?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfsck: add early code to handle corrupted block groups
[btrfs-progs-unstable/devel.git] / ctree.c
blob2d86b1e7b544a56aa39bc50e42aaf51371a598bd
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 #include "ctree.h"
19 #include "disk-io.h"
20 #include "transaction.h"
21 #include "print-tree.h"
22 #include "repair.h"
23
24 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
25 *root, struct btrfs_path *path, int level);
26 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_key *ins_key,
28 struct btrfs_path *path, int data_size, int extend);
29 static int push_node_left(struct btrfs_trans_handle *trans,
30 struct btrfs_root *root, struct extent_buffer *dst,
31 struct extent_buffer *src, int empty);
32 static int balance_node_right(struct btrfs_trans_handle *trans,
33 struct btrfs_root *root,
34 struct extent_buffer *dst_buf,
35 struct extent_buffer *src_buf);
36
37 inline void btrfs_init_path(struct btrfs_path *p)
38 {
39 memset(p, 0, sizeof(*p));
40 }
41
42 struct btrfs_path *btrfs_alloc_path(void)
43 {
44 struct btrfs_path *path;
45 path = kmalloc(sizeof(struct btrfs_path), GFP_NOFS);
46 if (path) {
47 btrfs_init_path(path);
48 path->reada = 0;
49 }
50 return path;
51 }
52
53 void btrfs_free_path(struct btrfs_path *p)
54 {
55 btrfs_release_path(NULL, p);
56 kfree(p);
57 }
58
59 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
60 {
61 int i;
62 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
63 if (!p->nodes[i])
64 continue;
65 free_extent_buffer(p->nodes[i]);
66 }
67 memset(p, 0, sizeof(*p));
68 }
69
70 static void add_root_to_dirty_list(struct btrfs_root *root)
71 {
72 if (root->track_dirty && list_empty(&root->dirty_list)) {
73 list_add(&root->dirty_list,
74 &root->fs_info->dirty_cowonly_roots);
75 }
76 }
77
78 int btrfs_copy_root(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root,
80 struct extent_buffer *buf,
81 struct extent_buffer **cow_ret, u64 new_root_objectid)
82 {
83 struct extent_buffer *cow;
84 int ret = 0;
85 int level;
86 struct btrfs_root *new_root;
87 struct btrfs_disk_key disk_key;
88
89 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
90 if (!new_root)
91 return -ENOMEM;
92
93 memcpy(new_root, root, sizeof(*new_root));
94 new_root->root_key.objectid = new_root_objectid;
95
96 WARN_ON(root->ref_cows && trans->transid !=
97 root->fs_info->running_transaction->transid);
98 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
99
100 level = btrfs_header_level(buf);
101 if (level == 0)
102 btrfs_item_key(buf, &disk_key, 0);
103 else
104 btrfs_node_key(buf, &disk_key, 0);
105 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
106 new_root_objectid, &disk_key,
107 level, buf->start, 0);
108 if (IS_ERR(cow)) {
109 kfree(new_root);
110 return PTR_ERR(cow);
111 }
112
113 copy_extent_buffer(cow, buf, 0, 0, cow->len);
114 btrfs_set_header_bytenr(cow, cow->start);
115 btrfs_set_header_generation(cow, trans->transid);
116 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
117 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
118 BTRFS_HEADER_FLAG_RELOC);
119 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
120 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
121 else
122 btrfs_set_header_owner(cow, new_root_objectid);
123
124 write_extent_buffer(cow, root->fs_info->fsid,
125 (unsigned long)btrfs_header_fsid(cow),
126 BTRFS_FSID_SIZE);
127
128 WARN_ON(btrfs_header_generation(buf) > trans->transid);
129 ret = btrfs_inc_ref(trans, new_root, cow, 0);
130 kfree(new_root);
131
132 if (ret)
133 return ret;
134
135 btrfs_mark_buffer_dirty(cow);
136 *cow_ret = cow;
137 return 0;
138 }
139
140 int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
141 struct btrfs_root *root)
142 {
143 struct extent_buffer *c;
144 struct extent_buffer *old = root->node;
145 int level;
146 struct btrfs_disk_key disk_key = {0,0,0};
147
148 level = 0;
149
150 c = btrfs_alloc_free_block(trans, root,
151 btrfs_level_size(root, 0),
152 root->root_key.objectid,
153 &disk_key, level, 0, 0);
154 if (IS_ERR(c)) {
155 c = old;
156 extent_buffer_get(c);
157 }
158
159 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
160 btrfs_set_header_level(c, level);
161 btrfs_set_header_bytenr(c, c->start);
162 btrfs_set_header_generation(c, trans->transid);
163 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
164 btrfs_set_header_owner(c, root->root_key.objectid);
165
166 write_extent_buffer(c, root->fs_info->fsid,
167 (unsigned long)btrfs_header_fsid(c),
168 BTRFS_FSID_SIZE);
169
170 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
171 (unsigned long)btrfs_header_chunk_tree_uuid(c),
172 BTRFS_UUID_SIZE);
173
174 btrfs_mark_buffer_dirty(c);
175
176 free_extent_buffer(old);
177 root->node = c;
178 add_root_to_dirty_list(root);
179 return 0;
180 }
181
182 /*
183 * check if the tree block can be shared by multiple trees
184 */
185 int btrfs_block_can_be_shared(struct btrfs_root *root,
186 struct extent_buffer *buf)
187 {
188 /*
189 * Tree blocks not in refernece counted trees and tree roots
190 * are never shared. If a block was allocated after the last
191 * snapshot and the block was not allocated by tree relocation,
192 * we know the block is not shared.
