?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Update converter for the new format
[btrfs-progs-unstable/devel.git] / extent-tree.c
blob6bc1cc80b86cc693980054390fd5fe137680ffb0
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 if (owner < BTRFS_FIRST_FREE_OBJECTID && insert &&
1065 item_size + extra_size >= BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1066 err = -EAGAIN;
1067 goto out;
1068 }
1069
1070 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1071 flags = btrfs_extent_flags(leaf, ei);
1072
1073 ptr = (unsigned long)(ei + 1);
1074 end = (unsigned long)ei + item_size;
1075
1076 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1077 ptr += sizeof(struct btrfs_tree_block_info);
1078 BUG_ON(ptr > end);
1079 } else {
1080 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1081 }
1082
1083 err = -ENOENT;
1084 while (1) {
1085 if (ptr >= end) {
1086 WARN_ON(ptr > end);
1087 break;
1088 }
1089 iref = (struct btrfs_extent_inline_ref *)ptr;
1090 type = btrfs_extent_inline_ref_type(leaf, iref);
1091 if (want < type)
1092 break;
1093 if (want > type) {
1094 ptr += btrfs_extent_inline_ref_size(type);
1095 continue;
1096 }
1097
1098 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1099 struct btrfs_extent_data_ref *dref;
1100 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1101 if (match_extent_data_ref(leaf, dref, root_objectid,
1102 owner, offset)) {
1103 err = 0;
1104 break;
1105 }
1106 if (hash_extent_data_ref_item(leaf, dref) <
1107 hash_extent_data_ref(root_objectid, owner, offset))
1108 break;
1109 } else {
1110 u64 ref_offset;
1111 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1112 if (parent > 0) {
1113 if (parent == ref_offset) {
1114 err = 0;
1115 break;
1116 }
1117 if (ref_offset < parent)
1118 break;
1119 } else {
1120 if (root_objectid == ref_offset) {
1121 err = 0;
1122 break;
1123 }
1124 if (ref_offset < root_objectid)
1125 break;
1126 }
1127 }
1128 ptr += btrfs_extent_inline_ref_size(type);
1129 }
1130 if (err == -ENOENT && insert) {
1131 if (item_size + extra_size >=
1132 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1133 err = -EAGAIN;
1134 goto out;
1135 }
1136 /*
1137 * To add new inline back ref, we have to make sure
1138 * there is no corresponding back ref item.
1139 * For simplicity, we just do not add new inline back
1140 * ref if there is any back ref item.
1141 */
1142 if (owner >= BTRFS_FIRST_FREE_OBJECTID &&
1143 find_next_key(path, &key) == 0 && key.objectid == bytenr) {
1144 err = -EAGAIN;
1145 goto out;
1146 }
1147 }
1148 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1149 out:
1150 return err;
1151 }
1152
1153 static int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
1154 struct btrfs_root *root,
1155 struct btrfs_path *path,
1156 struct btrfs_extent_inline_ref *iref,
1157 u64 parent, u64 root_objectid,
1158 u64 owner, u64 offset, int refs_to_add)
1159 {
1160 struct extent_buffer *leaf;
1161 struct btrfs_extent_item *ei;
1162 unsigned long ptr;
1163 unsigned long end;
1164 unsigned long item_offset;
1165 u64 refs;
1166 int size;
1167 int type;
1168 int ret;
1169
1170 leaf = path->nodes[0];
1171 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1172 item_offset = (unsigned long)iref - (unsigned long)ei;
1173
1174 type = extent_ref_type(parent, owner);
1175 size = btrfs_extent_inline_ref_size(type);
1176
1177 ret = btrfs_extend_item(trans, root, path, size);
1178 BUG_ON(ret);
1179
1180 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1181 refs = btrfs_extent_refs(leaf, ei);
1182 refs += refs_to_add;
1183 btrfs_set_extent_refs(leaf, ei, refs);
1184
1185 ptr = (unsigned long)ei + item_offset;
1186 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1187 if (ptr < end - size)
1188 memmove_extent_buffer(leaf, ptr + size, ptr,
1189 end - size - ptr);
1190
1191 iref = (struct btrfs_extent_inline_ref *)ptr;
1192 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1193 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1194 struct btrfs_extent_data_ref *dref;
1195 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1196 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1197 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1198 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1199 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1200 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1201 struct btrfs_shared_data_ref *sref;
1202 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1203 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1204 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1205 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1206 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1207 } else {
1208 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1209 }
1210 btrfs_mark_buffer_dirty(leaf);
1211 return 0;
1212 }
1213
1214 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1215 struct btrfs_root *root,
1216 struct btrfs_path *path,
1217 struct btrfs_extent_inline_ref **ref_ret,
1218 u64 bytenr, u64 num_bytes, u64 parent,
1219 u64 root_objectid, u64 owner, u64 offset)
1220 {
1221 int ret;
1222
1223 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1224 bytenr, num_bytes, parent,
1225 root_objectid, owner, offset, 0);
1226 if (ret != -ENOENT)
1227 return ret;
1228
1229 btrfs_release_path(root, path);
1230 *ref_ret = NULL;
1231
1232 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1233 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1234 root_objectid);
1235 } else {
1236 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1237 root_objectid, owner, offset);
1238 }
1239 return ret;
1240 }
1241
1242 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1243 struct btrfs_root *root,
1244 struct btrfs_path *path,
1245 struct btrfs_extent_inline_ref *iref,
1246 int refs_to_mod)
1247 {
1248 struct extent_buffer *leaf;
1249 struct btrfs_extent_item *ei;
1250 struct btrfs_extent_data_ref *dref = NULL;
1251 struct btrfs_shared_data_ref *sref = NULL;
1252 unsigned long ptr;
1253 unsigned long end;
1254 u32 item_size;
1255 int size;
1256 int type;
1257 int ret;
1258 u64 refs;
1259
1260 leaf = path->nodes[0];
1261 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1262 refs = btrfs_extent_refs(leaf, ei);
1263 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1264 refs += refs_to_mod;
1265 btrfs_set_extent_refs(leaf, ei, refs);
1266
1267 type = btrfs_extent_inline_ref_type(leaf, iref);
1268
1269 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1270 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1271 refs = btrfs_extent_data_ref_count(leaf, dref);
1272 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1273 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1274 refs = btrfs_shared_data_ref_count(leaf, sref);
1275 } else {
1276 refs = 1;
1277 BUG_ON(refs_to_mod != -1);
1278 }
1279
1280 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1281 refs += refs_to_mod;
1282
1283 if (refs > 0) {
1284 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1285 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1286 else
1287 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1288 } else {
1289 size = btrfs_extent_inline_ref_size(type);
1290 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1291 ptr = (unsigned long)iref;
1292 end = (unsigned long)ei + item_size;
1293 if (ptr + size < end)
1294 memmove_extent_buffer(leaf, ptr, ptr + size,
1295 end - ptr - size);
1296 item_size -= size;
1297 ret = btrfs_truncate_item(trans, root, path, item_size, 1);
1298 BUG_ON(ret);
1299 }
1300 btrfs_mark_buffer_dirty(leaf);
1301 return 0;
1302 }
1303
1304 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1305 struct btrfs_root *root,
1306 struct btrfs_path *path,
1307 u64 bytenr, u64 num_bytes, u64 parent,
1308 u64 root_objectid, u64 owner,
1309 u64 offset, int refs_to_add)
1310 {
1311 struct btrfs_extent_inline_ref *iref;
1312 int ret;
1313
1314 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1315 bytenr, num_bytes, parent,
1316 root_objectid, owner, offset, 1);
1317 if (ret == 0) {
1318 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1319 ret = update_inline_extent_backref(trans, root, path, iref,
1320 refs_to_add);
