_start);
1057
1058 btrfs_init_path(path);
1059 ins->objectid = search_start;
1060 ins->offset = 0;
1061 start_found = 0;
1062
1063 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1064 if (ret < 0)
1065 goto error;
1066
1067 if (path->slots[0] > 0) {
1068 path->slots[0]--;
1069 }
1070
1071 l = btrfs_buffer_leaf(path->nodes[0]);
1072 btrfs_disk_key_to_cpu(&key, &l->items[path->slots[0]].key);
1073 /*
1074 * a rare case, go back one key if we hit a block group item
1075 * instead of an extent item
1076 */
1077 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY &&
1078 key.objectid + key.offset >= search_start) {
1079 ins->objectid = key.objectid;
1080 ins->offset = key.offset - 1;
1081 btrfs_release_path(root, path);
1082 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
1083 if (ret < 0)
1084 goto error;
1085
1086 if (path->slots[0] > 0) {
1087 path->slots[0]--;
1088 }
1089 }
1090
1091 while (1) {
1092 l = btrfs_buffer_leaf(path->nodes[0]);
1093 slot = path->slots[0];
1094 if (slot >= btrfs_header_nritems(&l->header)) {
1095 if (fill_prealloc) {
1096 info->extent_tree_prealloc_nr = 0;
1097 total_found = 0;
1098 }
1099 if (start_found)
1100 limit = last_block +
1101 (block_group->key.offset >> 1);
1102 else
1103 limit = search_start +
1104 (block_group->key.offset >> 1);
1105 ret = btrfs_next_leaf(root, path);
1106 if (ret == 0)
1107 continue;
1108 if (ret < 0)
1109 goto error;
1110 if (!start_found) {
1111 ins->objectid = search_start;
1112 ins->offset = search_end - search_start;
1113 start_found = 1;
1114 goto check_pending;
1115 }
1116 ins->objectid = last_block > search_start ?
1117 last_block : search_start;
1118 ins->offset = search_end - ins->objectid;
1119 goto check_pending;
1120 }
1121
1122 btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
1123 if (key.objectid >= search_start && key.objectid > last_block &&
1124 start_found) {
1125 if (last_block < search_start)
1126 last_block = search_start;
1127 hole_size = key.objectid - last_block;
1128 if (hole_size >= num_blocks) {
1129 ins->objectid = last_block;
1130 ins->offset = hole_size;
1131 goto check_pending;
1132 }
1133 }
1134
1135 if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY)
1136 goto next;
1137
1138 start_found = 1;
1139 last_block = key.objectid + key.offset;
1140 if (!full_scan && last_block >= block_group->key.objectid +
1141 block_group->key.offset) {
1142 btrfs_release_path(root, path);
1143 search_start = block_group->key.objectid +
1144 block_group->key.offset * 2;
1145 goto new_group;
1146 }
1147next:
1148 path->slots[0]++;
1149 cond_resched();
1150 }
1151check_pending:
1152 /* we have to make sure we didn't find an extent that has already
1153 * been allocated by the map tree or the original allocation
1154 */
1155 btrfs_release_path(root, path);
1156 BUG_ON(ins->objectid < search_start);
1157
1158 if (ins->objectid + num_blocks >= search_end) {
1159 if (full_scan) {
1160 ret = -ENOSPC;
1161 goto error;
1162 }
1163 search_start = orig_search_start;
1164 if (wrapped)
1165 full_scan = 1;
1166 else
1167 wrapped = 1;
1168 goto new_group;
1169 }
1170 for (test_block = ins->objectid;
1171 test_block < ins->objectid + num_blocks; test_block++) {
1172 if (test_radix_bit(&info->pinned_radix, test_block)) {
1173 search_start = test_block + 1;
1174 goto new_group;
1175 }
1176 }
1177 if (!fill_prealloc && info->extent_tree_insert_nr) {
1178 u64 last =
1179 info->extent_tree_insert[info->extent_tree_insert_nr - 1];
1180 if (ins->objectid + num_blocks >
1181 info->extent_tree_insert[0] &&
1182 ins->objectid <= last) {
1183 search_start = last + 1;
1184 WARN_ON(!full_scan);
1185 goto new_group;
1186 }
1187 }
