?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Fix inode link count checks in btrfsck
[btrfs-progs-unstable.git] / extent-tree.c
blobb2f9bb22daf312f14da15767cb41a89eb120c28d
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 <stdio.h>
20 #include <stdlib.h>
21 #include "kerncompat.h"
22 #include "radix-tree.h"
23 #include "ctree.h"
24 #include "disk-io.h"
25 #include "print-tree.h"
26 #include "transaction.h"
27 #include "crc32c.h"
28 #include "volumes.h"
29
30 #define BLOCK_GROUP_DATA EXTENT_WRITEBACK
31 #define BLOCK_GROUP_METADATA EXTENT_UPTODATE
32 #define BLOCK_GROUP_SYSTEM EXTENT_NEW
33
34 #define BLOCK_GROUP_DIRTY EXTENT_DIRTY
35
36 #define PENDING_EXTENT_INSERT 0
37 #define PENDING_EXTENT_DELETE 1
38 #define PENDING_BACKREF_UPDATE 2
39
40 struct pending_extent_op {
41 int type;
42 u64 bytenr;
43 u64 num_bytes;
44 u64 flags;
45 struct btrfs_disk_key key;
46 int level;
47 };
48
49 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
50 struct btrfs_root *root,
51 u64 root_objectid, u64 generation,
52 u64 flags, struct btrfs_disk_key *key,
53 int level, struct btrfs_key *ins);
54 static int __free_extent(struct btrfs_trans_handle *trans,
55 struct btrfs_root *root,
56 u64 bytenr, u64 num_bytes, u64 parent,
57 u64 root_objectid, u64 owner_objectid,
58 u64 owner_offset, int refs_to_drop);
59 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
60 btrfs_root *extent_root);
61 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
62 btrfs_root *extent_root);
63
64 static int remove_sb_from_cache(struct btrfs_root *root,
65 struct btrfs_block_group_cache *cache)
66 {
67 u64 bytenr;
68 u64 *logical;
69 int stripe_len;
70 int i, nr, ret;
71 struct extent_io_tree *free_space_cache;
72
73 free_space_cache = &root->fs_info->free_space_cache;
74 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
75 bytenr = btrfs_sb_offset(i);
76 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
77 cache->key.objectid, bytenr, 0,
78 &logical, &nr, &stripe_len);
79 BUG_ON(ret);
80 while (nr--) {
81 clear_extent_dirty(free_space_cache, logical[nr],
82 logical[nr] + stripe_len - 1, GFP_NOFS);
83 }
84 kfree(logical);
85 }
86 return 0;
87 }
88
89 static int cache_block_group(struct btrfs_root *root,
90 struct btrfs_block_group_cache *block_group)
91 {
92 struct btrfs_path *path;
93 int ret;
94 struct btrfs_key key;
95 struct extent_buffer *leaf;
96 struct extent_io_tree *free_space_cache;
97 int slot;
98 u64 last;
99 u64 hole_size;
100
101 if (!block_group)
102 return 0;
103
104 root = root->fs_info->extent_root;
105 free_space_cache = &root->fs_info->free_space_cache;
106
107 if (block_group->cached)
108 return 0;
109
110 path = btrfs_alloc_path();
111 if (!path)
112 return -ENOMEM;
113
114 path->reada = 2;
115 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
116 key.objectid = last;
117 key.offset = 0;
118 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
119 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
120 if (ret < 0)
121 goto err;
122
123 while(1) {
124 leaf = path->nodes[0];
125 slot = path->slots[0];
126 if (slot >= btrfs_header_nritems(leaf)) {
127 ret = btrfs_next_leaf(root, path);
128 if (ret < 0)
129 goto err;
130 if (ret == 0) {
131 continue;
132 } else {
133 break;
134 }
135 }
136 btrfs_item_key_to_cpu(leaf, &key, slot);
137 if (key.objectid < block_group->key.objectid) {
138 goto next;
139 }
140 if (key.objectid >= block_group->key.objectid +
141 block_group->key.offset) {
142 break;
143 }
144
145 if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
146 if (key.objectid > last) {
147 hole_size = key.objectid - last;
148 set_extent_dirty(free_space_cache, last,
149 last + hole_size - 1,
150 GFP_NOFS);
151 }
152 last = key.objectid + key.offset;
153 }
154 next:
155 path->slots[0]++;
156 }
157
158 if (block_group->key.objectid +
159 block_group->key.offset > last) {
160 hole_size = block_group->key.objectid +
161 block_group->key.offset - last;
162 set_extent_dirty(free_space_cache, last,
163 last + hole_size - 1, GFP_NOFS);
164 }
165 remove_sb_from_cache(root, block_group);
166 block_group->cached = 1;
167 err:
168 btrfs_free_path(path);
169 return 0;
170 }
171
172 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
173 btrfs_fs_info *info,
174 u64 bytenr)
175 {
176 struct extent_io_tree *block_group_cache;
177 struct btrfs_block_group_cache *block_group = NULL;
178 u64 ptr;
179 u64 start;
180 u64 end;
181 int ret;
182
183 bytenr = max_t(u64, bytenr,
184 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
185 block_group_cache = &info->block_group_cache;
186 ret = find_first_extent_bit(block_group_cache,
187 bytenr, &start, &end,
188 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
189 BLOCK_GROUP_SYSTEM);
190 if (ret) {
191 return NULL;
192 }
193 ret = get_state_private(block_group_cache, start, &ptr);
194 if (ret)
195 return NULL;
196
197 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
198 return block_group;
199 }
200
201 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
202 btrfs_fs_info *info,
203 u64 bytenr)
204 {
205 struct extent_io_tree *block_group_cache;
206 struct btrfs_block_group_cache *block_group = NULL;
207 u64 ptr;
208 u64 start;
209 u64 end;
210 int ret;
211
212 block_group_cache = &info->block_group_cache;
213 ret = find_first_extent_bit(block_group_cache,
214 bytenr, &start, &end,
215 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
216 BLOCK_GROUP_SYSTEM);
217 if (ret) {
218 return NULL;
219 }
220 ret = get_state_private(block_group_cache, start, &ptr);
221 if (ret)
222 return NULL;
223
224 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
225 if (block_group->key.objectid <= bytenr && bytenr <
226 block_group->key.objectid + block_group->key.offset)
227 return block_group;
228 return NULL;
229 }
230
231 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
232 {
233 return (cache->flags & bits) == bits;
234 }
235
236 static int noinline find_search_start(struct btrfs_root *root,
237 struct btrfs_block_group_cache **cache_ret,
238 u64 *start_ret, int num, int data)
239 {
240 int ret;
241 struct btrfs_block_group_cache *cache = *cache_ret;
242 u64 last;
243 u64 start = 0;
244 u64 end = 0;
245 u64 search_start = *start_ret;
246 int wrapped = 0;
247
248 if (!cache) {
249 goto out;
250 }
251 again:
252 ret = cache_block_group(root, cache);
253 if (ret)
254 goto out;
255
256 last = max(search_start, cache->key.objectid);
257 if (cache->ro || !block_group_bits(cache, data)) {
258 goto new_group;
259 }
260
261 while(1) {
262 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
263 last, &start, &end, EXTENT_DIRTY);
264 if (ret) {
265 goto new_group;
266 }
267
268 start = max(last, start);
269 last = end + 1;
270 if (last - start < num) {
271 continue;
272 }
273 if (start + num > cache->key.objectid + cache->key.offset) {
274 goto new_group;
275 }
276 *start_ret = start;
277 return 0;
278 }
279 out:
280 cache = btrfs_lookup_block_group(root->fs_info, search_start);
281 if (!cache) {
282 printk("Unable to find block group for %llu\n",
283 (unsigned long long)search_start);
284 WARN_ON(1);
285 }
286 return -ENOSPC;
287
288 new_group:
289 last = cache->key.objectid + cache->key.offset;
290 wrapped:
291 cache = btrfs_lookup_first_block_group(root->fs_info, last);
292 if (!cache) {
293 no_cache:
294 if (!wrapped) {
295 wrapped = 1;
296 last = search_start;
297 goto wrapped;
298 }
299 goto out;
300 }
301 cache = btrfs_find_block_group(root, cache, last, data, 0);
302 cache = btrfs_find_block_group(root, cache, last, data, 0);
303 if (!cache)
304 goto no_cache;
305
306 *cache_ret = cache;
307 goto again;
308 }
309
310 static u64 div_factor(u64 num, int factor)
311 {
312 if (factor == 10)
313 return num;
314 num *= factor;
315 num /= 10;
316 return num;
317 }
318
319 static int block_group_state_bits(u64 flags)
320 {
321 int bits = 0;
322 if (flags & BTRFS_BLOCK_GROUP_DATA)
323 bits |= BLOCK_GROUP_DATA;
324 if (flags & BTRFS_BLOCK_GROUP_METADATA)
325 bits |= BLOCK_GROUP_METADATA;
326 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
327 bits |= BLOCK_GROUP_SYSTEM;
328 return bits;
329 }
330
331 struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root,
332 struct btrfs_block_group_cache
333 *hint, u64 search_start,
334 int data, int owner)
335 {
336 struct btrfs_block_group_cache *cache;
337 struct extent_io_tree *block_group_cache;
338 struct btrfs_block_group_cache *found_group = NULL;
339 struct btrfs_fs_info *info = root->fs_info;
340 u64 used;
341 u64 last = 0;
342 u64 hint_last;
343 u64 start;
344 u64 end;
345 u64 free_check;
346 u64 ptr;
347 int bit;
348 int ret;
349 int full_search = 0;
350 int factor = 10;
351
352 block_group_cache = &info->block_group_cache;
353
354 if (!owner)
355 factor = 10;
356
357 bit = block_group_state_bits(data);
358
359 if (search_start) {
360 struct btrfs_block_group_cache *shint;
361 shint = btrfs_lookup_block_group(info, search_start);
362 if (shint && !shint->ro && block_group_bits(shint, data)) {
363 used = btrfs_block_group_used(&shint->item);
364 if (used + shint->pinned <
365 div_factor(shint->key.offset, factor)) {
366 return shint;
367 }
368 }
369 }
370 if (hint && !hint->ro && block_group_bits(hint, data)) {
371 used = btrfs_block_group_used(&hint->item);
372 if (used + hint->pinned <
373 div_factor(hint->key.offset, factor)) {
374 return hint;
375 }
376 last = hint->key.objectid + hint->key.offset;
377 hint_last = last;
378 } else {
379 if (hint)
380 hint_last = max(hint->key.objectid, search_start);
381 else
382 hint_last = search_start;
383
384 last = hint_last;
385 }
386 again:
387 while(1) {
388 ret = find_first_extent_bit(block_group_cache, last,
389 &start, &end, bit);
390 if (ret)
391 break;
392
393 ret = get_state_private(block_group_cache, start, &ptr);
394 if (ret)
395 break;
396
397 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
398 last = cache->key.objectid + cache->key.offset;
399 used = btrfs_block_group_used(&cache->item);
400
401 if (!cache->ro && block_group_bits(cache, data)) {
402 if (full_search)
403 free_check = cache->key.offset;
404 else
405 free_check = div_factor(cache->key.offset,
406 factor);
407
408 if (used + cache->pinned < free_check) {
409 found_group = cache;
410 goto found;
411 }
412 }
413 cond_resched();
414 }
415 if (!full_search) {
416 last = search_start;
417 full_search = 1;
418 goto again;
419 }
420 found:
421 return found_group;
422 }
423
424 /*
425 * Back reference rules. Back refs have three main goals:
426 *
427 * 1) differentiate between all holders of references to an extent so that
428 * when a reference is dropped we can make sure it was a valid reference
429 * before freeing the extent.
430 *
431 * 2) Provide enough information to quickly find the holders of an extent
432 * if we notice a given block is corrupted or bad.
433 *
434 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
435 * maintenance. This is actually the same as #2, but with a slightly
436 * different use case.
437 *
438 * There are two kinds of back refs. The implicit back refs is optimized
439 * for pointers in non-shared tree blocks. For a given pointer in a block,
440 * back refs of this kind provide information about the block's owner tree
441 * and the pointer's key. These information allow us to find the block by
442 * b-tree searching. The full back refs is for pointers in tree blocks not
443 * referenced by their owner trees. The location of tree block is recorded
444 * in the back refs. Actually the full back refs is generic, and can be
445 * used in all cases the implicit back refs is used. The major shortcoming
446 * of the full back refs is its overhead. Every time a tree block gets
447 * COWed, we have to update back refs entry for all pointers in it.