1321 } else if (ret == -ENOENT) {
1322 ret = setup_inline_extent_backref(trans, root, path, iref,
1323 parent, root_objectid,
1324 owner, offset, refs_to_add);
1325 }
1326 return ret;
1327 }
1328
1329 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1330 struct btrfs_root *root,
1331 struct btrfs_path *path,
1332 u64 bytenr, u64 parent, u64 root_objectid,
1333 u64 owner, u64 offset, int refs_to_add)
1334 {
1335 int ret;
1336
1337 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1338 ret = insert_extent_data_ref(trans, root, path, bytenr,
1339 parent, root_objectid,
1340 owner, offset, refs_to_add);
1341 } else {
1342 BUG_ON(refs_to_add != 1);
1343 ret = insert_tree_block_ref(trans, root, path, bytenr,
1344 parent, root_objectid);
1345 }
1346 return ret;
1347 }
1348
1349 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1350 struct btrfs_root *root,
1351 struct btrfs_path *path,
1352 struct btrfs_extent_inline_ref *iref,
1353 int refs_to_drop, int is_data)
1354 {
1355 int ret;
1356
1357 BUG_ON(!is_data && refs_to_drop != 1);
1358 if (iref) {
1359 ret = update_inline_extent_backref(trans, root, path, iref,
1360 -refs_to_drop);
1361 } else if (is_data) {
1362 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1363 } else {
1364 ret = btrfs_del_item(trans, root, path);
1365 }
1366 return ret;
1367 }
1368
1369 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1370 struct btrfs_root *root,
1371 u64 bytenr, u64 num_bytes, u64 parent,
1372 u64 root_objectid, u64 owner, u64 offset)
1373 {
1374 struct btrfs_path *path;
1375 struct extent_buffer *leaf;
1376 struct btrfs_extent_item *item;
1377 u64 refs;
1378 int ret;
1379 int err = 0;
1380
1381 path = btrfs_alloc_path();
1382 if (!path)
1383 return -ENOMEM;
1384
1385 path->reada = 1;
1386 path->leave_spinning = 1;
1387
1388 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1389 path, bytenr, num_bytes, parent,
1390 root_objectid, owner, offset, 1);
1391 if (ret == 0)
1392 goto out;
1393
1394 if (ret != -EAGAIN) {
1395 err = ret;
1396 goto out;
1397 }
1398
1399 leaf = path->nodes[0];
1400 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1401 refs = btrfs_extent_refs(leaf, item);
1402 btrfs_set_extent_refs(leaf, item, refs + 1);
1403
1404 btrfs_mark_buffer_dirty(leaf);
1405 btrfs_release_path(root->fs_info->extent_root, path);
1406
1407 path->reada = 1;
1408 path->leave_spinning = 1;
1409
1410 /* now insert the actual backref */
1411 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1412 path, bytenr, parent, root_objectid,
1413 owner, offset, 1);
1414 if (ret)
1415 err = ret;
1416 out:
1417 btrfs_free_path(path);
1418 finish_current_insert(trans, root->fs_info->extent_root);
1419 del_pending_extents(trans, root->fs_info->extent_root);
1420 BUG_ON(err);
1421 return err;
1422 }
1423
1424 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1425 struct btrfs_root *root)
1426 {
1427 finish_current_insert(trans, root->fs_info->extent_root);
1428 del_pending_extents(trans, root->fs_info->extent_root);
1429 return 0;
1430 }
1431
1432 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1433 struct btrfs_root *root, u64 bytenr,
1434 u64 num_bytes, u64 *refs, u64 *flags)
1435 {
1436 struct btrfs_path *path;
1437 int ret;
1438 struct btrfs_key key;
1439 struct extent_buffer *l;
1440 struct btrfs_extent_item *item;
1441 u32 item_size;
1442 u64 num_refs;
1443 u64 extent_flags;
1444
1445 WARN_ON(num_bytes < root->sectorsize);
1446 path = btrfs_alloc_path();
1447 path->reada = 1;
1448 key.objectid = bytenr;
1449 key.offset = num_bytes;
1450 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1451 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1452 0, 0);
1453 if (ret < 0)
1454 goto out;
1455 if (ret != 0) {
1456 btrfs_print_leaf(root, path->nodes[0]);
1457 printk("failed to find block number %Lu\n", bytenr);
1458 BUG();
1459 }
1460
1461 l = path->nodes[0];
1462 item_size = btrfs_item_size_nr(l, path->slots[0]);
1463 if (item_size >= sizeof(*item)) {
1464 item = btrfs_item_ptr(l, path->slots[0],
1465 struct btrfs_extent_item);
1466 num_refs = btrfs_extent_refs(l, item);
1467 extent_flags = btrfs_extent_flags(l, item);
1468 } else {
1469 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1470 struct btrfs_extent_item_v0 *ei0;
1471 BUG_ON(item_size != sizeof(*ei0));
1472 ei0 = btrfs_item_ptr(l, path->slots[0],
1473 struct btrfs_extent_item_v0);
1474 num_refs = btrfs_extent_refs_v0(l, ei0);
1475 /* FIXME: this isn't correct for data */
1476 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1477 #else
1478 BUG();
1479 #endif
1480 }
1481 BUG_ON(num_refs == 0);
1482 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1483 if (refs)
1484 *refs = num_refs;
1485 if (flags)
1486 *flags = extent_flags;
1487 out:
1488 btrfs_free_path(path);
1489 return 0;
1490 }
1491
1492 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1493 struct btrfs_root *root,
1494 u64 bytenr, u64 num_bytes, u64 flags)
1495 {
1496 struct btrfs_path *path;
1497 int ret;
1498 struct btrfs_key key;
1499 struct extent_buffer *l;
1500 struct btrfs_extent_item *item;
1501 u32 item_size;
1502
1503 WARN_ON(num_bytes < root->sectorsize);
1504 path = btrfs_alloc_path();
1505 path->reada = 1;
1506 key.objectid = bytenr;
1507 key.offset = num_bytes;
1508 btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
1509 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1510 0, 0);
1511 if (ret < 0)
1512 goto out;
1513 if (ret != 0) {
1514 btrfs_print_leaf(root, path->nodes[0]);
1515 printk("failed to find block number %Lu\n",
1516 (unsigned long long)bytenr);
1517 BUG();
1518 }
1519 l = path->nodes[0];
1520 item_size = btrfs_item_size_nr(l, path->slots[0]);
1521 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1522 if (item_size < sizeof(*item)) {
1523 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1524 path, (u64)-1, 0);
1525 if (ret < 0)
1526 goto out;
1527
1528 l = path->nodes[0];
1529 item_size = btrfs_item_size_nr(l, path->slots[0]);
1530 }
1531 #endif
1532 BUG_ON(item_size < sizeof(*item));
1533 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1534 flags |= btrfs_extent_flags(l, item);
1535 btrfs_set_extent_flags(l, item, flags);
1536 out:
1537 btrfs_free_path(path);
1538 finish_current_insert(trans, root->fs_info->extent_root);
1539 del_pending_extents(trans, root->fs_info->extent_root);
1540 return ret;
1541 }
1542
1543 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1544 struct btrfs_root *root,
1545 struct extent_buffer *buf,
1546 int record_parent, int inc)
1547 {
1548 u64 bytenr;
1549 u64 num_bytes;
1550 u64 parent;
1551 u64 ref_root;
1552 u32 nritems;
1553 struct btrfs_key key;
1554 struct btrfs_file_extent_item *fi;
1555 int i;
1556 int level;
1557 int ret = 0;
1558 int faili = 0;
1559 int (*process_func)(struct btrfs_trans_handle *trans,
1560 struct btrfs_root *root,
1561 u64, u64, u64, u64, u64, u64);
1562
1563 ref_root = btrfs_header_owner(buf);
1564 nritems = btrfs_header_nritems(buf);
1565 level = btrfs_header_level(buf);
1566
1567 if (!root->ref_cows && level == 0)
1568 return 0;
1569
1570 if (inc)
1571 process_func = btrfs_inc_extent_ref;
1572 else
1573 process_func = btrfs_free_extent;
1574
1575 if (record_parent)
1576 parent = buf->start;
1577 else
1578 parent = 0;
1579
1580 for (i = 0; i < nritems; i++) {
1581 cond_resched();
1582 if (level == 0) {
1583 btrfs_item_key_to_cpu(buf, &key, i);
1584 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1585 continue;
1586 fi = btrfs_item_ptr(buf, i,
1587 struct btrfs_file_extent_item);
1588 if (btrfs_file_extent_type(buf, fi) ==
1589 BTRFS_FILE_EXTENT_INLINE)
1590 continue;
1591 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1592 if (bytenr == 0)
1593 continue;
1594
1595 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1596 key.offset -= btrfs_file_extent_offset(buf, fi);
1597 ret = process_func(trans, root, bytenr, num_bytes,
1598 parent, ref_root, key.objectid,
1599 key.offset);
1600 if (ret) {
1601 faili = i;
1602 WARN_ON(1);
1603 goto fail;
1604 }
1605 } else {
1606 bytenr = btrfs_node_blockptr(buf, i);
1607 num_bytes = btrfs_level_size(root, level - 1);
1608 ret = process_func(trans, root, bytenr, num_bytes,
1609 parent, ref_root, level - 1, 0);
1610 if (ret) {
1611 faili = i;
