?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: mkfs: fix regression preventing --rootdir to create file
[btrfs-progs-unstable/devel.git] / extent-tree.c
blob055582c36da6c7c257ce50d3113ac48c4984a5e3
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 <stdint.h>
22 #include <math.h>
23 #include "kerncompat.h"
24 #include "radix-tree.h"
25 #include "ctree.h"
26 #include "disk-io.h"
27 #include "print-tree.h"
28 #include "transaction.h"
29 #include "crc32c.h"
30 #include "volumes.h"
31 #include "free-space-cache.h"
32 #include "utils.h"
33
34 #define PENDING_EXTENT_INSERT 0
35 #define PENDING_EXTENT_DELETE 1
36 #define PENDING_BACKREF_UPDATE 2
37
38 struct pending_extent_op {
39 int type;
40 u64 bytenr;
41 u64 num_bytes;
42 u64 flags;
43 struct btrfs_disk_key key;
44 int level;
45 };
46
47 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
48 struct btrfs_root *root,
49 u64 root_objectid, u64 generation,
50 u64 flags, struct btrfs_disk_key *key,
51 int level, struct btrfs_key *ins);
52 static int __free_extent(struct btrfs_trans_handle *trans,
53 struct btrfs_root *root,
54 u64 bytenr, u64 num_bytes, u64 parent,
55 u64 root_objectid, u64 owner_objectid,
56 u64 owner_offset, int refs_to_drop);
57 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
58 btrfs_root *extent_root);
59 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
60 btrfs_root *extent_root);
61 static struct btrfs_block_group_cache *
62 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
63 *hint, u64 search_start, int data, int owner);
64
65 static int remove_sb_from_cache(struct btrfs_root *root,
66 struct btrfs_block_group_cache *cache)
67 {
68 u64 bytenr;
69 u64 *logical;
70 int stripe_len;
71 int i, nr, ret;
72 struct btrfs_fs_info *fs_info = root->fs_info;
73 struct extent_io_tree *free_space_cache;
74
75 free_space_cache = &fs_info->free_space_cache;
76 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
77 bytenr = btrfs_sb_offset(i);
78 ret = btrfs_rmap_block(fs_info, cache->key.objectid, bytenr, 0,
79 &logical, &nr, &stripe_len);
80 BUG_ON(ret);
81 while (nr--) {
82 clear_extent_dirty(free_space_cache, logical[nr],
83 logical[nr] + stripe_len - 1);
84 }
85 kfree(logical);
86 }
87 return 0;
88 }
89
90 static int cache_block_group(struct btrfs_root *root,
91 struct btrfs_block_group_cache *block_group)
92 {
93 struct btrfs_path *path;
94 int ret;
95 struct btrfs_key key;
96 struct extent_buffer *leaf;
97 struct extent_io_tree *free_space_cache;
98 int slot;
99 u64 last;
100 u64 hole_size;
101
102 if (!block_group)
103 return 0;
104
105 root = root->fs_info->extent_root;
106 free_space_cache = &root->fs_info->free_space_cache;
107
108 if (block_group->cached)
109 return 0;
110
111 path = btrfs_alloc_path();
112 if (!path)
113 return -ENOMEM;
114
115 path->reada = 2;
116 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
117 key.objectid = last;
118 key.offset = 0;
119 key.type = 0;
120
121 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
122 if (ret < 0)
123 goto err;
124
125 while(1) {
126 leaf = path->nodes[0];
127 slot = path->slots[0];
128 if (slot >= btrfs_header_nritems(leaf)) {
129 ret = btrfs_next_leaf(root, path);
130 if (ret < 0)
131 goto err;
132 if (ret == 0) {
133 continue;
134 } else {
135 break;
136 }
137 }
138 btrfs_item_key_to_cpu(leaf, &key, slot);
139 if (key.objectid < block_group->key.objectid) {
140 goto next;
141 }
142 if (key.objectid >= block_group->key.objectid +
143 block_group->key.offset) {
144 break;
145 }
146
147 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
148 key.type == BTRFS_METADATA_ITEM_KEY) {
149 if (key.objectid > last) {
150 hole_size = key.objectid - last;
151 set_extent_dirty(free_space_cache, last,
152 last + hole_size - 1);
153 }
154 if (key.type == BTRFS_METADATA_ITEM_KEY)
155 last = key.objectid + root->fs_info->nodesize;
156 else
157 last = key.objectid + key.offset;
158 }
159 next:
160 path->slots[0]++;
161 }
162
163 if (block_group->key.objectid +
164 block_group->key.offset > last) {
165 hole_size = block_group->key.objectid +
166 block_group->key.offset - last;
167 set_extent_dirty(free_space_cache, last, last + hole_size - 1);
168 }
169 remove_sb_from_cache(root, block_group);
170 block_group->cached = 1;
171 err:
172 btrfs_free_path(path);
173 return 0;
174 }
175
176 struct btrfs_block_group_cache *btrfs_lookup_first_block_group(struct
177 btrfs_fs_info *info,
178 u64 bytenr)
179 {
180 struct extent_io_tree *block_group_cache;
181 struct btrfs_block_group_cache *block_group = NULL;
182 u64 ptr;
183 u64 start;
184 u64 end;
185 int ret;
186
187 bytenr = max_t(u64, bytenr,
188 BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE);
189 block_group_cache = &info->block_group_cache;
190 ret = find_first_extent_bit(block_group_cache,
191 bytenr, &start, &end,
192 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
193 BLOCK_GROUP_SYSTEM);
194 if (ret) {
195 return NULL;
196 }
197 ret = get_state_private(block_group_cache, start, &ptr);
198 if (ret)
199 return NULL;
200
201 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
202 return block_group;
203 }
204
205 struct btrfs_block_group_cache *btrfs_lookup_block_group(struct
206 btrfs_fs_info *info,
207 u64 bytenr)
208 {
209 struct extent_io_tree *block_group_cache;
210 struct btrfs_block_group_cache *block_group = NULL;
211 u64 ptr;
212 u64 start;
213 u64 end;
214 int ret;
215
216 block_group_cache = &info->block_group_cache;
217 ret = find_first_extent_bit(block_group_cache,
218 bytenr, &start, &end,
219 BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA |
220 BLOCK_GROUP_SYSTEM);
221 if (ret) {
222 return NULL;
223 }
224 ret = get_state_private(block_group_cache, start, &ptr);
225 if (ret)
226 return NULL;
227
228 block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr;
229 if (block_group->key.objectid <= bytenr && bytenr <
230 block_group->key.objectid + block_group->key.offset)
231 return block_group;
232 return NULL;
233 }
234
235 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
236 {
237 return (cache->flags & bits) == bits;
238 }
239
240 static int noinline find_search_start(struct btrfs_root *root,
241 struct btrfs_block_group_cache **cache_ret,
242 u64 *start_ret, int num, int data)
243 {
244 int ret;
245 struct btrfs_block_group_cache *cache = *cache_ret;
246 u64 last = *start_ret;
247 u64 start = 0;
248 u64 end = 0;
249 u64 search_start = *start_ret;
250 int wrapped = 0;
251
252 if (!cache)
253 goto out;
254 again:
255 ret = cache_block_group(root, cache);
256 if (ret)
257 goto out;
258
259 last = max(search_start, cache->key.objectid);
260 if (cache->ro || !block_group_bits(cache, data))
261 goto new_group;
262
263 while(1) {
264 ret = find_first_extent_bit(&root->fs_info->free_space_cache,
265 last, &start, &end, EXTENT_DIRTY);
266 if (ret) {
267 goto new_group;
268 }
269
270 start = max(last, start);
271 last = end + 1;
272 if (last - start < num) {
273 continue;
274 }
275 if (start + num > cache->key.objectid + cache->key.offset) {
276 goto new_group;
277 }
278 *start_ret = start;
279 return 0;
280 }
281 out:
282 *start_ret = last;
283 cache = btrfs_lookup_block_group(root->fs_info, search_start);
284 if (!cache) {
285 printk("Unable to find block group for %llu\n",
286 (unsigned long long)search_start);
287 return -ENOENT;
288 }
289 return -ENOSPC;
290
291 new_group:
292 last = cache->key.objectid + cache->key.offset;
293 wrapped:
294 cache = btrfs_lookup_first_block_group(root->fs_info, last);
295 if (!cache) {
296 if (!wrapped) {
297 wrapped = 1;
298 last = search_start;
299 goto wrapped;
300 }
301 goto out;
302 }
303 *cache_ret = cache;
304 goto again;
305 }
306
307 static int block_group_state_bits(u64 flags)
308 {
309 int bits = 0;
310 if (flags & BTRFS_BLOCK_GROUP_DATA)
311 bits |= BLOCK_GROUP_DATA;
312 if (flags & BTRFS_BLOCK_GROUP_METADATA)
313 bits |= BLOCK_GROUP_METADATA;
314 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
315 bits |= BLOCK_GROUP_SYSTEM;
316 return bits;
317 }
318
319 static struct btrfs_block_group_cache *
320 btrfs_find_block_group(struct btrfs_root *root, struct btrfs_block_group_cache
321 *hint, u64 search_start, int data, int owner)
322 {
323 struct btrfs_block_group_cache *cache;
324 struct extent_io_tree *block_group_cache;
325 struct btrfs_block_group_cache *found_group = NULL;
326 struct btrfs_fs_info *info = root->fs_info;
327 u64 used;
328 u64 last = 0;
329 u64 hint_last;
330 u64 start;
331 u64 end;
332 u64 free_check;
333 u64 ptr;
334 int bit;
335 int ret;
336 int full_search = 0;
337 int factor = 10;
338
339 block_group_cache = &info->block_group_cache;
340
341 if (!owner)
342 factor = 10;
343
344 bit = block_group_state_bits(data);
345
346 if (search_start) {
347 struct btrfs_block_group_cache *shint;
348 shint = btrfs_lookup_block_group(info, search_start);
349 if (shint && !shint->ro && block_group_bits(shint, data)) {
350 used = btrfs_block_group_used(&shint->item);
351 if (used + shint->pinned <
352 div_factor(shint->key.offset, factor)) {
353 return shint;
354 }
355 }
356 }
357 if (hint && !hint->ro && block_group_bits(hint, data)) {
358 used = btrfs_block_group_used(&hint->item);
359 if (used + hint->pinned <
360 div_factor(hint->key.offset, factor)) {
361 return hint;
362 }
363 last = hint->key.objectid + hint->key.offset;
364 hint_last = last;
365 } else {
366 if (hint)
367 hint_last = max(hint->key.objectid, search_start);
368 else
369 hint_last = search_start;
370
371 last = hint_last;
372 }
373 again:
374 while(1) {
375 ret = find_first_extent_bit(block_group_cache, last,
376 &start, &end, bit);
377 if (ret)
378 break;
379
380 ret = get_state_private(block_group_cache, start, &ptr);
381 if (ret)
382 break;
383
384 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
385 last = cache->key.objectid + cache->key.offset;
386 used = btrfs_block_group_used(&cache->item);
387
388 if (!cache->ro && block_group_bits(cache, data)) {
389 if (full_search)
390 free_check = cache->key.offset;
391 else
392 free_check = div_factor(cache->key.offset,
393 factor);
394
395 if (used + cache->pinned < free_check) {
396 found_group = cache;
397 goto found;
398 }
399 }
400 cond_resched();
401 }
402 if (!full_search) {
403 last = search_start;
404 full_search = 1;
405 goto again;
406 }
407 found:
408 return found_group;
409 }
410
411 /*
412 * Back reference rules. Back refs have three main goals:
413 *
414 * 1) differentiate between all holders of references to an extent so that
415 * when a reference is dropped we can make sure it was a valid reference
416 * before freeing the extent.
417 *
418 * 2) Provide enough information to quickly find the holders of an extent
419 * if we notice a given block is corrupted or bad.
420 *
421 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
422 * maintenance. This is actually the same as #2, but with a slightly
423 * different use case.
424 *
425 * There are two kinds of back refs. The implicit back refs is optimized
426 * for pointers in non-shared tree blocks. For a given pointer in a block,
427 * back refs of this kind provide information about the block's owner tree
428 * and the pointer's key. These information allow us to find the block by
429 * b-tree searching. The full back refs is for pointers in tree blocks not
430 * referenced by their owner trees. The location of tree block is recorded
431 * in the back refs. Actually the full back refs is generic, and can be
432 * used in all cases the implicit back refs is used. The major shortcoming
433 * of the full back refs is its overhead. Every time a tree block gets
434 * COWed, we have to update back refs entry for all pointers in it.
435 *
436 * For a newly allocated tree block, we use implicit back refs for
437 * pointers in it. This means most tree related operations only involve
438 * implicit back refs. For a tree block created in old transaction, the
439 * only way to drop a reference to it is COW it. So we can detect the
440 * event that tree block loses its owner tree's reference and do the
441 * back refs conversion.
442 *
443 * When a tree block is COW'd through a tree, there are four cases:
444 *
445 * The reference count of the block is one and the tree is the block's
446 * owner tree. Nothing to do in this case.
447 *
448 * The reference count of the block is one and the tree is not the
449 * block's owner tree. In this case, full back refs is used for pointers
450 * in the block. Remove these full back refs, add implicit back refs for
451 * every pointers in the new block.
452 *
453 * The reference count of the block is greater than one and the tree is
454 * the block's owner tree. In this case, implicit back refs is used for
455 * pointers in the block. Add full back refs for every pointers in the
456 * block, increase lower level extents' reference counts. The original
457 * implicit back refs are entailed to the new block.
458 *
459 * The reference count of the block is greater than one and the tree is
460 * not the block's owner tree. Add implicit back refs for every pointer in
461 * the new block, increase lower level extents' reference count.
462 *
463 * Back Reference Key composing:
464 *
465 * The key objectid corresponds to the first byte in the extent,
466 * The key type is used to differentiate between types of back refs.
467 * There are different meanings of the key offset for different types
468 * of back refs.
469 *
470 * File extents can be referenced by:
471 *
472 * - multiple snapshots, subvolumes, or different generations in one subvol
473 * - different files inside a single subvolume
474 * - different offsets inside a file (bookend extents in file.c)
475 *
476 * The extent ref structure for the implicit back refs has fields for:
477 *
478 * - Objectid of the subvolume root
479 * - objectid of the file holding the reference
480 * - original offset in the file
481 * - how many bookend extents
482 *
483 * The key offset for the implicit back refs is hash of the first
484 * three fields.
485 *
486 * The extent ref structure for the full back refs has field for:
487 *
488 * - number of pointers in the tree leaf
489 *
490 * The key offset for the implicit back refs is the first byte of
491 * the tree leaf
492 *
493 * When a file extent is allocated, The implicit back refs is used.
494 * the fields are filled in:
495 *
496 * (root_key.objectid, inode objectid, offset in file, 1)
497 *
498 * When a file extent is removed file truncation, we find the
499 * corresponding implicit back refs and check the following fields:
500 *
501 * (btrfs_header_owner(leaf), inode objectid, offset in file)
502 *
503 * Btree extents can be referenced by:
504 *
505 * - Different subvolumes
506 *
507 * Both the implicit back refs and the full back refs for tree blocks
508 * only consist of key. The key offset for the implicit back refs is
509 * objectid of block's owner tree. The key offset for the full back refs
510 * is the first byte of parent block.
511 *
512 * When implicit back refs is used, information about the lowest key and
513 * level of the tree block are required. These information are stored in
514 * tree block info structure.