448 *
449 * For a newly allocated tree block, we use implicit back refs for
450 * pointers in it. This means most tree related operations only involve
451 * implicit back refs. For a tree block created in old transaction, the
452 * only way to drop a reference to it is COW it. So we can detect the
453 * event that tree block loses its owner tree's reference and do the
454 * back refs conversion.
455 *
456 * When a tree block is COW'd through a tree, there are four cases:
457 *
458 * The reference count of the block is one and the tree is the block's
459 * owner tree. Nothing to do in this case.
460 *
461 * The reference count of the block is one and the tree is not the
462 * block's owner tree. In this case, full back refs is used for pointers
463 * in the block. Remove these full back refs, add implicit back refs for
464 * every pointers in the new block.
465 *
466 * The reference count of the block is greater than one and the tree is
467 * the block's owner tree. In this case, implicit back refs is used for
468 * pointers in the block. Add full back refs for every pointers in the
469 * block, increase lower level extents' reference counts. The original
470 * implicit back refs are entailed to the new block.
471 *
472 * The reference count of the block is greater than one and the tree is
473 * not the block's owner tree. Add implicit back refs for every pointer in
474 * the new block, increase lower level extents' reference count.
475 *
476 * Back Reference Key composing:
477 *
478 * The key objectid corresponds to the first byte in the extent,
479 * The key type is used to differentiate between types of back refs.
480 * There are different meanings of the key offset for different types
481 * of back refs.
482 *
483 * File extents can be referenced by:
484 *
485 * - multiple snapshots, subvolumes, or different generations in one subvol
486 * - different files inside a single subvolume
487 * - different offsets inside a file (bookend extents in file.c)
488 *
489 * The extent ref structure for the implicit back refs has fields for:
490 *
491 * - Objectid of the subvolume root
492 * - objectid of the file holding the reference
493 * - original offset in the file
494 * - how many bookend extents
495 *
496 * The key offset for the implicit back refs is hash of the first
497 * three fields.
498 *
499 * The extent ref structure for the full back refs has field for:
500 *
501 * - number of pointers in the tree leaf
502 *
503 * The key offset for the implicit back refs is the first byte of
504 * the tree leaf
505 *
506 * When a file extent is allocated, The implicit back refs is used.
507 * the fields are filled in:
508 *
509 * (root_key.objectid, inode objectid, offset in file, 1)
510 *
511 * When a file extent is removed file truncation, we find the
512 * corresponding implicit back refs and check the following fields:
513 *
514 * (btrfs_header_owner(leaf), inode objectid, offset in file)
515 *
516 * Btree extents can be referenced by:
517 *
518 * - Different subvolumes
519 *
520 * Both the implicit back refs and the full back refs for tree blocks
521 * only consist of key. The key offset for the implicit back refs is
522 * objectid of block's owner tree. The key offset for the full back refs
523 * is the first byte of parent block.
524 *
525 * When implicit back refs is used, information about the lowest key and
526 * level of the tree block are required. These information are stored in
527 * tree block info structure.
528 */
529
530 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
531 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
532 struct btrfs_root *root,
533 struct btrfs_path *path,
534 u64 owner, u32 extra_size)
535 {
536 struct btrfs_extent_item *item;
537 struct btrfs_extent_item_v0 *ei0;
538 struct btrfs_extent_ref_v0 *ref0;
539 struct btrfs_tree_block_info *bi;
540 struct extent_buffer *leaf;
541 struct btrfs_key key;
542 struct btrfs_key found_key;
543 u32 new_size = sizeof(*item);
544 u64 refs;
545 int ret;
546
547 leaf = path->nodes[0];
548 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
549
550 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
551 ei0 = btrfs_item_ptr(leaf, path->slots[0],
552 struct btrfs_extent_item_v0);
553 refs = btrfs_extent_refs_v0(leaf, ei0);
554
555 if (owner == (u64)-1) {
556 while (1) {
557 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
558 ret = btrfs_next_leaf(root, path);
559 if (ret < 0)
560 return ret;
561 BUG_ON(ret > 0);
562 leaf = path->nodes[0];
563 }
564 btrfs_item_key_to_cpu(leaf, &found_key,
565 path->slots[0]);
566 BUG_ON(key.objectid != found_key.objectid);
567 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
568 path->slots[0]++;
569 continue;
570 }
571 ref0 = btrfs_item_ptr(leaf, path->slots[0],
572 struct btrfs_extent_ref_v0);
573 owner = btrfs_ref_objectid_v0(leaf, ref0);
574 break;
575 }
576 }
577 btrfs_release_path(root, path);
578
579 if (owner < BTRFS_FIRST_FREE_OBJECTID)
580 new_size += sizeof(*bi);
581
582 new_size -= sizeof(*ei0);
583 ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
584 if (ret < 0)
585 return ret;
586 BUG_ON(ret);
587
588 ret = btrfs_extend_item(trans, root, path, new_size);
589 BUG_ON(ret);
590
591 leaf = path->nodes[0];
592 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
593 btrfs_set_extent_refs(leaf, item, refs);
594 /* FIXME: get real generation */
595 btrfs_set_extent_generation(leaf, item, 0);
596 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
597 btrfs_set_extent_flags(leaf, item,
598 BTRFS_EXTENT_FLAG_TREE_BLOCK |
599 BTRFS_BLOCK_FLAG_FULL_BACKREF);
600 bi = (struct btrfs_tree_block_info *)(item + 1);
601 /* FIXME: get first key of the block */
602 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
603 btrfs_set_tree_block_level(leaf, bi, (int)owner);
604 } else {
605 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
606 }
607 btrfs_mark_buffer_dirty(leaf);
608 return 0;
609 }
610 #endif
611
612 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
613 {
614 u32 high_crc = ~(u32)0;
615 u32 low_crc = ~(u32)0;
616 __le64 lenum;
617
618 lenum = cpu_to_le64(root_objectid);
619 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
620 lenum = cpu_to_le64(owner);
621 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
622 lenum = cpu_to_le64(offset);
623 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
624
625 return ((u64)high_crc << 31) ^ (u64)low_crc;
626 }
627
628 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
629 struct btrfs_extent_data_ref *ref)
630 {
631 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
632 btrfs_extent_data_ref_objectid(leaf, ref),
633 btrfs_extent_data_ref_offset(leaf, ref));
634 }
635
636 static int match_extent_data_ref(struct extent_buffer *leaf,
637 struct btrfs_extent_data_ref *ref,
638 u64 root_objectid, u64 owner, u64 offset)
639 {
640 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
641 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
642 btrfs_extent_data_ref_offset(leaf, ref) != offset)
643 return 0;
644 return 1;
645 }
646
647 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
648 struct btrfs_root *root,
649 struct btrfs_path *path,
650 u64 bytenr, u64 parent,
651 u64 root_objectid,
652 u64 owner, u64 offset)
653 {
654 struct btrfs_key key;
655 struct btrfs_extent_data_ref *ref;
656 struct extent_buffer *leaf;
657 u32 nritems;
658 int ret;
659 int recow;
660 int err = -ENOENT;
661
662 key.objectid = bytenr;
663 if (parent) {
664 key.type = BTRFS_SHARED_DATA_REF_KEY;
665 key.offset = parent;
666 } else {
667 key.type = BTRFS_EXTENT_DATA_REF_KEY;
668 key.offset = hash_extent_data_ref(root_objectid,
669 owner, offset);
670 }
671 again:
672 recow = 0;
673 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
674 if (ret < 0) {
675 err = ret;
676 goto fail;
677 }
678
679 if (parent) {
680 if (!ret)
681 return 0;
682 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
683 key.type = BTRFS_EXTENT_REF_V0_KEY;
684 btrfs_release_path(root, path);
685 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
686 if (ret < 0) {
687 err = ret;
688 goto fail;
689 }
690 if (!ret)
691 return 0;
692 #endif
693 goto fail;
694 }
695
696 leaf = path->nodes[0];
697 nritems = btrfs_header_nritems(leaf);
698 while (1) {
699 if (path->slots[0] >= nritems) {
700 ret = btrfs_next_leaf(root, path);
701 if (ret < 0)
702 err = ret;
703 if (ret)
704 goto fail;
705
706 leaf = path->nodes[0];
707 nritems = btrfs_header_nritems(leaf);
708 recow = 1;
709 }
710
711 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
712 if (key.objectid != bytenr ||
713 key.type != BTRFS_EXTENT_DATA_REF_KEY)
714 goto fail;
715
716 ref = btrfs_item_ptr(leaf, path->slots[0],
717 struct btrfs_extent_data_ref);
718
719 if (match_extent_data_ref(leaf, ref, root_objectid,
720 owner, offset)) {
721 if (recow) {
722 btrfs_release_path(root, path);
723 goto again;
724 }
725 err = 0;
726 break;
727 }
728 path->slots[0]++;
729 }
730 fail:
731 return err;
732 }
733
734 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
735 struct btrfs_root *root,
736 struct btrfs_path *path,
737 u64 bytenr, u64 parent,
738 u64 root_objectid, u64 owner,
739 u64 offset, int refs_to_add)
740 {
741 struct btrfs_key key;
742 struct extent_buffer *leaf;
743 u32 size;
744 u32 num_refs;
745 int ret;
746
747 key.objectid = bytenr;
748 if (parent) {
749 key.type = BTRFS_SHARED_DATA_REF_KEY;
750 key.offset = parent;
751 size = sizeof(struct btrfs_shared_data_ref);
752 } else {
753 key.type = BTRFS_EXTENT_DATA_REF_KEY;
754 key.offset = hash_extent_data_ref(root_objectid,
755 owner, offset);
756 size = sizeof(struct btrfs_extent_data_ref);
757 }
758
759 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
760 if (ret && ret != -EEXIST)
761 goto fail;
762
763 leaf = path->nodes[0];
764 if (parent) {
765 struct btrfs_shared_data_ref *ref;
766 ref = btrfs_item_ptr(leaf, path->slots[0],
767 struct btrfs_shared_data_ref);
768 if (ret == 0) {
769 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
770 } else {
771 num_refs = btrfs_shared_data_ref_count(leaf, ref);
772 num_refs += refs_to_add;
773 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
774 }
775 } else {
776 struct btrfs_extent_data_ref *ref;
777 while (ret == -EEXIST) {
778 ref = btrfs_item_ptr(leaf, path->slots[0],
779 struct btrfs_extent_data_ref);
780 if (match_extent_data_ref(leaf, ref, root_objectid,
781 owner, offset))
782 break;
783 btrfs_release_path(root, path);
784
785 key.offset++;
786 ret = btrfs_insert_empty_item(trans, root, path, &key,
787 size);
788 if (ret && ret != -EEXIST)
789 goto fail;
790
791 leaf = path->nodes[0];
792 }
793 ref = btrfs_item_ptr(leaf, path->slots[0],
794 struct btrfs_extent_data_ref);
795 if (ret == 0) {
796 btrfs_set_extent_data_ref_root(leaf, ref,
797 root_objectid);
798 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
799 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
800 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
801 } else {
802 num_refs = btrfs_extent_data_ref_count(leaf, ref);
803 num_refs += refs_to_add;
804 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
805 }
806 }
807 btrfs_mark_buffer_dirty(leaf);
808 ret = 0;
809 fail:
810 btrfs_release_path(root, path);
811 return ret;
812 }
813
814 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
815 struct btrfs_root *root,
816 struct btrfs_path *path,
817 int refs_to_drop)
818 {
819 struct btrfs_key key;
820 struct btrfs_extent_data_ref *ref1 = NULL;