1612 WARN_ON(1);
1613 goto fail;
1614 }
1615 }
1616 }
1617 return 0;
1618 fail:
1619 WARN_ON(1);
1620 #if 0
1621 for (i =0; i < faili; i++) {
1622 if (level == 0) {
1623 u64 disk_bytenr;
1624 btrfs_item_key_to_cpu(buf, &key, i);
1625 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1626 continue;
1627 fi = btrfs_item_ptr(buf, i,
1628 struct btrfs_file_extent_item);
1629 if (btrfs_file_extent_type(buf, fi) ==
1630 BTRFS_FILE_EXTENT_INLINE)
1631 continue;
1632 disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1633 if (disk_bytenr == 0)
1634 continue;
1635 err = btrfs_free_extent(trans, root, disk_bytenr,
1636 btrfs_file_extent_disk_num_bytes(buf,
1637 fi), 0);
1638 BUG_ON(err);
1639 } else {
1640 bytenr = btrfs_node_blockptr(buf, i);
1641 err = btrfs_free_extent(trans, root, bytenr,
1642 btrfs_level_size(root, level - 1), 0);
1643 BUG_ON(err);
1644 }
1645 }
1646 #endif
1647 return ret;
1648 }
1649
1650 int btrfs_inc_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, 1);
1654 }
1655
1656 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1657 struct extent_buffer *buf, int record_parent)
1658 {
1659 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1660 }
1661
1662 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1663 struct btrfs_root *root,
1664 struct btrfs_path *path,
1665 struct btrfs_block_group_cache *cache)
1666 {
1667 int ret;
1668 int pending_ret;
1669 struct btrfs_root *extent_root = root->fs_info->extent_root;
1670 unsigned long bi;
1671 struct extent_buffer *leaf;
1672
1673 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1674 if (ret < 0)
1675 goto fail;
1676 BUG_ON(ret);
1677
1678 leaf = path->nodes[0];
1679 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1680 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1681 btrfs_mark_buffer_dirty(leaf);
1682 btrfs_release_path(extent_root, path);
1683 fail:
1684 finish_current_insert(trans, extent_root);
1685 pending_ret = del_pending_extents(trans, extent_root);
1686 if (ret)
1687 return ret;
1688 if (pending_ret)
1689 return pending_ret;
1690 return 0;
1691
1692 }
1693
1694 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1695 struct btrfs_root *root)
1696 {
1697 struct extent_io_tree *block_group_cache;
1698 struct btrfs_block_group_cache *cache;
1699 int ret;
1700 int err = 0;
1701 int werr = 0;
1702 struct btrfs_path *path;
1703 u64 last = 0;
1704 u64 start;
1705 u64 end;
1706 u64 ptr;
1707
1708 block_group_cache = &root->fs_info->block_group_cache;
1709 path = btrfs_alloc_path();
1710 if (!path)
1711 return -ENOMEM;
1712
1713 while(1) {
1714 ret = find_first_extent_bit(block_group_cache, last,
1715 &start, &end, BLOCK_GROUP_DIRTY);
1716 if (ret)
1717 break;
1718
1719 last = end + 1;
1720 ret = get_state_private(block_group_cache, start, &ptr);
1721 if (ret)
1722 break;
1723 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1724 err = write_one_cache_group(trans, root,
1725 path, cache);
1726 /*
1727 * if we fail to write the cache group, we want
1728 * to keep it marked dirty in hopes that a later
1729 * write will work
1730 */
1731 if (err) {
1732 werr = err;
1733 continue;
1734 }
1735 clear_extent_bits(block_group_cache, start, end,
1736 BLOCK_GROUP_DIRTY, GFP_NOFS);
1737 }
1738 btrfs_free_path(path);
1739 return werr;
1740 }
1741
1742 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1743 u64 flags)
1744 {
1745 struct list_head *head = &info->space_info;
1746 struct list_head *cur;
1747 struct btrfs_space_info *found;
1748 list_for_each(cur, head) {
1749 found = list_entry(cur, struct btrfs_space_info, list);
1750 if (found->flags == flags)
1751 return found;
1752 }
1753 return NULL;
1754
1755 }
1756
1757 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1758 u64 total_bytes, u64 bytes_used,
1759 struct btrfs_space_info **space_info)
1760 {
1761 struct btrfs_space_info *found;
1762
1763 found = __find_space_info(info, flags);
1764 if (found) {
1765 found->total_bytes += total_bytes;
1766 found->bytes_used += bytes_used;
1767 WARN_ON(found->total_bytes < found->bytes_used);
1768 *space_info = found;
1769 return 0;
1770 }
1771 found = kmalloc(sizeof(*found), GFP_NOFS);
1772 if (!found)
1773 return -ENOMEM;
1774
1775 list_add(&found->list, &info->space_info);
1776 found->flags = flags;
1777 found->total_bytes = total_bytes;
1778 found->bytes_used = bytes_used;
1779 found->bytes_pinned = 0;
1780 found->full = 0;
1781 *space_info = found;
1782 return 0;
1783 }
1784
1785
1786 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1787 {
1788 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1789 BTRFS_BLOCK_GROUP_RAID1 |
1790 BTRFS_BLOCK_GROUP_DUP);
1791 if (extra_flags) {
1792 if (flags & BTRFS_BLOCK_GROUP_DATA)
1793 fs_info->avail_data_alloc_bits |= extra_flags;
1794 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1795 fs_info->avail_metadata_alloc_bits |= extra_flags;
1796 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1797 fs_info->avail_system_alloc_bits |= extra_flags;
1798 }
1799 }
1800
1801 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1802 struct btrfs_root *extent_root, u64 alloc_bytes,
1803 u64 flags)
1804 {
1805 struct btrfs_space_info *space_info;
1806 u64 thresh;
1807 u64 start;
1808 u64 num_bytes;
1809 int ret;
1810
1811 space_info = __find_space_info(extent_root->fs_info, flags);
1812 if (!space_info) {
1813 ret = update_space_info(extent_root->fs_info, flags,
1814 0, 0, &space_info);
1815 BUG_ON(ret);
1816 }
1817 BUG_ON(!space_info);
1818
1819 if (space_info->full)
1820 return 0;
1821
1822 thresh = div_factor(space_info->total_bytes, 7);
1823 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1824 thresh)
1825 return 0;
1826
1827 ret = btrfs_alloc_chunk(trans, extent_root, &start, &num_bytes, flags);
1828 if (ret == -ENOSPC) {
1829 space_info->full = 1;
1830 return 0;
1831 }
1832
1833 BUG_ON(ret);
1834
1835 ret = btrfs_make_block_group(trans, extent_root, 0, flags,
1836 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1837 BUG_ON(ret);
1838 return 0;
1839 }
1840
1841 static int update_block_group(struct btrfs_trans_handle *trans,
1842 struct btrfs_root *root,
1843 u64 bytenr, u64 num_bytes, int alloc,
1844 int mark_free)
1845 {
1846 struct btrfs_block_group_cache *cache;
1847 struct btrfs_fs_info *info = root->fs_info;
1848 u64 total = num_bytes;
1849 u64 old_val;
1850 u64 byte_in_group;
1851 u64 start;
1852 u64 end;
1853
1854 /* block accounting for super block */
1855 old_val = btrfs_super_bytes_used(&info->super_copy);
1856 if (alloc)
1857 old_val += num_bytes;
1858 else
1859 old_val -= num_bytes;
1860 btrfs_set_super_bytes_used(&info->super_copy, old_val);
1861
1862 /* block accounting for root item */
1863 old_val = btrfs_root_used(&root->root_item);
1864 if (alloc)
1865 old_val += num_bytes;
1866 else
1867 old_val -= num_bytes;
1868 btrfs_set_root_used(&root->root_item, old_val);
1869
1870 while(total) {
1871 cache = btrfs_lookup_block_group(info, bytenr);
1872 if (!cache) {
1873 return -1;
1874 }
1875 byte_in_group = bytenr - cache->key.objectid;
1876 WARN_ON(byte_in_group > cache->key.offset);
1877 start = cache->key.objectid;
1878 end = start + cache->key.offset - 1;
1879 set_extent_bits(&info->block_group_cache, start, end,
1880 BLOCK_GROUP_DIRTY, GFP_NOFS);
1881
1882 old_val = btrfs_block_group_used(&cache->item);
1883 num_bytes = min(total, cache->key.offset - byte_in_group);
1884 if (alloc) {
1885 old_val += num_bytes;
1886 cache->space_info->bytes_used += num_bytes;
1887 } else {
1888 old_val -= num_bytes;
1889 cache->space_info->bytes_used -= num_bytes;
1890 if (mark_free) {
1891 set_extent_dirty(&info->free_space_cache,
1892 bytenr, bytenr + num_bytes - 1,
1893 GFP_NOFS);
1894 }
1895 }
1896 btrfs_set_block_group_used(&cache->item, old_val);
1897 total -= num_bytes;
1898 bytenr += num_bytes;
1899 }
1900 return 0;
1901 }
1902
1903 static int update_pinned_extents(struct btrfs_root *root,
1904 u64 bytenr, u64 num, int pin)
1905 {
1906 u64 len;
1907 struct btrfs_block_group_cache *cache;
1908 struct btrfs_fs_info *fs_info = root->fs_info;
1909
1910 if (pin) {
1911 set_extent_dirty(&fs_info->pinned_extents,
1912 bytenr, bytenr + num - 1, GFP_NOFS);
1913 } else {
1914 clear_extent_dirty(&fs_info->pinned_extents,