193 */
194 if (root->ref_cows &&
195 buf != root->node && buf != root->commit_root &&
196 (btrfs_header_generation(buf) <=
197 btrfs_root_last_snapshot(&root->root_item) ||
198 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
199 return 1;
200 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
201 if (root->ref_cows &&
202 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
203 return 1;
204 #endif
205 return 0;
206 }
207
208 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
209 struct btrfs_root *root,
210 struct extent_buffer *buf,
211 struct extent_buffer *cow)
212 {
213 u64 refs;
214 u64 owner;
215 u64 flags;
216 u64 new_flags = 0;
217 int ret;
218
219 /*
220 * Backrefs update rules:
221 *
222 * Always use full backrefs for extent pointers in tree block
223 * allocated by tree relocation.
224 *
225 * If a shared tree block is no longer referenced by its owner
226 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
227 * use full backrefs for extent pointers in tree block.
228 *
229 * If a tree block is been relocating
230 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
231 * use full backrefs for extent pointers in tree block.
232 * The reason for this is some operations (such as drop tree)
233 * are only allowed for blocks use full backrefs.
234 */
235
236 if (btrfs_block_can_be_shared(root, buf)) {
237 ret = btrfs_lookup_extent_info(trans, root, buf->start,
238 buf->len, &refs, &flags);
239 BUG_ON(ret);
240 BUG_ON(refs == 0);
241 } else {
242 refs = 1;
243 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
244 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
245 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
246 else
247 flags = 0;
248 }
249
250 owner = btrfs_header_owner(buf);
251 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
252 owner == BTRFS_TREE_RELOC_OBJECTID);
253
254 if (refs > 1) {
255 if ((owner == root->root_key.objectid ||
256 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
257 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
258 ret = btrfs_inc_ref(trans, root, buf, 1);
259 BUG_ON(ret);
260
261 if (root->root_key.objectid ==
262 BTRFS_TREE_RELOC_OBJECTID) {
263 ret = btrfs_dec_ref(trans, root, buf, 0);
264 BUG_ON(ret);
265 ret = btrfs_inc_ref(trans, root, cow, 1);
266 BUG_ON(ret);
267 }
268 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
269 } else {
270
271 if (root->root_key.objectid ==
272 BTRFS_TREE_RELOC_OBJECTID)
273 ret = btrfs_inc_ref(trans, root, cow, 1);
274 else
275 ret = btrfs_inc_ref(trans, root, cow, 0);
276 BUG_ON(ret);
277 }
278 if (new_flags != 0) {
279 ret = btrfs_set_block_flags(trans, root, buf->start,
280 buf->len, new_flags);
281 BUG_ON(ret);
282 }
283 } else {
284 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
285 if (root->root_key.objectid ==
286 BTRFS_TREE_RELOC_OBJECTID)
287 ret = btrfs_inc_ref(trans, root, cow, 1);
288 else
289 ret = btrfs_inc_ref(trans, root, cow, 0);
290 BUG_ON(ret);
291 ret = btrfs_dec_ref(trans, root, buf, 1);
292 BUG_ON(ret);
293 }
294 clean_tree_block(trans, root, buf);
295 }
296 return 0;
297 }
298
299 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
300 struct btrfs_root *root,
301 struct extent_buffer *buf,
302 struct extent_buffer *parent, int parent_slot,
303 struct extent_buffer **cow_ret,
304 u64 search_start, u64 empty_size)
305 {
306 struct extent_buffer *cow;
307 struct btrfs_disk_key disk_key;
308 int level;
309
310 WARN_ON(root->ref_cows && trans->transid !=
311 root->fs_info->running_transaction->transid);
312 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
313
314 level = btrfs_header_level(buf);
315
316 if (level == 0)
317 btrfs_item_key(buf, &disk_key, 0);
318 else
319 btrfs_node_key(buf, &disk_key, 0);
320
321 cow = btrfs_alloc_free_block(trans, root, buf->len,
322 root->root_key.objectid, &disk_key,
323 level, search_start, empty_size);
324 if (IS_ERR(cow))
325 return PTR_ERR(cow);
326
327 copy_extent_buffer(cow, buf, 0, 0, cow->len);
328 btrfs_set_header_bytenr(cow, cow->start);
329 btrfs_set_header_generation(cow, trans->transid);
330 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
331 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
332 BTRFS_HEADER_FLAG_RELOC);
333 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
334 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
335 else
336 btrfs_set_header_owner(cow, root->root_key.objectid);
337
338 write_extent_buffer(cow, root->fs_info->fsid,
339 (unsigned long)btrfs_header_fsid(cow),
340 BTRFS_FSID_SIZE);
341
342 WARN_ON(btrfs_header_generation(buf) > trans->transid);
343
344 update_ref_for_cow(trans, root, buf, cow);
345
346 if (buf == root->node) {
347 root->node = cow;
348 extent_buffer_get(cow);
349
350 btrfs_free_extent(trans, root, buf->start, buf->len,
351 0, root->root_key.objectid, level, 0);
352 free_extent_buffer(buf);
353 add_root_to_dirty_list(root);
354 } else {
355 btrfs_set_node_blockptr(parent, parent_slot,
356 cow->start);
357 WARN_ON(trans->transid == 0);
358 btrfs_set_node_ptr_generation(parent, parent_slot,
359 trans->transid);
360 btrfs_mark_buffer_dirty(parent);
361 WARN_ON(btrfs_header_generation(parent) != trans->transid);
362
363 btrfs_free_extent(trans, root, buf->start, buf->len,
364 0, root->root_key.objectid, level, 1);
365 }
366 free_extent_buffer(buf);
367 btrfs_mark_buffer_dirty(cow);
368 *cow_ret = cow;
369 return 0;
370 }
371
372 static inline int should_cow_block(struct btrfs_trans_handle *trans,