515 */
516
517 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
518 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
519 struct btrfs_root *root,
520 struct btrfs_path *path,
521 u64 owner, u32 extra_size)
522 {
523 struct btrfs_extent_item *item;
524 struct btrfs_extent_item_v0 *ei0;
525 struct btrfs_extent_ref_v0 *ref0;
526 struct btrfs_tree_block_info *bi;
527 struct extent_buffer *leaf;
528 struct btrfs_key key;
529 struct btrfs_key found_key;
530 u32 new_size = sizeof(*item);
531 u64 refs;
532 int ret;
533
534 leaf = path->nodes[0];
535 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
536
537 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
538 ei0 = btrfs_item_ptr(leaf, path->slots[0],
539 struct btrfs_extent_item_v0);
540 refs = btrfs_extent_refs_v0(leaf, ei0);
541
542 if (owner == (u64)-1) {
543 while (1) {
544 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
545 ret = btrfs_next_leaf(root, path);
546 if (ret < 0)
547 return ret;
548 BUG_ON(ret > 0);
549 leaf = path->nodes[0];
550 }
551 btrfs_item_key_to_cpu(leaf, &found_key,
552 path->slots[0]);
553 BUG_ON(key.objectid != found_key.objectid);
554 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
555 path->slots[0]++;
556 continue;
557 }
558 ref0 = btrfs_item_ptr(leaf, path->slots[0],
559 struct btrfs_extent_ref_v0);
560 owner = btrfs_ref_objectid_v0(leaf, ref0);
561 break;
562 }
563 }
564 btrfs_release_path(path);
565
566 if (owner < BTRFS_FIRST_FREE_OBJECTID)
567 new_size += sizeof(*bi);
568
569 new_size -= sizeof(*ei0);
570 ret = btrfs_search_slot(trans, root, &key, path, new_size, 1);
571 if (ret < 0)
572 return ret;
573 BUG_ON(ret);
574
575 ret = btrfs_extend_item(root, path, new_size);
576 BUG_ON(ret);
577
578 leaf = path->nodes[0];
579 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
580 btrfs_set_extent_refs(leaf, item, refs);
581 /* FIXME: get real generation */
582 btrfs_set_extent_generation(leaf, item, 0);
583 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
584 btrfs_set_extent_flags(leaf, item,
585 BTRFS_EXTENT_FLAG_TREE_BLOCK |
586 BTRFS_BLOCK_FLAG_FULL_BACKREF);
587 bi = (struct btrfs_tree_block_info *)(item + 1);
588 /* FIXME: get first key of the block */
589 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
590 btrfs_set_tree_block_level(leaf, bi, (int)owner);
591 } else {
592 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
593 }
594 btrfs_mark_buffer_dirty(leaf);
595 return 0;
596 }
597 #endif
598
599 u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
600 {
601 u32 high_crc = ~(u32)0;
602 u32 low_crc = ~(u32)0;
603 __le64 lenum;
604
605 lenum = cpu_to_le64(root_objectid);
606 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
607 lenum = cpu_to_le64(owner);
608 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
609 lenum = cpu_to_le64(offset);
610 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
611
612 return ((u64)high_crc << 31) ^ (u64)low_crc;
613 }
614
615 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
616 struct btrfs_extent_data_ref *ref)
617 {
618 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
619 btrfs_extent_data_ref_objectid(leaf, ref),
620 btrfs_extent_data_ref_offset(leaf, ref));
621 }
622
623 static int match_extent_data_ref(struct extent_buffer *leaf,
624 struct btrfs_extent_data_ref *ref,
625 u64 root_objectid, u64 owner, u64 offset)
626 {
627 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
628 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
629 btrfs_extent_data_ref_offset(leaf, ref) != offset)
630 return 0;
631 return 1;
632 }
633
634 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
635 struct btrfs_root *root,
636 struct btrfs_path *path,
637 u64 bytenr, u64 parent,
638 u64 root_objectid,
639 u64 owner, u64 offset)
640 {
641 struct btrfs_key key;
642 struct btrfs_extent_data_ref *ref;
643 struct extent_buffer *leaf;
644 u32 nritems;
645 int ret;
646 int recow;
647 int err = -ENOENT;
648
649 key.objectid = bytenr;
650 if (parent) {
651 key.type = BTRFS_SHARED_DATA_REF_KEY;
652 key.offset = parent;
653 } else {
654 key.type = BTRFS_EXTENT_DATA_REF_KEY;
655 key.offset = hash_extent_data_ref(root_objectid,
656 owner, offset);
657 }
658 again:
659 recow = 0;
660 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
661 if (ret < 0) {
662 err = ret;
663 goto fail;
664 }
665
666 if (parent) {
667 if (!ret)
668 return 0;
669 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
670 key.type = BTRFS_EXTENT_REF_V0_KEY;
671 btrfs_release_path(path);
672 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
673 if (ret < 0) {
674 err = ret;
675 goto fail;
676 }
677 if (!ret)
678 return 0;
679 #endif
680 goto fail;
681 }
682
683 leaf = path->nodes[0];
684 nritems = btrfs_header_nritems(leaf);
685 while (1) {
686 if (path->slots[0] >= nritems) {
687 ret = btrfs_next_leaf(root, path);
688 if (ret < 0)
689 err = ret;
690 if (ret)
691 goto fail;
692
693 leaf = path->nodes[0];
694 nritems = btrfs_header_nritems(leaf);
695 recow = 1;
696 }
697
698 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
699 if (key.objectid != bytenr ||
700 key.type != BTRFS_EXTENT_DATA_REF_KEY)
701 goto fail;
702
703 ref = btrfs_item_ptr(leaf, path->slots[0],
704 struct btrfs_extent_data_ref);
705
706 if (match_extent_data_ref(leaf, ref, root_objectid,
707 owner, offset)) {
708 if (recow) {
709 btrfs_release_path(path);
710 goto again;
711 }
712 err = 0;
713 break;
714 }
715 path->slots[0]++;
716 }
717 fail:
718 return err;
719 }
720
721 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
722 struct btrfs_root *root,
723 struct btrfs_path *path,
724 u64 bytenr, u64 parent,
725 u64 root_objectid, u64 owner,
726 u64 offset, int refs_to_add)
727 {
728 struct btrfs_key key;
729 struct extent_buffer *leaf;
730 u32 size;
731 u32 num_refs;
732 int ret;
733
734 key.objectid = bytenr;
735 if (parent) {
736 key.type = BTRFS_SHARED_DATA_REF_KEY;
737 key.offset = parent;
738 size = sizeof(struct btrfs_shared_data_ref);
739 } else {
740 key.type = BTRFS_EXTENT_DATA_REF_KEY;
741 key.offset = hash_extent_data_ref(root_objectid,
742 owner, offset);
743 size = sizeof(struct btrfs_extent_data_ref);
744 }
745
746 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
747 if (ret && ret != -EEXIST)
748 goto fail;
749
750 leaf = path->nodes[0];
751 if (parent) {
752 struct btrfs_shared_data_ref *ref;
753 ref = btrfs_item_ptr(leaf, path->slots[0],
754 struct btrfs_shared_data_ref);
755 if (ret == 0) {
756 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
757 } else {
758 num_refs = btrfs_shared_data_ref_count(leaf, ref);
759 num_refs += refs_to_add;
760 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
761 }
762 } else {
763 struct btrfs_extent_data_ref *ref;
764 while (ret == -EEXIST) {
765 ref = btrfs_item_ptr(leaf, path->slots[0],
766 struct btrfs_extent_data_ref);
767 if (match_extent_data_ref(leaf, ref, root_objectid,
768 owner, offset))
769 break;
770 btrfs_release_path(path);
771
772 key.offset++;
773 ret = btrfs_insert_empty_item(trans, root, path, &key,
774 size);
775 if (ret && ret != -EEXIST)
776 goto fail;
777
778 leaf = path->nodes[0];
779 }
780 ref = btrfs_item_ptr(leaf, path->slots[0],
781 struct btrfs_extent_data_ref);
782 if (ret == 0) {
783 btrfs_set_extent_data_ref_root(leaf, ref,
784 root_objectid);
785 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
786 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
787 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
788 } else {
789 num_refs = btrfs_extent_data_ref_count(leaf, ref);
790 num_refs += refs_to_add;
791 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
792 }
793 }
794 btrfs_mark_buffer_dirty(leaf);
795 ret = 0;
796 fail:
797 btrfs_release_path(path);
798 return ret;
799 }
800
801 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
802 struct btrfs_root *root,
803 struct btrfs_path *path,
804 int refs_to_drop)
805 {
806 struct btrfs_key key;
807 struct btrfs_extent_data_ref *ref1 = NULL;
808 struct btrfs_shared_data_ref *ref2 = NULL;
809 struct extent_buffer *leaf;
810 u32 num_refs = 0;
811 int ret = 0;
812
813 leaf = path->nodes[0];
814 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
815
816 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
817 ref1 = btrfs_item_ptr(leaf, path->slots[0],
818 struct btrfs_extent_data_ref);
819 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
820 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
821 ref2 = btrfs_item_ptr(leaf, path->slots[0],
822 struct btrfs_shared_data_ref);
823 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
824 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
825 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
826 struct btrfs_extent_ref_v0 *ref0;
827 ref0 = btrfs_item_ptr(leaf, path->slots[0],
828 struct btrfs_extent_ref_v0);
829 num_refs = btrfs_ref_count_v0(leaf, ref0);
830 #endif
831 } else {
832 BUG();
833 }
834
835 BUG_ON(num_refs < refs_to_drop);
836 num_refs -= refs_to_drop;
837
838 if (num_refs == 0) {
839 ret = btrfs_del_item(trans, root, path);
840 } else {
841 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
842 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
843 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
844 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
845 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
846 else {
847 struct btrfs_extent_ref_v0 *ref0;
848 ref0 = btrfs_item_ptr(leaf, path->slots[0],
849 struct btrfs_extent_ref_v0);
850 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
851 }
852 #endif
853 btrfs_mark_buffer_dirty(leaf);
854 }
855 return ret;
856 }
857
858 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
859 struct btrfs_extent_inline_ref *iref)
860 {
861 struct btrfs_key key;
862 struct extent_buffer *leaf;
863 struct btrfs_extent_data_ref *ref1;
864 struct btrfs_shared_data_ref *ref2;
865 u32 num_refs = 0;
866
867 leaf = path->nodes[0];
868 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
869 if (iref) {
870 if (btrfs_extent_inline_ref_type(leaf, iref) ==
871 BTRFS_EXTENT_DATA_REF_KEY) {
872 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
873 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
874 } else {
875 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
876 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
877 }
878 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
879 ref1 = btrfs_item_ptr(leaf, path->slots[0],
880 struct btrfs_extent_data_ref);
881 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
882 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
883 ref2 = btrfs_item_ptr(leaf, path->slots[0],
884 struct btrfs_shared_data_ref);
885 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
886 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
887 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
888 struct btrfs_extent_ref_v0 *ref0;
889 ref0 = btrfs_item_ptr(leaf, path->slots[0],
890 struct btrfs_extent_ref_v0);
891 num_refs = btrfs_ref_count_v0(leaf, ref0);
892 #endif
893 } else {
894 BUG();
895 }
896 return num_refs;
897 }
898
899 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
900 struct btrfs_root *root,
901 struct btrfs_path *path,
902 u64 bytenr, u64 parent,
903 u64 root_objectid)
904 {
905 struct btrfs_key key;
906 int ret;
907
908 key.objectid = bytenr;
909 if (parent) {
910 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
911 key.offset = parent;
912 } else {
913 key.type = BTRFS_TREE_BLOCK_REF_KEY;
914 key.offset = root_objectid;
915 }
916
917 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
918 if (ret > 0)
919 ret = -ENOENT;
920 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
921 if (ret == -ENOENT && parent) {
922 btrfs_release_path(path);
923 key.type = BTRFS_EXTENT_REF_V0_KEY;
924 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
925 if (ret > 0)
926 ret = -ENOENT;
927 }
928 #endif
929 return ret;
930 }
931
932 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
933 struct btrfs_root *root,
934 struct btrfs_path *path,
935 u64 bytenr, u64 parent,
936 u64 root_objectid)
937 {
938 struct btrfs_key key;
939 int ret;
940
941 key.objectid = bytenr;
942 if (parent) {
943 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
944 key.offset = parent;
945 } else {
946 key.type = BTRFS_TREE_BLOCK_REF_KEY;
947 key.offset = root_objectid;
948 }
949
950 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
951
952 btrfs_release_path(path);
953 return ret;
954 }
955
956 static inline int extent_ref_type(u64 parent, u64 owner)
957 {
958 int type;
959 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
960 if (parent > 0)
961 type = BTRFS_SHARED_BLOCK_REF_KEY;
962 else
963 type = BTRFS_TREE_BLOCK_REF_KEY;
964 } else {
965 if (parent > 0)
966 type = BTRFS_SHARED_DATA_REF_KEY;
967 else
968 type = BTRFS_EXTENT_DATA_REF_KEY;
969 }
970 return type;
971 }
972
973 static int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
974 struct btrfs_root *root,
975 struct btrfs_path *path,
976 struct btrfs_extent_inline_ref **ref_ret,
977 u64 bytenr, u64 num_bytes,
978 u64 parent, u64 root_objectid,
979 u64 owner, u64 offset, int insert)
980 {
981 struct btrfs_key key;
982 struct extent_buffer *leaf;
983 struct btrfs_extent_item *ei;
984 struct btrfs_extent_inline_ref *iref;
985 u64 flags;
986 u32 item_size;
987 unsigned long ptr;
988 unsigned long end;
989 int extra_size;
990 int type;
991 int want;
992 int ret;
993 int err = 0;
994 int skinny_metadata =
995 btrfs_fs_incompat(root->fs_info, SKINNY_METADATA);
996
997 key.objectid = bytenr;
998 key.type = BTRFS_EXTENT_ITEM_KEY;
999 key.offset = num_bytes;
1000
1001 want = extent_ref_type(parent, owner);
1002 if (insert)
1003 extra_size = btrfs_extent_inline_ref_size(want);
1004 else
1005 extra_size = -1;
1006
1007 if (owner < BTRFS_FIRST_FREE_OBJECTID && skinny_metadata) {
1008 skinny_metadata = 1;
1009 key.type = BTRFS_METADATA_ITEM_KEY;
1010 key.offset = owner;
1011 } else if (skinny_metadata) {
1012 skinny_metadata = 0;
1013 }
1014
1015 again:
1016 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1017 if (ret < 0) {
1018 err = ret;
1019 goto out;
1020 }
1021
1022 /*
1023 * We may be a newly converted file system which still has the old fat
1024 * extent entries for metadata, so try and see if we have one of those.
1025 */
1026 if (ret > 0 && skinny_metadata) {
1027 skinny_metadata = 0;
1028 if (path->slots[0]) {
1029 path->slots[0]--;
1030 btrfs_item_key_to_cpu(path->nodes[0], &key,
1031 path->slots[0]);
1032 if (key.objectid == bytenr &&
1033 key.type == BTRFS_EXTENT_ITEM_KEY &&
1034 key.offset == num_bytes)
1035 ret = 0;
1036 }
1037 if (ret) {
1038 key.type = BTRFS_EXTENT_ITEM_KEY;
1039 key.offset = num_bytes;
1040 btrfs_release_path(path);
1041 goto again;
1042 }
1043 }
1044
1045 if (ret) {
1046 printf("Failed to find [%llu, %u, %llu]\n", key.objectid, key.type, key.offset);
1047 return -ENOENT;
1048 }
1049
1050 BUG_ON(ret);
1051
1052 leaf = path->nodes[0];
1053 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1054 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1055 if (item_size < sizeof(*ei)) {
1056 if (!insert) {
1057 err = -ENOENT;
1058 goto out;
1059 }
1060 ret = convert_extent_item_v0(trans, root, path, owner,
1061 extra_size);
1062 if (ret < 0) {
1063 err = ret;
1064 goto out;
1065 }
1066 leaf = path->nodes[0];
1067 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1068 }
1069 #endif
1070 if (item_size < sizeof(*ei)) {
1071 printf("Size is %u, needs to be %u, slot %d\n",
1072 (unsigned)item_size,
1073 (unsigned)sizeof(*ei), path->slots[0]);
1074 btrfs_print_leaf(root, leaf);
1075 return -EINVAL;
1076 }
1077 BUG_ON(item_size < sizeof(*ei));
1078
1079 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1080 flags = btrfs_extent_flags(leaf, ei);
1081
1082 ptr = (unsigned long)(ei + 1);
1083 end = (unsigned long)ei + item_size;
1084
1085 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1086 ptr += sizeof(struct btrfs_tree_block_info);
1087 BUG_ON(ptr > end);
1088 } else if (!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) {
1089 if (!(flags & BTRFS_EXTENT_FLAG_DATA)) {
1090 return -EIO;
1091 }
1092 }
1093
1094 err = -ENOENT;
1095 while (1) {
1096 if (ptr >= end) {
1097 WARN_ON(ptr > end);
1098 break;
1099 }
1100 iref = (struct btrfs_extent_inline_ref *)ptr;
1101 type = btrfs_extent_inline_ref_type(leaf, iref);
1102 if (want < type)
1103 break;
1104 if (want > type) {
1105 ptr += btrfs_extent_inline_ref_size(type);
1106 continue;
1107 }
1108
1109 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1110 struct btrfs_extent_data_ref *dref;
1111 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1112 if (match_extent_data_ref(leaf, dref, root_objectid,
1113 owner, offset)) {
1114 err = 0;
1115 break;
1116 }
1117 if (hash_extent_data_ref_item(leaf, dref) <
1118 hash_extent_data_ref(root_objectid, owner, offset))
1119 break;
1120 } else {
1121 u64 ref_offset;
1122 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1123 if (parent > 0) {
1124 if (parent == ref_offset) {
1125 err = 0;
1126 break;
1127 }
1128 if (ref_offset < parent)
1129 break;
1130 } else {
1131 if (root_objectid == ref_offset) {
1132 err = 0;
1133 break;
1134 }
1135 if (ref_offset < root_objectid)
1136 break;
1137 }
1138 }
1139 ptr += btrfs_extent_inline_ref_size(type);
1140 }
1141 if (err == -ENOENT && insert) {
1142 if (item_size + extra_size >=
1143 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1144 err = -EAGAIN;
1145 goto out;
1146 }
1147 /*
1148 * To add new inline back ref, we have to make sure
1149 * there is no corresponding back ref item.
1150 * For simplicity, we just do not add new inline back
1151 * ref if there is any back ref item.