1915 bytenr, bytenr + num - 1, GFP_NOFS);
1916 }
1917 while (num > 0) {
1918 cache = btrfs_lookup_block_group(fs_info, bytenr);
1919 WARN_ON(!cache);
1920 len = min(num, cache->key.offset -
1921 (bytenr - cache->key.objectid));
1922 if (pin) {
1923 cache->pinned += len;
1924 cache->space_info->bytes_pinned += len;
1925 fs_info->total_pinned += len;
1926 } else {
1927 cache->pinned -= len;
1928 cache->space_info->bytes_pinned -= len;
1929 fs_info->total_pinned -= len;
1930 }
1931 bytenr += len;
1932 num -= len;
1933 }
1934 return 0;
1935 }
1936
1937 int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
1938 {
1939 u64 last = 0;
1940 u64 start;
1941 u64 end;
1942 struct extent_io_tree *pinned_extents = &root->fs_info->pinned_extents;
1943 int ret;
1944
1945 while(1) {
1946 ret = find_first_extent_bit(pinned_extents, last,
1947 &start, &end, EXTENT_DIRTY);
1948 if (ret)
1949 break;
1950 set_extent_dirty(copy, start, end, GFP_NOFS);
1951 last = end + 1;
1952 }
1953 return 0;
1954 }
1955
1956 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
1957 struct btrfs_root *root,
1958 struct extent_io_tree *unpin)
1959 {
1960 u64 start;
1961 u64 end;
1962 int ret;
1963 struct extent_io_tree *free_space_cache;
1964 free_space_cache = &root->fs_info->free_space_cache;
1965
1966 while(1) {
1967 ret = find_first_extent_bit(unpin, 0, &start, &end,
1968 EXTENT_DIRTY);
1969 if (ret)
1970 break;
1971 update_pinned_extents(root, start, end + 1 - start, 0);
1972 clear_extent_dirty(unpin, start, end, GFP_NOFS);
1973 set_extent_dirty(free_space_cache, start, end, GFP_NOFS);
1974 }
1975 return 0;
1976 }
1977
1978 static int finish_current_insert(struct btrfs_trans_handle *trans,
1979 struct btrfs_root *extent_root)
1980 {
1981 u64 start;
1982 u64 end;
1983 u64 priv;
1984 struct btrfs_fs_info *info = extent_root->fs_info;
1985 struct btrfs_path *path;
1986 struct pending_extent_op *extent_op;
1987 struct btrfs_key key;
1988 int ret;
1989
1990 path = btrfs_alloc_path();
1991
1992 while(1) {
1993 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
1994 &end, EXTENT_LOCKED);
1995 if (ret)
1996 break;
1997
1998 ret = get_state_private(&info->extent_ins, start, &priv);
1999 BUG_ON(ret);
2000 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2001
2002 if (extent_op->type == PENDING_EXTENT_INSERT) {
2003 key.objectid = start;
2004 key.offset = end + 1 - start;
2005 key.type = BTRFS_EXTENT_ITEM_KEY;
2006 ret = alloc_reserved_tree_block(trans, extent_root,
2007 extent_root->root_key.objectid,
2008 trans->transid,
2009 extent_op->flags,
2010 &extent_op->key,
2011 extent_op->level, &key);
2012 } else {
2013 BUG_ON(1);
2014 }
2015
2016 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED,
2017 GFP_NOFS);
2018 kfree(extent_op);
2019 }
2020 btrfs_free_path(path);
2021 return 0;
2022 }
2023
2024 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2025 struct btrfs_root *root,
2026 u64 bytenr, u64 num_bytes, int is_data)
2027 {
2028 int err = 0;
2029 struct extent_buffer *buf;
2030
2031 if (is_data)
2032 goto pinit;
2033
2034 buf = btrfs_find_tree_block(root, bytenr, num_bytes);
2035 if (!buf)
2036 goto pinit;
2037
2038 /* we can reuse a block if it hasn't been written
2039 * and it is from this transaction. We can't
2040 * reuse anything from the tree log root because
2041 * it has tiny sub-transactions.
2042 */
2043 if (btrfs_buffer_uptodate(buf, 0)) {
2044 u64 header_owner = btrfs_header_owner(buf);
2045 u64 header_transid = btrfs_header_generation(buf);
2046 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2047 header_transid == trans->transid &&
2048 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2049 clean_tree_block(NULL, root, buf);
2050 free_extent_buffer(buf);
2051 return 1;
2052 }
2053 }
2054 free_extent_buffer(buf);
2055 pinit:
2056 update_pinned_extents(root, bytenr, num_bytes, 1);
2057
2058 BUG_ON(err < 0);
2059 return 0;
2060 }
2061
2062 /*
2063 * remove an extent from the root, returns 0 on success
2064 */
2065 static int __free_extent(struct btrfs_trans_handle *trans,
2066 struct btrfs_root *root,
2067 u64 bytenr, u64 num_bytes, u64 parent,
2068 u64 root_objectid, u64 owner_objectid,
2069 u64 owner_offset, int refs_to_drop)
2070 {
2071
2072 struct btrfs_key key;
2073 struct btrfs_path *path;
2074 struct btrfs_extent_ops *ops = root->fs_info->extent_ops;
2075 struct btrfs_root *extent_root = root->fs_info->extent_root;
2076 struct extent_buffer *leaf;
2077 struct btrfs_extent_item *ei;
2078 struct btrfs_extent_inline_ref *iref;
2079 int ret;
2080 int is_data;
2081 int extent_slot = 0;
2082 int found_extent = 0;
2083 int num_to_del = 1;
2084 u32 item_size;
2085 u64 refs;
2086
2087 path = btrfs_alloc_path();
2088 if (!path)
2089 return -ENOMEM;
2090
2091 path->reada = 1;
2092 path->leave_spinning = 1;
2093
2094 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2095 BUG_ON(!is_data && refs_to_drop != 1);
2096
2097 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2098 bytenr, num_bytes, parent,
2099 root_objectid, owner_objectid,
2100 owner_offset);
2101 if (ret == 0) {
2102 extent_slot = path->slots[0];
2103 while (extent_slot >= 0) {
2104 btrfs_item_key_to_cpu(path->nodes[0], &key,
2105 extent_slot);
2106 if (key.objectid != bytenr)
2107 break;
2108 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2109 key.offset == num_bytes) {
2110 found_extent = 1;
2111 break;
2112 }
2113 if (path->slots[0] - extent_slot > 5)
2114 break;
2115 extent_slot--;
2116 }
2117 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2118 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2119 if (found_extent && item_size < sizeof(*ei))
2120 found_extent = 0;
2121 #endif
2122 if (!found_extent) {
2123 BUG_ON(iref);
2124 ret = remove_extent_backref(trans, extent_root, path,
2125 NULL, refs_to_drop,
2126 is_data);
2127 BUG_ON(ret);
2128 btrfs_release_path(extent_root, path);
2129 path->leave_spinning = 1;
2130
2131 key.objectid = bytenr;
2132 key.type = BTRFS_EXTENT_ITEM_KEY;
2133 key.offset = num_bytes;
2134
2135 ret = btrfs_search_slot(trans, extent_root,
2136 &key, path, -1, 1);
2137 if (ret) {
2138 printk(KERN_ERR "umm, got %d back from search"
2139 ", was looking for %llu\n", ret,
2140 (unsigned long long)bytenr);
2141 btrfs_print_leaf(extent_root, path->nodes[0]);
2142 }
2143 BUG_ON(ret);
2144 extent_slot = path->slots[0];
2145 }
2146 } else {
2147 btrfs_print_leaf(extent_root, path->nodes[0]);
2148 WARN_ON(1);
2149 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2150 "parent %llu root %llu owner %llu offset %llu\n",
2151 (unsigned long long)bytenr,
2152 (unsigned long long)parent,
2153 (unsigned long long)root_objectid,
2154 (unsigned long long)owner_objectid,
2155 (unsigned long long)owner_offset);
2156 }
2157
2158 leaf = path->nodes[0];
2159 item_size = btrfs_item_size_nr(leaf, extent_slot);
2160 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2161 if (item_size < sizeof(*ei)) {
2162 BUG_ON(found_extent || extent_slot != path->slots[0]);
2163 ret = convert_extent_item_v0(trans, extent_root, path,
2164 owner_objectid, 0);
2165 BUG_ON(ret < 0);
2166
2167 btrfs_release_path(extent_root, path);
2168 path->leave_spinning = 1;
2169
2170 key.objectid = bytenr;
2171 key.type = BTRFS_EXTENT_ITEM_KEY;
2172 key.offset = num_bytes;
2173
2174 ret = btrfs_search_slot(trans, extent_root, &key, path,
2175 -1, 1);
2176 if (ret) {
2177 printk(KERN_ERR "umm, got %d back from search"
2178 ", was looking for %llu\n", ret,
2179 (unsigned long long)bytenr);
2180 btrfs_print_leaf(extent_root, path->nodes[0]);
2181 }
2182 BUG_ON(ret);
2183 extent_slot = path->slots[0];
2184 leaf = path->nodes[0];
2185 item_size = btrfs_item_size_nr(leaf, extent_slot);
2186 }
2187 #endif
2188 BUG_ON(item_size < sizeof(*ei));
2189 ei = btrfs_item_ptr(leaf, extent_slot,
2190 struct btrfs_extent_item);
2191 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) {
2192 struct btrfs_tree_block_info *bi;
2193 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2194 bi = (struct btrfs_tree_block_info *)(ei + 1);
2195 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2196 }
2197
2198 refs = btrfs_extent_refs(leaf, ei);
2199 BUG_ON(refs < refs_to_drop);
2200 refs -= refs_to_drop;
2201
2202 if (refs > 0) {
2203 /*