373 struct btrfs_root *root,
374 struct extent_buffer *buf)
375 {
376 if (btrfs_header_generation(buf) == trans->transid &&
377 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
378 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
379 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
380 return 0;
381 return 1;
382 }
383
384 int btrfs_cow_block(struct btrfs_trans_handle *trans,
385 struct btrfs_root *root, struct extent_buffer *buf,
386 struct extent_buffer *parent, int parent_slot,
387 struct extent_buffer **cow_ret)
388 {
389 u64 search_start;
390 int ret;
391 /*
392 if (trans->transaction != root->fs_info->running_transaction) {
393 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
394 root->fs_info->running_transaction->transid);
395 WARN_ON(1);
396 }
397 */
398 if (trans->transid != root->fs_info->generation) {
399 printk(KERN_CRIT "trans %llu running %llu\n",
400 (unsigned long long)trans->transid,
401 (unsigned long long)root->fs_info->generation);
402 WARN_ON(1);
403 }
404 if (!should_cow_block(trans, root, buf)) {
405 *cow_ret = buf;
406 return 0;
407 }
408
409 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
410 ret = __btrfs_cow_block(trans, root, buf, parent,
411 parent_slot, cow_ret, search_start, 0);
412 return ret;
413 }
414
415 /*
416 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
417 {
418 if (blocknr < other && other - (blocknr + blocksize) < 32768)
419 return 1;
420 if (blocknr > other && blocknr - (other + blocksize) < 32768)
421 return 1;
422 return 0;
423 }
424 */
425
426 /*
427 * compare two keys in a memcmp fashion
428 */
429 int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
430 {
431 struct btrfs_key k1;
432
433 btrfs_disk_key_to_cpu(&k1, disk);
434
435 if (k1.objectid > k2->objectid)
436 return 1;
437 if (k1.objectid < k2->objectid)
438 return -1;
439 if (k1.type > k2->type)
440 return 1;
441 if (k1.type < k2->type)
442 return -1;
443 if (k1.offset > k2->offset)
444 return 1;
445 if (k1.offset < k2->offset)
446 return -1;
447 return 0;
448 }
449
450
451 #if 0
452 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
453 struct btrfs_root *root, struct extent_buffer *parent,
454 int start_slot, int cache_only, u64 *last_ret,
455 struct btrfs_key *progress)
456 {
457 struct extent_buffer *cur;
458 struct extent_buffer *tmp;
459 u64 blocknr;
460 u64 gen;
461 u64 search_start = *last_ret;
462 u64 last_block = 0;
463 u64 other;
464 u32 parent_nritems;
465 int end_slot;
466 int i;
467 int err = 0;
468 int parent_level;
469 int uptodate;
470 u32 blocksize;
471 int progress_passed = 0;
472 struct btrfs_disk_key disk_key;
473
474 parent_level = btrfs_header_level(parent);
475 if (cache_only && parent_level != 1)
476 return 0;
477
478 if (trans->transaction != root->fs_info->running_transaction) {
479 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
480 root->fs_info->running_transaction->transid);
481 WARN_ON(1);
482 }
483 if (trans->transid != root->fs_info->generation) {
484 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
485 root->fs_info->generation);
486 WARN_ON(1);
487 }
488
489 parent_nritems = btrfs_header_nritems(parent);
490 blocksize = btrfs_level_size(root, parent_level - 1);
491 end_slot = parent_nritems;
492
493 if (parent_nritems == 1)
494 return 0;
495
496 for (i = start_slot; i < end_slot; i++) {
497 int close = 1;
498
499 if (!parent->map_token) {
500 map_extent_buffer(parent,
501 btrfs_node_key_ptr_offset(i),
502 sizeof(struct btrfs_key_ptr),
503 &parent->map_token, &parent->kaddr,
504 &parent->map_start, &parent->map_len,
505 KM_USER1);
506 }
507 btrfs_node_key(parent, &disk_key, i);
508 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
509 continue;
510
511 progress_passed = 1;
512 blocknr = btrfs_node_blockptr(parent, i);
513 gen = btrfs_node_ptr_generation(parent, i);
514 if (last_block == 0)
515 last_block = blocknr;
516
517 if (i > 0) {
518 other = btrfs_node_blockptr(parent, i - 1);
519 close = close_blocks(blocknr, other, blocksize);
520 }
521 if (close && i < end_slot - 2) {
522 other = btrfs_node_blockptr(parent, i + 1);
523 close = close_blocks(blocknr, other, blocksize);
524 }
525 if (close) {
526 last_block = blocknr;
527 continue;
528 }
529 if (parent->map_token) {
530 unmap_extent_buffer(parent, parent->map_token,
531 KM_USER1);
532 parent->map_token = NULL;
533 }
534
535 cur = btrfs_find_tree_block(root, blocknr, blocksize);
536 if (cur)
537 uptodate = btrfs_buffer_uptodate(cur, gen);
538 else
539 uptodate = 0;
540 if (!cur || !uptodate) {
541 if (cache_only) {
542 free_extent_buffer(cur);
543 continue;
544 }
545 if (!cur) {
546 cur = read_tree_block(root, blocknr,
547 blocksize, gen);
548 } else if (!uptodate) {
549 btrfs_read_buffer(cur, gen);
550 }
551 }
552 if (search_start == 0)
553 search_start = last_block;
554
555 err = __btrfs_cow_block(trans, root, cur, parent, i,
556 &tmp, search_start,
557 min(16 * blocksize,
558 (end_slot - i) * blocksize));
559 if (err) {
560 free_extent_buffer(cur);
561 break;
562 }
563 search_start = tmp->start;
564 last_block = tmp->start;
565 *last_ret = search_start;
566 if (parent_level == 1)
567 btrfs_clear_buffer_defrag(tmp);
568 free_extent_buffer(tmp);
569 }
570 if (parent->map_token) {
571 unmap_extent_buffer(parent, parent->map_token,
572 KM_USER1);
573 parent->map_token = NULL;
574 }
575 return err;
576 }
577 #endif
578
579 /*
580 * The leaf data grows from end-to-front in the node.