1152 */
1153 if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
1154 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1155 err = -EAGAIN;
1156 goto out;
1157 }
1158 }
1159 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1160 out:
1161 return err;
1162 }
1163
1164 static int setup_inline_extent_backref(struct btrfs_root *root,
1165 struct btrfs_path *path,
1166 struct btrfs_extent_inline_ref *iref,
1167 u64 parent, u64 root_objectid,
1168 u64 owner, u64 offset, int refs_to_add)
1169 {
1170 struct extent_buffer *leaf;
1171 struct btrfs_extent_item *ei;
1172 unsigned long ptr;
1173 unsigned long end;
1174 unsigned long item_offset;
1175 u64 refs;
1176 int size;
1177 int type;
1178 int ret;
1179
1180 leaf = path->nodes[0];
1181 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1182 item_offset = (unsigned long)iref - (unsigned long)ei;
1183
1184 type = extent_ref_type(parent, owner);
1185 size = btrfs_extent_inline_ref_size(type);
1186
1187 ret = btrfs_extend_item(root, path, size);
1188 BUG_ON(ret);
1189
1190 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1191 refs = btrfs_extent_refs(leaf, ei);
1192 refs += refs_to_add;
1193 btrfs_set_extent_refs(leaf, ei, refs);
1194
1195 ptr = (unsigned long)ei + item_offset;
1196 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1197 if (ptr < end - size)
1198 memmove_extent_buffer(leaf, ptr + size, ptr,
1199 end - size - ptr);
1200
1201 iref = (struct btrfs_extent_inline_ref *)ptr;
1202 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1203 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1204 struct btrfs_extent_data_ref *dref;
1205 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1206 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1207 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1208 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1209 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1210 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1211 struct btrfs_shared_data_ref *sref;
1212 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1213 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1214 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1215 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1216 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1217 } else {
1218 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1219 }
1220 btrfs_mark_buffer_dirty(leaf);
1221 return 0;
1222 }
1223
1224 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1225 struct btrfs_root *root,
1226 struct btrfs_path *path,
1227 struct btrfs_extent_inline_ref **ref_ret,
1228 u64 bytenr, u64 num_bytes, u64 parent,
1229 u64 root_objectid, u64 owner, u64 offset)
1230 {
1231 int ret;
1232
1233 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1234 bytenr, num_bytes, parent,
1235 root_objectid, owner, offset, 0);
1236 if (ret != -ENOENT)
1237 return ret;
1238
1239 btrfs_release_path(path);
1240 *ref_ret = NULL;
1241
1242 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1243 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1244 root_objectid);
1245 } else {
1246 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1247 root_objectid, owner, offset);
1248 }
1249 return ret;
1250 }
1251
1252 static int update_inline_extent_backref(struct btrfs_trans_handle *trans,
1253 struct btrfs_root *root,
1254 struct btrfs_path *path,
1255 struct btrfs_extent_inline_ref *iref,
1256 int refs_to_mod)
1257 {
1258 struct extent_buffer *leaf;
1259 struct btrfs_extent_item *ei;
1260 struct btrfs_extent_data_ref *dref = NULL;
1261 struct btrfs_shared_data_ref *sref = NULL;
1262 unsigned long ptr;
1263 unsigned long end;
1264 u32 item_size;
1265 int size;
1266 int type;
1267 int ret;
1268 u64 refs;
1269
1270 leaf = path->nodes[0];
1271 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1272 refs = btrfs_extent_refs(leaf, ei);
1273 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1274 refs += refs_to_mod;
1275 btrfs_set_extent_refs(leaf, ei, refs);
1276
1277 type = btrfs_extent_inline_ref_type(leaf, iref);
1278
1279 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1280 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1281 refs = btrfs_extent_data_ref_count(leaf, dref);
1282 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1283 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1284 refs = btrfs_shared_data_ref_count(leaf, sref);
1285 } else {
1286 refs = 1;
1287 BUG_ON(refs_to_mod != -1);
1288 }
1289
1290 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1291 refs += refs_to_mod;
1292
1293 if (refs > 0) {
1294 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1295 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1296 else
1297 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1298 } else {
1299 size = btrfs_extent_inline_ref_size(type);
1300 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1301 ptr = (unsigned long)iref;
1302 end = (unsigned long)ei + item_size;
1303 if (ptr + size < end)
1304 memmove_extent_buffer(leaf, ptr, ptr + size,
1305 end - ptr - size);
1306 item_size -= size;
1307 ret = btrfs_truncate_item(root, path, item_size, 1);
1308 BUG_ON(ret);
1309 }
1310 btrfs_mark_buffer_dirty(leaf);
1311 return 0;
1312 }
1313
1314 static int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1315 struct btrfs_root *root,
1316 struct btrfs_path *path,
1317 u64 bytenr, u64 num_bytes, u64 parent,
1318 u64 root_objectid, u64 owner,
1319 u64 offset, int refs_to_add)
1320 {
1321 struct btrfs_extent_inline_ref *iref;
1322 int ret;
1323
1324 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1325 bytenr, num_bytes, parent,
1326 root_objectid, owner, offset, 1);
1327 if (ret == 0) {
1328 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1329 ret = update_inline_extent_backref(trans, root, path, iref,
1330 refs_to_add);
1331 } else if (ret == -ENOENT) {
1332 ret = setup_inline_extent_backref(root, path, iref,
1333 parent, root_objectid,
1334 owner, offset, refs_to_add);
1335 }
1336 return ret;
1337 }
1338
1339 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1340 struct btrfs_root *root,
1341 struct btrfs_path *path,
1342 u64 bytenr, u64 parent, u64 root_objectid,
1343 u64 owner, u64 offset, int refs_to_add)
1344 {
1345 int ret;
1346
1347 if (owner >= BTRFS_FIRST_FREE_OBJECTID) {
1348 ret = insert_extent_data_ref(trans, root, path, bytenr,
1349 parent, root_objectid,
1350 owner, offset, refs_to_add);
1351 } else {
1352 BUG_ON(refs_to_add != 1);
1353 ret = insert_tree_block_ref(trans, root, path, bytenr,
1354 parent, root_objectid);
1355 }
1356 return ret;
1357 }
1358
1359 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1360 struct btrfs_root *root,
1361 struct btrfs_path *path,
1362 struct btrfs_extent_inline_ref *iref,
1363 int refs_to_drop, int is_data)
1364 {
1365 int ret;
1366
1367 BUG_ON(!is_data && refs_to_drop != 1);
1368 if (iref) {
1369 ret = update_inline_extent_backref(trans, root, path, iref,
1370 -refs_to_drop);
1371 } else if (is_data) {
1372 ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
1373 } else {
1374 ret = btrfs_del_item(trans, root, path);
1375 }
1376 return ret;
1377 }
1378
1379 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1380 struct btrfs_root *root,
1381 u64 bytenr, u64 num_bytes, u64 parent,
1382 u64 root_objectid, u64 owner, u64 offset)
1383 {
1384 struct btrfs_path *path;
1385 struct extent_buffer *leaf;
1386 struct btrfs_extent_item *item;
1387 u64 refs;
1388 int ret;
1389 int err = 0;
1390
1391 path = btrfs_alloc_path();
1392 if (!path)
1393 return -ENOMEM;
1394
1395 path->reada = 1;
1396
1397 ret = insert_inline_extent_backref(trans, root->fs_info->extent_root,
1398 path, bytenr, num_bytes, parent,
1399 root_objectid, owner, offset, 1);
1400 if (ret == 0)
1401 goto out;
1402
1403 if (ret != -EAGAIN) {
1404 err = ret;
1405 goto out;
1406 }
1407
1408 leaf = path->nodes[0];
1409 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1410 refs = btrfs_extent_refs(leaf, item);
1411 btrfs_set_extent_refs(leaf, item, refs + 1);
1412
1413 btrfs_mark_buffer_dirty(leaf);
1414 btrfs_release_path(path);
1415
1416 path->reada = 1;
1417
1418 /* now insert the actual backref */
1419 ret = insert_extent_backref(trans, root->fs_info->extent_root,
1420 path, bytenr, parent, root_objectid,
1421 owner, offset, 1);
1422 if (ret)
1423 err = ret;
1424 out:
1425 btrfs_free_path(path);
1426 finish_current_insert(trans, root->fs_info->extent_root);
1427 del_pending_extents(trans, root->fs_info->extent_root);
1428 BUG_ON(err);
1429 return err;
1430 }
1431
1432 int btrfs_extent_post_op(struct btrfs_trans_handle *trans,
1433 struct btrfs_root *root)
1434 {
1435 finish_current_insert(trans, root->fs_info->extent_root);
1436 del_pending_extents(trans, root->fs_info->extent_root);
1437 return 0;
1438 }
1439
1440 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
1441 struct btrfs_root *root, u64 bytenr,
1442 u64 offset, int metadata, u64 *refs, u64 *flags)
1443 {
1444 struct btrfs_path *path;
1445 int ret;
1446 struct btrfs_key key;
1447 struct extent_buffer *l;
1448 struct btrfs_extent_item *item;
1449 u32 item_size;
1450 u64 num_refs;
1451 u64 extent_flags;
1452
1453 if (metadata &&
1454 !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
1455 offset = root->fs_info->nodesize;
1456 metadata = 0;
1457 }
1458
1459 path = btrfs_alloc_path();
1460 if (!path)
1461 return -ENOMEM;
1462 path->reada = 1;
1463
1464 key.objectid = bytenr;
1465 key.offset = offset;
1466 if (metadata)
1467 key.type = BTRFS_METADATA_ITEM_KEY;
1468 else
1469 key.type = BTRFS_EXTENT_ITEM_KEY;
1470
1471 again:
1472 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1473 0, 0);
1474 if (ret < 0)
1475 goto out;
1476
1477 /*
1478 * Deal with the fact that we may have mixed SKINNY and normal refs. If
1479 * we didn't find what we wanted check and see if we have a normal ref
1480 * right next to us, or re-search if we are on the edge of the leaf just
1481 * to make sure.
1482 */
1483 if (ret > 0 && metadata) {
1484 if (path->slots[0]) {
1485 path->slots[0]--;
1486 btrfs_item_key_to_cpu(path->nodes[0], &key,
1487 path->slots[0]);
1488 if (key.objectid == bytenr &&
1489 key.type == BTRFS_EXTENT_ITEM_KEY &&
1490 key.offset == root->fs_info->nodesize)
1491 ret = 0;
1492 }
1493
1494 if (ret) {
1495 btrfs_release_path(path);
1496 key.type = BTRFS_EXTENT_ITEM_KEY;
1497 key.offset = root->fs_info->nodesize;
1498 metadata = 0;
1499 goto again;
1500 }
1501 }
1502
1503 if (ret != 0) {
1504 ret = -EIO;
1505 goto out;
1506 }
1507
1508 l = path->nodes[0];
1509 item_size = btrfs_item_size_nr(l, path->slots[0]);
1510 if (item_size >= sizeof(*item)) {
1511 item = btrfs_item_ptr(l, path->slots[0],
1512 struct btrfs_extent_item);
1513 num_refs = btrfs_extent_refs(l, item);
1514 extent_flags = btrfs_extent_flags(l, item);
1515 } else {
1516 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1517 struct btrfs_extent_item_v0 *ei0;
1518 BUG_ON(item_size != sizeof(*ei0));
1519 ei0 = btrfs_item_ptr(l, path->slots[0],
1520 struct btrfs_extent_item_v0);
1521 num_refs = btrfs_extent_refs_v0(l, ei0);
1522 /* FIXME: this isn't correct for data */
1523 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
1524 #else
1525 BUG();
1526 #endif
1527 }
1528 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1529 if (refs)
1530 *refs = num_refs;
1531 if (flags)
1532 *flags = extent_flags;
1533 out:
1534 btrfs_free_path(path);
1535 return ret;
1536 }
1537
1538 int btrfs_set_block_flags(struct btrfs_trans_handle *trans,
1539 struct btrfs_root *root,
1540 u64 bytenr, int level, u64 flags)
1541 {
1542 struct btrfs_path *path;
1543 int ret;
1544 struct btrfs_key key;
1545 struct extent_buffer *l;
1546 struct btrfs_extent_item *item;
1547 u32 item_size;
1548 int skinny_metadata =
1549 btrfs_fs_incompat(root->fs_info, SKINNY_METADATA);
1550
1551 path = btrfs_alloc_path();
1552 if (!path)
1553 return -ENOMEM;
1554 path->reada = 1;
1555
1556 key.objectid = bytenr;
1557 if (skinny_metadata) {
1558 key.offset = level;
1559 key.type = BTRFS_METADATA_ITEM_KEY;
1560 } else {
1561 key.offset = root->fs_info->nodesize;
1562 key.type = BTRFS_EXTENT_ITEM_KEY;
1563 }
1564
1565 again:
1566 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
1567 0, 0);
1568 if (ret < 0)
1569 goto out;
1570
1571 if (ret > 0 && skinny_metadata) {
1572 skinny_metadata = 0;
1573 if (path->slots[0]) {
1574 path->slots[0]--;
1575 btrfs_item_key_to_cpu(path->nodes[0], &key,
1576 path->slots[0]);
1577 if (key.objectid == bytenr &&
1578 key.offset == root->fs_info->nodesize &&
1579 key.type == BTRFS_EXTENT_ITEM_KEY)
1580 ret = 0;
1581 }
1582 if (ret) {
1583 btrfs_release_path(path);
1584 key.offset = root->fs_info->nodesize;
1585 key.type = BTRFS_EXTENT_ITEM_KEY;
1586 goto again;
1587 }
1588 }
1589
1590 if (ret != 0) {
1591 btrfs_print_leaf(root, path->nodes[0]);
1592 printk("failed to find block number %Lu\n",
1593 (unsigned long long)bytenr);
1594 BUG();
1595 }
1596 l = path->nodes[0];
1597 item_size = btrfs_item_size_nr(l, path->slots[0]);
1598 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1599 if (item_size < sizeof(*item)) {
1600 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
1601 path, (u64)-1, 0);
1602 if (ret < 0)
1603 goto out;
1604
1605 l = path->nodes[0];
1606 item_size = btrfs_item_size_nr(l, path->slots[0]);
1607 }
1608 #endif
1609 BUG_ON(item_size < sizeof(*item));
1610 item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
1611 flags |= btrfs_extent_flags(l, item);
1612 btrfs_set_extent_flags(l, item, flags);
1613 out:
1614 btrfs_free_path(path);
1615 finish_current_insert(trans, root->fs_info->extent_root);
1616 del_pending_extents(trans, root->fs_info->extent_root);
1617 return ret;
1618 }
1619
1620 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