2204 * In the case of inline back ref, reference count will
2205 * be updated by remove_extent_backref
2206 */
2207 if (iref) {
2208 BUG_ON(!found_extent);
2209 } else {
2210 btrfs_set_extent_refs(leaf, ei, refs);
2211 btrfs_mark_buffer_dirty(leaf);
2212 }
2213 if (found_extent) {
2214 ret = remove_extent_backref(trans, extent_root, path,
2215 iref, refs_to_drop,
2216 is_data);
2217 BUG_ON(ret);
2218 }
2219 } else {
2220 int mark_free = 0;
2221 int pin = 1;
2222
2223 if (found_extent) {
2224 BUG_ON(is_data && refs_to_drop !=
2225 extent_data_ref_count(root, path, iref));
2226 if (iref) {
2227 BUG_ON(path->slots[0] != extent_slot);
2228 } else {
2229 BUG_ON(path->slots[0] != extent_slot + 1);
2230 path->slots[0] = extent_slot;
2231 num_to_del = 2;
2232 }
2233 }
2234
2235 if (ops && ops->free_extent) {
2236 ret = ops->free_extent(root, bytenr, num_bytes);
2237 if (ret > 0) {
2238 pin = 0;
2239 mark_free = 0;
2240 }
2241 }
2242
2243 if (pin) {
2244 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2245 is_data);
2246 if (ret > 0)
2247 mark_free = 1;
2248 BUG_ON(ret < 0);
2249 }
2250
2251 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2252 num_to_del);
2253 BUG_ON(ret);
2254 btrfs_release_path(extent_root, path);
2255
2256 if (is_data) {
2257 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2258 BUG_ON(ret);
2259 }
2260
2261 ret = update_block_group(trans, root, bytenr, num_bytes, 0,
2262 mark_free);
2263 BUG_ON(ret);
2264 }
2265 btrfs_free_path(path);
2266 finish_current_insert(trans, extent_root);
2267 return ret;
2268 }
2269
2270 /*
2271 * find all the blocks marked as pending in the radix tree and remove
2272 * them from the extent map
2273 */
2274 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2275 btrfs_root *extent_root)
2276 {
2277 int ret;
2278 int err = 0;
2279 u64 start;
2280 u64 end;
2281 u64 priv;
2282 struct extent_io_tree *pending_del;
2283 struct extent_io_tree *extent_ins;
2284 struct pending_extent_op *extent_op;
2285
2286 extent_ins = &extent_root->fs_info->extent_ins;
2287 pending_del = &extent_root->fs_info->pending_del;
2288
2289 while(1) {
2290 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2291 EXTENT_LOCKED);
2292 if (ret)
2293 break;
2294
2295 ret = get_state_private(pending_del, start, &priv);
2296 BUG_ON(ret);
2297 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2298
2299 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED,
2300 GFP_NOFS);
2301
2302 if (!test_range_bit(extent_ins, start, end,
2303 EXTENT_LOCKED, 0)) {
2304 ret = __free_extent(trans, extent_root,
2305 start, end + 1 - start, 0,
2306 extent_root->root_key.objectid,
2307 extent_op->level, 0, 1);
2308 kfree(extent_op);
2309 } else {
2310 kfree(extent_op);
2311 ret = get_state_private(extent_ins, start, &priv);
2312 BUG_ON(ret);
2313 extent_op = (struct pending_extent_op *)
2314 (unsigned long)priv;
2315
2316 clear_extent_bits(extent_ins, start, end,
2317 EXTENT_LOCKED, GFP_NOFS);
2318
2319 if (extent_op->type == PENDING_BACKREF_UPDATE)
2320 BUG_ON(1);
2321
2322 kfree(extent_op);
2323 }
2324 if (ret)
2325 err = ret;
2326 }
2327 return err;
2328 }
2329
2330 /*
2331 * remove an extent from the root, returns 0 on success
2332 */
2333
2334 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2335 struct btrfs_root *root,
2336 u64 bytenr, u64 num_bytes, u64 parent,
2337 u64 root_objectid, u64 owner, u64 offset)
2338 {
2339 struct btrfs_root *extent_root = root->fs_info->extent_root;
2340 int pending_ret;
2341 int ret;
2342
2343 WARN_ON(num_bytes < root->sectorsize);
2344 if (root == extent_root) {
2345 struct pending_extent_op *extent_op;
2346
2347 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2348 BUG_ON(!extent_op);
2349
2350 extent_op->type = PENDING_EXTENT_DELETE;
2351 extent_op->bytenr = bytenr;
2352 extent_op->num_bytes = num_bytes;
2353 extent_op->level = (int)owner;
2354
2355 set_extent_bits(&root->fs_info->pending_del,
2356 bytenr, bytenr + num_bytes - 1,
2357 EXTENT_LOCKED, GFP_NOFS);
2358 set_state_private(&root->fs_info->pending_del,
2359 bytenr, (unsigned long)extent_op);
2360 return 0;
2361 }
2362 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2363 root_objectid, owner, offset, 1);
2364 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2365 return ret ? ret : pending_ret;
2366 }
2367
2368 static u64 stripe_align(struct btrfs_root *root, u64 val)
2369 {
2370 u64 mask = ((u64)root->stripesize - 1);
2371 u64 ret = (val + mask) & ~mask;
2372 return ret;
2373 }
2374
2375 /*
2376 * walks the btree of allocated extents and find a hole of a given size.
2377 * The key ins is changed to record the hole:
2378 * ins->objectid == block start
2379 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2380 * ins->offset == number of blocks
2381 * Any available blocks before search_start are skipped.
2382 */
2383 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2384 struct btrfs_root *orig_root,
2385 u64 num_bytes, u64 empty_size,
2386 u64 search_start, u64 search_end,
2387 u64 hint_byte, struct btrfs_key *ins,
2388 u64 exclude_start, u64 exclude_nr,
2389 int data)
2390 {
2391 int ret;
2392 u64 orig_search_start = search_start;
2393 struct btrfs_root * root = orig_root->fs_info->extent_root;
2394 struct btrfs_fs_info *info = root->fs_info;
2395 u64 total_needed = num_bytes;
2396 struct btrfs_block_group_cache *block_group;
2397 int full_scan = 0;
2398 int wrapped = 0;
2399
2400 WARN_ON(num_bytes < root->sectorsize);
2401 btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
2402
2403 if (hint_byte) {
2404 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2405 if (!block_group)
2406 hint_byte = search_start;
2407 block_group = btrfs_find_block_group(root, block_group,
2408 hint_byte, data, 1);
2409 } else {
2410 block_group = btrfs_find_block_group(root,
2411 trans->block_group,
2412 search_start, data, 1);
2413 }
2414
2415 total_needed += empty_size;
2416
2417 check_failed:
2418 if (!block_group) {
2419 block_group = btrfs_lookup_first_block_group(info,
2420 search_start);
2421 if (!block_group)
2422 block_group = btrfs_lookup_first_block_group(info,
2423 orig_search_start);
2424 }
2425 ret = find_search_start(root, &block_group, &search_start,
2426 total_needed, data);
2427 if (ret)
2428 goto error;
2429
2430 search_start = stripe_align(root, search_start);
2431 ins->objectid = search_start;
2432 ins->offset = num_bytes;
2433
2434 if (ins->objectid + num_bytes >
2435 block_group->key.objectid + block_group->key.offset) {
2436 search_start = block_group->key.objectid +
2437 block_group->key.offset;
2438 goto new_group;
2439 }
2440
2441 if (test_range_bit(&info->extent_ins, ins->objectid,
2442 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2443 search_start = ins->objectid + num_bytes;
2444 goto new_group;
2445 }
2446
2447 if (test_range_bit(&info->pinned_extents, ins->objectid,
2448 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2449 search_start = ins->objectid + num_bytes;
2450 goto new_group;
2451 }
2452
2453 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2454 ins->objectid < exclude_start + exclude_nr)) {
2455 search_start = exclude_start + exclude_nr;
2456 goto new_group;
2457 }
2458
2459 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2460 block_group = btrfs_lookup_block_group(info, ins->objectid);
2461 if (block_group)
2462 trans->block_group = block_group;
2463 }
2464 ins->offset = num_bytes;
2465 return 0;
2466
2467 new_group:
2468 block_group = btrfs_lookup_first_block_group(info, search_start);
2469 if (!block_group) {
2470 search_start = orig_search_start;
2471 if (full_scan) {
2472 ret = -ENOSPC;
2473 goto error;
2474 }
2475 if (wrapped) {
2476 if (!full_scan)
2477 total_needed -= empty_size;
2478 full_scan = 1;
2479 } else
2480 wrapped = 1;
2481 }
2482 cond_resched();
2483 block_group = btrfs_find_block_group(root, block_group,
2484 search_start, data, 0);
2485 goto check_failed;
2486
2487 error:
2488 return ret;
2489 }
2490
2491 static int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2492 struct btrfs_root *root,
2493 u64 num_bytes, u64 empty_size,
2494 u64 hint_byte, u64 search_end,
2495 struct btrfs_key *ins, int data)
2496 {
2497 int ret;
2498 u64 search_start = 0;
2499 u64 alloc_profile;