581 * this returns the address of the start of the last item,
582 * which is the stop of the leaf data stack
583 */
584 static inline unsigned int leaf_data_end(struct btrfs_root *root,
585 struct extent_buffer *leaf)
586 {
587 u32 nr = btrfs_header_nritems(leaf);
588 if (nr == 0)
589 return BTRFS_LEAF_DATA_SIZE(root);
590 return btrfs_item_offset_nr(leaf, nr - 1);
591 }
592
593 int btrfs_check_node(struct btrfs_root *root,
594 struct btrfs_disk_key *parent_key,
595 struct extent_buffer *buf)
596 {
597 int i;
598 struct btrfs_key cpukey;
599 struct btrfs_disk_key key;
600 u32 nritems = btrfs_header_nritems(buf);
601
602 if (nritems == 0 || nritems > BTRFS_NODEPTRS_PER_BLOCK(root))
603 goto fail;
604
605 if (parent_key && parent_key->type) {
606 btrfs_node_key(buf, &key, 0);
607 if (memcmp(parent_key, &key, sizeof(key)))
608 goto fail;
609 }
610 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
611 btrfs_node_key(buf, &key, i);
612 btrfs_node_key_to_cpu(buf, &cpukey, i + 1);
613 if (btrfs_comp_keys(&key, &cpukey) >= 0)
614 goto fail;
615 }
616 return 0;
617 fail:
618 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
619 if (parent_key)
620 btrfs_disk_key_to_cpu(&cpukey, parent_key);
621 else
622 btrfs_node_key_to_cpu(buf, &cpukey, 0);
623 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
624 buf->start, buf->len,
625 btrfs_header_level(buf));
626 }
627 return -EIO;
628 }
629
630 int btrfs_check_leaf(struct btrfs_root *root,
631 struct btrfs_disk_key *parent_key,
632 struct extent_buffer *buf)
633 {
634 int i;
635 struct btrfs_key cpukey;
636 struct btrfs_disk_key key;
637 u32 nritems = btrfs_header_nritems(buf);
638
639 if (btrfs_header_level(buf) != 0) {
640 fprintf(stderr, "leaf is not a leaf %llu\n",
641 (unsigned long long)btrfs_header_bytenr(buf));
642 goto fail;
643 }
644 if (btrfs_leaf_free_space(root, buf) < 0) {
645 fprintf(stderr, "leaf free space incorrect %llu %d\n",
646 (unsigned long long)btrfs_header_bytenr(buf),
647 btrfs_leaf_free_space(root, buf));
648 goto fail;
649 }
650
651 if (nritems == 0)
652 return 0;
653
654 btrfs_item_key(buf, &key, 0);
655 if (parent_key && parent_key->type &&
656 memcmp(parent_key, &key, sizeof(key))) {
657 fprintf(stderr, "leaf parent key incorrect %llu\n",
658 (unsigned long long)btrfs_header_bytenr(buf));
659 goto fail;
660 }
661 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
662 btrfs_item_key(buf, &key, i);
663 btrfs_item_key_to_cpu(buf, &cpukey, i + 1);
664 if (btrfs_comp_keys(&key, &cpukey) >= 0) {
665 fprintf(stderr, "bad key ordering %d %d\n", i, i+1);
666 goto fail;
667 }
668 if (btrfs_item_offset_nr(buf, i) !=
669 btrfs_item_end_nr(buf, i + 1)) {
670 fprintf(stderr, "incorrect offsets %u %u\n",
671 btrfs_item_offset_nr(buf, i),
672 btrfs_item_end_nr(buf, i + 1));
673 goto fail;
674 }
675 if (i == 0 && btrfs_item_end_nr(buf, i) !=
676 BTRFS_LEAF_DATA_SIZE(root)) {
677 fprintf(stderr, "bad item end %u wanted %u\n",
678 btrfs_item_end_nr(buf, i),
679 (unsigned)BTRFS_LEAF_DATA_SIZE(root));
680 goto fail;
681 }
682 }
683 return 0;
684 fail:
685 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
686 if (parent_key)
687 btrfs_disk_key_to_cpu(&cpukey, parent_key);
688 else
689 btrfs_item_key_to_cpu(buf, &cpukey, 0);
690
691 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
692 buf->start, buf->len, 0);
693 }
694 return -EIO;
695 }
696
697 static int noinline check_block(struct btrfs_root *root,
698 struct btrfs_path *path, int level)
699 {
700 struct btrfs_disk_key key;
701 struct btrfs_disk_key *key_ptr = NULL;
702 struct extent_buffer *parent;
703
704 if (path->nodes[level + 1]) {
705 parent = path->nodes[level + 1];
706 btrfs_node_key(parent, &key, path->slots[level + 1]);
707 key_ptr = &key;
708 }
709 if (level == 0)
710 return btrfs_check_leaf(root, key_ptr, path->nodes[0]);
711 return btrfs_check_node(root, key_ptr, path->nodes[level]);
712 }
713
714 /*
715 * search for key in the extent_buffer. The items start at offset p,
716 * and they are item_size apart. There are 'max' items in p.
717 *
718 * the slot in the array is returned via slot, and it points to
719 * the place where you would insert key if it is not found in
720 * the array.