1188 if (!fill_prealloc && info->extent_tree_prealloc_nr) {
1189 u64 first =
1190 info->extent_tree_prealloc[info->extent_tree_prealloc_nr - 1];
1191 if (ins->objectid + num_blocks > first &&
1192 ins->objectid <= info->extent_tree_prealloc[0]) {
1193 search_start = info->extent_tree_prealloc[0] + 1;
1194 goto new_group;
1195 }
1196 }
1197 if (exclude_nr > 0 && (ins->objectid + num_blocks > exclude_start &&
1198 ins->objectid < exclude_start + exclude_nr)) {
1199 search_start = exclude_start + exclude_nr;
1200 goto new_group;
1201 }
1202 if (fill_prealloc) {
1203 int nr;
1204 test_block = ins->objectid;
1205 if (test_block - info->extent_tree_prealloc[total_needed - 1] >=
1206 leaf_range(root)) {
1207 total_found = 0;
1208 info->extent_tree_prealloc_nr = total_found;
1209 }
1210 while(test_block < ins->objectid + ins->offset &&
1211 total_found < total_needed) {
1212 nr = total_needed - total_found - 1;
1213 BUG_ON(nr < 0);
1214 info->extent_tree_prealloc[nr] = test_block;
1215 total_found++;
1216 test_block++;
1217 }
1218 if (total_found < total_needed) {
1219 search_start = test_block;
1220 goto new_group;
1221 }
1222 info->extent_tree_prealloc_nr = total_found;
1223 }
1224 if (!data) {
1225 block_group = btrfs_lookup_block_group(info, ins->objectid);
1226 if (block_group) {
1227 if (fill_prealloc)
1228 block_group->last_prealloc =
1229 info->extent_tree_prealloc[total_needed-1];
1230 else
1231 trans->block_group = block_group;
1232 }
1233 }
1234 ins->offset = num_blocks;
1235 btrfs_free_path(path);
1236 return 0;
1237
1238new_group:
1239 if (search_start + num_blocks >= search_end) {
1240 search_start = orig_search_start;
1241 if (full_scan) {
1242 ret = -ENOSPC;
1243 goto error;
1244 }
1245 if (wrapped)
1246 full_scan = 1;
1247 else
1248 wrapped = 1;
1249 }
1250 block_group = btrfs_lookup_block_group(info, search_start);
1251 cond_resched();
1252 if (!full_scan)
1253 block_group = btrfs_find_block_group(root, block_group,
1254 search_start, data, 0);
1255 goto check_failed;
1256
1257error:
1258 btrfs_release_path(root, path);
1259 btrfs_free_path(path);
1260 return ret;
1261}
1262/*
1263 * finds a free extent and does all the dirty work required for allocation
1264 * returns the key for the extent through ins, and a tree buffer for
1265 * the first block of the extent through buf.
1266 *
1267 * returns 0 if everything worked, non-zero otherwise.
1268 */
1269int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
1270 struct btrfs_root *root, u64 owner,
1271 u64 num_blocks, u64 hint_block,
1272 u64 search_end, struct btrfs_key *ins, int data)
1273{
1274 int ret;
1275 int pending_ret;
1276 u64 super_blocks_used;
1277 u64 search_start = 0;
1278 u64 exclude_start = 0;
1279 u64 exclude_nr = 0;
1280 struct btrfs_fs_info *info = root->fs_info;
1281 struct btrfs_root *extent_root = info->extent_root;
1282 struct btrfs_extent_item extent_item;
1283 struct btrfs_key prealloc_key;
1284
1285 btrfs_set_extent_refs(&extent_item, 1);
1286 btrfs_set_extent_owner(&extent_item, owner);
1287
1288 if (root == extent_root) {
1289 int nr;
1290 BUG_ON(info->extent_tree_prealloc_nr == 0);
1291 BUG_ON(num_blocks != 1);
1292 ins->offset = 1;
1293 info->extent_tree_prealloc_nr--;
1294 nr = info->extent_tree_prealloc_nr;
1295 ins->objectid = info->extent_tree_prealloc[nr];
1296 info->extent_tree_insert[info->extent_tree_insert_nr++] =
1297 ins->objectid;
1298 ret = update_block_group(trans, root,
1299 ins->objectid, ins->offset, 1, 0, 0);
1300 BUG_ON(ret);
1301 return 0;
1302 }
1303
1304 /*
1305 * if we're doing a data allocation, preallocate room in the
1306 * extent tree first. This way the extent tree blocks end up
1307 * in the correct block group.