2500 struct btrfs_fs_info *info = root->fs_info;
2501
2502 if (info->extent_ops) {
2503 struct btrfs_extent_ops *ops = info->extent_ops;
2504 ret = ops->alloc_extent(root, num_bytes, hint_byte, ins);
2505 BUG_ON(ret);
2506 goto found;
2507 }
2508
2509 if (data) {
2510 alloc_profile = info->avail_data_alloc_bits &
2511 info->data_alloc_profile;
2512 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2513 } else if ((info->system_allocs > 0 || root == info->chunk_root) &&
2514 info->system_allocs >= 0) {
2515 alloc_profile = info->avail_system_alloc_bits &
2516 info->system_alloc_profile;
2517 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2518 } else {
2519 alloc_profile = info->avail_metadata_alloc_bits &
2520 info->metadata_alloc_profile;
2521 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2522 }
2523
2524 if (root->ref_cows) {
2525 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2526 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2527 num_bytes,
2528 BTRFS_BLOCK_GROUP_METADATA);
2529 BUG_ON(ret);
2530 }
2531 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
2532 num_bytes + 2 * 1024 * 1024, data);
2533 BUG_ON(ret);
2534 }
2535
2536 WARN_ON(num_bytes < root->sectorsize);
2537 ret = find_free_extent(trans, root, num_bytes, empty_size,
2538 search_start, search_end, hint_byte, ins,
2539 trans->alloc_exclude_start,
2540 trans->alloc_exclude_nr, data);
2541 BUG_ON(ret);
2542 found:
2543 clear_extent_dirty(&root->fs_info->free_space_cache,
2544 ins->objectid, ins->objectid + ins->offset - 1,
2545 GFP_NOFS);
2546 return ret;
2547 }
2548
2549 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2550 struct btrfs_root *root,
2551 u64 root_objectid, u64 generation,
2552 u64 flags, struct btrfs_disk_key *key,
2553 int level, struct btrfs_key *ins)
2554 {
2555 int ret;
2556 struct btrfs_fs_info *fs_info = root->fs_info;
2557 struct btrfs_extent_item *extent_item;
2558 struct btrfs_tree_block_info *block_info;
2559 struct btrfs_extent_inline_ref *iref;
2560 struct btrfs_path *path;
2561 struct extent_buffer *leaf;
2562 u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref);
2563
2564 path = btrfs_alloc_path();
2565 BUG_ON(!path);
2566
2567 path->leave_spinning = 1;
2568 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2569 ins, size);
2570 BUG_ON(ret);
2571
2572 leaf = path->nodes[0];
2573 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2574 struct btrfs_extent_item);
2575 btrfs_set_extent_refs(leaf, extent_item, 1);
2576 btrfs_set_extent_generation(leaf, extent_item, generation);
2577 btrfs_set_extent_flags(leaf, extent_item,
2578 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2579 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2580
2581 btrfs_set_tree_block_key(leaf, block_info, key);
2582 btrfs_set_tree_block_level(leaf, block_info, level);
2583
2584 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2585 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2586 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2587
2588 btrfs_mark_buffer_dirty(leaf);
2589 btrfs_free_path(path);
2590
2591 ret = update_block_group(trans, root, ins->objectid, ins->offset,
2592 1, 0);
2593 if (ret) {
2594 printk(KERN_ERR "btrfs update block group failed for %llu "
2595 "%llu\n", (unsigned long long)ins->objectid,
2596 (unsigned long long)ins->offset);
2597 BUG();
2598 }
2599 return ret;
2600 }
2601
2602 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2603 struct btrfs_root *root, u64 num_bytes,
2604 u64 root_objectid, u64 generation,
2605 u64 flags, struct btrfs_disk_key *key,
2606 int level, u64 empty_size, u64 hint_byte,
2607 u64 search_end, struct btrfs_key *ins)
2608 {
2609 int ret;
2610 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2611 hint_byte, search_end, ins, 0);
2612 BUG_ON(ret);
2613
2614 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2615 struct pending_extent_op *extent_op;
2616
2617 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2618 BUG_ON(!extent_op);
2619
2620 extent_op->type = PENDING_EXTENT_INSERT;
2621 extent_op->bytenr = ins->objectid;
2622 extent_op->num_bytes = ins->offset;
2623 extent_op->level = level;
2624 extent_op->flags = flags;
2625 memcpy(&extent_op->key, key, sizeof(*key));
2626
2627 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2628 ins->objectid + ins->offset - 1,
2629 EXTENT_LOCKED, GFP_NOFS);
2630 set_state_private(&root->fs_info->extent_ins,
2631 ins->objectid, (unsigned long)extent_op);
2632 } else {
2633 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2634 generation, flags,
2635 key, level, ins);
2636 finish_current_insert(trans, root->fs_info->extent_root);
2637 del_pending_extents(trans, root->fs_info->extent_root);
2638 }
2639 return ret;
2640 }
2641
2642 /*
2643 * helper function to allocate a block for a given tree
2644 * returns the tree buffer or NULL.
2645 */
2646 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2647 struct btrfs_root *root,
2648 u32 blocksize, u64 root_objectid,
2649 struct btrfs_disk_key *key, int level,
2650 u64 hint, u64 empty_size)
2651 {
2652 struct btrfs_key ins;
2653 int ret;
2654 struct extent_buffer *buf;
2655
2656 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2657 trans->transid, 0, key, level,
2658 empty_size, hint, (u64)-1, &ins);
2659 if (ret) {
2660 BUG_ON(ret > 0);
2661 return ERR_PTR(ret);
2662 }
2663
2664 buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize);
2665 if (!buf) {
2666 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2667 0, root->root_key.objectid, level, 0);
2668 BUG_ON(1);
2669 return ERR_PTR(-ENOMEM);
2670 }
2671 btrfs_set_buffer_uptodate(buf);
2672 trans->blocks_used++;
2673
2674 return buf;
2675 }
2676
2677 #if 0
2678
2679 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2680 struct btrfs_root *root,
2681 struct extent_buffer *leaf)
2682 {
2683 u64 leaf_owner;
2684 u64 leaf_generation;
2685 struct btrfs_key key;
2686 struct btrfs_file_extent_item *fi;
2687 int i;
2688 int nritems;
2689 int ret;
2690
2691 BUG_ON(!btrfs_is_leaf(leaf));
2692 nritems = btrfs_header_nritems(leaf);
2693 leaf_owner = btrfs_header_owner(leaf);
2694 leaf_generation = btrfs_header_generation(leaf);
2695
2696 for (i = 0; i < nritems; i++) {
2697 u64 disk_bytenr;
2698
2699 btrfs_item_key_to_cpu(leaf, &key, i);
2700 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2701 continue;
2702 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2703 if (btrfs_file_extent_type(leaf, fi) ==
2704 BTRFS_FILE_EXTENT_INLINE)
2705 continue;
2706 /*
2707 * FIXME make sure to insert a trans record that
2708 * repeats the snapshot del on crash
2709 */
2710 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2711 if (disk_bytenr == 0)
2712 continue;
2713 ret = btrfs_free_extent(trans, root, disk_bytenr,
2714 btrfs_file_extent_disk_num_bytes(leaf, fi),
2715 leaf->start, leaf_owner, leaf_generation,
2716 key.objectid, 0);
2717 BUG_ON(ret);
2718 }
2719 return 0;
2720 }
2721
2722 static void noinline reada_walk_down(struct btrfs_root *root,
2723 struct extent_buffer *node,
2724 int slot)
2725 {
2726 u64 bytenr;
2727 u64 last = 0;
2728 u32 nritems;
2729 u32 refs;
2730 u32 blocksize;
2731 int ret;
2732 int i;
2733 int level;
2734 int skipped = 0;
2735
2736 nritems = btrfs_header_nritems(node);
2737 level = btrfs_header_level(node);
2738 if (level)
2739 return;
2740
2741 for (i = slot; i < nritems && skipped < 32; i++) {
2742 bytenr = btrfs_node_blockptr(node, i);
2743 if (last && ((bytenr > last && bytenr - last > 32 * 1024) ||
2744 (last > bytenr && last - bytenr > 32 * 1024))) {
2745 skipped++;
2746 continue;
2747 }
2748 blocksize = btrfs_level_size(root, level - 1);
2749 if (i != slot) {
2750 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2751 blocksize, &refs);
2752 BUG_ON(ret);
2753 if (refs != 1) {
2754 skipped++;
2755 continue;
2756 }
2757 }
2758 mutex_unlock(&root->fs_info->fs_mutex);
2759 ret = readahead_tree_block(root, bytenr, blocksize,
2760 btrfs_node_ptr_generation(node, i));
2761 last = bytenr + blocksize;
2762 cond_resched();
2763 mutex_lock(&root->fs_info->fs_mutex);
2764 if (ret)
2765 break;
2766 }
2767 }
2768
2769 /*
2770 * helper function for drop_snapshot, this walks down the tree dropping ref
2771 * counts as it goes.