821 struct btrfs_shared_data_ref *ref2 = NULL;
822 struct extent_buffer *leaf;
823 u32 num_refs = 0;
824 int ret = 0;
825
826 leaf = path->nodes[0];
827 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
828
829 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
830 ref1 = btrfs_item_ptr(leaf, path->slots[0],
831 struct btrfs_extent_data_ref);
832 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
833 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
834 ref2 = btrfs_item_ptr(leaf, path->slots[0],
835 struct btrfs_shared_data_ref);
836 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
837 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
838 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
839 struct btrfs_extent_ref_v0 *ref0;
840 ref0 = btrfs_item_ptr(leaf, path->slots[0],
841 struct btrfs_extent_ref_v0);
842 num_refs = btrfs_ref_count_v0(leaf, ref0);
843 #endif
844 } else {
845 BUG();
846 }
847
848 BUG_ON(num_refs < refs_to_drop);
849 num_refs -= refs_to_drop;
850
851 if (num_refs == 0) {
852 ret = btrfs_del_item(trans, root, path);
853 } else {
854 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
855 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
856 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
857 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
858 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
859 else {
860 struct btrfs_extent_ref_v0 *ref0;
861 ref0 = btrfs_item_ptr(leaf, path->slots[0],
862 struct btrfs_extent_ref_v0);
863 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
864 }
865 #endif
866 btrfs_mark_buffer_dirty(leaf);
867 }
868 return ret;
869 }
870
871 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
872 struct btrfs_path *path,
873 struct btrfs_extent_inline_ref *iref)
874 {
875 struct btrfs_key key;
876 struct extent_buffer *leaf;
877 struct btrfs_extent_data_ref *ref1;
878 struct btrfs_shared_data_ref *ref2;
879 u32 num_refs = 0;
880
881 leaf = path->nodes[0];
882 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
883 if (iref) {
884 if (btrfs_extent_inline_ref_type(leaf, iref) ==
885 BTRFS_EXTENT_DATA_REF_KEY) {
886 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
887 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
888 } else {
889 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
890 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
891 }
892 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
893 ref1 = btrfs_item_ptr(leaf, path->slots[0],
894 struct btrfs_extent_data_ref);
895 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
896 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
897 ref2 = btrfs_item_ptr(leaf, path->slots[0],
898 struct btrfs_shared_data_ref);
899 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
900 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
901 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
902 struct btrfs_extent_ref_v0 *ref0;
903 ref0 = btrfs_item_ptr(leaf, path->slots[0],
904 struct btrfs_extent_ref_v0);
905 num_refs = btrfs_ref_count_v0(leaf, ref0);
906 #endif
907 } else {
908 BUG();
909 }
910 return num_refs;
911 }
912
913 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
914 struct btrfs_root *root,
915 struct btrfs_path *path,
916 u64 bytenr, u64 parent,
917 u64 root_objectid)
918 {
919 struct btrfs_key key;
920 int ret;
921
922 key.objectid = bytenr;
923 if (parent) {
924 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
925 key.offset = parent;
926 } else {
927 key.type = BTRFS_TREE_BLOCK_REF_KEY;
928 key.offset = root_objectid;
929 }
930
931 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
932 if (ret > 0)
933 ret = -ENOENT;
934 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
935 if (ret == -ENOENT && parent) {
936 btrfs_release_path(root, path);
937 key.type = BTRFS_EXTENT_REF_V0_KEY;
938 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
939 if (ret > 0)
940 ret = -ENOENT;
941 }
942 #endif
943 return ret;
944 }
945
946 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
947 struct btrfs_root *root,
948 struct btrfs_path *path,
949 u64 bytenr, u64 parent,
950 u64 root_objectid)
951 {
952 struct btrfs_key key;
953 int ret;
954
955 key.objectid = bytenr;
956 if (parent) {
957 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
958 key.offset = parent;
959 } else {
960 key.type = BTRFS_TREE_BLOCK_REF_KEY;
961 key.offset = root_objectid;
962 }
963
964 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
965
966 btrfs_release_path(root, path);
967 return ret;
968 }
969
970 static inline int extent_ref_type(u64 parent, u64 owner)
971 {
972 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
973 if (parent > 0)
974 return BTRFS_SHARED_BLOCK_REF_KEY;
975 else
976 return BTRFS_TREE_BLOCK_REF_KEY;
977 } else {
978 if (parent > 0)
979 return BTRFS_SHARED_DATA_REF_KEY;
980 else
981 return BTRFS_EXTENT_DATA_REF_KEY;
982 }
983 }
984
985 static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
986
987 {
988 int level;
989 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
990 if (!path->nodes[level])
991 break;
992 if (path->slots[level] + 1 >=
993 btrfs_header_nritems(path->nodes[level]))
994 continue;
995 if (level == 0)
996 btrfs_item_key_to_cpu(path->nodes[level], key,
997 path->slots[level] + 1);
998 else
999 btrfs_node_key_to_cpu(path->nodes[level], key,
1000 path->slots[level] + 1);
1001 return 0;
1002 }
1003 return 1;
1004 }
1005
1006 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1007 struct btrfs_root *root,
1008 struct btrfs_path *path,
1009 struct btrfs_extent_inline_ref **ref_ret,
1010 u64 bytenr, u64 num_bytes,
1011 u64 parent, u64 root_objectid,
1012 u64 owner, u64 offset, int insert)
1013 {
1014 struct btrfs_key key;
1015 struct extent_buffer *leaf;
1016 struct btrfs_extent_item *ei;
1017 struct btrfs_extent_inline_ref *iref;
1018 u64 flags;
1019 u32 item_size;
1020 unsigned long ptr;
1021 unsigned long end;
1022 int extra_size;
1023 int type;
1024 int want;
1025 int ret;
1026 int err = 0;
1027
1028 key.objectid = bytenr;
1029 key.type = BTRFS_EXTENT_ITEM_KEY;
1030 key.offset = num_bytes;
1031
1032 want = extent_ref_type(parent, owner);
1033 if (insert)
1034 extra_size = btrfs_extent_inline_ref_size(want);
1035 else
1036 extra_size = -1;
1037 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1038 if (ret < 0) {
1039 err = ret;
1040 goto out;
1041 }
1042 BUG_ON(ret);
1043
1044 leaf = path->nodes[0];
1045 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1046 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1047 if (item_size < sizeof(*ei)) {
1048 if (!insert) {
1049 err = -ENOENT;
1050 goto out;
1051 }
1052 ret = convert_extent_item_v0(trans, root, path, owner,
1053 extra_size);
1054 if (ret < 0) {
1055 err = ret;
1056 goto out;
1057 }
1058 leaf = path->nodes[0];
1059 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1060 }
1061 #endif
1062 BUG_ON(item_size < sizeof(*ei));
1063
1064 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1065 flags = btrfs_extent_flags(leaf, ei);
1066
1067 ptr = (unsigned long)(ei + 1);
1068 end = (unsigned long)ei + item_size;
1069
1070 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1071 ptr += sizeof(struct btrfs_tree_block_info);
1072 BUG_ON(ptr > end);
1073 } else {
1074 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1075 }
1076
1077 err = -ENOENT;
1078 while (1) {
1079 if (ptr >= end) {
1080 WARN_ON(ptr > end);
1081 break;
1082 }
1083 iref = (struct btrfs_extent_inline_ref *)ptr;
1084 type = btrfs_extent_inline_ref_type(leaf, iref);
1085 if (want < type)
1086 break;
1087 if (want > type) {
1088 ptr += btrfs_extent_inline_ref_size(type);
1089 continue;
1090 }
1091
1092 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1093 struct btrfs_extent_data_ref *dref;
1094 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1095 if (match_extent_data_ref(leaf, dref, root_objectid,
1096 owner, offset)) {
1097 err = 0;
1098 break;
1099 }
1100 if (hash_extent_data_ref_item(leaf, dref) <
1101 hash_extent_data_ref(root_objectid, owner, offset))
1102 break;
1103 } else {
1104 u64 ref_offset;
1105 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1106 if (parent > 0) {
1107 if (parent == ref_offset) {
1108 err = 0;
1109 break;
1110 }
1111 if (ref_offset < parent)
1112 break;
1113 } else {
1114 if (root_objectid == ref_offset) {
1115 err = 0;
1116 break;
1117 }
1118 if (ref_offset < root_objectid)
1119 break;
1120 }
1121 }
1122 ptr += btrfs_extent_inline_ref_size(type);
1123 }
1124 if (err == -ENOENT && insert) {
1125 if (item_size + extra_size >=
1126 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1127 err = -EAGAIN;
1128 goto out;
1129 }
1130 /*
1131 * To add new inline back ref, we have to make sure
1132 * there is no corresponding back ref item.
1133 * For simplicity, we just do not add new inline back
1134 * ref if there is any back ref item.
1135 */
1136 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1137 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1138 err = -EAGAIN;
1139 goto out;
1140 }
1141 }
1142 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1143 out:
1144 return err;
1145 }
1146
1147 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1148 struct btrfs_root *root,
1149 struct btrfs_path *path,
1150 struct btrfs_extent_inline_ref *iref,
1151 u64 parent, u64 root_objectid,
1152 u64 owner, u64 offset, int refs_to_add)
1153 {
1154 struct extent_buffer *leaf;
1155 struct btrfs_extent_item *ei;
1156 unsigned long ptr;
1157 unsigned long end;
1158 unsigned long item_offset;
1159 u64 refs;
1160 int size;
1161 int type;
1162 int ret;
1163
1164 leaf = path->nodes[0];
1165 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1166 item_offset = (unsigned long)iref - (unsigned long)ei;
1167
1168 type = extent_ref_type(parent, owner);
1169 size = btrfs_extent_inline_ref_size(type);
1170
1171 ret = btrfs_extend_item(trans, root, path, size);
1172 BUG_ON(ret);
1173
1174 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1175 refs = btrfs_extent_refs(leaf, ei);
1176 refs += refs_to_add;
1177 btrfs_set_extent_refs(leaf, ei, refs);
1178
1179 ptr = (unsigned long)ei + item_offset;
1180 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1181 if (ptr < end - size)
1182 memmove_extent_buffer(leaf, ptr + size, ptr,
1183 end - size - ptr);
1184
1185 iref = (struct btrfs_extent_inline_ref *)ptr;
1186 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1187 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1188 struct btrfs_extent_data_ref *dref;
1189 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1190 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1191 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1192 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1193 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1194 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1195 struct btrfs_shared_data_ref *sref;
1196 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1197 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1198 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1199 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1200 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1201 } else {
1202 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1203 }
1204 btrfs_mark_buffer_dirty(leaf);
1205 return 0;
1206 }
1207
1208 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root,
1210 struct btrfs_path *path,