721 *
722 * slot may point to max if the key is bigger than all of the keys
723 */
724 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
725 int item_size, struct btrfs_key *key,
726 int max, int *slot)
727 {
728 int low = 0;
729 int high = max;
730 int mid;
731 int ret;
732 unsigned long offset;
733 struct btrfs_disk_key *tmp;
734
735 while(low < high) {
736 mid = (low + high) / 2;
737 offset = p + mid * item_size;
738
739 tmp = (struct btrfs_disk_key *)(eb->data + offset);
740 ret = btrfs_comp_keys(tmp, key);
741
742 if (ret < 0)
743 low = mid + 1;
744 else if (ret > 0)
745 high = mid;
746 else {
747 *slot = mid;
748 return 0;
749 }
750 }
751 *slot = low;
752 return 1;
753 }
754
755 /*
756 * simple bin_search frontend that does the right thing for
757 * leaves vs nodes
758 */
759 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
760 int level, int *slot)
761 {
762 if (level == 0) {
763 return generic_bin_search(eb,
764 offsetof(struct btrfs_leaf, items),
765 sizeof(struct btrfs_item),
766 key, btrfs_header_nritems(eb),
767 slot);
768 } else {
769 return generic_bin_search(eb,
770 offsetof(struct btrfs_node, ptrs),
771 sizeof(struct btrfs_key_ptr),
772 key, btrfs_header_nritems(eb),
773 slot);
774 }
775 return -1;
776 }
777
778 struct extent_buffer *read_node_slot(struct btrfs_root *root,
779 struct extent_buffer *parent, int slot)
780 {
781 int level = btrfs_header_level(parent);
782 if (slot < 0)
783 return NULL;
784 if (slot >= btrfs_header_nritems(parent))
785 return NULL;
786
787 BUG_ON(level == 0);
788
789 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
790 btrfs_level_size(root, level - 1),
791 btrfs_node_ptr_generation(parent, slot));
792 }
793
794 static int balance_level(struct btrfs_trans_handle *trans,
795 struct btrfs_root *root,
796 struct btrfs_path *path, int level)
797 {
798 struct extent_buffer *right = NULL;
799 struct extent_buffer *mid;
800 struct extent_buffer *left = NULL;
801 struct extent_buffer *parent = NULL;
802 int ret = 0;
803 int wret;
804 int pslot;
805 int orig_slot = path->slots[level];
806 u64 orig_ptr;
807
808 if (level == 0)
809 return 0;
810
811 mid = path->nodes[level];
812 WARN_ON(btrfs_header_generation(mid) != trans->transid);
813
814 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
815
816 if (level < BTRFS_MAX_LEVEL - 1)
817 parent = path->nodes[level + 1];
818 pslot = path->slots[level + 1];
819
820 /*
821 * deal with the case where there is only one pointer in the root
822 * by promoting the node below to a root
823 */
824 if (!parent) {
825 struct extent_buffer *child;
826
827 if (btrfs_header_nritems(mid) != 1)
828 return 0;
829
830 /* promote the child to a root */
831 child = read_node_slot(root, mid, 0);
832 BUG_ON(!child);
833 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
834 BUG_ON(ret);
835
836 root->node = child;
837 add_root_to_dirty_list(root);
838 path->nodes[level] = NULL;
839 clean_tree_block(trans, root, mid);
840 wait_on_tree_block_writeback(root, mid);
841 /* once for the path */
842 free_extent_buffer(mid);
843
844 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
845 0, root->root_key.objectid,
846 level, 1);
847 /* once for the root ptr */
848 free_extent_buffer(mid);
849 return ret;
850 }
851 if (btrfs_header_nritems(mid) >
852 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
853 return 0;
854
855 left = read_node_slot(root, parent, pslot - 1);
856 if (left) {
857 wret = btrfs_cow_block(trans, root, left,
858 parent, pslot - 1, &left);
859 if (wret) {
860 ret = wret;
861 goto enospc;
862 }
863 }
864 right = read_node_slot(root, parent, pslot + 1);
865 if (right) {
866 wret = btrfs_cow_block(trans, root, right,
867 parent, pslot + 1, &right);
868 if (wret) {
869 ret = wret;
870 goto enospc;
871 }
872 }
873
874 /* first, try to make some room in the middle buffer */
875 if (left) {
876 orig_slot += btrfs_header_nritems(left);
877 wret = push_node_left(trans, root, left, mid, 1);
878 if (wret < 0)
879 ret = wret;
880 }
881
882 /*
883 * then try to empty the right most buffer into the middle
884 */
885 if (right) {
886 wret = push_node_left(trans, root, mid, right, 1);
887 if (wret < 0 && wret != -ENOSPC)
888 ret = wret;
889 if (btrfs_header_nritems(right) == 0) {
890 u64 bytenr = right->start;
891 u32 blocksize = right->len;
892
893 clean_tree_block(trans, root, right);
894 wait_on_tree_block_writeback(root, right);
895 free_extent_buffer(right);
896 right = NULL;
897 wret = btrfs_del_ptr(trans, root, path,
898 level + 1, pslot + 1);
899 if (wret)
900 ret = wret;
901 wret = btrfs_free_extent(trans, root, bytenr,
902 blocksize, 0,
903 root->root_key.objectid,
904 level, 0);
905 if (wret)
906 ret = wret;
907 } else {
908 struct btrfs_disk_key right_key;
909 btrfs_node_key(right, &right_key, 0);
910 btrfs_set_node_key(parent, &right_key, pslot + 1);
911 btrfs_mark_buffer_dirty(parent);
912 }
913 }
914 if (btrfs_header_nritems(mid) == 1) {
915 /*
916 * we're not allowed to leave a node with one item in the
917 * tree during a delete. A deletion from lower in the tree
918 * could try to delete the only pointer in this node.
919 * So, pull some keys from the left.