1308 */
1309 if (data) {
1310 ret = find_free_extent(trans, root, 0, 0,
1311 search_end, 0, &prealloc_key, 0, 0, 0);
1312 if (ret)
1313 return ret;
1314 exclude_nr = info->extent_tree_prealloc_nr;
1315 exclude_start = info->extent_tree_prealloc[exclude_nr - 1];
1316 }
1317
1318 /* do the real allocation */
1319 ret = find_free_extent(trans, root, num_blocks, search_start,
1320 search_end, hint_block, ins,
1321 exclude_start, exclude_nr, data);
1322 if (ret)
1323 return ret;
1324
1325 /*
1326 * if we're doing a metadata allocation, preallocate space in the
1327 * extent tree second. This way, we don't create a tiny hole
1328 * in the allocation map between any unused preallocation blocks
1329 * and the metadata block we're actually allocating. On disk,
1330 * it'll go:
1331 * [block we've allocated], [used prealloc 1], [ unused prealloc ]
1332 * The unused prealloc will get reused the next time around.
1333 */
1334 if (!data) {
1335 exclude_start = ins->objectid;
1336 exclude_nr = ins->offset;
1337 ret = find_free_extent(trans, root, 0, search_start,
1338 search_end, hint_block,
1339 &prealloc_key, exclude_start,
1340 exclude_nr, 0);
1341 if (ret)
1342 return ret;
1343 }
1344
1345 super_blocks_used = btrfs_super_blocks_used(&info->super_copy);
1346 btrfs_set_super_blocks_used(&info->super_copy, super_blocks_used +
1347 num_blocks);
1348 ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
1349 sizeof(extent_item));
1350
1351 finish_current_insert(trans, extent_root);
1352 pending_ret = del_pending_extents(trans, extent_root);
1353 if (ret) {
1354 return ret;
1355 }
1356 if (pending_ret) {
1357 return pending_ret;
1358 }
1359 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0,
1360 data);
1361 BUG_ON(ret);
1362 return 0;
1363}
1364
1365/*
1366 * helper function to allocate a block for a given tree
1367 * returns the tree buffer or NULL.
1368 */
1369struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root, u64 hint)
1371{
1372 struct btrfs_key ins;
1373 int ret;
1374 struct buffer_head *buf;
1375
1376 ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
1377 1, hint, (unsigned long)-1, &ins, 0);
1378 if (ret) {
1379 BUG_ON(ret > 0);
1380 return ERR_PTR(ret);
1381 }
1382 buf = btrfs_find_create_tree_block(root, ins.objectid);
1383 if (!buf) {
1384 btrfs_free_extent(trans, root, ins.objectid, 1, 0);
1385 return ERR_PTR(-ENOMEM);
1386 }
1387 set_buffer_uptodate(buf);
1388 set_buffer_checked(buf);
1389 set_radix_bit(&trans->transaction->dirty_pages, buf->b_page->index);
1390 return buf;
1391}
1392
1393static int drop_leaf_ref(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root, struct buffer_head *cur)
1395{
1396 struct btrfs_disk_key *key;
1397 struct btrfs_leaf *leaf;
1398 struct btrfs_file_extent_item *fi;
1399 int i;
1400 int nritems;
1401 int ret;
1402
1403 BUG_ON(!btrfs_is_leaf(btrfs_buffer_node(cur)));
1404 leaf = btrfs_buffer_leaf(cur);
1405 nritems = btrfs_header_nritems(&leaf->header);
1406 for (i = 0; i < nritems; i++) {
1407 u64 disk_blocknr;
1408 key = &leaf->items[i].key;
1409 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
1410 continue;
1411 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1412 if (btrfs_file_extent_type(fi) == BTRFS_FILE_EXTENT_INLINE)
1413 continue;
1414 /*
1415 * FIXME make sure to insert a trans record that
1416 * repeats the snapshot del on crash
1417 */
1418 disk_blocknr = btrfs_file_extent_disk_blocknr(fi);
1419 if (disk_blocknr == 0)
1420 continue;
1421 ret = btrfs_free_extent(trans, root, disk_blocknr,
1422 btrfs_file_extent_disk_num_blocks(fi),
1423 0);
1424 BUG_ON(ret);
1425 }
1426 return 0;
1427}
1428
1429static void reada_walk_down(struct btrfs_root *root,
1430 struct btrfs_node *node)
1431{
1432 int i;
1433 u32 nritems;
1434 u64 blocknr;
1435 int ret;
1436 u32 refs;
1437
1438 nritems = btrfs_header_nritems(&node->header);
1439 for (i = 0; i < nritems; i++) {
1440 blocknr = btrfs_node_blockptr(node, i);
1441 ret = lookup_extent_ref(NULL, root, blocknr, 1, &refs);
1442 BUG_ON(ret);
1443 if (refs != 1)
1444 continue;
1445 ret = readahead_tree_block(root, blocknr);
1446 if (ret)
1447 break;
1448 }
1449}
1450
1451/*
1452 * helper function for drop_snapshot, this walks down the tree dropping ref
1453 * counts as it goes.