1621 struct btrfs_root *root,
1622 struct extent_buffer *buf,
1623 int record_parent, int inc)
1624 {
1625 u64 bytenr;
1626 u64 num_bytes;
1627 u64 parent;
1628 u64 ref_root;
1629 u32 nritems;
1630 struct btrfs_key key;
1631 struct btrfs_file_extent_item *fi;
1632 int i;
1633 int level;
1634 int ret = 0;
1635 int (*process_func)(struct btrfs_trans_handle *trans,
1636 struct btrfs_root *root,
1637 u64, u64, u64, u64, u64, u64);
1638
1639 ref_root = btrfs_header_owner(buf);
1640 nritems = btrfs_header_nritems(buf);
1641 level = btrfs_header_level(buf);
1642
1643 if (!root->ref_cows && level == 0)
1644 return 0;
1645
1646 if (inc)
1647 process_func = btrfs_inc_extent_ref;
1648 else
1649 process_func = btrfs_free_extent;
1650
1651 if (record_parent)
1652 parent = buf->start;
1653 else
1654 parent = 0;
1655
1656 for (i = 0; i < nritems; i++) {
1657 cond_resched();
1658 if (level == 0) {
1659 btrfs_item_key_to_cpu(buf, &key, i);
1660 if (key.type != BTRFS_EXTENT_DATA_KEY)
1661 continue;
1662 fi = btrfs_item_ptr(buf, i,
1663 struct btrfs_file_extent_item);
1664 if (btrfs_file_extent_type(buf, fi) ==
1665 BTRFS_FILE_EXTENT_INLINE)
1666 continue;
1667 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
1668 if (bytenr == 0)
1669 continue;
1670
1671 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
1672 key.offset -= btrfs_file_extent_offset(buf, fi);
1673 ret = process_func(trans, root, bytenr, num_bytes,
1674 parent, ref_root, key.objectid,
1675 key.offset);
1676 if (ret) {
1677 WARN_ON(1);
1678 goto fail;
1679 }
1680 } else {
1681 bytenr = btrfs_node_blockptr(buf, i);
1682 num_bytes = root->fs_info->nodesize;
1683 ret = process_func(trans, root, bytenr, num_bytes,
1684 parent, ref_root, level - 1, 0);
1685 if (ret) {
1686 WARN_ON(1);
1687 goto fail;
1688 }
1689 }
1690 }
1691 return 0;
1692 fail:
1693 WARN_ON(1);
1694 return ret;
1695 }
1696
1697 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1698 struct extent_buffer *buf, int record_parent)
1699 {
1700 return __btrfs_mod_ref(trans, root, buf, record_parent, 1);
1701 }
1702
1703 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1704 struct extent_buffer *buf, int record_parent)
1705 {
1706 return __btrfs_mod_ref(trans, root, buf, record_parent, 0);
1707 }
1708
1709 static int write_one_cache_group(struct btrfs_trans_handle *trans,
1710 struct btrfs_root *root,
1711 struct btrfs_path *path,
1712 struct btrfs_block_group_cache *cache)
1713 {
1714 int ret;
1715 int pending_ret;
1716 struct btrfs_root *extent_root = root->fs_info->extent_root;
1717 unsigned long bi;
1718 struct extent_buffer *leaf;
1719
1720 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
1721 if (ret < 0)
1722 goto fail;
1723 BUG_ON(ret);
1724
1725 leaf = path->nodes[0];
1726 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
1727 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
1728 btrfs_mark_buffer_dirty(leaf);
1729 btrfs_release_path(path);
1730 fail:
1731 finish_current_insert(trans, extent_root);
1732 pending_ret = del_pending_extents(trans, extent_root);
1733 if (ret)
1734 return ret;
1735 if (pending_ret)
1736 return pending_ret;
1737 return 0;
1738
1739 }
1740
1741 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
1742 struct btrfs_root *root)
1743 {
1744 struct extent_io_tree *block_group_cache;
1745 struct btrfs_block_group_cache *cache;
1746 int ret;
1747 struct btrfs_path *path;
1748 u64 last = 0;
1749 u64 start;
1750 u64 end;
1751 u64 ptr;
1752
1753 block_group_cache = &root->fs_info->block_group_cache;
1754 path = btrfs_alloc_path();
1755 if (!path)
1756 return -ENOMEM;
1757
1758 while(1) {
1759 ret = find_first_extent_bit(block_group_cache, last,
1760 &start, &end, BLOCK_GROUP_DIRTY);
1761 if (ret) {
1762 if (last == 0)
1763 break;
1764 last = 0;
1765 continue;
1766 }
1767
1768 last = end + 1;
1769 ret = get_state_private(block_group_cache, start, &ptr);
1770 BUG_ON(ret);
1771
1772 clear_extent_bits(block_group_cache, start, end,
1773 BLOCK_GROUP_DIRTY);
1774
1775 cache = (struct btrfs_block_group_cache *)(unsigned long)ptr;
1776 ret = write_one_cache_group(trans, root, path, cache);
1777 }
1778 btrfs_free_path(path);
1779 return 0;
1780 }
1781
1782 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
1783 u64 flags)
1784 {
1785 struct btrfs_space_info *found;
1786
1787 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
1788
1789 list_for_each_entry(found, &info->space_info, list) {
1790 if (found->flags & flags)
1791 return found;
1792 }
1793 return NULL;
1794
1795 }
1796
1797 static int free_space_info(struct btrfs_fs_info *fs_info, u64 flags,
1798 u64 total_bytes, u64 bytes_used,
1799 struct btrfs_space_info **space_info)
1800 {
1801 struct btrfs_space_info *found;
1802
1803 /* only support free block group which is empty */
1804 if (bytes_used)
1805 return -ENOTEMPTY;
1806
1807 found = __find_space_info(fs_info, flags);
1808 if (!found)
1809 return -ENOENT;
1810 if (found->total_bytes < total_bytes) {
1811 fprintf(stderr,
1812 "WARNING: bad space info to free %llu only have %llu\n",
1813 total_bytes, found->total_bytes);
1814 return -EINVAL;
1815 }
1816 found->total_bytes -= total_bytes;
1817 if (space_info)
1818 *space_info = found;
1819 return 0;
1820 }
1821
1822 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
1823 u64 total_bytes, u64 bytes_used,
1824 struct btrfs_space_info **space_info)
1825 {
1826 struct btrfs_space_info *found;
1827
1828 found = __find_space_info(info, flags);
1829 if (found) {
1830 found->total_bytes += total_bytes;
1831 found->bytes_used += bytes_used;
1832 if (found->total_bytes < found->bytes_used) {
1833 fprintf(stderr, "warning, bad space info total_bytes "
1834 "%llu used %llu\n",
1835 (unsigned long long)found->total_bytes,
1836 (unsigned long long)found->bytes_used);
1837 }
1838 *space_info = found;
1839 return 0;
1840 }
1841 found = kmalloc(sizeof(*found), GFP_NOFS);
1842 if (!found)
1843 return -ENOMEM;
1844
1845 list_add(&found->list, &info->space_info);
1846 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
1847 found->total_bytes = total_bytes;
1848 found->bytes_used = bytes_used;
1849 found->bytes_pinned = 0;
1850 found->full = 0;
1851 *space_info = found;
1852 return 0;
1853 }
1854
1855
1856 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
1857 {
1858 u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 |
1859 BTRFS_BLOCK_GROUP_RAID1 |
1860 BTRFS_BLOCK_GROUP_RAID10 |
1861 BTRFS_BLOCK_GROUP_RAID5 |
1862 BTRFS_BLOCK_GROUP_RAID6 |
1863 BTRFS_BLOCK_GROUP_DUP);
1864 if (extra_flags) {
1865 if (flags & BTRFS_BLOCK_GROUP_DATA)
1866 fs_info->avail_data_alloc_bits |= extra_flags;
1867 if (flags & BTRFS_BLOCK_GROUP_METADATA)
1868 fs_info->avail_metadata_alloc_bits |= extra_flags;
1869 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
1870 fs_info->avail_system_alloc_bits |= extra_flags;
1871 }
1872 }
1873
1874 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
1875 struct btrfs_fs_info *fs_info, u64 alloc_bytes,
1876 u64 flags)
1877 {
1878 struct btrfs_space_info *space_info;
1879 u64 thresh;
1880 u64 start;
1881 u64 num_bytes;
1882 int ret;
1883
1884 space_info = __find_space_info(fs_info, flags);
1885 if (!space_info) {
1886 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
1887 BUG_ON(ret);
1888 }
1889 BUG_ON(!space_info);
1890
1891 if (space_info->full)
1892 return 0;
1893
1894 thresh = div_factor(space_info->total_bytes, 7);
1895 if ((space_info->bytes_used + space_info->bytes_pinned + alloc_bytes) <
1896 thresh)
1897 return 0;
1898
1899 /*
1900 * Avoid allocating given chunk type
1901 */
1902 if (fs_info->avoid_meta_chunk_alloc &&
1903 (flags & BTRFS_BLOCK_GROUP_METADATA))
1904 return 0;
1905 if (fs_info->avoid_sys_chunk_alloc &&
1906 (flags & BTRFS_BLOCK_GROUP_SYSTEM))
1907 return 0;
1908
1909 ret = btrfs_alloc_chunk(trans, fs_info, &start, &num_bytes,
1910 space_info->flags);
1911 if (ret == -ENOSPC) {
1912 space_info->full = 1;
1913 return 0;
1914 }
1915
1916 BUG_ON(ret);
1917
1918 ret = btrfs_make_block_group(trans, fs_info, 0, space_info->flags,
1919 BTRFS_FIRST_CHUNK_TREE_OBJECTID, start, num_bytes);
1920 BUG_ON(ret);
1921 return 0;
1922 }
1923
1924 static int update_block_group(struct btrfs_trans_handle *trans,
1925 struct btrfs_root *root,
1926 u64 bytenr, u64 num_bytes, int alloc,
1927 int mark_free)
1928 {
1929 struct btrfs_block_group_cache *cache;
1930 struct btrfs_fs_info *info = root->fs_info;
1931 u64 total = num_bytes;
1932 u64 old_val;
1933 u64 byte_in_group;
1934 u64 start;
1935 u64 end;
1936
1937 /* block accounting for super block */
1938 old_val = btrfs_super_bytes_used(info->super_copy);
1939 if (alloc)
1940 old_val += num_bytes;
1941 else
1942 old_val -= num_bytes;
1943 btrfs_set_super_bytes_used(info->super_copy, old_val);
1944
1945 /* block accounting for root item */
1946 old_val = btrfs_root_used(&root->root_item);
1947 if (alloc)
1948 old_val += num_bytes;
1949 else
1950 old_val -= num_bytes;
1951 btrfs_set_root_used(&root->root_item, old_val);
1952
1953 while(total) {
1954 cache = btrfs_lookup_block_group(info, bytenr);
1955 if (!cache) {
1956 return -1;
1957 }
1958 byte_in_group = bytenr - cache->key.objectid;
1959 WARN_ON(byte_in_group > cache->key.offset);
1960 start = cache->key.objectid;
1961 end = start + cache->key.offset - 1;
1962 set_extent_bits(&info->block_group_cache, start, end,
1963 BLOCK_GROUP_DIRTY);
1964
1965 old_val = btrfs_block_group_used(&cache->item);
1966 num_bytes = min(total, cache->key.offset - byte_in_group);
1967
1968 if (alloc) {
1969 old_val += num_bytes;
1970 cache->space_info->bytes_used += num_bytes;
1971 } else {
1972 old_val -= num_bytes;
1973 cache->space_info->bytes_used -= num_bytes;
1974 if (mark_free) {
1975 set_extent_dirty(&info->free_space_cache,
1976 bytenr, bytenr + num_bytes - 1);
1977 }
1978 }
1979 btrfs_set_block_group_used(&cache->item, old_val);
1980 total -= num_bytes;
1981 bytenr += num_bytes;
1982 }
1983 return 0;
1984 }
1985
1986 static int update_pinned_extents(struct btrfs_root *root,
1987 u64 bytenr, u64 num, int pin)
1988 {
1989 u64 len;
1990 struct btrfs_block_group_cache *cache;
1991 struct btrfs_fs_info *fs_info = root->fs_info;
1992
1993 if (pin) {
1994 set_extent_dirty(&fs_info->pinned_extents,
1995 bytenr, bytenr + num - 1);
1996 } else {
1997 clear_extent_dirty(&fs_info->pinned_extents,
1998 bytenr, bytenr + num - 1);
1999 }
2000 while (num > 0) {
2001 cache = btrfs_lookup_block_group(fs_info, bytenr);
2002 if (!cache) {
2003 len = min((u64)fs_info->sectorsize, num);
2004 goto next;
2005 }
2006 WARN_ON(!cache);
2007 len = min(num, cache->key.offset -
2008 (bytenr - cache->key.objectid));
2009 if (pin) {
2010 cache->pinned += len;
2011 cache->space_info->bytes_pinned += len;
2012 fs_info->total_pinned += len;
2013 } else {
2014 cache->pinned -= len;
2015 cache->space_info->bytes_pinned -= len;
2016 fs_info->total_pinned -= len;
2017 }
2018 next:
2019 bytenr += len;
2020 num -= len;
2021 }
2022 return 0;
2023 }
2024
2025 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
2026 struct btrfs_root *root,
2027 struct extent_io_tree *unpin)
2028 {
2029 u64 start;
2030 u64 end;
2031 int ret;
2032 struct extent_io_tree *free_space_cache;
2033 free_space_cache = &root->fs_info->free_space_cache;
2034
2035 while(1) {
2036 ret = find_first_extent_bit(unpin, 0, &start, &end,
2037 EXTENT_DIRTY);
2038 if (ret)
2039 break;
2040 update_pinned_extents(root, start, end + 1 - start, 0);
2041 clear_extent_dirty(unpin, start, end);
2042 set_extent_dirty(free_space_cache, start, end);
2043 }
2044 return 0;
2045 }
2046
2047 static int extent_root_pending_ops(struct btrfs_fs_info *info)
2048 {
2049 u64 start;
2050 u64 end;
2051 int ret;
2052
2053 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2054 &end, EXTENT_LOCKED);
2055 if (!ret) {
2056 ret = find_first_extent_bit(&info->pending_del, 0, &start, &end,
2057 EXTENT_LOCKED);
2058 }
2059 return ret == 0;
2060
2061 }
2062 static int finish_current_insert(struct btrfs_trans_handle *trans,
2063 struct btrfs_root *extent_root)
2064 {
2065 u64 start;
2066 u64 end;
2067 u64 priv;
2068 struct btrfs_fs_info *info = extent_root->fs_info;
2069 struct pending_extent_op *extent_op;
2070 struct btrfs_key key;
2071 int ret;
2072 int skinny_metadata =
2073 btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
2074
2075 while(1) {
2076 ret = find_first_extent_bit(&info->extent_ins, 0, &start,
2077 &end, EXTENT_LOCKED);
2078 if (ret)
2079 break;
2080
2081 ret = get_state_private(&info->extent_ins, start, &priv);
2082 BUG_ON(ret);
2083 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2084
2085 if (extent_op->type == PENDING_EXTENT_INSERT) {
2086 key.objectid = start;
2087 if (skinny_metadata) {
2088 key.offset = extent_op->level;
2089 key.type = BTRFS_METADATA_ITEM_KEY;
2090 } else {
2091 key.offset = extent_op->num_bytes;
2092 key.type = BTRFS_EXTENT_ITEM_KEY;
2093 }
2094 ret = alloc_reserved_tree_block(trans, extent_root,
2095 extent_root->root_key.objectid,
2096 trans->transid,
2097 extent_op->flags,
2098 &extent_op->key,
2099 extent_op->level, &key);
2100 BUG_ON(ret);
2101 } else {
2102 BUG_ON(1);
2103 }
2104
2105 clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED);
2106 kfree(extent_op);
2107 }
2108 return 0;
2109 }
2110
2111 static int pin_down_bytes(struct btrfs_trans_handle *trans,
2112 struct btrfs_root *root,
2113 u64 bytenr, u64 num_bytes, int is_data)
2114 {
2115 int err = 0;
2116 struct extent_buffer *buf;
2117
2118 if (is_data)
2119 goto pinit;
2120
2121 buf = btrfs_find_tree_block(root->fs_info, bytenr, num_bytes);
2122 if (!buf)
2123 goto pinit;
2124
2125 /* we can reuse a block if it hasn't been written
2126 * and it is from this transaction. We can't
2127 * reuse anything from the tree log root because
2128 * it has tiny sub-transactions.