1211 struct btrfs_extent_inline_ref **ref_ret,
1212 u64 bytenr, u64 num_bytes, u64 parent,
1213 u64 root_objectid, u64 owner, u64 offset)
1214 {
1215 int ret;
1216
1217 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1218 bytenr, num_bytes, parent,
1219 root_objectid, owner, offset, 0);
1220 if (ret != -ENOENT)
1221 return ret;
1222
1223 btrfs_release_path(root, path);
1224 *ref_ret = NULL;
1225
1226 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1227 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1228 root_objectid);
1229 } else {
1230 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1231 root_objectid, owner, offset);
1232 }
1233 return ret;
1234 }
1235
1236 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1237 struct btrfs_root *root,
1238 struct btrfs_path *path,
1239 struct btrfs_extent_inline_ref *iref,
1240 int refs_to_mod)
1241 {
1242 struct extent_buffer *leaf;
1243 struct btrfs_extent_item *ei;
1244 struct btrfs_extent_data_ref *dref = NULL;
1245 struct btrfs_shared_data_ref *sref = NULL;
1246 unsigned long ptr;
1247 unsigned long end;
1248 u32 item_size;
1249 int size;
1250 int type;
1251 int ret;
1252 u64 refs;
1253
1254 leaf = path->nodes[0];
1255 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1256 refs = btrfs_extent_refs(leaf, ei);
1257 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1258 refs += refs_to_mod;
1259 btrfs_set_extent_refs(leaf, ei, refs);
1260
1261 type = btrfs_extent_inline_ref_type(leaf, iref);
1262
1263 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1264 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1265 refs = btrfs_extent_data_ref_count(leaf, dref);
1266 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1267 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1268 refs = btrfs_shared_data_ref_count(leaf, sref);
1269 } else {
1270 refs = 1;
1271 BUG_ON(refs_to_mod != -1);
1272 }
1273
1274 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1275 refs += refs_to_mod;
1276
1277 if (refs > 0) {
1278 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1279 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1280 else
1281 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1282 } else {
1283 size = btrfs_extent_inline_ref_size(type);
1284 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1285 ptr = (unsigned long)iref;
1286 end = (unsigned long)ei + item_size;
1287 if (ptr + size < end)
1288 memmove_extent_buffer(leaf, ptr, ptr + size,
1289 end - ptr - size);
1290 item_size -= size;
1291 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1292 BUG_ON(ret);
1293 }
1294 btrfs_mark_buffer_dirty(leaf);
1295 return 0;
1296 }
1297
1298 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1299 struct btrfs_root *root,
1300 struct btrfs_path *path,
1301 u64 bytenr, u64 num_bytes, u64 parent,
1302 u64 root_objectid, u64 owner,
1303 u64 offset, int refs_to_add)
1304 {
1305 struct btrfs_extent_inline_ref *iref;
1306 int ret;
1307
1308 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1309 bytenr, num_bytes, parent,
1310 root_objectid, owner, offset, 1);
1311 if (ret == 0) {
1312 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1313 ret = update_inline_extent_backref(trans, root, path, iref,
1314 refs_to_add);
1315 } else if (ret == -ENOENT) {
1316 ret = setup_inline_extent_backref(trans, root, path, iref,
1317 parent, root_objectid,
1318 owner, offset, refs_to_add);
1319 }
1320 return ret;
1321 }
1322
1323 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1324 struct btrfs_root *root,
1325 struct btrfs_path *path,
1326 u64 bytenr, u64 parent, u64 root_objectid,
1327 u64 owner, u64 offset, int refs_to_add)
1328 {
1329 int ret;
1330
1331 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1332 ret = insert_extent_data_ref(trans, root, path, bytenr,
1333 parent, root_objectid,
1334 owner, offset, refs_to_add);
1335 } else {
1336 BUG_ON(refs_to_add != 1);
1337 ret = insert_tree_block_ref(trans, root, path, bytenr,
1338 parent, root_objectid);
1339 }
1340 return ret;
1341 }
1342
1343 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1344 struct btrfs_root *root,
1345 struct btrfs_path *path,
1346 struct btrfs_extent_inline_ref *iref,
1347 int refs_to_drop, int is_data)
1348 {
1349 int ret;
1350
1351 BUG_ON(!is_data && refs_to_drop != 1);
1352 if (iref) {
1353 ret = update_inline_extent_backref(trans, root, path, iref,
1354 -refs_to_drop);
1355 } else if (is_data) {
1356 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1357 } else {
1358 ret = btrfs_del_item(trans, root, path);
1359 }
1360 return ret;
1361 }
1362
1363 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1364 struct btrfs_root *root,
1365 u64 bytenr, u64 num_bytes, u64 parent,
1366 u64 root_objectid, u64 owner, u64 offset)
1367 {
1368 struct btrfs_path *path;
1369 struct extent_buffer *leaf;
1370 struct btrfs_extent_item *item;
1371 u64 refs;
1372 int ret;
1373 int err = 0;
1374
1375 path = btrfs_alloc_path();
1376 if (!path)
1377 return -ENOMEM;
1378
1379 path->reada = 1;
1380 path->leave_spinning = 1;
1381
1382 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1383 path, bytenr, num_bytes, parent,
1384 root_objectid, owner, offset, 1);
1385 if (ret == 0)
1386 goto out;
1387
1388 if (ret != -EAGAIN) {
1389 err = ret;
1390 goto out;
1391 }
1392
1393 leaf = path->nodes[0];
1394 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1395 refs = btrfs_extent_refs(leaf, item);
1396 btrfs_set_extent_refs(leaf, item, refs + 1);
1397
1398 btrfs_mark_buffer_dirty(leaf);
1399 btrfs_release_path(root->fs_info->extent_root, path);
1400
1401 path->reada = 1;
1402 path->leave_spinning = 1;
1403
1404 /* now insert the actual backref */
1405 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1406 path, bytenr, parent, root_objectid,
1407 owner, offset, 1);
1408 if (ret)
1409 err = ret;
1410 out:
1411 btrfs_free_path(path);
1412 finish_current_insert(trans, root->fs_info->extent_root);
1413 del_pending_extents(trans, root->fs_info->extent_root);
1414 BUG_ON(err);
1415 return err;
1416 }
1417
1418 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1419 struct btrfs_root *root)
1420 {
1421 finish_current_insert(trans, root->fs_info->extent_root);
1422 del_pending_extents(trans, root->fs_info->extent_root);
1423 return 0;
1424 }
1425
1426 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1427 struct btrfs_root *root, u64 bytenr,
1428 u64 num_bytes, u64 *refs, u64 *flags)
1429 {
1430 struct btrfs_path *path;
1431 int ret;
1432 struct btrfs_key key;
1433 struct extent_buffer *l;
1434 struct btrfs_extent_item *item;
1435 u32 item_size;
1436 u64 num_refs;
1437 u64 extent_flags;
1438
1439 WARN_ON(num_bytes < root->sectorsize);
1440 path = btrfs_alloc_path();
1441 path->reada = 1;
1442 key.objectid = bytenr;
1443 key.offset = num_bytes;
1444 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1445 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1446 0, 0);
1447 if (ret < 0)
1448 goto out;
1449 if (ret != 0) {
1450 btrfs_print_leaf(root, path->nodes[0]);
1451 printk("failed to find block number %Lu\n", bytenr);
1452 BUG();
1453 }
1454
1455 l = path->nodes[0];
1456 item_size = btrfs_item_size_nr(l, path->slots[0]);
1457 if (item_size >= sizeof(*item)) {
1458 item = btrfs_item_ptr(l, path->slots[0],
1459 struct btrfs_extent_item);
1460 num_refs = btrfs_extent_refs(l, item);
1461 extent_flags = btrfs_extent_flags(l, item);
1462 } else {
1463 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1464 struct btrfs_extent_item_v0 *ei0;
1465 BUG_ON(item_size != sizeof(*ei0));
1466 ei0 = btrfs_item_ptr(l, path->slots[0],
1467 struct btrfs_extent_item_v0);
1468 num_refs = btrfs_extent_refs_v0(l, ei0);
1469 /* FIXME: this isn't correct for data */
1470 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1471 #else
1472 BUG();
1473 #endif
1474 }
1475 BUG_ON(num_refs == 0);
1476 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1477 if (refs)
1478 *refs = num_refs;
1479 if (flags)
1480 *flags = extent_flags;
1481 out:
1482 btrfs_free_path(path);
1483 return 0;
1484 }
1485
1486 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1487 struct btrfs_root *root,
1488 u64 bytenr, u64 num_bytes, u64 flags)
1489 {
1490 struct btrfs_path *path;
1491 int ret;
1492 struct btrfs_key key;
1493 struct extent_buffer *l;
1494 struct btrfs_extent_item *item;
1495 u32 item_size;
1496
1497 WARN_ON(num_bytes < root->sectorsize);
1498 path = btrfs_alloc_path();
1499 path->reada = 1;
1500 key.objectid = bytenr;
1501 key.offset = num_bytes;
1502 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1503 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1504 0, 0);
1505 if (ret < 0)
1506 goto out;
1507 if (ret != 0) {
1508 btrfs_print_leaf(root, path->nodes[0]);
1509 printk("failed to find block number %Lu\n",
1510 (unsigned long long)bytenr);
1511 BUG();
1512 }
1513 l = path->nodes[0];
1514 item_size = btrfs_item_size_nr(l, path->slots[0]);
1515 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1516 if (item_size < sizeof(*item)) {
1517 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1518 path, (u64)-1, 0);
1519 if (ret < 0)
1520 goto out;
1521
1522 l = path->nodes[0];
1523 item_size = btrfs_item_size_nr(l, path->slots[0]);
1524 }
1525 #endif
1526 BUG_ON(item_size < sizeof(*item));
1527 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1528 flags |= btrfs_extent_flags(l, item);
1529 btrfs_set_extent_flags(l, item, flags);
1530 out:
1531 btrfs_free_path(path);
1532 finish_current_insert(trans, root->fs_info->extent_root);
1533 del_pending_extents(trans, root->fs_info->extent_root);
1534 return ret;
1535 }
1536
1537 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1538 struct btrfs_root *root,
1539 struct extent_buffer *buf,
1540 int record_parent, int inc)
1541 {
1542 u64 bytenr;
1543 u64 num_bytes;
1544 u64 parent;
1545 u64 ref_root;
1546 u32 nritems;
1547 struct btrfs_key key;
1548 struct btrfs_file_extent_item *fi;
1549 int i;
1550 int level;
1551 int ret = 0;
1552 int faili = 0;
1553 int (*process_func)(struct btrfs_trans_handle *trans,
1554 struct btrfs_root *root,
1555 u64, u64, u64, u64, u64, u64);
1556
1557 ref_root = btrfs_header_owner(buf);
1558 nritems = btrfs_header_nritems(buf);
1559 level = btrfs_header_level(buf);
1560
1561 if (!root->ref_cows && level == 0)
1562 return 0;
1563
1564 if (inc)
1565 process_func = btrfs_inc_extent_ref;
1566 else
1567 process_func = btrfs_free_extent;
1568
1569 if (record_parent)
1570 parent = buf->start;
1571 else
1572 parent = 0;
1573
1574 for (i = 0; i < nritems; i++) {
1575 cond_resched();
1576 if (level == 0) {
1577 btrfs_item_key_to_cpu(buf, &key, i);
1578 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1579 continue;
1580 fi = btrfs_item_ptr(buf, i,
1581 struct btrfs_file_extent_item);
1582 if (btrfs_file_extent_type(buf, fi) ==
1583 BTRFS_FILE_EXTENT_INLINE)
1584 continue;
1585 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1586 if (bytenr == 0)
1587 continue;
1588
1589 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1590 key.offset -= btrfs_file_extent_offset(buf, fi);
1591 ret = process_func(trans, root, bytenr, num_bytes,
1592 parent, ref_root, key.objectid,
1593 key.offset);
1594 if (ret) {
1595 faili = i;
1596 WARN_ON(1);
1597 goto fail;
1598 }
1599 } else {
1600 bytenr = btrfs_node_blockptr(buf, i);
1601 num_bytes = btrfs_level_size(root, level - 1);
1602 ret = process_func(trans, root, bytenr, num_bytes,
1603 parent, ref_root, level - 1, 0);
1604 if (ret) {