1454 */
1455static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1456 *root, struct btrfs_path *path, int *level)
1457{
1458 struct buffer_head *next;
1459 struct buffer_head *cur;
1460 u64 blocknr;
1461 int ret;
1462 u32 refs;
1463
1464 WARN_ON(*level < 0);
1465 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1466 ret = lookup_extent_ref(trans, root, bh_blocknr(path->nodes[*level]),
1467 1, &refs);
1468 BUG_ON(ret);
1469 if (refs > 1)
1470 goto out;
1471
1472 /*
1473 * walk down to the last node level and free all the leaves
1474 */
1475 while(*level >= 0) {
1476 WARN_ON(*level < 0);
1477 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1478 cur = path->nodes[*level];
1479
1480 if (*level > 0 && path->slots[*level] == 0)
1481 reada_walk_down(root, btrfs_buffer_node(cur));
1482
1483 if (btrfs_header_level(btrfs_buffer_header(cur)) != *level)
1484 WARN_ON(1);
1485
1486 if (path->slots[*level] >=
1487 btrfs_header_nritems(btrfs_buffer_header(cur)))
1488 break;
1489 if (*level == 0) {
1490 ret = drop_leaf_ref(trans, root, cur);
1491 BUG_ON(ret);
1492 break;
1493 }
1494 blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur),
1495 path->slots[*level]);
1496 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
1497 BUG_ON(ret);
1498 if (refs != 1) {
1499 path->slots[*level]++;
1500 ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
1501 BUG_ON(ret);
1502 continue;
1503 }
1504 next = read_tree_block(root, blocknr);
1505 WARN_ON(*level <= 0);
1506 if (path->nodes[*level-1])
1507 btrfs_block_release(root, path->nodes[*level-1]);
1508 path->nodes[*level-1] = next;
1509 *level = btrfs_header_level(btrfs_buffer_header(next));
1510 path->slots[*level] = 0;
1511 }
1512out:
1513 WARN_ON(*level < 0);
1514 WARN_ON(*level >= BTRFS_MAX_LEVEL);
1515 ret = btrfs_free_extent(trans, root,
1516 bh_blocknr(path->nodes[*level]), 1, 1);
1517 btrfs_block_release(root, path->nodes[*level]);
1518 path->nodes[*level] = NULL;
1519 *level += 1;
1520 BUG_ON(ret);
1521 return 0;
1522}
1523
1524/*
1525 * helper for dropping snapshots. This walks back up the tree in the path
1526 * to find the first node higher up where we haven't yet gone through
1527 * all the slots
1528 */
1529static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
1530 *root, struct btrfs_path *path, int *level)
1531{
1532 int i;
1533 int slot;
1534 int ret;
1535 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
1536 slot = path->slots[i];
1537 if (slot < btrfs_header_nritems(
1538 btrfs_buffer_header(path->nodes[i])) - 1) {
1539 path->slots[i]++;
1540 *level = i;
1541 return 0;
1542 } else {
1543 ret = btrfs_free_extent(trans, root,
1544 bh_blocknr(path->nodes[*level]),
1545 1, 1);
1546 BUG_ON(ret);
1547 btrfs_block_release(root, path->nodes[*level]);
1548 path->nodes[*level] = NULL;
1549 *level = i + 1;
1550 }
1551 }
1552 return 1;
1553}
1554
1555/*
1556 * drop the reference count on the tree rooted at 'snap'. This traverses
1557 * the tree freeing any blocks that have a ref count of zero after being
1558 * decremented.