2772 */
2773 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2774 struct btrfs_root *root,
2775 struct btrfs_path *path, int *level)
2776 {
2777 u64 root_owner;
2778 u64 root_gen;
2779 u64 bytenr;
2780 u64 ptr_gen;
2781 struct extent_buffer *next;
2782 struct extent_buffer *cur;
2783 struct extent_buffer *parent;
2784 u32 blocksize;
2785 int ret;
2786 u32 refs;
2787
2788 WARN_ON(*level < 0);
2789 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2790 ret = btrfs_lookup_extent_ref(trans, root,
2791 path->nodes[*level]->start,
2792 path->nodes[*level]->len, &refs);
2793 BUG_ON(ret);
2794 if (refs > 1)
2795 goto out;
2796
2797 /*
2798 * walk down to the last node level and free all the leaves
2799 */
2800 while(*level >= 0) {
2801 WARN_ON(*level < 0);
2802 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2803 cur = path->nodes[*level];
2804
2805 if (btrfs_header_level(cur) != *level)
2806 WARN_ON(1);
2807
2808 if (path->slots[*level] >=
2809 btrfs_header_nritems(cur))
2810 break;
2811 if (*level == 0) {
2812 ret = drop_leaf_ref(trans, root, cur);
2813 BUG_ON(ret);
2814 break;
2815 }
2816 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2817 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2818 blocksize = btrfs_level_size(root, *level - 1);
2819 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2820 &refs);
2821 BUG_ON(ret);
2822 if (refs != 1) {
2823 parent = path->nodes[*level];
2824 root_owner = btrfs_header_owner(parent);
2825 root_gen = btrfs_header_generation(parent);
2826 path->slots[*level]++;
2827 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2828 parent->start, root_owner,
2829 root_gen, *level - 1, 1);
2830 BUG_ON(ret);
2831 continue;
2832 }
2833 next = btrfs_find_tree_block(root, bytenr, blocksize);
2834 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2835 free_extent_buffer(next);
2836 reada_walk_down(root, cur, path->slots[*level]);
2837 mutex_unlock(&root->fs_info->fs_mutex);
2838 next = read_tree_block(root, bytenr, blocksize,
2839 ptr_gen);
2840 mutex_lock(&root->fs_info->fs_mutex);
2841 }
2842 WARN_ON(*level <= 0);
2843 if (path->nodes[*level-1])
2844 free_extent_buffer(path->nodes[*level-1]);
2845 path->nodes[*level-1] = next;
2846 *level = btrfs_header_level(next);
2847 path->slots[*level] = 0;
2848 }
2849 out:
2850 WARN_ON(*level < 0);
2851 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2852
2853 if (path->nodes[*level] == root->node) {
2854 root_owner = root->root_key.objectid;
2855 parent = path->nodes[*level];
2856 } else {
2857 parent = path->nodes[*level + 1];
2858 root_owner = btrfs_header_owner(parent);
2859 }
2860
2861 root_gen = btrfs_header_generation(parent);
2862 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
2863 path->nodes[*level]->len, parent->start,
2864 root_owner, root_gen, *level, 1);
2865 free_extent_buffer(path->nodes[*level]);
2866 path->nodes[*level] = NULL;
2867 *level += 1;
2868 BUG_ON(ret);
2869 return 0;
2870 }
2871
2872 /*
2873 * helper for dropping snapshots. This walks back up the tree in the path
2874 * to find the first node higher up where we haven't yet gone through
2875 * all the slots
2876 */
2877 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
2878 struct btrfs_root *root,
2879 struct btrfs_path *path, int *level)
2880 {
2881 u64 root_owner;
2882 u64 root_gen;
2883 struct btrfs_root_item *root_item = &root->root_item;
2884 int i;
2885 int slot;
2886 int ret;
2887
2888 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
2889 slot = path->slots[i];
2890 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
2891 struct extent_buffer *node;
2892 struct btrfs_disk_key disk_key;
2893 node = path->nodes[i];
2894 path->slots[i]++;
2895 *level = i;
2896 WARN_ON(*level == 0);
2897 btrfs_node_key(node, &disk_key, path->slots[i]);
2898 memcpy(&root_item->drop_progress,
2899 &disk_key, sizeof(disk_key));
2900 root_item->drop_level = i;
2901 return 0;
2902 } else {
2903 struct extent_buffer *parent;
2904 if (path->nodes[*level] == root->node)
2905 parent = path->nodes[*level];
2906 else
2907 parent = path->nodes[*level + 1];
2908
2909 root_owner = btrfs_header_owner(parent);
2910 root_gen = btrfs_header_generation(parent);
2911 ret = btrfs_free_extent(trans, root,
2912 path->nodes[*level]->start,
2913 path->nodes[*level]->len,
2914 parent->start, root_owner,
2915 root_gen, *level, 1);
2916 BUG_ON(ret);
2917 free_extent_buffer(path->nodes[*level]);
2918 path->nodes[*level] = NULL;
2919 *level = i + 1;
2920 }
2921 }
2922 return 1;
2923 }
2924
2925 /*
2926 * drop the reference count on the tree rooted at 'snap'. This traverses
2927 * the tree freeing any blocks that have a ref count of zero after being
2928 * decremented.
2929 */
2930 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
2931 *root)
2932 {
2933 int ret = 0;
2934 int wret;
2935 int level;
2936 struct btrfs_path *path;
2937 int i;
2938 int orig_level;
2939 struct btrfs_root_item *root_item = &root->root_item;
2940
2941 path = btrfs_alloc_path();
2942 BUG_ON(!path);
2943
2944 level = btrfs_header_level(root->node);
2945 orig_level = level;
2946 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2947 path->nodes[level] = root->node;
2948 extent_buffer_get(root->node);
2949 path->slots[level] = 0;
2950 } else {
2951 struct btrfs_key key;
2952 struct btrfs_disk_key found_key;
2953 struct extent_buffer *node;
2954
2955 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2956 level = root_item->drop_level;
2957 path->lowest_level = level;
2958 wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2959 if (wret < 0) {
2960 ret = wret;
2961 goto out;
2962 }
2963 node = path->nodes[level];
2964 btrfs_node_key(node, &found_key, path->slots[level]);
2965 WARN_ON(memcmp(&found_key, &root_item->drop_progress,
2966 sizeof(found_key)));
2967 }
2968 while(1) {
2969 wret = walk_down_tree(trans, root, path, &level);
2970 if (wret < 0)
2971 ret = wret;
2972 if (wret != 0)
2973 break;
2974
2975 wret = walk_up_tree(trans, root, path, &level);
2976 if (wret < 0)
2977 ret = wret;
2978 if (wret != 0)
2979 break;
2980 /*
2981 ret = -EAGAIN;
2982 break;
2983 */
2984 }
2985 for (i = 0; i <= orig_level; i++) {
2986 if (path->nodes[i]) {
2987 free_extent_buffer(path->nodes[i]);
2988 path->nodes[i] = NULL;
2989 }
2990 }
2991 out:
2992 btrfs_free_path(path);
2993 return ret;
2994 }
2995
2996 #endif
2997
2998 int btrfs_free_block_groups(struct btrfs_fs_info *info)
2999 {
3000 u64 start;
3001 u64 end;
3002 u64 ptr;
3003 int ret;
3004 while(1) {
3005 ret = find_first_extent_bit(&info->block_group_cache, 0,
3006 &start, &end, (unsigned int)-1);
3007 if (ret)
3008 break;
3009 ret = get_state_private(&info->block_group_cache, start, &ptr);
3010 if (!ret)
3011 kfree((void *)(unsigned long)ptr);
3012 clear_extent_bits(&info->block_group_cache, start,
3013 end, (unsigned int)-1, GFP_NOFS);
3014 }
3015 while(1) {
3016 ret = find_first_extent_bit(&info->free_space_cache, 0,
3017 &start, &end, EXTENT_DIRTY);
3018 if (ret)
3019 break;
3020 clear_extent_dirty(&info->free_space_cache, start,
3021 end, GFP_NOFS);
3022 }
3023 return 0;
3024 }
3025
3026 int find_first_block_group(struct btrfs_root *root, struct btrfs_path *path,
3027 struct btrfs_key *key)
3028 {
3029 int ret;
3030 struct btrfs_key found_key;
3031 struct extent_buffer *leaf;
3032 int slot;
3033
3034 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3035 if (ret < 0)
3036 return ret;
3037 while(1) {
3038 slot = path->slots[0];
3039 leaf = path->nodes[0];
3040 if (slot >= btrfs_header_nritems(leaf)) {
3041 ret = btrfs_next_leaf(root, path);
3042 if (ret == 0)
3043 continue;
3044 if (ret < 0)
3045 goto error;
3046 break;
3047 }
3048 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3049
3050 if (found_key.objectid >= key->objectid &&