920 * There has to be a left pointer at this point because
921 * otherwise we would have pulled some pointers from the
922 * right
923 */
924 BUG_ON(!left);
925 wret = balance_node_right(trans, root, mid, left);
926 if (wret < 0) {
927 ret = wret;
928 goto enospc;
929 }
930 if (wret == 1) {
931 wret = push_node_left(trans, root, left, mid, 1);
932 if (wret < 0)
933 ret = wret;
934 }
935 BUG_ON(wret == 1);
936 }
937 if (btrfs_header_nritems(mid) == 0) {
938 /* we've managed to empty the middle node, drop it */
939 u64 bytenr = mid->start;
940 u32 blocksize = mid->len;
941 clean_tree_block(trans, root, mid);
942 wait_on_tree_block_writeback(root, mid);
943 free_extent_buffer(mid);
944 mid = NULL;
945 wret = btrfs_del_ptr(trans, root, path, level + 1, pslot);
946 if (wret)
947 ret = wret;
948 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
949 0, root->root_key.objectid,
950 level, 0);
951 if (wret)
952 ret = wret;
953 } else {
954 /* update the parent key to reflect our changes */
955 struct btrfs_disk_key mid_key;
956 btrfs_node_key(mid, &mid_key, 0);
957 btrfs_set_node_key(parent, &mid_key, pslot);
958 btrfs_mark_buffer_dirty(parent);
959 }
960
961 /* update the path */
962 if (left) {
963 if (btrfs_header_nritems(left) > orig_slot) {
964 extent_buffer_get(left);
965 path->nodes[level] = left;
966 path->slots[level + 1] -= 1;
967 path->slots[level] = orig_slot;
968 if (mid)
969 free_extent_buffer(mid);
970 } else {
971 orig_slot -= btrfs_header_nritems(left);
972 path->slots[level] = orig_slot;
973 }
974 }
975 /* double check we haven't messed things up */
976 check_block(root, path, level);
977 if (orig_ptr !=
978 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
979 BUG();
980 enospc:
981 if (right)
982 free_extent_buffer(right);
983 if (left)
984 free_extent_buffer(left);
985 return ret;
986 }
987
988 /* returns zero if the push worked, non-zero otherwise */
989 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_path *path, int level)
992 {
993 struct extent_buffer *right = NULL;
994 struct extent_buffer *mid;
995 struct extent_buffer *left = NULL;
996 struct extent_buffer *parent = NULL;
997 int ret = 0;
998 int wret;
999 int pslot;
1000 int orig_slot = path->slots[level];
1001
1002 if (level == 0)
1003 return 1;
1004
1005 mid = path->nodes[level];
1006 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1007
1008 if (level < BTRFS_MAX_LEVEL - 1)
1009 parent = path->nodes[level + 1];
1010 pslot = path->slots[level + 1];
1011
1012 if (!parent)
1013 return 1;
1014
1015 left = read_node_slot(root, parent, pslot - 1);
1016
1017 /* first, try to make some room in the middle buffer */
1018 if (left) {
1019 u32 left_nr;
1020 left_nr = btrfs_header_nritems(left);
1021 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1022 wret = 1;
1023 } else {
1024 ret = btrfs_cow_block(trans, root, left, parent,
1025 pslot - 1, &left);
1026 if (ret)
1027 wret = 1;
1028 else {
1029 wret = push_node_left(trans, root,
1030 left, mid, 0);
1031 }
1032 }
1033 if (wret < 0)
1034 ret = wret;
1035 if (wret == 0) {
1036 struct btrfs_disk_key disk_key;
1037 orig_slot += left_nr;
1038 btrfs_node_key(mid, &disk_key, 0);
1039 btrfs_set_node_key(parent, &disk_key, pslot);
1040 btrfs_mark_buffer_dirty(parent);
1041 if (btrfs_header_nritems(left) > orig_slot) {
1042 path->nodes[level] = left;
1043 path->slots[level + 1] -= 1;
1044 path->slots[level] = orig_slot;
1045 free_extent_buffer(mid);
1046 } else {
1047 orig_slot -=
1048 btrfs_header_nritems(left);
1049 path->slots[level] = orig_slot;
1050 free_extent_buffer(left);
1051 }
1052 return 0;
1053 }
1054 free_extent_buffer(left);
1055 }
1056 right= read_node_slot(root, parent, pslot + 1);
1057
1058 /*
1059 * then try to empty the right most buffer into the middle
1060 */
1061 if (right) {
1062 u32 right_nr;
1063 right_nr = btrfs_header_nritems(right);
1064 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1065 wret = 1;
1066 } else {
1067 ret = btrfs_cow_block(trans, root, right,
1068 parent, pslot + 1,
1069 &right);
1070 if (ret)
1071 wret = 1;
1072 else {
1073 wret = balance_node_right(trans, root,
1074 right, mid);
1075 }
1076 }
1077 if (wret < 0)
1078 ret = wret;
1079 if (wret == 0) {
1080 struct btrfs_disk_key disk_key;
1081
1082 btrfs_node_key(right, &disk_key, 0);
1083 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1084 btrfs_mark_buffer_dirty(parent);
1085
1086 if (btrfs_header_nritems(mid) <= orig_slot) {
1087 path->nodes[level] = right;
1088 path->slots[level + 1] += 1;
1089 path->slots[level] = orig_slot -
1090 btrfs_header_nritems(mid);
1091 free_extent_buffer(mid);
1092 } else {
1093 free_extent_buffer(right);
1094 }
1095 return 0;
1096 }
1097 free_extent_buffer(right);
1098 }
1099 return 1;
1100 }
1101
1102 /*
1103 * readahead one full node of leaves
1104 */
1105 void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
1106 int level, int slot, u64 objectid)
1107 {
1108 struct extent_buffer *node;
1109 struct btrfs_disk_key disk_key;
1110 u32 nritems;
1111 u64 search;