2129 */
2130 if (btrfs_buffer_uptodate(buf, 0)) {
2131 u64 header_owner = btrfs_header_owner(buf);
2132 u64 header_transid = btrfs_header_generation(buf);
2133 if (header_owner != BTRFS_TREE_LOG_OBJECTID &&
2134 header_transid == trans->transid &&
2135 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
2136 clean_tree_block(NULL, root, buf);
2137 free_extent_buffer(buf);
2138 return 1;
2139 }
2140 }
2141 free_extent_buffer(buf);
2142 pinit:
2143 update_pinned_extents(root, bytenr, num_bytes, 1);
2144
2145 BUG_ON(err < 0);
2146 return 0;
2147 }
2148
2149 void btrfs_pin_extent(struct btrfs_fs_info *fs_info,
2150 u64 bytenr, u64 num_bytes)
2151 {
2152 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 1);
2153 }
2154
2155 void btrfs_unpin_extent(struct btrfs_fs_info *fs_info,
2156 u64 bytenr, u64 num_bytes)
2157 {
2158 update_pinned_extents(fs_info->extent_root, bytenr, num_bytes, 0);
2159 }
2160
2161 /*
2162 * remove an extent from the root, returns 0 on success
2163 */
2164 static int __free_extent(struct btrfs_trans_handle *trans,
2165 struct btrfs_root *root,
2166 u64 bytenr, u64 num_bytes, u64 parent,
2167 u64 root_objectid, u64 owner_objectid,
2168 u64 owner_offset, int refs_to_drop)
2169 {
2170
2171 struct btrfs_key key;
2172 struct btrfs_path *path;
2173 struct btrfs_root *extent_root = root->fs_info->extent_root;
2174 struct extent_buffer *leaf;
2175 struct btrfs_extent_item *ei;
2176 struct btrfs_extent_inline_ref *iref;
2177 int ret;
2178 int is_data;
2179 int extent_slot = 0;
2180 int found_extent = 0;
2181 int num_to_del = 1;
2182 u32 item_size;
2183 u64 refs;
2184 int skinny_metadata =
2185 btrfs_fs_incompat(extent_root->fs_info, SKINNY_METADATA);
2186
2187 if (root->fs_info->free_extent_hook) {
2188 root->fs_info->free_extent_hook(trans, root, bytenr, num_bytes,
2189 parent, root_objectid, owner_objectid,
2190 owner_offset, refs_to_drop);
2191
2192 }
2193 path = btrfs_alloc_path();
2194 if (!path)
2195 return -ENOMEM;
2196
2197 path->reada = 1;
2198
2199 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
2200 if (is_data)
2201 skinny_metadata = 0;
2202 BUG_ON(!is_data && refs_to_drop != 1);
2203
2204 ret = lookup_extent_backref(trans, extent_root, path, &iref,
2205 bytenr, num_bytes, parent,
2206 root_objectid, owner_objectid,
2207 owner_offset);
2208 if (ret == 0) {
2209 extent_slot = path->slots[0];
2210 while (extent_slot >= 0) {
2211 btrfs_item_key_to_cpu(path->nodes[0], &key,
2212 extent_slot);
2213 if (key.objectid != bytenr)
2214 break;
2215 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
2216 key.offset == num_bytes) {
2217 found_extent = 1;
2218 break;
2219 }
2220 if (key.type == BTRFS_METADATA_ITEM_KEY &&
2221 key.offset == owner_objectid) {
2222 found_extent = 1;
2223 break;
2224 }
2225 if (path->slots[0] - extent_slot > 5)
2226 break;
2227 extent_slot--;
2228 }
2229 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2230 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
2231 if (found_extent && item_size < sizeof(*ei))
2232 found_extent = 0;
2233 #endif
2234 if (!found_extent) {
2235 BUG_ON(iref);
2236 ret = remove_extent_backref(trans, extent_root, path,
2237 NULL, refs_to_drop,
2238 is_data);
2239 BUG_ON(ret);
2240 btrfs_release_path(path);
2241
2242 key.objectid = bytenr;
2243
2244 if (skinny_metadata) {
2245 key.type = BTRFS_METADATA_ITEM_KEY;
2246 key.offset = owner_objectid;
2247 } else {
2248 key.type = BTRFS_EXTENT_ITEM_KEY;
2249 key.offset = num_bytes;
2250 }
2251
2252 ret = btrfs_search_slot(trans, extent_root,
2253 &key, path, -1, 1);
2254 if (ret > 0 && skinny_metadata && path->slots[0]) {
2255 path->slots[0]--;
2256 btrfs_item_key_to_cpu(path->nodes[0],
2257 &key,
2258 path->slots[0]);
2259 if (key.objectid == bytenr &&
2260 key.type == BTRFS_EXTENT_ITEM_KEY &&
2261 key.offset == num_bytes)
2262 ret = 0;
2263 }
2264
2265 if (ret > 0 && skinny_metadata) {
2266 skinny_metadata = 0;
2267 btrfs_release_path(path);
2268 key.type = BTRFS_EXTENT_ITEM_KEY;
2269 key.offset = num_bytes;
2270 ret = btrfs_search_slot(trans, extent_root,
2271 &key, path, -1, 1);
2272 }
2273
2274 if (ret) {
2275 printk(KERN_ERR "umm, got %d back from search"
2276 ", was looking for %llu\n", ret,
2277 (unsigned long long)bytenr);
2278 btrfs_print_leaf(extent_root, path->nodes[0]);
2279 }
2280 BUG_ON(ret);
2281 extent_slot = path->slots[0];
2282 }
2283 } else {
2284 printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
2285 "parent %llu root %llu owner %llu offset %llu\n",
2286 (unsigned long long)bytenr,
2287 (unsigned long long)parent,
2288 (unsigned long long)root_objectid,
2289 (unsigned long long)owner_objectid,
2290 (unsigned long long)owner_offset);
2291 ret = -EIO;
2292 goto fail;
2293 }
2294
2295 leaf = path->nodes[0];
2296 item_size = btrfs_item_size_nr(leaf, extent_slot);
2297 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2298 if (item_size < sizeof(*ei)) {
2299 BUG_ON(found_extent || extent_slot != path->slots[0]);
2300 ret = convert_extent_item_v0(trans, extent_root, path,
2301 owner_objectid, 0);
2302 BUG_ON(ret < 0);
2303
2304 btrfs_release_path(path);
2305
2306 key.objectid = bytenr;
2307 key.type = BTRFS_EXTENT_ITEM_KEY;
2308 key.offset = num_bytes;
2309
2310 ret = btrfs_search_slot(trans, extent_root, &key, path,
2311 -1, 1);
2312 if (ret) {
2313 printk(KERN_ERR "umm, got %d back from search"
2314 ", was looking for %llu\n", ret,
2315 (unsigned long long)bytenr);
2316 btrfs_print_leaf(extent_root, path->nodes[0]);
2317 }
2318 BUG_ON(ret);
2319 extent_slot = path->slots[0];
2320 leaf = path->nodes[0];
2321 item_size = btrfs_item_size_nr(leaf, extent_slot);
2322 }
2323 #endif
2324 BUG_ON(item_size < sizeof(*ei));
2325 ei = btrfs_item_ptr(leaf, extent_slot,
2326 struct btrfs_extent_item);
2327 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
2328 key.type == BTRFS_EXTENT_ITEM_KEY) {
2329 struct btrfs_tree_block_info *bi;
2330 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
2331 bi = (struct btrfs_tree_block_info *)(ei + 1);
2332 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
2333 }
2334
2335 refs = btrfs_extent_refs(leaf, ei);
2336 BUG_ON(refs < refs_to_drop);
2337 refs -= refs_to_drop;
2338
2339 if (refs > 0) {
2340 /*
2341 * In the case of inline back ref, reference count will
2342 * be updated by remove_extent_backref
2343 */
2344 if (iref) {
2345 BUG_ON(!found_extent);
2346 } else {
2347 btrfs_set_extent_refs(leaf, ei, refs);
2348 btrfs_mark_buffer_dirty(leaf);
2349 }
2350 if (found_extent) {
2351 ret = remove_extent_backref(trans, extent_root, path,
2352 iref, refs_to_drop,
2353 is_data);
2354 BUG_ON(ret);
2355 }
2356 } else {
2357 int mark_free = 0;
2358 int pin = 1;
2359
2360 if (found_extent) {
2361 BUG_ON(is_data && refs_to_drop !=
2362 extent_data_ref_count(path, iref));
2363 if (iref) {
2364 BUG_ON(path->slots[0] != extent_slot);
2365 } else {
2366 BUG_ON(path->slots[0] != extent_slot + 1);
2367 path->slots[0] = extent_slot;
2368 num_to_del = 2;
2369 }
2370 }
2371
2372 if (pin) {
2373 ret = pin_down_bytes(trans, root, bytenr, num_bytes,
2374 is_data);
2375 if (ret > 0)
2376 mark_free = 1;
2377 BUG_ON(ret < 0);
2378 }
2379
2380 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
2381 num_to_del);
2382 BUG_ON(ret);
2383 btrfs_release_path(path);
2384
2385 if (is_data) {
2386 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
2387 BUG_ON(ret);
2388 }
2389
2390 update_block_group(trans, root, bytenr, num_bytes, 0, mark_free);
2391 }
2392 fail:
2393 btrfs_free_path(path);
2394 finish_current_insert(trans, extent_root);
2395 return ret;
2396 }
2397
2398 /*
2399 * find all the blocks marked as pending in the radix tree and remove
2400 * them from the extent map
2401 */
2402 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
2403 btrfs_root *extent_root)
2404 {
2405 int ret;
2406 int err = 0;
2407 u64 start;
2408 u64 end;
2409 u64 priv;
2410 struct extent_io_tree *pending_del;
2411 struct extent_io_tree *extent_ins;
2412 struct pending_extent_op *extent_op;
2413
2414 extent_ins = &extent_root->fs_info->extent_ins;
2415 pending_del = &extent_root->fs_info->pending_del;
2416
2417 while(1) {
2418 ret = find_first_extent_bit(pending_del, 0, &start, &end,
2419 EXTENT_LOCKED);
2420 if (ret)
2421 break;
2422
2423 ret = get_state_private(pending_del, start, &priv);
2424 BUG_ON(ret);
2425 extent_op = (struct pending_extent_op *)(unsigned long)priv;
2426
2427 clear_extent_bits(pending_del, start, end, EXTENT_LOCKED);
2428
2429 if (!test_range_bit(extent_ins, start, end,
2430 EXTENT_LOCKED, 0)) {
2431 ret = __free_extent(trans, extent_root,
2432 start, end + 1 - start, 0,
2433 extent_root->root_key.objectid,
2434 extent_op->level, 0, 1);
2435 kfree(extent_op);
2436 } else {
2437 kfree(extent_op);
2438 ret = get_state_private(extent_ins, start, &priv);
2439 BUG_ON(ret);
2440 extent_op = (struct pending_extent_op *)
2441 (unsigned long)priv;
2442
2443 clear_extent_bits(extent_ins, start, end,
2444 EXTENT_LOCKED);
2445
2446 if (extent_op->type == PENDING_BACKREF_UPDATE)
2447 BUG_ON(1);
2448
2449 kfree(extent_op);
2450 }
2451 if (ret)
2452 err = ret;
2453 }
2454 return err;
2455 }
2456
2457
2458 int btrfs_free_tree_block(struct btrfs_trans_handle *trans,
2459 struct btrfs_root *root,
2460 struct extent_buffer *buf,
2461 u64 parent, int last_ref)
2462 {
2463 return btrfs_free_extent(trans, root, buf->start, buf->len, parent,
2464 root->root_key.objectid,
2465 btrfs_header_level(buf), 0);
2466 }
2467
2468 /*
2469 * remove an extent from the root, returns 0 on success
2470 */
2471
2472 int btrfs_free_extent(struct btrfs_trans_handle *trans,
2473 struct btrfs_root *root,
2474 u64 bytenr, u64 num_bytes, u64 parent,
2475 u64 root_objectid, u64 owner, u64 offset)
2476 {
2477 struct btrfs_root *extent_root = root->fs_info->extent_root;
2478 int pending_ret;
2479 int ret;
2480
2481 WARN_ON(num_bytes < root->fs_info->sectorsize);
2482 if (root == extent_root) {
2483 struct pending_extent_op *extent_op;
2484
2485 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2486 BUG_ON(!extent_op);
2487
2488 extent_op->type = PENDING_EXTENT_DELETE;
2489 extent_op->bytenr = bytenr;
2490 extent_op->num_bytes = num_bytes;
2491 extent_op->level = (int)owner;
2492
2493 set_extent_bits(&root->fs_info->pending_del,
2494 bytenr, bytenr + num_bytes - 1,
2495 EXTENT_LOCKED);
2496 set_state_private(&root->fs_info->pending_del,
2497 bytenr, (unsigned long)extent_op);
2498 return 0;
2499 }
2500 ret = __free_extent(trans, root, bytenr, num_bytes, parent,
2501 root_objectid, owner, offset, 1);
2502 pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
2503 return ret ? ret : pending_ret;
2504 }
2505
2506 static u64 stripe_align(struct btrfs_root *root, u64 val)
2507 {
2508 return round_up(val, (u64)root->fs_info->stripesize);
2509 }
2510
2511 /*
2512 * walks the btree of allocated extents and find a hole of a given size.
2513 * The key ins is changed to record the hole:
2514 * ins->objectid == block start
2515 * ins->flags = BTRFS_EXTENT_ITEM_KEY
2516 * ins->offset == number of blocks
2517 * Any available blocks before search_start are skipped.
2518 */
2519 static int noinline find_free_extent(struct btrfs_trans_handle *trans,
2520 struct btrfs_root *orig_root,
2521 u64 num_bytes, u64 empty_size,
2522 u64 search_start, u64 search_end,
2523 u64 hint_byte, struct btrfs_key *ins,
2524 u64 exclude_start, u64 exclude_nr,
2525 int data)
2526 {
2527 int ret;
2528 u64 orig_search_start = search_start;
2529 struct btrfs_root * root = orig_root->fs_info->extent_root;
2530 struct btrfs_fs_info *info = root->fs_info;
2531 u64 total_needed = num_bytes;
2532 struct btrfs_block_group_cache *block_group;
2533 int full_scan = 0;
2534 int wrapped = 0;
2535
2536 WARN_ON(num_bytes < info->sectorsize);
2537 ins->type = BTRFS_EXTENT_ITEM_KEY;
2538
2539 search_start = stripe_align(root, search_start);
2540
2541 if (hint_byte) {
2542 block_group = btrfs_lookup_first_block_group(info, hint_byte);
2543 if (!block_group)
2544 hint_byte = search_start;
2545 block_group = btrfs_find_block_group(root, block_group,
2546 hint_byte, data, 1);
2547 } else {
2548 block_group = btrfs_find_block_group(root,
2549 trans->block_group,
2550 search_start, data, 1);
2551 }
2552
2553 total_needed += empty_size;
2554
2555 check_failed:
2556 search_start = stripe_align(root, search_start);
2557 if (!block_group) {
2558 block_group = btrfs_lookup_first_block_group(info,
2559 search_start);
2560 if (!block_group)
2561 block_group = btrfs_lookup_first_block_group(info,
2562 orig_search_start);
2563 }
2564 ret = find_search_start(root, &block_group, &search_start,
2565 total_needed, data);
2566 if (ret)
2567 goto new_group;
2568
2569 ins->objectid = search_start;
2570 ins->offset = num_bytes;
2571
2572 if (ins->objectid + num_bytes >
2573 block_group->key.objectid + block_group->key.offset) {
2574 search_start = block_group->key.objectid +
2575 block_group->key.offset;
2576 goto new_group;
2577 }
2578
2579 if (test_range_bit(&info->extent_ins, ins->objectid,
2580 ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) {
2581 search_start = ins->objectid + num_bytes;
2582 goto new_group;
2583 }
2584
2585 if (test_range_bit(&info->pinned_extents, ins->objectid,
2586 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2587 search_start = ins->objectid + num_bytes;
2588 goto new_group;
2589 }
2590
2591 if (info->excluded_extents &&
2592 test_range_bit(info->excluded_extents, ins->objectid,
2593 ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) {
2594 search_start = ins->objectid + num_bytes;
2595 goto new_group;
2596 }
2597
2598 if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start &&
2599 ins->objectid < exclude_start + exclude_nr)) {
2600 search_start = exclude_start + exclude_nr;
2601 goto new_group;
2602 }
2603
2604 if (!(data & BTRFS_BLOCK_GROUP_DATA)) {
2605 if (check_crossing_stripes(info, ins->objectid, num_bytes)) {
2606 struct btrfs_block_group_cache *bg_cache;
2607 u64 bg_offset;
2608
2609 bg_cache = btrfs_lookup_block_group(info, ins->objectid);
2610 if (!bg_cache)
2611 goto no_bg_cache;
2612 bg_offset = ins->objectid - bg_cache->key.objectid;
2613
2614 search_start = round_up(
2615 bg_offset + num_bytes, BTRFS_STRIPE_LEN) +
2616 bg_cache->key.objectid;
2617 goto new_group;
2618 }
2619 no_bg_cache:
2620 block_group = btrfs_lookup_block_group(info, ins->objectid);
2621 if (block_group)
2622 trans->block_group = block_group;
2623 }
2624 ins->offset = num_bytes;
2625 return 0;
2626
2627 new_group:
2628 block_group = btrfs_lookup_first_block_group(info, search_start);
2629 if (!block_group) {
2630 search_start = orig_search_start;
2631 if (full_scan) {
2632 ret = -ENOSPC;
2633 goto error;
2634 }
2635 if (wrapped) {
2636 if (!full_scan)
2637 total_needed -= empty_size;
2638 full_scan = 1;
2639 } else
2640 wrapped = 1;
2641 }
2642 cond_resched();
2643 block_group = btrfs_find_block_group(root, block_group,
2644 search_start, data, 0);
2645 goto check_failed;
2646
2647 error:
2648 return ret;
2649 }
2650
2651 int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
2652 struct btrfs_root *root,
2653 u64 num_bytes, u64 empty_size,
2654 u64 hint_byte, u64 search_end,
2655 struct btrfs_key *ins, int data)
2656 {
2657 int ret;
2658 u64 search_start = 0;
2659 u64 alloc_profile;
2660 struct btrfs_fs_info *info = root->fs_info;
2661
2662 if (data) {
2663 alloc_profile = info->avail_data_alloc_bits &
2664 info->data_alloc_profile;
2665 data = BTRFS_BLOCK_GROUP_DATA | alloc_profile;
2666 } else if (info->system_allocs == 1 || root == info->chunk_root) {
2667 alloc_profile = info->avail_system_alloc_bits &
2668 info->system_alloc_profile;
2669 data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile;
2670 } else {
2671 alloc_profile = info->avail_metadata_alloc_bits &
2672 info->metadata_alloc_profile;
2673 data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
2674 }
2675
2676 if (root->ref_cows) {
2677 if (!(data & BTRFS_BLOCK_GROUP_METADATA)) {
2678 ret = do_chunk_alloc(trans, info,
2679 num_bytes,
2680 BTRFS_BLOCK_GROUP_METADATA);
2681 BUG_ON(ret);
2682 }
2683 ret = do_chunk_alloc(trans, info,
2684 num_bytes + SZ_2M, data);
2685 BUG_ON(ret);
2686 }
2687
2688 WARN_ON(num_bytes < info->sectorsize);
2689 ret = find_free_extent(trans, root, num_bytes, empty_size,
2690 search_start, search_end, hint_byte, ins,
2691 trans->alloc_exclude_start,
2692 trans->alloc_exclude_nr, data);
2693 if (ret < 0)
2694 return ret;
2695 clear_extent_dirty(&info->free_space_cache,
2696 ins->objectid, ins->objectid + ins->offset - 1);
2697 return ret;
2698 }
2699
2700 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
2701 struct btrfs_root *root,
2702 u64 root_objectid, u64 generation,
2703 u64 flags, struct btrfs_disk_key *key,
2704 int level, struct btrfs_key *ins)
2705 {
2706 int ret;
2707 struct btrfs_fs_info *fs_info = root->fs_info;
2708 struct btrfs_extent_item *extent_item;
2709 struct btrfs_tree_block_info *block_info;
2710 struct btrfs_extent_inline_ref *iref;
2711 struct btrfs_path *path;
2712 struct extent_buffer *leaf;
2713 u32 size = sizeof(*extent_item) + sizeof(*iref);
2714 int skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2715
2716 if (!skinny_metadata)
2717 size += sizeof(*block_info);
2718
2719 path = btrfs_alloc_path();
2720 if (!path)
2721 return -ENOMEM;
2722
2723 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
2724 ins, size);
2725 BUG_ON(ret);
2726
2727 leaf = path->nodes[0];
2728 extent_item = btrfs_item_ptr(leaf, path->slots[0],
2729 struct btrfs_extent_item);
2730 btrfs_set_extent_refs(leaf, extent_item, 1);
2731 btrfs_set_extent_generation(leaf, extent_item, generation);
2732 btrfs_set_extent_flags(leaf, extent_item,
2733 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
2734
2735 if (skinny_metadata) {
2736 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
2737 } else {
2738 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
2739 btrfs_set_tree_block_key(leaf, block_info, key);
2740 btrfs_set_tree_block_level(leaf, block_info, level);
2741 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
2742 }
2743
2744 btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_TREE_BLOCK_REF_KEY);
2745 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
2746
2747 btrfs_mark_buffer_dirty(leaf);
2748 btrfs_free_path(path);
2749
2750 ret = update_block_group(trans, root, ins->objectid, fs_info->nodesize,
2751 1, 0);
2752 return ret;
2753 }
2754
2755 static int alloc_tree_block(struct btrfs_trans_handle *trans,
2756 struct btrfs_root *root, u64 num_bytes,
2757 u64 root_objectid, u64 generation,
2758 u64 flags, struct btrfs_disk_key *key,
2759 int level, u64 empty_size, u64 hint_byte,
2760 u64 search_end, struct btrfs_key *ins)
2761 {
2762 int ret;
2763 ret = btrfs_reserve_extent(trans, root, num_bytes, empty_size,
2764 hint_byte, search_end, ins, 0);
2765 BUG_ON(ret);
2766
2767 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID) {
2768 struct pending_extent_op *extent_op;
2769
2770 extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
2771 BUG_ON(!extent_op);
2772
2773 extent_op->type = PENDING_EXTENT_INSERT;
2774 extent_op->bytenr = ins->objectid;
2775 extent_op->num_bytes = ins->offset;
2776 extent_op->level = level;
2777 extent_op->flags = flags;
2778 memcpy(&extent_op->key, key, sizeof(*key));
2779
2780 set_extent_bits(&root->fs_info->extent_ins, ins->objectid,
2781 ins->objectid + ins->offset - 1,
2782 EXTENT_LOCKED);
2783 set_state_private(&root->fs_info->extent_ins,
2784 ins->objectid, (unsigned long)extent_op);
2785 } else {
2786 if (btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
2787 ins->offset = level;
2788 ins->type = BTRFS_METADATA_ITEM_KEY;
2789 }
2790 ret = alloc_reserved_tree_block(trans, root, root_objectid,
2791 generation, flags,
2792 key, level, ins);
2793 finish_current_insert(trans, root->fs_info->extent_root);
2794 del_pending_extents(trans, root->fs_info->extent_root);
2795 }
2796 return ret;
2797 }
2798
2799 /*
2800 * helper function to allocate a block for a given tree
2801 * returns the tree buffer or NULL.