1559 */
1560int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
1561 *root, struct buffer_head *snap)
1562{
1563 int ret = 0;
1564 int wret;
1565 int level;
1566 struct btrfs_path *path;
1567 int i;
1568 int orig_level;
1569
1570 path = btrfs_alloc_path();
1571 BUG_ON(!path);
1572
1573 level = btrfs_header_level(btrfs_buffer_header(snap));
1574 orig_level = level;
1575 path->nodes[level] = snap;
1576 path->slots[level] = 0;
1577 while(1) {
1578 wret = walk_down_tree(trans, root, path, &level);
1579 if (wret > 0)
1580 break;
1581 if (wret < 0)
1582 ret = wret;
1583
1584 wret = walk_up_tree(trans, root, path, &level);
1585 if (wret > 0)
1586 break;
1587 if (wret < 0)
1588 ret = wret;
1589 }
1590 for (i = 0; i <= orig_level; i++) {
1591 if (path->nodes[i]) {
1592 btrfs_block_release(root, path->nodes[i]);
1593 }
1594 }
1595 btrfs_free_path(path);
1596 return ret;
1597}
1598
1599static int free_block_group_radix(struct radix_tree_root *radix)
1600{
1601 int ret;
1602 struct btrfs_block_group_cache *cache[8];
1603 int i;
1604
1605 while(1) {
1606 ret = radix_tree_gang_lookup(radix, (void **)cache, 0,
1607 ARRAY_SIZE(cache));
1608 if (!ret)
1609 break;
1610 for (i = 0; i < ret; i++) {
1611 radix_tree_delete(radix, cache[i]->key.objectid +
1612 cache[i]->key.offset - 1);
1613 kfree(cache[i]);
1614 }
1615 }
1616 return 0;
1617}
1618
1619int btrfs_free_block_groups(struct btrfs_fs_info *info)
1620{
1621 int ret;
1622 int ret2;
1623 unsigned long gang[16];
1624 int i;
1625
1626 ret = free_block_group_radix(&info->block_group_radix);
1627 ret2 = free_block_group_radix(&info->block_group_data_radix);
1628 if (ret)
1629 return ret;
1630 if (ret2)
1631 return ret2;
1632
1633 while(1) {
1634 ret = find_first_radix_bit(&info->extent_map_radix,
1635 gang, 0, ARRAY_SIZE(gang));
1636 if (!ret)
1637 break;
1638 for (i = 0; i < ret; i++) {
1639 clear_radix_bit(&info->extent_map_radix, gang[i]);
1640 }
1641 }
1642 return 0;
1643}
1644
1645int btrfs_read_block_groups(struct btrfs_root *root)
1646{
1647 struct btrfs_path *path;
1648 int ret;
1649 int err = 0;
1650 struct btrfs_block_group_item *bi;
1651 struct btrfs_block_group_cache *cache;
1652 struct btrfs_fs_info *info = root->fs_info;
1653 struct radix_tree_root *radix;
1654 struct btrfs_key key;
1655 struct btrfs_key found_key;
1656 struct btrfs_leaf *leaf;
1657 u64 group_size_blocks;
1658 u64 used;
1659
1660 group_size_blocks = BTRFS_BLOCK_GROUP_SIZE >>
1661 root->fs_info->sb->s_blocksize_bits;
1662 root = info->extent_root;
1663 key.objectid = 0;
1664 key.offset = group_size_blocks;
1665 key.flags = 0;
1666 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
1667
1668 path = btrfs_alloc_path();
1669 if (!path)
1670 return -ENOMEM;
1671
1672 while(1) {
1673 ret = btrfs_search_slot(NULL, info->extent_root,
1674 &key, path, 0, 0);
1675 if (ret != 0) {
1676 err = ret;
1677 break;
1678 }
1679 leaf = btrfs_buffer_leaf(path->nodes[0]);
1680 btrfs_disk_key_to_cpu(&found_key,
1681 &leaf->items[path->slots[0]].key);
1682 cache = kmalloc(sizeof(*cache), GFP_NOFS);
1683 if (!cache) {
1684 err = -1;
1685 break;
1686 }
1687
1688 bi = btrfs_item_ptr(leaf, path->slots[0],
1689 struct btrfs_block_group_item);
1690 if (bi->flags & BTRFS_BLOCK_GROUP_DATA) {
1691 radix = &info->block_group_data_radix;
1692 cache->data = 1;
1693 } else {
1694 radix = &info->block_group_radix;
1695 cache->data = 0;
1696 }
1697
1698 memcpy(&cache->item, bi, sizeof(*bi));
1699 memcpy(&cache->key, &found_key, sizeof(found_key));
1700 cache->last_alloc = cache->key.objectid;
1701 cache->first_free = cache->key.objectid;
1702 cache->last_prealloc = cache->key.objectid;
1703 cache->pinned = 0;
1704 cache->cached = 0;
1705
1706 cache->radix = radix;
1707
1708 key.objectid = found_key.objectid + found_key.offset;
1709 btrfs_release_path(root, path);
1710 ret = radix_tree_insert(radix, found_key.objectid +
1711 found_key.offset - 1,
1712 (void *)cache);
1713 BUG_ON(ret);
1714 used = btrfs_block_group_used(bi);
1715 if (used < div_factor(key.offset, 8)) {
1716 radix_tree_tag_set(radix, found_key.objectid +
1717 found_key.offset - 1,
1718 BTRFS_BLOCK_GROUP_AVAIL);
1719 }
1720 if (key.objectid >=
1721 btrfs_super_total_blocks(&info->super_copy))
1722 break;
1723 }
1724
1725 btrfs_free_path(path);
1726 return 0;
1727}
Linux 6.x block layer and io_uring tree(s)