1112 u64 lowest_read;
1113 u64 highest_read;
1114 u64 nread = 0;
1115 int direction = path->reada;
1116 struct extent_buffer *eb;
1117 u32 nr;
1118 u32 blocksize;
1119 u32 nscan = 0;
1120
1121 if (level != 1)
1122 return;
1123
1124 if (!path->nodes[level])
1125 return;
1126
1127 node = path->nodes[level];
1128 search = btrfs_node_blockptr(node, slot);
1129 blocksize = btrfs_level_size(root, level - 1);
1130 eb = btrfs_find_tree_block(root, search, blocksize);
1131 if (eb) {
1132 free_extent_buffer(eb);
1133 return;
1134 }
1135
1136 highest_read = search;
1137 lowest_read = search;
1138
1139 nritems = btrfs_header_nritems(node);
1140 nr = slot;
1141 while(1) {
1142 if (direction < 0) {
1143 if (nr == 0)
1144 break;
1145 nr--;
1146 } else if (direction > 0) {
1147 nr++;
1148 if (nr >= nritems)
1149 break;
1150 }
1151 if (path->reada < 0 && objectid) {
1152 btrfs_node_key(node, &disk_key, nr);
1153 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1154 break;
1155 }
1156 search = btrfs_node_blockptr(node, nr);
1157 if ((search >= lowest_read && search <= highest_read) ||
1158 (search < lowest_read && lowest_read - search <= 32768) ||
1159 (search > highest_read && search - highest_read <= 32768)) {
1160 readahead_tree_block(root, search, blocksize,
1161 btrfs_node_ptr_generation(node, nr));
1162 nread += blocksize;
1163 }
1164 nscan++;
1165 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1166 break;
1167 if(nread > (1024 * 1024) || nscan > 128)
1168 break;
1169
1170 if (search < lowest_read)
1171 lowest_read = search;
1172 if (search > highest_read)
1173 highest_read = search;
1174 }
1175 }
1176
1177 /*
1178 * look for key in the tree. path is filled in with nodes along the way
1179 * if key is found, we return zero and you can find the item in the leaf
1180 * level of the path (level 0)
1181 *
1182 * If the key isn't found, the path points to the slot where it should
1183 * be inserted, and 1 is returned. If there are other errors during the
1184 * search a negative error number is returned.
1185 *
1186 * if ins_len > 0, nodes and leaves will be split as we walk down the
1187 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1188 * possible)
1189 */
1190 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1191 *root, struct btrfs_key *key, struct btrfs_path *p, int
1192 ins_len, int cow)
1193 {
1194 struct extent_buffer *b;
1195 int slot;
1196 int ret;
1197 int level;
1198 int should_reada = p->reada;
1199 u8 lowest_level = 0;
1200
1201 lowest_level = p->lowest_level;
1202 WARN_ON(lowest_level && ins_len > 0);
1203 WARN_ON(p->nodes[0] != NULL);
1204 /*
1205 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1206 */
1207 again:
1208 b = root->node;
1209 extent_buffer_get(b);
1210 while (b) {
1211 level = btrfs_header_level(b);
1212 if (cow) {
1213 int wret;
1214 wret = btrfs_cow_block(trans, root, b,
1215 p->nodes[level + 1],
1216 p->slots[level + 1],
1217 &b);
1218 if (wret) {
1219 free_extent_buffer(b);
1220 return wret;
1221 }
1222 }
1223 BUG_ON(!cow && ins_len);
1224 if (level != btrfs_header_level(b))
1225 WARN_ON(1);
1226 level = btrfs_header_level(b);
1227 p->nodes[level] = b;
1228 ret = check_block(root, p, level);
1229 if (ret)
1230 return -1;
1231 ret = bin_search(b, key, level, &slot);
1232 if (level != 0) {
1233 if (ret && slot > 0)
1234 slot -= 1;
1235 p->slots[level] = slot;
1236 if ((p->search_for_split || ins_len > 0) &&
1237 btrfs_header_nritems(b) >=
1238 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1239 int sret = split_node(trans, root, p, level);
1240 BUG_ON(sret > 0);
1241 if (sret)
1242 return sret;
1243 b = p->nodes[level];
1244 slot = p->slots[level];
1245 } else if (ins_len < 0) {
1246 int sret = balance_level(trans, root, p,
1247 level);
1248 if (sret)
1249 return sret;
1250 b = p->nodes[level];
1251 if (!b) {
1252 btrfs_release_path(NULL, p);
1253 goto again;
1254 }
1255 slot = p->slots[level];
1256 BUG_ON(btrfs_header_nritems(b) == 1);
1257 }
1258 /* this is only true while dropping a snapshot */
1259 if (level == lowest_level)
1260 break;
1261
1262 if (should_reada)
1263 reada_for_search(root, p, level, slot,
1264 key->objectid);
1265
1266 b = read_node_slot(root, b, slot);
1267 if (!extent_buffer_uptodate(b))
1268 return -EIO;
1269 } else {
1270 p->slots[level] = slot;
1271 if (ins_len > 0 &&
1272 ins_len > btrfs_leaf_free_space(root, b)) {
1273 int sret = split_leaf(trans, root, key,
1274 p, ins_len, ret == 0);
1275 BUG_ON(sret > 0);
1276 if (sret)
1277 return sret;
1278 }
1279 return ret;
1280 }
1281 }
1282 return 1;
1283 }
1284
1285 /*
1286 * adjust the pointers going up the tree, starting at level
1287 * making sure the right key of each node is points to 'key'.
1288 * This is used after shifting pointers to the left, so it stops
1289 * fixing up pointers when a given leaf/node is not in slot 0 of the
1290 * higher levels
1291 *
1292 * If this fails to write a tree block, it returns -1, but continues
1293 * fixing up the blocks in ram so the tree is consistent.