2802 */
2803 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
2804 struct btrfs_root *root,
2805 u32 blocksize, u64 root_objectid,
2806 struct btrfs_disk_key *key, int level,
2807 u64 hint, u64 empty_size)
2808 {
2809 struct btrfs_key ins;
2810 int ret;
2811 struct extent_buffer *buf;
2812
2813 ret = alloc_tree_block(trans, root, blocksize, root_objectid,
2814 trans->transid, 0, key, level,
2815 empty_size, hint, (u64)-1, &ins);
2816 if (ret) {
2817 BUG_ON(ret > 0);
2818 return ERR_PTR(ret);
2819 }
2820
2821 buf = btrfs_find_create_tree_block(root->fs_info, ins.objectid);
2822 if (!buf) {
2823 btrfs_free_extent(trans, root, ins.objectid, ins.offset,
2824 0, root->root_key.objectid, level, 0);
2825 BUG_ON(1);
2826 return ERR_PTR(-ENOMEM);
2827 }
2828 btrfs_set_buffer_uptodate(buf);
2829 trans->blocks_used++;
2830
2831 return buf;
2832 }
2833
2834 #if 0
2835
2836 static int noinline drop_leaf_ref(struct btrfs_trans_handle *trans,
2837 struct btrfs_root *root,
2838 struct extent_buffer *leaf)
2839 {
2840 u64 leaf_owner;
2841 u64 leaf_generation;
2842 struct btrfs_key key;
2843 struct btrfs_file_extent_item *fi;
2844 int i;
2845 int nritems;
2846 int ret;
2847
2848 BUG_ON(!btrfs_is_leaf(leaf));
2849 nritems = btrfs_header_nritems(leaf);
2850 leaf_owner = btrfs_header_owner(leaf);
2851 leaf_generation = btrfs_header_generation(leaf);
2852
2853 for (i = 0; i < nritems; i++) {
2854 u64 disk_bytenr;
2855
2856 btrfs_item_key_to_cpu(leaf, &key, i);
2857 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2858 continue;
2859 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
2860 if (btrfs_file_extent_type(leaf, fi) ==
2861 BTRFS_FILE_EXTENT_INLINE)
2862 continue;
2863 /*
2864 * FIXME make sure to insert a trans record that
2865 * repeats the snapshot del on crash
2866 */
2867 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2868 if (disk_bytenr == 0)
2869 continue;
2870 ret = btrfs_free_extent(trans, root, disk_bytenr,
2871 btrfs_file_extent_disk_num_bytes(leaf, fi),
2872 leaf->start, leaf_owner, leaf_generation,
2873 key.objectid, 0);
2874 BUG_ON(ret);
2875 }
2876 return 0;
2877 }
2878
2879 static void noinline reada_walk_down(struct btrfs_root *root,
2880 struct extent_buffer *node,
2881 int slot)
2882 {
2883 u64 bytenr;
2884 u64 last = 0;
2885 u32 nritems;
2886 u32 refs;
2887 u32 blocksize;
2888 int ret;
2889 int i;
2890 int level;
2891 int skipped = 0;
2892
2893 nritems = btrfs_header_nritems(node);
2894 level = btrfs_header_level(node);
2895 if (level)
2896 return;
2897
2898 for (i = slot; i < nritems && skipped < 32; i++) {
2899 bytenr = btrfs_node_blockptr(node, i);
2900 if (last && ((bytenr > last && bytenr - last > SZ_32K) ||
2901 (last > bytenr && last - bytenr > SZ_32K))) {
2902 skipped++;
2903 continue;
2904 }
2905 blocksize = btrfs_level_size(root, level - 1);
2906 if (i != slot) {
2907 ret = btrfs_lookup_extent_ref(NULL, root, bytenr,
2908 blocksize, &refs);
2909 BUG_ON(ret);
2910 if (refs != 1) {
2911 skipped++;
2912 continue;
2913 }
2914 }
2915 mutex_unlock(&root->fs_info->fs_mutex);
2916 ret = readahead_tree_block(root, bytenr, blocksize,
2917 btrfs_node_ptr_generation(node, i));
2918 last = bytenr + blocksize;
2919 cond_resched();
2920 mutex_lock(&root->fs_info->fs_mutex);
2921 if (ret)
2922 break;
2923 }
2924 }
2925
2926 /*
2927 * helper function for drop_snapshot, this walks down the tree dropping ref
2928 * counts as it goes.
2929 */
2930 static int noinline walk_down_tree(struct btrfs_trans_handle *trans,
2931 struct btrfs_root *root,
2932 struct btrfs_path *path, int *level)
2933 {
2934 u64 root_owner;
2935 u64 root_gen;
2936 u64 bytenr;
2937 u64 ptr_gen;
2938 struct extent_buffer *next;
2939 struct extent_buffer *cur;
2940 struct extent_buffer *parent;
2941 u32 blocksize;
2942 int ret;
2943 u32 refs;
2944
2945 WARN_ON(*level < 0);
2946 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2947 ret = btrfs_lookup_extent_ref(trans, root,
2948 path->nodes[*level]->start,
2949 path->nodes[*level]->len, &refs);
2950 BUG_ON(ret);
2951 if (refs > 1)
2952 goto out;
2953
2954 /*
2955 * walk down to the last node level and free all the leaves
2956 */
2957 while(*level >= 0) {
2958 WARN_ON(*level < 0);
2959 WARN_ON(*level >= BTRFS_MAX_LEVEL);
2960 cur = path->nodes[*level];
2961
2962 if (btrfs_header_level(cur) != *level)
2963 WARN_ON(1);
2964
2965 if (path->slots[*level] >=
2966 btrfs_header_nritems(cur))
2967 break;
2968 if (*level == 0) {
2969 ret = drop_leaf_ref(trans, root, cur);
2970 BUG_ON(ret);
2971 break;
2972 }
2973 bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
2974 ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
2975 blocksize = btrfs_level_size(root, *level - 1);
2976 ret = btrfs_lookup_extent_ref(trans, root, bytenr, blocksize,
2977 &refs);
2978 BUG_ON(ret);
2979 if (refs != 1) {
2980 parent = path->nodes[*level];
2981 root_owner = btrfs_header_owner(parent);
2982 root_gen = btrfs_header_generation(parent);
2983 path->slots[*level]++;
2984 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
2985 parent->start, root_owner,
2986 root_gen, *level - 1, 1);
2987 BUG_ON(ret);
2988 continue;
2989 }
2990 next = btrfs_find_tree_block(root, bytenr, blocksize);
2991 if (!next || !btrfs_buffer_uptodate(next, ptr_gen)) {
2992 free_extent_buffer(next);
2993 reada_walk_down(root, cur, path->slots[*level]);
2994 mutex_unlock(&root->fs_info->fs_mutex);
2995 next = read_tree_block(root, bytenr, blocksize,
2996 ptr_gen);
2997 mutex_lock(&root->fs_info->fs_mutex);
2998 if (!extent_buffer_uptodate(next)) {
2999 if (IS_ERR(next))
3000 ret = PTR_ERR(next);
3001 else
3002 ret = -EIO;
3003 break;
3004 }
3005 }
3006 WARN_ON(*level <= 0);
3007 if (path->nodes[*level-1])
3008 free_extent_buffer(path->nodes[*level-1]);
3009 path->nodes[*level-1] = next;
3010 *level = btrfs_header_level(next);
3011 path->slots[*level] = 0;
3012 }
3013 out:
3014 WARN_ON(*level < 0);
3015 WARN_ON(*level >= BTRFS_MAX_LEVEL);
3016
3017 if (path->nodes[*level] == root->node) {
3018 root_owner = root->root_key.objectid;
3019 parent = path->nodes[*level];
3020 } else {
3021 parent = path->nodes[*level + 1];
3022 root_owner = btrfs_header_owner(parent);
3023 }
3024
3025 root_gen = btrfs_header_generation(parent);
3026 ret = btrfs_free_extent(trans, root, path->nodes[*level]->start,
3027 path->nodes[*level]->len, parent->start,
3028 root_owner, root_gen, *level, 1);
3029 free_extent_buffer(path->nodes[*level]);
3030 path->nodes[*level] = NULL;
3031 *level += 1;
3032 BUG_ON(ret);
3033 return 0;
3034 }
3035
3036 /*
3037 * helper for dropping snapshots. This walks back up the tree in the path
3038 * to find the first node higher up where we haven't yet gone through
3039 * all the slots
3040 */
3041 static int noinline walk_up_tree(struct btrfs_trans_handle *trans,
3042 struct btrfs_root *root,
3043 struct btrfs_path *path, int *level)
3044 {
3045 u64 root_owner;
3046 u64 root_gen;
3047 struct btrfs_root_item *root_item = &root->root_item;
3048 int i;
3049 int slot;
3050 int ret;
3051
3052 for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
3053 slot = path->slots[i];
3054 if (slot < btrfs_header_nritems(path->nodes[i]) - 1) {
3055 struct extent_buffer *node;
3056 struct btrfs_disk_key disk_key;
3057 node = path->nodes[i];
3058 path->slots[i]++;
3059 *level = i;
3060 WARN_ON(*level == 0);
3061 btrfs_node_key(node, &disk_key, path->slots[i]);
3062 memcpy(&root_item->drop_progress,
3063 &disk_key, sizeof(disk_key));
3064 root_item->drop_level = i;
3065 return 0;
3066 } else {
3067 struct extent_buffer *parent;
3068 if (path->nodes[*level] == root->node)
3069 parent = path->nodes[*level];
3070 else
3071 parent = path->nodes[*level + 1];
3072
3073 root_owner = btrfs_header_owner(parent);
3074 root_gen = btrfs_header_generation(parent);
3075 ret = btrfs_free_extent(trans, root,
3076 path->nodes[*level]->start,
3077 path->nodes[*level]->len,
3078 parent->start, root_owner,
3079 root_gen, *level, 1);
3080 BUG_ON(ret);
3081 free_extent_buffer(path->nodes[*level]);
3082 path->nodes[*level] = NULL;
3083 *level = i + 1;
3084 }
3085 }
3086 return 1;
3087 }
3088
3089 #endif
3090
3091 int btrfs_free_block_groups(struct btrfs_fs_info *info)
3092 {
3093 struct btrfs_space_info *sinfo;
3094 struct btrfs_block_group_cache *cache;
3095 u64 start;
3096 u64 end;
3097 u64 ptr;
3098 int ret;
3099
3100 while(1) {
3101 ret = find_first_extent_bit(&info->block_group_cache, 0,
3102 &start, &end, (unsigned int)-1);
3103 if (ret)
3104 break;
3105 ret = get_state_private(&info->block_group_cache, start, &ptr);
3106 if (!ret) {
3107 cache = u64_to_ptr(ptr);
3108 if (cache->free_space_ctl) {
3109 btrfs_remove_free_space_cache(cache);
3110 kfree(cache->free_space_ctl);
3111 }
3112 kfree(cache);
3113 }
3114 clear_extent_bits(&info->block_group_cache, start,
3115 end, (unsigned int)-1);
3116 }
3117 while(1) {
3118 ret = find_first_extent_bit(&info->free_space_cache, 0,
3119 &start, &end, EXTENT_DIRTY);
3120 if (ret)
3121 break;
3122 clear_extent_dirty(&info->free_space_cache, start, end);
3123 }
3124
3125 while (!list_empty(&info->space_info)) {
3126 sinfo = list_entry(info->space_info.next,
3127 struct btrfs_space_info, list);
3128 list_del_init(&sinfo->list);
3129 kfree(sinfo);
3130 }
3131 return 0;
3132 }
3133
3134 static int find_first_block_group(struct btrfs_root *root,
3135 struct btrfs_path *path, struct btrfs_key *key)
3136 {
3137 int ret;
3138 struct btrfs_key found_key;
3139 struct extent_buffer *leaf;
3140 int slot;
3141
3142 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
3143 if (ret < 0)
3144 return ret;
3145 while(1) {
3146 slot = path->slots[0];
3147 leaf = path->nodes[0];
3148 if (slot >= btrfs_header_nritems(leaf)) {
3149 ret = btrfs_next_leaf(root, path);
3150 if (ret == 0)
3151 continue;
3152 if (ret < 0)
3153 goto error;
3154 break;
3155 }
3156 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3157
3158 if (found_key.objectid >= key->objectid &&
3159 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY)
3160 return 0;
3161 path->slots[0]++;
3162 }
3163 ret = -ENOENT;
3164 error:
3165 return ret;
3166 }
3167
3168 static void account_super_bytes(struct btrfs_fs_info *fs_info,
3169 struct btrfs_block_group_cache *cache)
3170 {
3171 u64 bytenr;
3172 u64 *logical;
3173 int stripe_len;
3174 int i, nr, ret;
3175
3176 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
3177 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
3178 cache->bytes_super += stripe_len;
3179 }
3180
3181 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
3182 bytenr = btrfs_sb_offset(i);
3183 ret = btrfs_rmap_block(fs_info,
3184 cache->key.objectid, bytenr,
3185 0, &logical, &nr, &stripe_len);
3186 if (ret)
3187 return;
3188
3189 while (nr--) {
3190 u64 start, len;
3191
3192 if (logical[nr] > cache->key.objectid +
3193 cache->key.offset)
3194 continue;
3195
3196 if (logical[nr] + stripe_len <= cache->key.objectid)
3197 continue;
3198
3199 start = logical[nr];
3200 if (start < cache->key.objectid) {
3201 start = cache->key.objectid;
3202 len = (logical[nr] + stripe_len) - start;
3203 } else {
3204 len = min_t(u64, stripe_len,
3205 cache->key.objectid +
3206 cache->key.offset - start);
3207 }
3208
3209 cache->bytes_super += len;
3210 }
3211
3212 kfree(logical);
3213 }
3214 }
3215
3216 int btrfs_read_block_groups(struct btrfs_root *root)
3217 {
3218 struct btrfs_path *path;
3219 int ret;
3220 int bit;
3221 struct btrfs_block_group_cache *cache;
3222 struct btrfs_fs_info *info = root->fs_info;
3223 struct btrfs_space_info *space_info;
3224 struct extent_io_tree *block_group_cache;
3225 struct btrfs_key key;
3226 struct btrfs_key found_key;
3227 struct extent_buffer *leaf;
3228
3229 block_group_cache = &info->block_group_cache;
3230
3231 root = info->extent_root;
3232 key.objectid = 0;
3233 key.offset = 0;
3234 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3235 path = btrfs_alloc_path();
3236 if (!path)
3237 return -ENOMEM;
3238
3239 while(1) {
3240 ret = find_first_block_group(root, path, &key);
3241 if (ret > 0) {
3242 ret = 0;
3243 goto error;
3244 }
3245 if (ret != 0) {
3246 goto error;
3247 }
3248 leaf = path->nodes[0];
3249 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3250
3251 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3252 if (!cache) {
3253 ret = -ENOMEM;
3254 goto error;
3255 }
3256
3257 read_extent_buffer(leaf, &cache->item,
3258 btrfs_item_ptr_offset(leaf, path->slots[0]),
3259 sizeof(cache->item));
3260 memcpy(&cache->key, &found_key, sizeof(found_key));
3261 cache->cached = 0;
3262 cache->pinned = 0;
3263 key.objectid = found_key.objectid + found_key.offset;
3264 if (found_key.offset == 0)
3265 key.objectid++;
3266 btrfs_release_path(path);
3267
3268 /*
3269 * Skip 0 sized block group, don't insert them into block
3270 * group cache tree, as its length is 0, it won't get
3271 * freed at close_ctree() time.