1605 faili = i;
1606 WARN_ON(1);
1607 goto fail;
1608 }
1609 }
1610 }
1611 return 0;
1612 fail:
1613 WARN_ON(1);
1614 #if 0
1615 for (i =0; i < faili; i++) {
1616 if (level == 0) {
1617 u64 disk_bytenr;
1618 btrfs_item_key_to_cpu(buf, &key, i);
1619 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1620 continue;
1621 fi = btrfs_item_ptr(buf, i,
1622 struct btrfs_file_extent_item);
1623 if (btrfs_file_extent_type(buf, fi) ==
1624 BTRFS_FILE_EXTENT_INLINE)
1625 continue;
1626 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1627 if (disk_bytenr == 0)
1628 continue;
1629 err = btrfs_free_extent(trans, root, disk_bytenr,
1630 btrfs_file_extent_disk_num_bytes(buf,
1631 fi), 0);
1632 BUG_ON(err);
1633 } else {
1634 bytenr = btrfs_node_blockptr(buf, i);
1635 err = btrfs_free_extent(trans, root, bytenr,
1636 btrfs_level_size(root, level - 1), 0);
1637 BUG_ON(err);
1638 }
1639 }
1640 #endif
1641 return ret;
1642 }
1643
1644 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1645 struct extent_buffer *buf, int record_parent)
1646 {
1647 return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1648 }
1649
1650 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1651 struct extent_buffer *buf, int record_parent)
1652 {
1653 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1654 }
1655
1656 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1657 struct btrfs_root *root,
1658 struct btrfs_path *path,
1659 struct btrfs_block_group_cache *cache)
1660 {
1661 int ret;
1662 int pending_ret;
1663 struct btrfs_root *extent_root = root->fs_info->extent_root;
1664 unsigned long bi;
1665 struct extent_buffer *leaf;
1666
1667 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1668 if (ret < 0)
1669 goto fail;
1670 BUG_ON(ret);
1671
1672 leaf = path->nodes[0];
1673 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1674 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1675 btrfs_mark_buffer_dirty(leaf);
1676 btrfs_release_path(extent_root, path);
1677 fail:
1678 finish_current_insert(trans, extent_root);
1679 pending_ret = del_pending_extents(trans, extent_root);
1680 if (ret)
1681 return ret;
1682 if (pending_ret)
1683 return pending_ret;
1684 return 0;
1685
1686 }
1687
1688 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1689 struct btrfs_root *root)
1690 {
1691 struct extent_io_tree *block_group_cache;
1692 struct btrfs_block_group_cache *cache;
1693 int ret;
1694 struct btrfs_path *path;
1695 u64 last = 0;
1696 u64 start;
1697 u64 end;
1698 u64 ptr;
1699
1700 block_group_cache = &root->fs_info->block_group_cache;
1701 path = btrfs_alloc_path();
1702 if (!path)
1703 return -ENOMEM;
1704
1705 while(1) {
1706 ret = find_first_extent_bit(block_group_cache, last,
1707 &start, &end, BLOCK_GROUP_DIRTY);
1708 if (ret) {
1709 if (last == 0)
1710 break;
1711 last = 0;
1712 continue;
1713 }
1714
1715 last = end + 1;
1716 ret = get_state_private(block_group_cache, start, &ptr);
1717 BUG_ON(ret);
1718
1719 clear_extent_bits(block_group_cache, start, end,
1720 BLOCK_GROUP_DIRTY, GFP_NOFS);
1721
1722 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1723 ret = write_one_cache_group(trans, root, path, cache);
1724 BUG_ON(ret);
1725 }
1726 btrfs_free_path(path);
1727 return 0;
1728 }
1729
1730 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1731 u64 flags)
1732 {
1733 struct list_head *head = &info->space_info;
1734 struct list_head *cur;
1735 struct btrfs_space_info *found;
1736 list_for_each(cur, head) {
1737 found = list_entry(cur, struct btrfs_space_info, list);
1738 if (found->flags == flags)
1739 return found;
1740 }
1741 return NULL;
1742
1743 }
1744
1745 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1746 u64 total_bytes, u64 bytes_used,
1747 struct btrfs_space_info **space_info)
1748 {
1749 struct btrfs_space_info *found;
1750
1751 found = __find_space_info(info, flags);
1752 if (found) {
1753 found->total_bytes += total_bytes;
1754 found->bytes_used += bytes_used;
1755 WARN_ON(found->total_bytes < found->bytes_used);
1756 *space_info = found;
1757 return 0;
1758 }
1759 found = kmalloc(sizeof(*found), GFP_NOFS);
1760 if (!found)
1761 return -ENOMEM;
1762
1763 list_add(&found->list, &info->space_info);
1764 found->flags = flags;
1765 found->total_bytes = total_bytes;
1766 found->bytes_used = bytes_used;
1767 found->bytes_pinned = 0;
1768 found->full = 0;
1769 *space_info = found;
1770 return 0;
1771 }
1772
1773
1774 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1775 {
1776 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1777 BTRFS_BLOCK_GROUP_RAID1 |
1778 BTRFS_BLOCK_GROUP_DUP);
1779 if (extra_flags) {
1780 if (flags & BTRFS_BLOCK_GROUP_DATA)
1781 fs_info->avail_data_alloc_bits |= extra_flags;
1782 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1783 fs_info->avail_metadata_alloc_bits |= extra_flags;
1784 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1785 fs_info->avail_system_alloc_bits |= extra_flags;
1786 }
1787 }
1788
1789 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1790 struct btrfs_root *extent_root, u64 alloc_bytes,
1791 u64 flags)
1792 {
1793 struct btrfs_space_info *space_info;
1794 u64 thresh;
1795 u64 start;
1796 u64 num_bytes;
1797 int ret;
1798
1799 space_info = __find_space_info(extent_root->fs_info, flags);
1800 if (!space_info) {
1801 ret = update_space_info(extent_root->fs_info, flags,
1802 0, 0, &space_info);
1803 BUG_ON(ret);
1804 }
1805 BUG_ON(!space_info);
1806
1807 if (space_info->full)
1808 return 0;
1809
1810 thresh = div_factor(space_info->total_bytes, 7);
1811 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1812 thresh)
1813 return 0;
1814
1815 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1816 if (ret == -ENOSPC) {
1817 space_info->full = 1;
1818 return 0;
1819 }
1820
1821 BUG_ON(ret);
1822
1823 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1824 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1825 BUG_ON(ret);
1826 return 0;
1827 }
1828
1829 static int update_block_group(struct btrfs_trans_handle *trans,
1830 struct btrfs_root *root,
1831 u64 bytenr, u64 num_bytes, int alloc,
1832 int mark_free)
1833 {
1834 struct btrfs_block_group_cache *cache;
1835 struct btrfs_fs_info *info = root->fs_info;
1836 u64 total = num_bytes;
1837 u64 old_val;
1838 u64 byte_in_group;
1839 u64 start;
1840 u64 end;
1841
1842 /* block accounting for super block */
1843 old_val = btrfs_super_bytes_used(&info->super_copy);
1844 if (alloc)
1845 old_val += num_bytes;
1846 else
1847 old_val -= num_bytes;
1848 btrfs_set_super_bytes_used(&info->super_copy, old_val);
1849
1850 /* block accounting for root item */
1851 old_val = btrfs_root_used(&root->root_item);
1852 if (alloc)
1853 old_val += num_bytes;
1854 else
1855 old_val -= num_bytes;
1856 btrfs_set_root_used(&root->root_item, old_val);
1857
1858 while(total) {
1859 cache = btrfs_lookup_block_group(info, bytenr);
1860 if (!cache) {
1861 return -1;
1862 }
1863 byte_in_group = bytenr - cache->key.objectid;
1864 WARN_ON(byte_in_group > cache->key.offset);
1865 start = cache->key.objectid;
1866 end = start + cache->key.offset - 1;
1867 set_extent_bits(&info->block_group_cache, start, end,
1868 BLOCK_GROUP_DIRTY, GFP_NOFS);
1869
1870 old_val = btrfs_block_group_used(&cache->item);
1871 num_bytes = min(total, cache->key.offset - byte_in_group);
1872 if (alloc) {
1873 old_val += num_bytes;
1874 cache->space_info->bytes_used += num_bytes;
1875 } else {
1876 old_val -= num_bytes;
1877 cache->space_info->bytes_used -= num_bytes;
1878 if (mark_free) {
1879 set_extent_dirty(&info->free_space_cache,
1880 bytenr, bytenr + num_bytes - 1,
1881 GFP_NOFS);
1882 }
1883 }
1884 btrfs_set_block_group_used(&cache->item, old_val);
1885 total -= num_bytes;
1886 bytenr += num_bytes;
1887 }
1888 return 0;
1889 }
1890
1891 static int update_pinned_extents(struct btrfs_root *root,
1892 u64 bytenr, u64 num, int pin)
1893 {
1894 u64 len;
1895 struct btrfs_block_group_cache *cache;
1896 struct btrfs_fs_info *fs_info = root->fs_info;
1897
1898 if (pin) {
1899 set_extent_dirty(&fs_info->pinned_extents,
1900 bytenr, bytenr + num - 1, GFP_NOFS);
1901 } else {
1902 clear_extent_dirty(&fs_info->pinned_extents,
1903 bytenr, bytenr + num - 1, GFP_NOFS);
1904 }
1905 while (num > 0) {
1906 cache = btrfs_lookup_block_group(fs_info, bytenr);
1907 WARN_ON(!cache);
1908 len = min(num, cache->key.offset -
1909 (bytenr - cache->key.objectid));
1910 if (pin) {
1911 cache->pinned += len;
1912 cache->space_info->bytes_pinned += len;
1913 fs_info->total_pinned += len;
1914 } else {
1915 cache->pinned -= len;
1916 cache->space_info->bytes_pinned -= len;
1917 fs_info->total_pinned -= len;
1918 }
1919 bytenr += len;
1920 num -= len;
1921 }
1922 return 0;
1923 }
1924
1925 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1926 {
1927 u64 last = 0;
1928 u64 start;
1929 u64 end;
1930 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1931 int ret;
1932
1933 while(1) {
1934 ret = find_first_extent_bit(pinned_extents, last,
1935 &start, &end, EXTENT_DIRTY);
1936 if (ret)
1937 break;
1938 set_extent_dirty(copy, start, end, GFP_NOFS);
1939 last = end + 1;
1940 }
1941 return 0;
1942 }
1943
1944 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1945 struct btrfs_root *root,
1946 struct extent_io_tree *unpin)
1947 {
1948 u64 start;
1949 u64 end;
1950 int ret;
1951 struct extent_io_tree *free_space_cache;
1952 free_space_cache = &root->fs_info->free_space_cache;
1953
1954 while(1) {
1955 ret = find_first_extent_bit(unpin, 0, &start, &end,
1956 EXTENT_DIRTY);
1957 if (ret)
1958 break;
1959 update_pinned_extents(root, start, end + 1 - start, 0);
1960 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1961 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1962 }
1963 return 0;
1964 }
1965
1966 static int finish_current_insert(struct btrfs_trans_handle *trans,
1967 struct btrfs_root *extent_root)
1968 {
1969 u64 start;
1970 u64 end;
1971 u64 priv;
1972 struct btrfs_fs_info *info = extent_root->fs_info;
1973 struct btrfs_path *path;
1974 struct pending_extent_op *extent_op;
1975 struct btrfs_key key;
1976 int ret;
1977
1978 path = btrfs_alloc_path();
1979
1980 while(1) {
1981 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1982 &end, EXTENT_LOCKED);
1983 if (ret)
1984 break;
1985
1986 ret = get_state_private(&info->extent_ins, start, &priv);
1987 BUG_ON(ret);
1988 extent_op = (struct pending_extent_op *)(unsigned long)priv;
1989
1990 if (extent_op->type == PENDING_EXTENT_INSERT) {
1991 key.objectid = start;
1992 key.offset = end + 1 - start;
1993 key.type = BTRFS_EXTENT_ITEM_KEY;
1994 ret = alloc_reserved_tree_block(trans, extent_root,
1995 extent_root->root_key.objectid,
1996 trans->transid,
1997 extent_op->flags,
1998 &extent_op->key,
1999 extent_op->level, &key);
2000 } else {
2001 BUG_ON(1);
2002 }
2003
2004 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2005 GFP_NOFS);
2006 kfree(extent_op);
2007 }
2008 btrfs_free_path(path);
2009 return 0;
2010 }
2011
2012 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2013 struct btrfs_root *root,
2014 u64 bytenr, u64 num_bytes, int is_data)
2015 {
2016 int err = 0;
2017 struct extent_buffer *buf;
2018
2019 if (is_data)
2020 goto pinit;
2021
2022 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2023 if (!buf)
2024 goto pinit;
2025
2026 /* we can reuse a block if it hasn't been written
2027 * and it is from this transaction. We can't
2028 * reuse anything from the tree log root because
2029 * it has tiny sub-transactions.