3051 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3052 return 0;
3053 path->slots[0]++;
3054 }
3055 ret = -ENOENT;
3056 error:
3057 return ret;
3058 }
3059
3060 int btrfs_read_block_groups(struct btrfs_root *root)
3061 {
3062 struct btrfs_path *path;
3063 int ret;
3064 int bit;
3065 struct btrfs_block_group_cache *cache;
3066 struct btrfs_fs_info *info = root->fs_info;
3067 struct btrfs_space_info *space_info;
3068 struct extent_io_tree *block_group_cache;
3069 struct btrfs_key key;
3070 struct btrfs_key found_key;
3071 struct extent_buffer *leaf;
3072
3073 block_group_cache = &info->block_group_cache;
3074
3075 root = info->extent_root;
3076 key.objectid = 0;
3077 key.offset = 0;
3078 btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3079 path = btrfs_alloc_path();
3080 if (!path)
3081 return -ENOMEM;
3082
3083 while(1) {
3084 ret = find_first_block_group(root, path, &key);
3085 if (ret > 0) {
3086 ret = 0;
3087 goto error;
3088 }
3089 if (ret != 0) {
3090 goto error;
3091 }
3092 leaf = path->nodes[0];
3093 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3094 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3095 if (!cache) {
3096 ret = -ENOMEM;
3097 break;
3098 }
3099
3100 read_extent_buffer(leaf, &cache->item,
3101 btrfs_item_ptr_offset(leaf, path->slots[0]),
3102 sizeof(cache->item));
3103 memcpy(&cache->key, &found_key, sizeof(found_key));
3104 cache->cached = 0;
3105 cache->pinned = 0;
3106 key.objectid = found_key.objectid + found_key.offset;
3107 btrfs_release_path(root, path);
3108 cache->flags = btrfs_block_group_flags(&cache->item);
3109 bit = 0;
3110 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3111 bit = BLOCK_GROUP_DATA;
3112 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3113 bit = BLOCK_GROUP_SYSTEM;
3114 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3115 bit = BLOCK_GROUP_METADATA;
3116 }
3117 set_avail_alloc_bits(info, cache->flags);
3118 if (btrfs_chunk_readonly(root, cache->key.objectid))
3119 cache->ro = 1;
3120
3121 ret = update_space_info(info, cache->flags, found_key.offset,
3122 btrfs_block_group_used(&cache->item),
3123 &space_info);
3124 BUG_ON(ret);
3125 cache->space_info = space_info;
3126
3127 /* use EXTENT_LOCKED to prevent merging */
3128 set_extent_bits(block_group_cache, found_key.objectid,
3129 found_key.objectid + found_key.offset - 1,
3130 bit | EXTENT_LOCKED, GFP_NOFS);
3131 set_state_private(block_group_cache, found_key.objectid,
3132 (unsigned long)cache);
3133 }
3134 ret = 0;
3135 error:
3136 btrfs_free_path(path);
3137 return ret;
3138 }
3139
3140 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3141 struct btrfs_root *root, u64 bytes_used,
3142 u64 type, u64 chunk_objectid, u64 chunk_offset,
3143 u64 size)
3144 {
3145 int ret;
3146 int bit = 0;
3147 struct btrfs_root *extent_root;
3148 struct btrfs_block_group_cache *cache;
3149 struct extent_io_tree *block_group_cache;
3150
3151 extent_root = root->fs_info->extent_root;
3152 block_group_cache = &root->fs_info->block_group_cache;
3153
3154 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3155 BUG_ON(!cache);
3156 cache->key.objectid = chunk_offset;
3157 cache->key.offset = size;
3158
3159 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3160 btrfs_set_block_group_used(&cache->item, bytes_used);
3161 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3162 cache->flags = type;
3163 btrfs_set_block_group_flags(&cache->item, type);
3164
3165 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
3166 &cache->space_info);
3167 BUG_ON(ret);
3168
3169 bit = block_group_state_bits(type);
3170 set_extent_bits(block_group_cache, chunk_offset,
3171 chunk_offset + size - 1,
3172 bit | EXTENT_LOCKED, GFP_NOFS);
3173
3174 set_state_private(block_group_cache, chunk_offset,
3175 (unsigned long)cache);
3176 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3177 sizeof(cache->item));
3178 BUG_ON(ret);
3179
3180 finish_current_insert(trans, extent_root);
3181 ret = del_pending_extents(trans, extent_root);
3182 BUG_ON(ret);
3183 set_avail_alloc_bits(extent_root->fs_info, type);
3184 return 0;
3185 }
3186
3187 /*
3188 * This is for converter use only.
3189 *
3190 * In that case, we don't know where are free blocks located.
3191 * Therefore all block group cache entries must be setup properly
3192 * before doing any block allocation.
3193 */
3194 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3195 struct btrfs_root *root)
3196 {
3197 u64 total_bytes;
3198 u64 cur_start;
3199 u64 group_type;
3200 u64 group_size;
3201 u64 group_align;
3202 u64 total_data = 0;
3203 u64 total_metadata = 0;
3204 u64 chunk_objectid;
3205 int ret;
3206 int bit;
3207 struct btrfs_root *extent_root;
3208 struct btrfs_block_group_cache *cache;
3209 struct extent_io_tree *block_group_cache;
3210
3211 extent_root = root->fs_info->extent_root;
3212 block_group_cache = &root->fs_info->block_group_cache;
3213 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3214 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
3215 group_align = 64 * root->sectorsize;
3216
3217 cur_start = 0;
3218 while (cur_start < total_bytes) {
3219 group_size = total_bytes / 12;
3220 group_size = min_t(u64, group_size, total_bytes - cur_start);
3221 if (cur_start == 0) {
3222 bit = BLOCK_GROUP_SYSTEM;
3223 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3224 group_size /= 4;
3225 group_size &= ~(group_align - 1);
3226 group_size = max_t(u64, group_size, 8 * 1024 * 1024);
3227 group_size = min_t(u64, group_size, 32 * 1024 * 1024);
3228 } else {
3229 group_size &= ~(group_align - 1);
3230 if (total_data >= total_metadata * 2) {
3231 group_type = BTRFS_BLOCK_GROUP_METADATA;
3232 group_size = min_t(u64, group_size,
3233 1ULL * 1024 * 1024 * 1024);
3234 total_metadata += group_size;
3235 } else {
3236 group_type = BTRFS_BLOCK_GROUP_DATA;
3237 group_size = min_t(u64, group_size,
3238 5ULL * 1024 * 1024 * 1024);
3239 total_data += group_size;
3240 }
3241 if ((total_bytes - cur_start) * 4 < group_size * 5)
3242 group_size = total_bytes - cur_start;
3243 }
3244
3245 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3246 BUG_ON(!cache);
3247
3248 cache->key.objectid = cur_start;
3249 cache->key.offset = group_size;
3250 btrfs_set_key_type(&cache->key, BTRFS_BLOCK_GROUP_ITEM_KEY);
3251
3252 btrfs_set_block_group_used(&cache->item, 0);
3253 btrfs_set_block_group_chunk_objectid(&cache->item,
3254 chunk_objectid);
3255 btrfs_set_block_group_flags(&cache->item, group_type);
3256
3257 cache->flags = group_type;
3258
3259 ret = update_space_info(root->fs_info, group_type, group_size,
3260 0, &cache->space_info);
3261 BUG_ON(ret);
3262 set_avail_alloc_bits(extent_root->fs_info, group_type);
3263
3264 set_extent_bits(block_group_cache, cur_start,
3265 cur_start + group_size - 1,
3266 bit | EXTENT_LOCKED, GFP_NOFS);
3267 set_state_private(block_group_cache, cur_start,
3268 (unsigned long)cache);
3269 cur_start += group_size;
3270 }
3271 /* then insert all the items */
3272 cur_start = 0;
3273 while(cur_start < total_bytes) {
3274 cache = btrfs_lookup_block_group(root->fs_info, cur_start);
3275 BUG_ON(!cache);
3276
3277 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3278 sizeof(cache->item));
3279 BUG_ON(ret);
3280
3281 finish_current_insert(trans, extent_root);
3282 ret = del_pending_extents(trans, extent_root);
3283 BUG_ON(ret);
3284
3285 cur_start = cache->key.objectid + cache->key.offset;
3286 }
3287 return 0;
3288 }
3289
3290 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3291 struct btrfs_root *root,
3292 u64 bytenr, u64 num_bytes, int alloc,
3293 int mark_free)
3294 {
3295 return update_block_group(trans, root, bytenr, num_bytes,
3296 alloc, mark_free);
3297 }
(deprecated) Btrfs progs developments and patch integration