3272 */
3273 if (found_key.offset == 0) {
3274 free(cache);
3275 continue;
3276 }
3277
3278 cache->flags = btrfs_block_group_flags(&cache->item);
3279 bit = 0;
3280 if (cache->flags & BTRFS_BLOCK_GROUP_DATA) {
3281 bit = BLOCK_GROUP_DATA;
3282 } else if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
3283 bit = BLOCK_GROUP_SYSTEM;
3284 } else if (cache->flags & BTRFS_BLOCK_GROUP_METADATA) {
3285 bit = BLOCK_GROUP_METADATA;
3286 }
3287 set_avail_alloc_bits(info, cache->flags);
3288 if (btrfs_chunk_readonly(info, cache->key.objectid))
3289 cache->ro = 1;
3290
3291 account_super_bytes(info, cache);
3292
3293 ret = update_space_info(info, cache->flags, found_key.offset,
3294 btrfs_block_group_used(&cache->item),
3295 &space_info);
3296 BUG_ON(ret);
3297 cache->space_info = space_info;
3298
3299 /* use EXTENT_LOCKED to prevent merging */
3300 set_extent_bits(block_group_cache, found_key.objectid,
3301 found_key.objectid + found_key.offset - 1,
3302 bit | EXTENT_LOCKED);
3303 set_state_private(block_group_cache, found_key.objectid,
3304 (unsigned long)cache);
3305 }
3306 ret = 0;
3307 error:
3308 btrfs_free_path(path);
3309 return ret;
3310 }
3311
3312 struct btrfs_block_group_cache *
3313 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
3314 u64 chunk_objectid, u64 chunk_offset, u64 size)
3315 {
3316 int ret;
3317 int bit = 0;
3318 struct btrfs_block_group_cache *cache;
3319 struct extent_io_tree *block_group_cache;
3320
3321 block_group_cache = &fs_info->block_group_cache;
3322
3323 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3324 BUG_ON(!cache);
3325 cache->key.objectid = chunk_offset;
3326 cache->key.offset = size;
3327
3328 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3329 btrfs_set_block_group_used(&cache->item, bytes_used);
3330 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
3331 cache->flags = type;
3332 btrfs_set_block_group_flags(&cache->item, type);
3333
3334 account_super_bytes(fs_info, cache);
3335 ret = update_space_info(fs_info, cache->flags, size, bytes_used,
3336 &cache->space_info);
3337 BUG_ON(ret);
3338
3339 bit = block_group_state_bits(type);
3340 ret = set_extent_bits(block_group_cache, chunk_offset,
3341 chunk_offset + size - 1,
3342 bit | EXTENT_LOCKED);
3343 BUG_ON(ret);
3344
3345 ret = set_state_private(block_group_cache, chunk_offset,
3346 (unsigned long)cache);
3347 BUG_ON(ret);
3348 set_avail_alloc_bits(fs_info, type);
3349
3350 return cache;
3351 }
3352
3353 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
3354 struct btrfs_fs_info *fs_info, u64 bytes_used,
3355 u64 type, u64 chunk_objectid, u64 chunk_offset,
3356 u64 size)
3357 {
3358 int ret;
3359 struct btrfs_root *extent_root = fs_info->extent_root;
3360 struct btrfs_block_group_cache *cache;
3361
3362 cache = btrfs_add_block_group(fs_info, bytes_used, type,
3363 chunk_objectid, chunk_offset, size);
3364 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3365 sizeof(cache->item));
3366 BUG_ON(ret);
3367
3368 ret = finish_current_insert(trans, extent_root);
3369 BUG_ON(ret);
3370 ret = del_pending_extents(trans, extent_root);
3371 BUG_ON(ret);
3372
3373 return 0;
3374 }
3375
3376 /*
3377 * This is for converter use only.
3378 *
3379 * In that case, we don't know where are free blocks located.
3380 * Therefore all block group cache entries must be setup properly
3381 * before doing any block allocation.
3382 */
3383 int btrfs_make_block_groups(struct btrfs_trans_handle *trans,
3384 struct btrfs_fs_info *fs_info)
3385 {
3386 u64 total_bytes;
3387 u64 cur_start;
3388 u64 group_type;
3389 u64 group_size;
3390 u64 group_align;
3391 u64 total_data = 0;
3392 u64 total_metadata = 0;
3393 u64 chunk_objectid;
3394 int ret;
3395 int bit;
3396 struct btrfs_root *extent_root = fs_info->extent_root;
3397 struct btrfs_block_group_cache *cache;
3398 struct extent_io_tree *block_group_cache;
3399
3400 block_group_cache = &fs_info->block_group_cache;
3401 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3402 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
3403 group_align = 64 * fs_info->sectorsize;
3404
3405 cur_start = 0;
3406 while (cur_start < total_bytes) {
3407 group_size = total_bytes / 12;
3408 group_size = min_t(u64, group_size, total_bytes - cur_start);
3409 if (cur_start == 0) {
3410 bit = BLOCK_GROUP_SYSTEM;
3411 group_type = BTRFS_BLOCK_GROUP_SYSTEM;
3412 group_size /= 4;
3413 group_size &= ~(group_align - 1);
3414 group_size = max_t(u64, group_size, SZ_8M);
3415 group_size = min_t(u64, group_size, SZ_32M);
3416 } else {
3417 group_size &= ~(group_align - 1);
3418 if (total_data >= total_metadata * 2) {
3419 group_type = BTRFS_BLOCK_GROUP_METADATA;
3420 group_size = min_t(u64, group_size, SZ_1G);
3421 total_metadata += group_size;
3422 } else {
3423 group_type = BTRFS_BLOCK_GROUP_DATA;
3424 group_size = min_t(u64, group_size,
3425 5ULL * SZ_1G);
3426 total_data += group_size;
3427 }
3428 if ((total_bytes - cur_start) * 4 < group_size * 5)
3429 group_size = total_bytes - cur_start;
3430 }
3431
3432 cache = kzalloc(sizeof(*cache), GFP_NOFS);
3433 BUG_ON(!cache);
3434
3435 cache->key.objectid = cur_start;
3436 cache->key.offset = group_size;
3437 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3438
3439 btrfs_set_block_group_used(&cache->item, 0);
3440 btrfs_set_block_group_chunk_objectid(&cache->item,
3441 chunk_objectid);
3442 btrfs_set_block_group_flags(&cache->item, group_type);
3443
3444 cache->flags = group_type;
3445
3446 ret = update_space_info(fs_info, group_type, group_size,
3447 0, &cache->space_info);
3448 BUG_ON(ret);
3449 set_avail_alloc_bits(fs_info, group_type);
3450
3451 set_extent_bits(block_group_cache, cur_start,
3452 cur_start + group_size - 1,
3453 bit | EXTENT_LOCKED);
3454 set_state_private(block_group_cache, cur_start,
3455 (unsigned long)cache);
3456 cur_start += group_size;
3457 }
3458 /* then insert all the items */
3459 cur_start = 0;
3460 while(cur_start < total_bytes) {
3461 cache = btrfs_lookup_block_group(fs_info, cur_start);
3462 BUG_ON(!cache);
3463
3464 ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
3465 sizeof(cache->item));
3466 BUG_ON(ret);
3467
3468 finish_current_insert(trans, extent_root);
3469 ret = del_pending_extents(trans, extent_root);
3470 BUG_ON(ret);
3471
3472 cur_start = cache->key.objectid + cache->key.offset;
3473 }
3474 return 0;
3475 }
3476
3477 int btrfs_update_block_group(struct btrfs_trans_handle *trans,
3478 struct btrfs_root *root,
3479 u64 bytenr, u64 num_bytes, int alloc,
3480 int mark_free)
3481 {
3482 return update_block_group(trans, root, bytenr, num_bytes,
3483 alloc, mark_free);
3484 }
3485
3486 /*
3487 * Just remove a block group item in extent tree
3488 * Caller should ensure the block group is empty and all space is pinned.
3489 * Or new tree block/data may be allocated into it.
3490 */
3491 static int free_block_group_item(struct btrfs_trans_handle *trans,
3492 struct btrfs_fs_info *fs_info,
3493 u64 bytenr, u64 len)
3494 {
3495 struct btrfs_path *path;
3496 struct btrfs_key key;
3497 struct btrfs_root *root = fs_info->extent_root;
3498 int ret = 0;
3499
3500 key.objectid = bytenr;
3501 key.offset = len;
3502 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3503
3504 path = btrfs_alloc_path();
3505 if (!path)
3506 return -ENOMEM;
3507
3508 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3509 if (ret > 0) {
3510 ret = -ENOENT;
3511 goto out;
3512 }
3513 if (ret < 0)
3514 goto out;
3515
3516 ret = btrfs_del_item(trans, root, path);
3517 out:
3518 btrfs_free_path(path);
3519 return ret;
3520 }
3521
3522 static int free_dev_extent_item(struct btrfs_trans_handle *trans,
3523 struct btrfs_fs_info *fs_info,
3524 u64 devid, u64 dev_offset)
3525 {
3526 struct btrfs_root *root = fs_info->dev_root;
3527 struct btrfs_path *path;
3528 struct btrfs_key key;
3529 int ret;
3530
3531 path = btrfs_alloc_path();
3532 if (!path)
3533 return -ENOMEM;
3534
3535 key.objectid = devid;
3536 key.type = BTRFS_DEV_EXTENT_KEY;
3537 key.offset = dev_offset;
3538
3539 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3540 if (ret < 0)
3541 goto out;
3542 if (ret > 0) {
3543 ret = -ENOENT;
3544 goto out;
3545 }
3546
3547 ret = btrfs_del_item(trans, root, path);
3548 out:
3549 btrfs_free_path(path);
3550 return ret;
3551 }
3552
3553 static int free_chunk_dev_extent_items(struct btrfs_trans_handle *trans,
3554 struct btrfs_fs_info *fs_info,
3555 u64 chunk_offset)
3556 {
3557 struct btrfs_chunk *chunk = NULL;
3558 struct btrfs_root *root= fs_info->chunk_root;
3559 struct btrfs_path *path;
3560 struct btrfs_key key;
3561 u16 num_stripes;
3562 int i;
3563 int ret;
3564
3565 path = btrfs_alloc_path();
3566 if (!path)
3567 return -ENOMEM;
3568
3569 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3570 key.type = BTRFS_CHUNK_ITEM_KEY;
3571 key.offset = chunk_offset;
3572
3573 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3574 if (ret < 0)
3575 goto out;
3576 if (ret > 0) {
3577 ret = -ENOENT;
3578 goto out;
3579 }
3580 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3581 struct btrfs_chunk);
3582 num_stripes = btrfs_chunk_num_stripes(path->nodes[0], chunk);
3583 for (i = 0; i < num_stripes; i++) {
3584 ret = free_dev_extent_item(trans, fs_info,
3585 btrfs_stripe_devid_nr(path->nodes[0], chunk, i),
3586 btrfs_stripe_offset_nr(path->nodes[0], chunk, i));
3587 if (ret < 0)
3588 goto out;
3589 }
3590 out:
3591 btrfs_free_path(path);
3592 return ret;
3593 }
3594
3595 static int free_system_chunk_item(struct btrfs_super_block *super,
3596 struct btrfs_key *key)
3597 {
3598 struct btrfs_disk_key *disk_key;
3599 struct btrfs_key cpu_key;
3600 u32 array_size = btrfs_super_sys_array_size(super);
3601 char *ptr = (char *)super->sys_chunk_array;
3602 int cur = 0;
3603 int ret = -ENOENT;
3604
3605 while (cur < btrfs_super_sys_array_size(super)) {
3606 struct btrfs_chunk *chunk;
3607 u32 num_stripes;
3608 u32 chunk_len;
3609
3610 disk_key = (struct btrfs_disk_key *)(ptr + cur);
3611 btrfs_disk_key_to_cpu(&cpu_key, disk_key);
3612 if (cpu_key.type != BTRFS_CHUNK_ITEM_KEY) {
3613 /* just in case */
3614 ret = -EIO;
3615 goto out;
3616 }
3617
3618 chunk = (struct btrfs_chunk *)(ptr + cur + sizeof(*disk_key));
3619 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3620 chunk_len = btrfs_chunk_item_size(num_stripes) +
3621 sizeof(*disk_key);
3622
3623 if (key->objectid == cpu_key.objectid &&
3624 key->offset == cpu_key.offset &&
3625 key->type == cpu_key.type) {
3626 memmove(ptr + cur, ptr + cur + chunk_len,
3627 array_size - cur - chunk_len);
3628 array_size -= chunk_len;
3629 btrfs_set_super_sys_array_size(super, array_size);
3630 ret = 0;
3631 goto out;
3632 }
3633
3634 cur += chunk_len;
3635 }
3636 out:
3637 return ret;
3638 }
3639
3640 static int free_chunk_item(struct btrfs_trans_handle *trans,
3641 struct btrfs_fs_info *fs_info,
3642 u64 bytenr)
3643 {
3644 struct btrfs_path *path;
3645 struct btrfs_key key;
3646 struct btrfs_root *root = fs_info->chunk_root;
3647 struct btrfs_chunk *chunk;
3648 u64 chunk_type;
3649 int ret;
3650
3651 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3652 key.offset = bytenr;
3653 key.type = BTRFS_CHUNK_ITEM_KEY;
3654
3655 path = btrfs_alloc_path();
3656 if (!path)
3657 return -ENOMEM;
3658
3659 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3660 if (ret > 0) {
3661 ret = -ENOENT;
3662 goto out;
3663 }
3664 if (ret < 0)
3665 goto out;
3666 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
3667 struct btrfs_chunk);
3668 chunk_type = btrfs_chunk_type(path->nodes[0], chunk);
3669
3670 ret = btrfs_del_item(trans, root, path);
3671 if (ret < 0)
3672 goto out;
3673
3674 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3675 ret = free_system_chunk_item(fs_info->super_copy, &key);
3676 out:
3677 btrfs_free_path(path);
3678 return ret;
3679 }
3680
3681 static u64 get_dev_extent_len(struct map_lookup *map)
3682 {
3683 int div;
3684
3685 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3686 case 0: /* Single */
3687 case BTRFS_BLOCK_GROUP_DUP:
3688 case BTRFS_BLOCK_GROUP_RAID1:
3689 div = 1;
3690 break;
3691 case BTRFS_BLOCK_GROUP_RAID5:
3692 div = (map->num_stripes - 1);
3693 break;
3694 case BTRFS_BLOCK_GROUP_RAID6:
3695 div = (map->num_stripes - 2);
3696 break;
3697 case BTRFS_BLOCK_GROUP_RAID10:
3698 div = (map->num_stripes / map->sub_stripes);
3699 break;
3700 default:
3701 /* normally, read chunk security hook should handled it */
3702 BUG_ON(1);
3703 }
3704 return map->ce.size / div;
3705 }
3706
3707 /* free block group/chunk related caches */
3708 static int free_block_group_cache(struct btrfs_trans_handle *trans,
3709 struct btrfs_fs_info *fs_info,
3710 u64 bytenr, u64 len)
3711 {
3712 struct btrfs_block_group_cache *cache;
3713 struct cache_extent *ce;
3714 struct map_lookup *map;
3715 int ret;
3716 int i;
3717 u64 flags;
3718
3719 /* Free block group cache first */
3720 cache = btrfs_lookup_block_group(fs_info, bytenr);
3721 if (!cache)
3722 return -ENOENT;
3723 flags = cache->flags;
3724 if (cache->free_space_ctl) {
3725 btrfs_remove_free_space_cache(cache);
3726 kfree(cache->free_space_ctl);
3727 }
3728 clear_extent_bits(&fs_info->block_group_cache, bytenr, bytenr + len - 1,
3729 (unsigned int)-1);
3730 ret = free_space_info(fs_info, flags, len, 0, NULL);
3731 if (ret < 0)
3732 goto out;
3733 kfree(cache);
3734
3735 /* Then free mapping info and dev usage info */
3736 ce = search_cache_extent(&fs_info->mapping_tree.cache_tree, bytenr);
3737 if (!ce || ce->start != bytenr) {
3738 ret = -ENOENT;
3739 goto out;
3740 }
3741 map = container_of(ce, struct map_lookup, ce);
3742 for (i = 0; i < map->num_stripes; i++) {
3743 struct btrfs_device *device;
3744
3745 device = map->stripes[i].dev;
3746 device->bytes_used -= get_dev_extent_len(map);
3747 ret = btrfs_update_device(trans, device);
3748 if (ret < 0)
3749 goto out;
3750 }
3751 remove_cache_extent(&fs_info->mapping_tree.cache_tree, ce);
3752 free(map);
3753 out:
3754 return ret;
3755 }
3756
3757 int btrfs_free_block_group(struct btrfs_trans_handle *trans,
3758 struct btrfs_fs_info *fs_info, u64 bytenr, u64 len)
3759 {
3760 struct btrfs_root *extent_root = fs_info->extent_root;
3761 struct btrfs_path *path;
3762 struct btrfs_block_group_item *bgi;
3763 struct btrfs_key key;
3764 int ret = 0;
3765
3766 path = btrfs_alloc_path();
3767 if (!path)
3768 return -ENOMEM;
3769
3770 key.objectid = bytenr;
3771 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
3772 key.offset = len;
3773
3774 /* Double check the block group to ensure it's empty */
3775 ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 0);
3776 if (ret > 0) {
3777 ret = -ENONET;
3778 goto out;
3779 }
3780 if (ret < 0)
3781 goto out;
3782
3783 bgi = btrfs_item_ptr(path->nodes[0], path->slots[0],
3784 struct btrfs_block_group_item);
3785 if (btrfs_disk_block_group_used(path->nodes[0], bgi)) {
3786 fprintf(stderr,
3787 "WARNING: block group [%llu,%llu) is not empty\n",
3788 bytenr, bytenr + len);
3789 ret = -EINVAL;
3790 goto out;
3791 }
3792 btrfs_release_path(path);
3793
3794 /*
3795 * Now pin all space in the block group, to prevent further transaction
3796 * allocate space from it.