2030 */
2031 if (btrfs_buffer_uptodate(buf, 0)) {
2032 u64 header_owner = btrfs_header_owner(buf);
2033 u64 header_transid = btrfs_header_generation(buf);
2034 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2035 header_transid == trans->transid &&
2036 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2037 clean_tree_block(NULL, root, buf);
2038 free_extent_buffer(buf);
2039 return 1;
2040 }
2041 }
2042 free_extent_buffer(buf);
2043 pinit:
2044 update_pinned_extents(root, bytenr, num_bytes, 1);
2045
2046 BUG_ON(err < 0);
2047 return 0;
2048 }
2049
2050 /*
2051 * remove an extent from the root, returns 0 on success
2052 */
2053 static int __free_extent(struct btrfs_trans_handle *trans,
2054 struct btrfs_root *root,
2055 u64 bytenr, u64 num_bytes, u64 parent,
2056 u64 root_objectid, u64 owner_objectid,
2057 u64 owner_offset, int refs_to_drop)
2058 {
2059
2060 struct btrfs_key key;
2061 struct btrfs_path *path;
2062 struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2063 struct btrfs_root *extent_root = root->fs_info->extent_root;
2064 struct extent_buffer *leaf;
2065 struct btrfs_extent_item *ei;
2066 struct btrfs_extent_inline_ref *iref;
2067 int ret;
2068 int is_data;
2069 int extent_slot = 0;
2070 int found_extent = 0;
2071 int num_to_del = 1;
2072 u32 item_size;
2073 u64 refs;
2074
2075 path = btrfs_alloc_path();
2076 if (!path)
2077 return -ENOMEM;
2078
2079 path->reada = 1;
2080 path->leave_spinning = 1;
2081
2082 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2083 BUG_ON(!is_data && refs_to_drop != 1);
2084
2085 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2086 bytenr, num_bytes, parent,
2087 root_objectid, owner_objectid,
2088 owner_offset);
2089 if (ret == 0) {
2090 extent_slot = path->slots[0];
2091 while (extent_slot >= 0) {
2092 btrfs_item_key_to_cpu(path->nodes[0], &key,
2093 extent_slot);
2094 if (key.objectid != bytenr)
2095 break;
2096 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2097 key.offset == num_bytes) {
2098 found_extent = 1;
2099 break;
2100 }
2101 if (path->slots[0] - extent_slot > 5)
2102 break;
2103 extent_slot--;
2104 }
2105 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2106 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2107 if (found_extent && item_size < sizeof(*ei))
2108 found_extent = 0;
2109 #endif
2110 if (!found_extent) {
2111 BUG_ON(iref);
2112 ret = remove_extent_backref(trans, extent_root, path,
2113 NULL, refs_to_drop,
2114 is_data);
2115 BUG_ON(ret);
2116 btrfs_release_path(extent_root, path);
2117 path->leave_spinning = 1;
2118
2119 key.objectid = bytenr;
2120 key.type = BTRFS_EXTENT_ITEM_KEY;
2121 key.offset = num_bytes;
2122
2123 ret = btrfs_search_slot(trans, extent_root,
2124 &key, path, -1, 1);
2125 if (ret) {
2126 printk(KERN_ERR "umm, got %d back from search"
2127 ", was looking for %llu\n", ret,
2128 (unsigned long long)bytenr);
2129 btrfs_print_leaf(extent_root, path->nodes[0]);
2130 }
2131 BUG_ON(ret);
2132 extent_slot = path->slots[0];
2133 }
2134 } else {
2135 btrfs_print_leaf(extent_root, path->nodes[0]);
2136 WARN_ON(1);
2137 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2138 "parent %llu root %llu owner %llu offset %llu\n",
2139 (unsigned long long)bytenr,
2140 (unsigned long long)parent,
2141 (unsigned long long)root_objectid,
2142 (unsigned long long)owner_objectid,
2143 (unsigned long long)owner_offset);
2144 }
2145
2146 leaf = path->nodes[0];
2147 item_size = btrfs_item_size_nr(leaf, extent_slot);
2148 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2149 if (item_size < sizeof(*ei)) {
2150 BUG_ON(found_extent || extent_slot != path->slots[0]);
2151 ret = convert_extent_item_v0(trans, extent_root, path,
2152 owner_objectid, 0);
2153 BUG_ON(ret < 0);
2154
2155 btrfs_release_path(extent_root, path);
2156 path->leave_spinning = 1;
2157
2158 key.objectid = bytenr;
2159 key.type = BTRFS_EXTENT_ITEM_KEY;
2160 key.offset = num_bytes;
2161
2162 ret = btrfs_search_slot(trans, extent_root, &key, path,
2163 -1, 1);
2164 if (ret) {
2165 printk(KERN_ERR "umm, got %d back from search"
2166 ", was looking for %llu\n", ret,
2167 (unsigned long long)bytenr);
2168 btrfs_print_leaf(extent_root, path->nodes[0]);
2169 }
2170 BUG_ON(ret);
2171 extent_slot = path->slots[0];
2172 leaf = path->nodes[0];
2173 item_size = btrfs_item_size_nr(leaf, extent_slot);
2174 }
2175 #endif
2176 BUG_ON(item_size < sizeof(*ei));
2177 ei = btrfs_item_ptr(leaf, extent_slot,
2178 struct btrfs_extent_item);
2179 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2180 struct btrfs_tree_block_info *bi;
2181 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2182 bi = (struct btrfs_tree_block_info *)(ei + 1);
2183 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2184 }
2185
2186 refs = btrfs_extent_refs(leaf, ei);
2187 BUG_ON(refs < refs_to_drop);
2188 refs -= refs_to_drop;
2189
2190 if (refs > 0) {
2191 /*
2192 * In the case of inline back ref, reference count will
2193 * be updated by remove_extent_backref
2194 */
2195 if (iref) {
2196 BUG_ON(!found_extent);
2197 } else {
2198 btrfs_set_extent_refs(leaf, ei, refs);
2199 btrfs_mark_buffer_dirty(leaf);
2200 }
2201 if (found_extent) {
2202 ret = remove_extent_backref(trans, extent_root, path,
2203 iref, refs_to_drop,
2204 is_data);
2205 BUG_ON(ret);
2206 }
2207 } else {
2208 int mark_free = 0;
2209 int pin = 1;
2210
2211 if (found_extent) {
2212 BUG_ON(is_data && refs_to_drop !=
2213 extent_data_ref_count(root, path, iref));
2214 if (iref) {
2215 BUG_ON(path->slots[0] != extent_slot);
2216 } else {
2217 BUG_ON(path->slots[0] != extent_slot + 1);
2218 path->slots[0] = extent_slot;
2219 num_to_del = 2;
2220 }
2221 }
2222
2223 if (ops && ops->free_extent) {
2224 ret = ops->free_extent(root, bytenr, num_bytes);
2225 if (ret > 0) {
2226 pin = 0;
2227 mark_free = 0;
2228 }
2229 }
2230
2231 if (pin) {
2232 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2233 is_data);
2234 if (ret > 0)
2235 mark_free = 1;
2236 BUG_ON(ret < 0);
2237 }
2238
2239 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2240 num_to_del);
2241 BUG_ON(ret);
2242 btrfs_release_path(extent_root, path);
2243
2244 if (is_data) {
2245 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2246 BUG_ON(ret);
2247 }
2248
2249 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2250 mark_free);
2251 BUG_ON(ret);
2252 }
2253 btrfs_free_path(path);
2254 finish_current_insert(trans, extent_root);
2255 return ret;
2256 }
2257
2258 /*
2259 * find all the blocks marked as pending in the radix tree and remove
2260 * them from the extent map
2261 */
2262 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2263 btrfs_root *extent_root)
2264 {
2265 int ret;
2266 int err = 0;
2267 u64 start;
2268 u64 end;
2269 u64 priv;
2270 struct extent_io_tree *pending_del;
2271 struct extent_io_tree *extent_ins;
2272 struct pending_extent_op *extent_op;
2273
2274 extent_ins = &extent_root->fs_info->extent_ins;
2275 pending_del = &extent_root->fs_info->pending_del;
2276
2277 while(1) {
2278 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2279 EXTENT_LOCKED);
2280 if (ret)
2281 break;
2282
2283 ret = get_state_private(pending_del, start, &priv);
2284 BUG_ON(ret);
2285 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2286
2287 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2288 GFP_NOFS);
2289
2290 if (!test_range_bit(extent_ins, start, end,
2291 EXTENT_LOCKED, 0)) {
2292 ret = __free_extent(trans, extent_root,
2293 start, end + 1 - start, 0,
2294 extent_root->root_key.objectid,
2295 extent_op->level, 0, 1);
2296 kfree(extent_op);
2297 } else {
2298 kfree(extent_op);
2299 ret = get_state_private(extent_ins, start, &priv);
2300 BUG_ON(ret);
2301 extent_op = (struct pending_extent_op *)
2302 (unsigned long)priv;
2303
2304 clear_extent_bits(extent_ins, start, end,
2305 EXTENT_LOCKED, GFP_NOFS);
2306
2307 if (extent_op->type == PENDING_BACKREF_UPDATE)
2308 BUG_ON(1);
2309
2310 kfree(extent_op);
2311 }
2312 if (ret)
2313 err = ret;
2314 }
2315 return err;
2316 }
2317
2318 /*
2319 * remove an extent from the root, returns 0 on success
2320 */
2321
2322 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2323 struct btrfs_root *root,
2324 u64 bytenr, u64 num_bytes, u64 parent,
2325 u64 root_objectid, u64 owner, u64 offset)
2326 {
2327 struct btrfs_root *extent_root = root->fs_info->extent_root;
2328 int pending_ret;
2329 int ret;
2330
2331 WARN_ON(num_bytes < root->sectorsize);
2332 if (root == extent_root) {
2333 struct pending_extent_op *extent_op;
2334
2335 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2336 BUG_ON(!extent_op);
2337
2338 extent_op->type = PENDING_EXTENT_DELETE;
2339 extent_op->bytenr = bytenr;
2340 extent_op->num_bytes = num_bytes;
2341 extent_op->level = (int)owner;
2342
2343 set_extent_bits(&root->fs_info->pending_del,
2344 bytenr, bytenr + num_bytes - 1,
2345 EXTENT_LOCKED, GFP_NOFS);
2346 set_state_private(&root->fs_info->pending_del,
2347 bytenr, (unsigned long)extent_op);
2348 return 0;
2349 }
2350 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2351 root_objectid, owner, offset, 1);
2352 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2353 return ret ? ret : pending_ret;
2354 }
2355
2356 static u64 stripe_align(struct btrfs_root *root, u64 val)
2357 {
2358 u64 mask = ((u64)root->stripesize - 1);
2359 u64 ret = (val + mask) & ~mask;
2360 return ret;
2361 }
2362
2363 /*
2364 * walks the btree of allocated extents and find a hole of a given size.
2365 * The key ins is changed to record the hole:
2366 * ins->objectid == block start
2367 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2368 * ins->offset == number of blocks
2369 * Any available blocks before search_start are skipped.
2370 */
2371 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2372 struct btrfs_root *orig_root,
2373 u64 num_bytes, u64 empty_size,
2374 u64 search_start, u64 search_end,
2375 u64 hint_byte, struct btrfs_key *ins,
2376 u64 exclude_start, u64 exclude_nr,
2377 int data)
2378 {
2379 int ret;
2380 u64 orig_search_start = search_start;
2381 struct btrfs_root * root = orig_root->fs_info->extent_root;
2382 struct btrfs_fs_info *info = root->fs_info;
2383 u64 total_needed = num_bytes;
2384 struct btrfs_block_group_cache *block_group;
2385 int full_scan = 0;
2386 int wrapped = 0;
2387
2388 WARN_ON(num_bytes < root->sectorsize);
2389 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2390
2391 if (hint_byte) {
2392 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2393 if (!block_group)
2394 hint_byte = search_start;
2395 block_group = btrfs_find_block_group(root, block_group,
2396 hint_byte, data, 1);
2397 } else {
2398 block_group = btrfs_find_block_group(root,
2399 trans->block_group,
2400 search_start, data, 1);
2401 }
2402
2403 total_needed += empty_size;
2404
2405 check_failed:
2406 if (!block_group) {
2407 block_group = btrfs_lookup_first_block_group(info,
2408 search_start);
2409 if (!block_group)
2410 block_group = btrfs_lookup_first_block_group(info,
2411 orig_search_start);
2412 }
2413 ret = find_search_start(root, &block_group, &search_start,
2414 total_needed, data);
2415 if (ret)
2416 goto error;
2417
2418 search_start = stripe_align(root, search_start);
2419 ins->objectid = search_start;
2420 ins->offset = num_bytes;
2421
2422 if (ins->objectid + num_bytes >
2423 block_group->key.objectid + block_group->key.offset) {
2424 search_start = block_group->key.objectid +
2425 block_group->key.offset;
2426 goto new_group;
2427 }
2428
2429 if (test_range_bit(&info->extent_ins, ins->objectid,
2430 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2431 search_start = ins->objectid + num_bytes;
2432 goto new_group;
2433 }
2434
2435 if (test_range_bit(&info->pinned_extents, ins->objectid,
2436 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2437 search_start = ins->objectid + num_bytes;
2438 goto new_group;
2439 }
2440
2441 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2442 ins->objectid < exclude_start + exclude_nr)) {
2443 search_start = exclude_start + exclude_nr;
2444 goto new_group;
2445 }
2446
2447 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2448 block_group = btrfs_lookup_block_group(info, ins->objectid);
2449 if (block_group)
2450 trans->block_group = block_group;
2451 }
2452 ins->offset = num_bytes;
2453 return 0;
2454
2455 new_group:
2456 block_group = btrfs_lookup_first_block_group(info, search_start);
2457 if (!block_group) {
2458 search_start = orig_search_start;
2459 if (full_scan) {
2460 ret = -ENOSPC;
2461 goto error;
2462 }
2463 if (wrapped) {
2464 if (!full_scan)
2465 total_needed -= empty_size;
2466 full_scan = 1;
2467 } else
2468 wrapped = 1;
2469 }
2470 cond_resched();
2471 block_group = btrfs_find_block_group(root, block_group,
2472 search_start, data, 0);
2473 goto check_failed;
2474
2475 error:
2476 return ret;
2477 }
2478
2479 static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2480 struct btrfs_root *root,
2481 u64 num_bytes, u64 empty_size,
2482 u64 hint_byte, u64 search_end,