3797 * Every operation needs a transaction must be in the range.
3798 */
3799 btrfs_pin_extent(fs_info, bytenr, len);
3800
3801 /* delete block group item and chunk item */
3802 ret = free_block_group_item(trans, fs_info, bytenr, len);
3803 if (ret < 0) {
3804 fprintf(stderr,
3805 "failed to free block group item for [%llu,%llu)\n",
3806 bytenr, bytenr + len);
3807 btrfs_unpin_extent(fs_info, bytenr, len);
3808 goto out;
3809 }
3810
3811 ret = free_chunk_dev_extent_items(trans, fs_info, bytenr);
3812 if (ret < 0) {
3813 fprintf(stderr,
3814 "failed to dev extents belongs to [%llu,%llu)\n",
3815 bytenr, bytenr + len);
3816 btrfs_unpin_extent(fs_info, bytenr, len);
3817 goto out;
3818 }
3819 ret = free_chunk_item(trans, fs_info, bytenr);
3820 if (ret < 0) {
3821 fprintf(stderr,
3822 "failed to free chunk for [%llu,%llu)\n",
3823 bytenr, bytenr + len);
3824 btrfs_unpin_extent(fs_info, bytenr, len);
3825 goto out;
3826 }
3827
3828 /* Now release the block_group_cache */
3829 ret = free_block_group_cache(trans, fs_info, bytenr, len);
3830 btrfs_unpin_extent(fs_info, bytenr, len);
3831
3832 out:
3833 btrfs_free_path(path);
3834 return ret;
3835 }
3836
3837 /*
3838 * Fixup block accounting. The initial block accounting created by
3839 * make_block_groups isn't accuracy in this case.
3840 */
3841 int btrfs_fix_block_accounting(struct btrfs_trans_handle *trans,
3842 struct btrfs_root *root)
3843 {
3844 int ret = 0;
3845 int slot;
3846 u64 start = 0;
3847 u64 bytes_used = 0;
3848 struct btrfs_path path;
3849 struct btrfs_key key;
3850 struct extent_buffer *leaf;
3851 struct btrfs_block_group_cache *cache;
3852 struct btrfs_fs_info *fs_info = root->fs_info;
3853
3854 root = root->fs_info->extent_root;
3855
3856 while(extent_root_pending_ops(fs_info)) {
3857 ret = finish_current_insert(trans, root);
3858 if (ret)
3859 return ret;
3860 ret = del_pending_extents(trans, root);
3861 if (ret)
3862 return ret;
3863 }
3864
3865 while(1) {
3866 cache = btrfs_lookup_first_block_group(fs_info, start);
3867 if (!cache)
3868 break;
3869 start = cache->key.objectid + cache->key.offset;
3870 btrfs_set_block_group_used(&cache->item, 0);
3871 cache->space_info->bytes_used = 0;
3872 set_extent_bits(&root->fs_info->block_group_cache,
3873 cache->key.objectid,
3874 cache->key.objectid + cache->key.offset -1,
3875 BLOCK_GROUP_DIRTY);
3876 }
3877
3878 btrfs_init_path(&path);
3879 key.offset = 0;
3880 key.objectid = 0;
3881 key.type = BTRFS_EXTENT_ITEM_KEY;
3882 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
3883 &key, &path, 0, 0);
3884 if (ret < 0)
3885 return ret;
3886 while(1) {
3887 leaf = path.nodes[0];
3888 slot = path.slots[0];
3889 if (slot >= btrfs_header_nritems(leaf)) {
3890 ret = btrfs_next_leaf(root, &path);
3891 if (ret < 0)
3892 return ret;
3893 if (ret > 0)
3894 break;
3895 leaf = path.nodes[0];
3896 slot = path.slots[0];
3897 }
3898 btrfs_item_key_to_cpu(leaf, &key, slot);
3899 if (key.type == BTRFS_EXTENT_ITEM_KEY) {
3900 bytes_used += key.offset;
3901 ret = btrfs_update_block_group(trans, root,
3902 key.objectid, key.offset, 1, 0);
3903 BUG_ON(ret);
3904 } else if (key.type == BTRFS_METADATA_ITEM_KEY) {
3905 bytes_used += fs_info->nodesize;
3906 ret = btrfs_update_block_group(trans, root,
3907 key.objectid, fs_info->nodesize, 1, 0);
3908 if (ret)
3909 goto out;
3910 }
3911 path.slots[0]++;
3912 }
3913 btrfs_set_super_bytes_used(root->fs_info->super_copy, bytes_used);
3914 ret = 0;
3915 out:
3916 btrfs_release_path(&path);
3917 return ret;
3918 }
3919
3920 static void __get_extent_size(struct btrfs_root *root, struct btrfs_path *path,
3921 u64 *start, u64 *len)
3922 {
3923 struct btrfs_key key;
3924
3925 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
3926 BUG_ON(!(key.type == BTRFS_EXTENT_ITEM_KEY ||
3927 key.type == BTRFS_METADATA_ITEM_KEY));
3928 *start = key.objectid;
3929 if (key.type == BTRFS_EXTENT_ITEM_KEY)
3930 *len = key.offset;
3931 else
3932 *len = root->fs_info->nodesize;
3933 }
3934
3935 /*
3936 * Find first overlap extent for range [bytenr, bytenr + len)
3937 * Return 0 for found and point path to it.
3938 * Return >0 for not found.
3939 * Return <0 for err
3940 */
3941 int btrfs_search_overlap_extent(struct btrfs_root *root,
3942 struct btrfs_path *path, u64 bytenr, u64 len)
3943 {
3944 struct btrfs_key key;
3945 u64 cur_start;
3946 u64 cur_len;
3947 int ret;
3948
3949 key.objectid = bytenr;
3950 key.type = BTRFS_EXTENT_DATA_KEY;
3951 key.offset = (u64)-1;
3952
3953 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3954 if (ret < 0)
3955 return ret;
3956 BUG_ON(ret == 0);
3957
3958 ret = btrfs_previous_extent_item(root, path, 0);
3959 if (ret < 0)
3960 return ret;
3961 /* no previous, check next extent */
3962 if (ret > 0)
3963 goto next;
3964 __get_extent_size(root, path, &cur_start, &cur_len);
3965 /* Tail overlap */
3966 if (cur_start + cur_len > bytenr)
3967 return 1;
3968
3969 next:
3970 ret = btrfs_next_extent_item(root, path, bytenr + len);
3971 if (ret < 0)
3972 return ret;
3973 /* No next, prev already checked, no overlap */
3974 if (ret > 0)
3975 return 0;
3976 __get_extent_size(root, path, &cur_start, &cur_len);
3977 /* head overlap*/
3978 if (cur_start < bytenr + len)
3979 return 1;
3980 return 0;
3981 }
3982
3983 static int __btrfs_record_file_extent(struct btrfs_trans_handle *trans,
3984 struct btrfs_root *root, u64 objectid,
3985 struct btrfs_inode_item *inode,
3986 u64 file_pos, u64 disk_bytenr,
3987 u64 *ret_num_bytes)
3988 {
3989 int ret;
3990 struct btrfs_fs_info *info = root->fs_info;
3991 struct btrfs_root *extent_root = info->extent_root;
3992 struct extent_buffer *leaf;
3993 struct btrfs_file_extent_item *fi;
3994 struct btrfs_key ins_key;
3995 struct btrfs_path *path;
3996 struct btrfs_extent_item *ei;
3997 u64 nbytes;
3998 u64 extent_num_bytes;
3999 u64 extent_bytenr;
4000 u64 extent_offset;
4001 u64 num_bytes = *ret_num_bytes;
4002
4003 /*
4004 * All supported file system should not use its 0 extent.
4005 * As it's for hole
4006 *
4007 * And hole extent has no size limit, no need to loop.
4008 */
4009 if (disk_bytenr == 0) {
4010 ret = btrfs_insert_file_extent(trans, root, objectid,
4011 file_pos, disk_bytenr,
4012 num_bytes, num_bytes);
4013 return ret;
4014 }
4015 num_bytes = min_t(u64, num_bytes, BTRFS_MAX_EXTENT_SIZE);
4016
4017 path = btrfs_alloc_path();
4018 if (!path)
4019 return -ENOMEM;
4020
4021 /* First to check extent overlap */
4022 ret = btrfs_search_overlap_extent(extent_root, path, disk_bytenr,
4023 num_bytes);
4024 if (ret < 0)
4025 goto fail;
4026 if (ret > 0) {
4027 /* Found overlap */
4028 u64 cur_start;
4029 u64 cur_len;
4030
4031 __get_extent_size(extent_root, path, &cur_start, &cur_len);
4032 /*
4033 * For convert case, this extent should be a subset of
4034 * existing one.
4035 */
4036 BUG_ON(disk_bytenr < cur_start);
4037
4038 extent_bytenr = cur_start;
4039 extent_num_bytes = cur_len;
4040 extent_offset = disk_bytenr - extent_bytenr;
4041 } else {
4042 /* No overlap, create new extent */
4043 btrfs_release_path(path);
4044 ins_key.objectid = disk_bytenr;
4045 ins_key.offset = num_bytes;
4046 ins_key.type = BTRFS_EXTENT_ITEM_KEY;
4047
4048 ret = btrfs_insert_empty_item(trans, extent_root, path,
4049 &ins_key, sizeof(*ei));
4050 if (ret == 0) {
4051 leaf = path->nodes[0];
4052 ei = btrfs_item_ptr(leaf, path->slots[0],
4053 struct btrfs_extent_item);
4054
4055 btrfs_set_extent_refs(leaf, ei, 0);
4056 btrfs_set_extent_generation(leaf, ei, 0);
4057 btrfs_set_extent_flags(leaf, ei,
4058 BTRFS_EXTENT_FLAG_DATA);
4059 btrfs_mark_buffer_dirty(leaf);
4060
4061 ret = btrfs_update_block_group(trans, root, disk_bytenr,
4062 num_bytes, 1, 0);
4063 if (ret)
4064 goto fail;
4065 } else if (ret != -EEXIST) {
4066 goto fail;
4067 }
4068 btrfs_extent_post_op(trans, extent_root);
4069 extent_bytenr = disk_bytenr;
4070 extent_num_bytes = num_bytes;
4071 extent_offset = 0;
4072 }
4073 btrfs_release_path(path);
4074 ins_key.objectid = objectid;
4075 ins_key.offset = file_pos;
4076 ins_key.type = BTRFS_EXTENT_DATA_KEY;
4077 ret = btrfs_insert_empty_item(trans, root, path, &ins_key,
4078 sizeof(*fi));
4079 if (ret)
4080 goto fail;
4081 leaf = path->nodes[0];
4082 fi = btrfs_item_ptr(leaf, path->slots[0],
4083 struct btrfs_file_extent_item);
4084 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
4085 btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
4086 btrfs_set_file_extent_disk_bytenr(leaf, fi, extent_bytenr);
4087 btrfs_set_file_extent_disk_num_bytes(leaf, fi, extent_num_bytes);
4088 btrfs_set_file_extent_offset(leaf, fi, extent_offset);
4089 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
4090 btrfs_set_file_extent_ram_bytes(leaf, fi, extent_num_bytes);
4091 btrfs_set_file_extent_compression(leaf, fi, 0);
4092 btrfs_set_file_extent_encryption(leaf, fi, 0);
4093 btrfs_set_file_extent_other_encoding(leaf, fi, 0);
4094 btrfs_mark_buffer_dirty(leaf);
4095
4096 nbytes = btrfs_stack_inode_nbytes(inode) + num_bytes;
4097 btrfs_set_stack_inode_nbytes(inode, nbytes);
4098 btrfs_release_path(path);
4099
4100 ret = btrfs_inc_extent_ref(trans, root, extent_bytenr, extent_num_bytes,
4101 0, root->root_key.objectid, objectid,
4102 file_pos - extent_offset);
4103 if (ret)
4104 goto fail;
4105 ret = 0;
4106 *ret_num_bytes = min(extent_num_bytes - extent_offset, num_bytes);
4107 fail:
4108 btrfs_free_path(path);
4109 return ret;
4110 }
4111
4112 /*
4113 * Record a file extent. Do all the required works, such as inserting
4114 * file extent item, inserting extent item and backref item into extent
4115 * tree and updating block accounting.
4116 */
4117 int btrfs_record_file_extent(struct btrfs_trans_handle *trans,
4118 struct btrfs_root *root, u64 objectid,
4119 struct btrfs_inode_item *inode,
4120 u64 file_pos, u64 disk_bytenr,
4121 u64 num_bytes)
4122 {
4123 u64 cur_disk_bytenr = disk_bytenr;
4124 u64 cur_file_pos = file_pos;
4125 u64 cur_num_bytes = num_bytes;
4126 int ret = 0;
4127
4128 while (num_bytes > 0) {
4129 ret = __btrfs_record_file_extent(trans, root, objectid,
4130 inode, cur_file_pos,
4131 cur_disk_bytenr,
4132 &cur_num_bytes);
4133 if (ret < 0)
4134 break;
4135 cur_disk_bytenr += cur_num_bytes;
4136 cur_file_pos += cur_num_bytes;
4137 num_bytes -= cur_num_bytes;
4138 }
4139 return ret;
4140 }
4141
4142
4143 static int add_excluded_extent(struct btrfs_root *root,
4144 u64 start, u64 num_bytes)
4145 {
4146 u64 end = start + num_bytes - 1;
4147 set_extent_bits(&root->fs_info->pinned_extents,
4148 start, end, EXTENT_UPTODATE);
4149 return 0;
4150 }
4151
4152 void free_excluded_extents(struct btrfs_root *root,
4153 struct btrfs_block_group_cache *cache)
4154 {
4155 u64 start, end;
4156
4157 start = cache->key.objectid;
4158 end = start + cache->key.offset - 1;
4159
4160 clear_extent_bits(&root->fs_info->pinned_extents,
4161 start, end, EXTENT_UPTODATE);
4162 }
4163
4164 int exclude_super_stripes(struct btrfs_root *root,
4165 struct btrfs_block_group_cache *cache)
4166 {
4167 u64 bytenr;
4168 u64 *logical;
4169 int stripe_len;
4170 int i, nr, ret;
4171
4172 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
4173 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
4174 cache->bytes_super += stripe_len;
4175 ret = add_excluded_extent(root, cache->key.objectid,
4176 stripe_len);
4177 if (ret)
4178 return ret;
4179 }
4180
4181 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
4182 bytenr = btrfs_sb_offset(i);
4183 ret = btrfs_rmap_block(root->fs_info,
4184 cache->key.objectid, bytenr,
4185 0, &logical, &nr, &stripe_len);
4186 if (ret)
4187 return ret;
4188
4189 while (nr--) {
4190 u64 start, len;
4191
4192 if (logical[nr] > cache->key.objectid +
4193 cache->key.offset)
4194 continue;
4195
4196 if (logical[nr] + stripe_len <= cache->key.objectid)
4197 continue;
4198
4199 start = logical[nr];
4200 if (start < cache->key.objectid) {
4201 start = cache->key.objectid;
4202 len = (logical[nr] + stripe_len) - start;
4203 } else {
4204 len = min_t(u64, stripe_len,
4205 cache->key.objectid +
4206 cache->key.offset - start);
4207 }
4208
4209 cache->bytes_super += len;
4210 ret = add_excluded_extent(root, start, len);
4211 if (ret) {
4212 kfree(logical);
4213 return ret;
4214 }
4215 }
4216
4217 kfree(logical);
4218 }
4219 return 0;
4220 }
4221
4222 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
4223 struct btrfs_fs_info *info, u64 start, u64 end)
4224 {
4225 u64 extent_start, extent_end, size, total_added = 0;
4226 int ret;
4227
4228 while (start < end) {
4229 ret = find_first_extent_bit(&info->pinned_extents, start,
4230 &extent_start, &extent_end,
4231 EXTENT_DIRTY | EXTENT_UPTODATE);
4232 if (ret)
4233 break;
4234
4235 if (extent_start <= start) {
4236 start = extent_end + 1;
4237 } else if (extent_start > start && extent_start < end) {
4238 size = extent_start - start;
4239 total_added += size;
4240 ret = btrfs_add_free_space(block_group->free_space_ctl,
4241 start, size);
4242 BUG_ON(ret); /* -ENOMEM or logic error */
4243 start = extent_end + 1;
4244 } else {
4245 break;
4246 }
4247 }
4248
4249 if (start < end) {
4250 size = end - start;
4251 total_added += size;
4252 ret = btrfs_add_free_space(block_group->free_space_ctl, start,
4253 size);
4254 BUG_ON(ret); /* -ENOMEM or logic error */
4255 }
4256
4257 return total_added;
4258 }
(deprecated) Btrfs progs developments and patch integration