2483 struct btrfs_key *ins, int data)
2484 {
2485 int ret;
2486 u64 search_start = 0;
2487 u64 alloc_profile;
2488 struct btrfs_fs_info *info = root->fs_info;
2489
2490 if (info->extent_ops) {
2491 struct btrfs_extent_ops *ops = info->extent_ops;
2492 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2493 BUG_ON(ret);
2494 goto found;
2495 }
2496
2497 if (data) {
2498 alloc_profile = info->avail_data_alloc_bits &
2499 info->data_alloc_profile;
2500 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2501 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2502 info->system_allocs >= 0) {
2503 alloc_profile = info->avail_system_alloc_bits &
2504 info->system_alloc_profile;
2505 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2506 } else {
2507 alloc_profile = info->avail_metadata_alloc_bits &
2508 info->metadata_alloc_profile;
2509 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2510 }
2511
2512 if (root->ref_cows) {
2513 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2514 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2515 num_bytes,
2516 BTRFS_BLOCK_GROUP_METADATA);
2517 BUG_ON(ret);
2518 }
2519 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2520 num_bytes + 2 * 1024 * 1024, data);
2521 BUG_ON(ret);
2522 }
2523
2524 WARN_ON(num_bytes < root->sectorsize);
2525 ret = find_free_extent(trans, root, num_bytes, empty_size,
2526 search_start, search_end, hint_byte, ins,
2527 trans->alloc_exclude_start,
2528 trans->alloc_exclude_nr, data);
2529 BUG_ON(ret);
2530 found:
2531 clear_extent_dirty(&root->fs_info->free_space_cache,
2532 ins->objectid, ins->objectid + ins->offset - 1,
2533 GFP_NOFS);
2534 return ret;
2535 }
2536
2537 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2538 struct btrfs_root *root,
2539 u64 root_objectid, u64 generation,
2540 u64 flags, struct btrfs_disk_key *key,
2541 int level, struct btrfs_key *ins)
2542 {
2543 int ret;
2544 struct btrfs_fs_info *fs_info = root->fs_info;
2545 struct btrfs_extent_item *extent_item;
2546 struct btrfs_tree_block_info *block_info;
2547 struct btrfs_extent_inline_ref *iref;
2548 struct btrfs_path *path;
2549 struct extent_buffer *leaf;
2550 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
2551
2552 path = btrfs_alloc_path();
2553 BUG_ON(!path);
2554
2555 path->leave_spinning = 1;
2556 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2557 ins, size);
2558 BUG_ON(ret);
2559
2560 leaf = path->nodes[0];
2561 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2562 struct btrfs_extent_item);
2563 btrfs_set_extent_refs(leaf, extent_item, 1);
2564 btrfs_set_extent_generation(leaf, extent_item, generation);
2565 btrfs_set_extent_flags(leaf, extent_item,
2566 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2567 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2568
2569 btrfs_set_tree_block_key(leaf, block_info, key);
2570 btrfs_set_tree_block_level(leaf, block_info, level);
2571
2572 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2573 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2574 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2575
2576 btrfs_mark_buffer_dirty(leaf);
2577 btrfs_free_path(path);
2578
2579 ret = update_block_group(trans, root, ins->objectid, ins->offset,
2580 1, 0);
2581 if (ret) {
2582 printk(KERN_ERR "btrfs update block group failed for %llu "
2583 "%llu\n", (unsigned long long)ins->objectid,
2584 (unsigned long long)ins->offset);
2585 BUG();
2586 }
2587 return ret;
2588 }
2589
2590 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2591 struct btrfs_root *root, u64 num_bytes,
2592 u64 root_objectid, u64 generation,
2593 u64 flags, struct btrfs_disk_key *key,
2594 int level, u64 empty_size, u64 hint_byte,
2595 u64 search_end, struct btrfs_key *ins)
2596 {
2597 int ret;
2598 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2599 hint_byte, search_end, ins, 0);
2600 BUG_ON(ret);
2601
2602 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2603 struct pending_extent_op *extent_op;
2604
2605 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2606 BUG_ON(!extent_op);
2607
2608 extent_op->type = PENDING_EXTENT_INSERT;
2609 extent_op->bytenr = ins->objectid;
2610 extent_op->num_bytes = ins->offset;
2611 extent_op->level = level;
2612 extent_op->flags = flags;
2613 memcpy(&extent_op->key, key, sizeof(*key));
2614
2615 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2616 ins->objectid + ins->offset - 1,
2617 EXTENT_LOCKED, GFP_NOFS);
2618 set_state_private(&root->fs_info->extent_ins,
2619 ins->objectid, (unsigned long)extent_op);
2620 } else {
2621 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2622 generation, flags,
2623 key, level, ins);
2624 finish_current_insert(trans, root->fs_info->extent_root);
2625 del_pending_extents(trans, root->fs_info->extent_root);
2626 }
2627 return ret;
2628 }
2629
2630 /*
2631 * helper function to allocate a block for a given tree
2632 * returns the tree buffer or NULL.
2633 */
2634 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2635 struct btrfs_root *root,
2636 u32 blocksize, u64 root_objectid,
2637 struct btrfs_disk_key *key, int level,
2638 u64 hint, u64 empty_size)
2639 {
2640 struct btrfs_key ins;
2641 int ret;
2642 struct extent_buffer *buf;
2643
2644 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2645 trans->transid, 0, key, level,
2646 empty_size, hint, (u64)-1, &ins);
2647 if (ret) {
2648 BUG_ON(ret > 0);
2649 return ERR_PTR(ret);
2650 }
2651
2652 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2653 if (!buf) {
2654 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2655 0, root->root_key.objectid, level, 0);
2656 BUG_ON(1);
2657 return ERR_PTR(-ENOMEM);
2658 }
2659 btrfs_set_buffer_uptodate(buf);
2660 trans->blocks_used++;
2661
2662 return buf;
2663 }
2664
2665 #if 0
2666
2667 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2668 struct btrfs_root *root,
2669 struct extent_buffer *leaf)
2670 {
2671 u64 leaf_owner;
2672 u64 leaf_generation;
2673 struct btrfs_key key;
2674 struct btrfs_file_extent_item *fi;
2675 int i;
2676 int nritems;
2677 int ret;
2678
2679 BUG_ON(!btrfs_is_leaf(leaf));
2680 nritems = btrfs_header_nritems(leaf);
2681 leaf_owner = btrfs_header_owner(leaf);
2682 leaf_generation = btrfs_header_generation(leaf);
2683
2684 for (i = 0; i < nritems; i++) {
2685 u64 disk_bytenr;
2686
2687 btrfs_item_key_to_cpu(leaf, &key, i);
2688 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2689 continue;
2690 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2691 if (btrfs_file_extent_type(leaf, fi) ==
2692 BTRFS_FILE_EXTENT_INLINE)
2693 continue;
2694 /*
2695 * FIXME make sure to insert a trans record that
2696 * repeats the snapshot del on crash
2697 */
2698 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2699 if (disk_bytenr == 0)
2700 continue;
2701 ret = btrfs_free_extent(trans, root, disk_bytenr,
2702 btrfs_file_extent_disk_num_bytes(leaf, fi),
2703 leaf->start, leaf_owner, leaf_generation,
2704 key.objectid, 0);
2705 BUG_ON(ret);
2706 }
2707 return 0;
2708 }
2709
2710 static void noinline reada_walk_down(struct btrfs_root *root,
2711 struct extent_buffer *node,
2712 int slot)
2713 {
2714 u64 bytenr;
2715 u64 last = 0;
2716 u32 nritems;
2717 u32 refs;
2718 u32 blocksize;
2719 int ret;
2720 int i;
2721 int level;
2722 int skipped = 0;
2723
2724 nritems = btrfs_header_nritems(node);
2725 level = btrfs_header_level(node);
2726 if (level)
2727 return;
2728
2729 for (i = slot; i < nritems && skipped < 32; i++) {
2730 bytenr = btrfs_node_blockptr(node, i);
2731 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2732 (last > bytenr && last - bytenr > 32 * 1024))) {
2733 skipped++;
2734 continue;
2735 }
2736 blocksize = btrfs_level_size(root, level - 1);
2737 if (i != slot) {
2738 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2739 blocksize, &refs);
2740 BUG_ON(ret);
2741 if (refs != 1) {
2742 skipped++;
2743 continue;
2744 }
2745 }
2746 mutex_unlock(&root->fs_info->fs_mutex);
2747 ret = readahead_tree_block(root, bytenr, blocksize,
2748 btrfs_node_ptr_generation(node, i));
2749 last = bytenr + blocksize;
2750 cond_resched();
2751 mutex_lock(&root->fs_info->fs_mutex);
2752 if (ret)
2753 break;
2754 }
2755 }
2756
2757 /*
2758 * helper function for drop_snapshot, this walks down the tree dropping ref
2759 * counts as it goes.
2760 */
2761 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2762 struct btrfs_root *root,
2763 struct btrfs_path *path, int *level)
2764 {
2765 u64 root_owner;
2766 u64 root_gen;
2767 u64 bytenr;
2768 u64 ptr_gen;
2769 struct extent_buffer *next;
2770 struct extent_buffer *cur;
2771 struct extent_buffer *parent;
2772 u32 blocksize;
2773 int ret;
2774 u32 refs;
2775
2776 WARN_ON(*level < 0);
2777 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2778 ret = btrfs_lookup_extent_ref(trans, root,
2779 path->nodes[*level]->start,
2780 path->nodes[*level]->len, &refs);
2781 BUG_ON(ret);
2782 if (refs > 1)
2783 goto out;
2784
2785 /*
2786 * walk down to the last node level and free all the leaves
2787 */
2788 while(*level >= 0) {
2789 WARN_ON(*level < 0);
2790 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2791 cur = path->nodes[*level];
2792
2793 if (btrfs_header_level(cur) != *level)
2794 WARN_ON(1);
2795
2796 if (path->slots[*level] >=
2797 btrfs_header_nritems(cur))
2798 break;
2799 if (*level == 0) {
2800 ret = drop_leaf_ref(trans, root, cur);
2801 BUG_ON(ret);
2802 break;
2803 }
2804 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2805 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2806 blocksize = btrfs_level_size(root, *level - 1);
2807 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2808 &refs);
2809 BUG_ON(ret);
2810 if (refs != 1) {
2811 parent = path->nodes[*level];
2812 root_owner = btrfs_header_owner(parent);
2813 root_gen = btrfs_header_generation(parent);
2814 path->slots[*level]++;
2815 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2816 parent->start, root_owner,
2817 root_gen, *level - 1, 1);
2818 BUG_ON(ret);
2819 continue;
2820 }
2821 next = btrfs_find_tree_block(root, bytenr, blocksize);
2822 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2823 free_extent_buffer(next);
2824 reada_walk_down(root, cur, path->slots[*level]);
2825 mutex_unlock(&root->fs_info->fs_mutex);
2826 next = read_tree_block(root, bytenr, blocksize,
2827 ptr_gen);
2828 mutex_lock(&root->fs_info->fs_mutex);
2829 }
2830 WARN_ON(*level <= 0);
2831 if (path->nodes[*level-1])
2832 free_extent_buffer(path->nodes[*level-1]);
2833 path->nodes[*level-1] = next;
2834 *level = btrfs_header_level(next);
2835 path->slots[*level] = 0;
2836 }
2837 out:
2838 WARN_ON(*level < 0);
2839 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2840
2841 if (path->nodes[*level] == root->node) {
2842 root_owner = root->root_key.objectid;
2843 parent = path->nodes[*level];
2844 } else {
2845 parent = path->nodes[*level + 1];
2846 root_owner = btrfs_header_owner(parent);
2847 }
2848
2849 root_gen = btrfs_header_generation(parent);
2850 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2851 path->nodes[*level]->len, parent->start,
2852 root_owner, root_gen, *level, 1);
2853 free_extent_buffer(path->nodes[*level]);
2854 path->nodes[*level] = NULL;
2855 *level += 1;
2856 BUG_ON(ret);
2857 return 0;
2858 }
2859
2860 /*
2861 * helper for dropping snapshots. This walks back up the tree in the path
2862 * to find the first node higher up where we haven't yet gone through
2863 * all the slots
2864 */
2865 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2866 struct btrfs_root *root,
2867 struct btrfs_path *path, int *level)
2868 {
2869 u64 root_owner;
2870 u64 root_gen;
2871 struct btrfs_root_item *root_item = &root->root_item;
2872 int i;
2873 int slot;
2874 int ret;
2875
2876 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2877 slot = path->slots[i];
2878 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2879 struct extent_buffer *node;
2880 struct btrfs_disk_key disk_key;
2881 node = path->nodes[i];
2882 path->slots[i]++;
2883 *level = i;
2884 WARN_ON(*level == 0);
2885 btrfs_node_key(node, &disk_key, path->slots[i]);
2886 memcpy(&root_item->drop_progress,
2887 &disk_key, sizeof(disk_key));
2888 root_item->drop_level = i;
2889 return 0;
2890 } else {
2891 struct extent_buffer *parent;
2892 if (path->nodes[*level] == root->node)
2893 parent = path->nodes[*level];
2894 else
2895 parent = path->nodes[*level + 1];
2896
2897 root_owner = btrfs_header_owner(parent);
2898 root_gen = btrfs_header_generation(parent);
2899 ret = btrfs_free_extent(trans, root,
2900 path->nodes[*level]->start,
2901 path->nodes[*level]->len,
2902 parent->start, root_owner,
2903 root_gen, *level, 1);
2904 BUG_ON(ret);
2905 free_extent_buffer(path->nodes[*level]);
2906 path->nodes[*level] = NULL;
2907 *level = i + 1;
2908 }
2909 }
2910 return 1;
2911 }
2912
2913 /*
2914 * drop the reference count on the tree rooted at 'snap'. This traverses
2915 * the tree freeing any blocks that have a ref count of zero after being
2916 * decremented.
2917 */
2918 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2919 *root)
2920 {
2921 int ret = 0;
2922 int wret;
2923 int level;
2924 struct btrfs_path *path;
2925 int i;
2926 int orig_level;
2927 struct btrfs_root_item *root_item = &root->root_item;
2928
2929 path = btrfs_alloc_path();
2930 BUG_ON(!path);
2931
2932 level = btrfs_header_level(root->node);
2933 orig_level = level;
2934 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2935 path->nodes[level] = root->node;
2936 extent_buffer_get(root->node);
2937 path->slots[level] = 0;
2938 } else {
2939 struct btrfs_key key;
2940 struct btrfs_disk_key found_key;
2941 struct extent_buffer *node;
2942
2943 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2944 level = root_item->drop_level;
2945 path->lowest_level = level;
2946 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2947 if (wret < 0) {
2948 ret = wret;
2949 goto out;
2950 }
2951 node = path->nodes[level];
2952 btrfs_node_key(node, &found_key, path->slots[level]);
2953 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2954 sizeof(found_key)));
2955 }
2956 while(1) {
2957 wret = walk_down_tree(trans, root, path, &level);
2958 if (wret < 0)
2959 ret = wret;
2960 if (wret != 0)
2961 break;
2962
2963 wret = walk_up_tree(trans, root, path, &level);
2964 if (wret < 0)
2965 ret = wret;
2966 if (wret != 0)
2967 break;
2968 /*
2969 ret = -EAGAIN;
2970 break;
2971 */
2972 }
2973 for (i = 0; i <= orig_level; i++) {
2974 if (path->nodes[i]) {
2975 free_extent_buffer(path->nodes[i]);
2976 path->nodes[i] = NULL;
2977 }
2978 }
2979 out:
2980 btrfs_free_path(path);
2981 return ret;
2982 }
2983
2984 #endif
2985
2986 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2987 {
2988 u64 start;
2989 u64 end;
2990 u64 ptr;
2991 int ret;
2992 while(1) {
2993 ret = find_first_extent_bit(&info->block_group_cache, 0,
2994 &start, &end, (unsigned int)-1);
2995 if (ret)
2996 break;
2997 ret = get_state_private(&info->block_group_cache, start, &ptr);
2998 if (!ret)
2999 kfree((void *)(unsigned long)ptr);
3000 clear_extent_bits(&info->block_group_cache, start,
3001 end, (unsigned int)-1, GFP_NOFS);
3002 }
3003 while(1) {
3004 ret = find_first_extent_bit(&info->free_space_cache, 0,
3005 &start, &end, EXTENT_DIRTY);
3006 if (ret)
3007 break;
3008 clear_extent_dirty(&info->free_space_cache, start,
3009 end, GFP_NOFS);
3010 }
3011 return 0;
3012 }
3013
3014 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3015 struct btrfs_key *key)
3016 {
3017 int ret;
3018 struct btrfs_key found_key;
3019 struct extent_buffer *leaf;
3020 int slot;
3021
3022 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3023 if (ret < 0)
3024 return ret;
3025 while(1) {
3026 slot = path->slots[0];
3027 leaf = path->nodes[0];
3028 if (slot >= btrfs_header_nritems(leaf)) {
3029 ret = btrfs_next_leaf(root, path);
3030 if (ret == 0)
3031 continue;
3032 if (ret < 0)
3033 goto error;
3034 break;
3035 }
3036 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3037
3038 if (found_key.objectid >= key->objectid &&
3039 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3040 return 0;
3041 path->slots[0]++;
3042 }
3043 ret = -ENOENT;
3044 error:
3045 return ret;
3046 }
3047
3048 int btrfs_read_block_groups(struct btrfs_root *root)
3049 {
3050 struct btrfs_path *path;
3051 int ret;
3052 int bit;
3053 struct btrfs_block_group_cache *cache;
3054 struct btrfs_fs_info *info = root->fs_info;
3055 struct btrfs_space_info *space_info;
3056 struct extent_io_tree *block_group_cache;
3057 struct btrfs_key key;
3058 struct btrfs_key found_key;
3059 struct extent_buffer *leaf;
3060
3061 block_group_cache = &info->block_group_cache;
3062
3063 root = info->extent_root;
3064 key.objectid = 0;
3065 key.offset = 0;
3066 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3067 path = btrfs_alloc_path();
3068 if (!path)
3069 return -ENOMEM;
3070
3071 while(1) {
3072 ret = find_first_block_group(root, path, &key);
3073 if (ret > 0) {
3074 ret = 0;
3075 goto error;
3076 }
3077 if (ret != 0) {
3078 goto error;
3079 }
3080 leaf = path->nodes[0];
3081 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3082 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3083 if (!cache) {
3084 ret = -ENOMEM;
3085 break;
3086 }
3087
3088 read_extent_buffer(leaf, &cache->item,
3089 btrfs_item_ptr_offset(leaf, path->slots[0]),
3090 sizeof(cache->item));
3091 memcpy(&cache->key, &found_key, sizeof(found_key));
3092 cache->cached = 0;
3093 cache->pinned = 0;
3094 key.objectid = found_key.objectid + found_key.offset;
3095 btrfs_release_path(root, path);
3096 cache->flags = btrfs_block_group_flags(&cache->item);
3097 bit = 0;
3098 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3099 bit = BLOCK_GROUP_DATA;
3100 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3101 bit = BLOCK_GROUP_SYSTEM;
3102 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3103 bit = BLOCK_GROUP_METADATA;
3104 }
3105 set_avail_alloc_bits(info, cache->flags);
3106 if (btrfs_chunk_readonly(root, cache->key.objectid))
3107 cache->ro = 1;
3108
3109 ret = update_space_info(info, cache->flags, found_key.offset,
3110 btrfs_block_group_used(&cache->item),
3111 &space_info);
3112 BUG_ON(ret);
3113 cache->space_info = space_info;
3114
3115 /* use EXTENT_LOCKED to prevent merging */
3116 set_extent_bits(block_group_cache, found_key.objectid,
3117 found_key.objectid + found_key.offset - 1,
3118 bit | EXTENT_LOCKED, GFP_NOFS);
3119 set_state_private(block_group_cache, found_key.objectid,
3120 (unsigned long)cache);
3121 }
3122 ret = 0;
3123 error:
3124 btrfs_free_path(path);
3125 return ret;
3126 }
3127
3128 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3129 struct btrfs_root *root, u64 bytes_used,
3130 u64 type, u64 chunk_objectid, u64 chunk_offset,
3131 u64 size)
3132 {
3133 int ret;
3134 int bit = 0;
3135 struct btrfs_root *extent_root;
3136 struct btrfs_block_group_cache *cache;
3137 struct extent_io_tree *block_group_cache;
3138
3139 extent_root = root->fs_info->extent_root;
3140 block_group_cache = &root->fs_info->block_group_cache;
3141
3142 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3143 BUG_ON(!cache);
3144 cache->key.objectid = chunk_offset;
3145 cache->key.offset = size;
3146
3147 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3148 btrfs_set_block_group_used(&cache->item, bytes_used);
3149 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3150 cache->flags = type;
3151 btrfs_set_block_group_flags(&cache->item, type);
3152
3153 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3154 &cache->space_info);
3155 BUG_ON(ret);
3156
3157 bit = block_group_state_bits(type);
3158 set_extent_bits(block_group_cache, chunk_offset,
3159 chunk_offset + size - 1,
3160 bit | EXTENT_LOCKED, GFP_NOFS);
3161
3162 set_state_private(block_group_cache, chunk_offset,
3163 (unsigned long)cache);
3164 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3165 sizeof(cache->item));
3166 BUG_ON(ret);
3167
3168 finish_current_insert(trans, extent_root);
3169 ret = del_pending_extents(trans, extent_root);
3170 BUG_ON(ret);
3171 set_avail_alloc_bits(extent_root->fs_info, type);
3172 return 0;
3173 }
3174
3175 /*
3176 * This is for converter use only.
3177 *
3178 * In that case, we don't know where are free blocks located.
3179 * Therefore all block group cache entries must be setup properly
3180 * before doing any block allocation.
3181 */
3182 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3183 struct btrfs_root *root)
3184 {
3185 u64 total_bytes;
3186 u64 cur_start;
3187 u64 group_type;
3188 u64 group_size;
3189 u64 group_align;
3190 u64 total_data = 0;
3191 u64 total_metadata = 0;
3192 u64 chunk_objectid;
3193 int ret;
3194 int bit;
3195 struct btrfs_root *extent_root;
3196 struct btrfs_block_group_cache *cache;
3197 struct extent_io_tree *block_group_cache;
3198
3199 extent_root = root->fs_info->extent_root;
3200 block_group_cache = &root->fs_info->block_group_cache;
3201 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3202 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
3203 group_align = 64 * root->sectorsize;
3204
3205 cur_start = 0;
3206 while (cur_start < total_bytes) {
3207 group_size = total_bytes / 12;
3208 group_size = min_t(u64, group_size, total_bytes - cur_start);
3209 if (cur_start == 0) {
3210 bit = BLOCK_GROUP_SYSTEM;
3211 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3212 group_size /= 4;
3213 group_size &= ~(group_align - 1);
3214 group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3215 group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3216 } else {
3217 group_size &= ~(group_align - 1);
3218 if (total_data >= total_metadata * 2) {
3219 group_type = BTRFS_BLOCK_GROUP_METADATA;
3220 group_size = min_t(u64, group_size,
3221 1ULL * 1024 * 1024 * 1024);
3222 total_metadata += group_size;
3223 } else {
3224 group_type = BTRFS_BLOCK_GROUP_DATA;
3225 group_size = min_t(u64, group_size,
3226 5ULL * 1024 * 1024 * 1024);
3227 total_data += group_size;
3228 }
3229 if ((total_bytes - cur_start) * 4 < group_size * 5)
3230 group_size = total_bytes - cur_start;
3231 }
3232
3233 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3234 BUG_ON(!cache);
3235
3236 cache->key.objectid = cur_start;
3237 cache->key.offset = group_size;
3238 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3239
3240 btrfs_set_block_group_used(&cache->item, 0);
3241 btrfs_set_block_group_chunk_objectid(&cache->item,
3242 chunk_objectid);
3243 btrfs_set_block_group_flags(&cache->item, group_type);
3244
3245 cache->flags = group_type;
3246
3247 ret = update_space_info(root->fs_info, group_type, group_size,
3248 0, &cache->space_info);
3249 BUG_ON(ret);
3250 set_avail_alloc_bits(extent_root->fs_info, group_type);
3251
3252 set_extent_bits(block_group_cache, cur_start,
3253 cur_start + group_size - 1,
3254 bit | EXTENT_LOCKED, GFP_NOFS);
3255 set_state_private(block_group_cache, cur_start,
3256 (unsigned long)cache);
3257 cur_start += group_size;
3258 }
3259 /* then insert all the items */
3260 cur_start = 0;
3261 while(cur_start < total_bytes) {
3262 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3263 BUG_ON(!cache);
3264
3265 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3266 sizeof(cache->item));
3267 BUG_ON(ret);
3268
3269 finish_current_insert(trans, extent_root);
3270 ret = del_pending_extents(trans, extent_root);
3271 BUG_ON(ret);
3272
3273 cur_start = cache->key.objectid + cache->key.offset;
3274 }
3275 return 0;
3276 }
3277
3278 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3279 struct btrfs_root *root,
3280 u64 bytenr, u64 num_bytes, int alloc,
3281 int mark_free)
3282 {
3283 return update_block_group(trans, root, bytenr, num_bytes,
3284 alloc, mark_free);
3285 }
Btrfs-progs-unstable mirror