1294 */
1295 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1296 struct btrfs_root *root, struct btrfs_path *path,
1297 struct btrfs_disk_key *key, int level)
1298 {
1299 int i;
1300 int ret = 0;
1301 struct extent_buffer *t;
1302
1303 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1304 int tslot = path->slots[i];
1305 if (!path->nodes[i])
1306 break;
1307 t = path->nodes[i];
1308 btrfs_set_node_key(t, key, tslot);
1309 btrfs_mark_buffer_dirty(path->nodes[i]);
1310 if (tslot != 0)
1311 break;
1312 }
1313 return ret;
1314 }
1315
1316 /*
1317 * update item key.
1318 *
1319 * This function isn't completely safe. It's the caller's responsibility
1320 * that the new key won't break the order
1321 */
1322 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1323 struct btrfs_root *root, struct btrfs_path *path,
1324 struct btrfs_key *new_key)
1325 {
1326 struct btrfs_disk_key disk_key;
1327 struct extent_buffer *eb;
1328 int slot;
1329
1330 eb = path->nodes[0];
1331 slot = path->slots[0];
1332 if (slot > 0) {
1333 btrfs_item_key(eb, &disk_key, slot - 1);
1334 if (btrfs_comp_keys(&disk_key, new_key) >= 0)
1335 return -1;
1336 }
1337 if (slot < btrfs_header_nritems(eb) - 1) {
1338 btrfs_item_key(eb, &disk_key, slot + 1);
1339 if (btrfs_comp_keys(&disk_key, new_key) <= 0)
1340 return -1;
1341 }
1342
1343 btrfs_cpu_key_to_disk(&disk_key, new_key);
1344 btrfs_set_item_key(eb, &disk_key, slot);
1345 btrfs_mark_buffer_dirty(eb);
1346 if (slot == 0)
1347 fixup_low_keys(trans, root, path, &disk_key, 1);
1348 return 0;
1349 }
1350
1351 /*
1352 * try to push data from one node into the next node left in the
1353 * tree.
1354 *
1355 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1356 * error, and > 0 if there was no room in the left hand block.
1357 */
1358 static int push_node_left(struct btrfs_trans_handle *trans,
1359 struct btrfs_root *root, struct extent_buffer *dst,
1360 struct extent_buffer *src, int empty)
1361 {
1362 int push_items = 0;
1363 int src_nritems;
1364 int dst_nritems;
1365 int ret = 0;
1366
1367 src_nritems = btrfs_header_nritems(src);
1368 dst_nritems = btrfs_header_nritems(dst);
1369 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1370 WARN_ON(btrfs_header_generation(src) != trans->transid);
1371 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1372
1373 if (!empty && src_nritems <= 8)
1374 return 1;
1375
1376 if (push_items <= 0) {
1377 return 1;
1378 }
1379
1380 if (empty) {
1381 push_items = min(src_nritems, push_items);
1382 if (push_items < src_nritems) {
1383 /* leave at least 8 pointers in the node if
1384 * we aren't going to empty it
1385 */
1386 if (src_nritems - push_items < 8) {
1387 if (push_items <= 8)
1388 return 1;
1389 push_items -= 8;
1390 }
1391 }
1392 } else
1393 push_items = min(src_nritems - 8, push_items);
1394
1395 copy_extent_buffer(dst, src,
1396 btrfs_node_key_ptr_offset(dst_nritems),
1397 btrfs_node_key_ptr_offset(0),
1398 push_items * sizeof(struct btrfs_key_ptr));
1399
1400 if (push_items < src_nritems) {
1401 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1402 btrfs_node_key_ptr_offset(push_items),
1403 (src_nritems - push_items) *
1404 sizeof(struct btrfs_key_ptr));
1405 }
1406 btrfs_set_header_nritems(src, src_nritems - push_items);
1407 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1408 btrfs_mark_buffer_dirty(src);
1409 btrfs_mark_buffer_dirty(dst);
1410
1411 return ret;
1412 }
1413
1414 /*
1415 * try to push data from one node into the next node right in the
1416 * tree.
1417 *
1418 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1419 * error, and > 0 if there was no room in the right hand block.
1420 *
1421 * this will only push up to 1/2 the contents of the left node over
1422 */
1423 static int balance_node_right(struct btrfs_trans_handle *trans,
1424 struct btrfs_root *root,
1425 struct extent_buffer *dst,
1426 struct extent_buffer *src)
1427 {
1428 int push_items = 0;
1429 int max_push;
1430 int src_nritems;
1431 int dst_nritems;
1432 int ret = 0;
1433
1434 WARN_ON(btrfs_header_generation(src) != trans->transid);
1435 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1436
1437 src_nritems = btrfs_header_nritems(src);
1438 dst_nritems = btrfs_header_nritems(dst);
1439 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1440 if (push_items <= 0) {
1441 return 1;
1442 }
1443
1444 if (src_nritems < 4) {
1445 return 1;
1446 }
1447
1448 max_push = src_nritems / 2 + 1;
1449 /* don't try to empty the node */
1450 if (max_push >= src_nritems) {
1451 return 1;
1452 }
1453
1454 if (max_push < push_items)
1455 push_items = max_push;
1456
1457 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1458 btrfs_node_key_ptr_offset(0),
1459 (dst_nritems) *
1460 sizeof(struct btrfs_key_ptr));
1461
1462 copy_extent_buffer(dst, src,
1463 btrfs_node_key_ptr_offset(0),
1464 btrfs_node_key_ptr_offset(src_nritems - push_items),
1465 push_items * sizeof(struct btrfs_key_ptr));
1466
1467 btrfs_set_header_nritems(src, src_nritems - push_items);
1468 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1469
1470 btrfs_mark_buffer_dirty(src);
1471 btrfs_mark_buffer_dirty(dst);
1472
1473 return ret;
1474 }
1475
1476 /*
1477 * helper function to insert a new root level in the tree.
1478 * A new node is allocated, and a single item is inserted to
1479 * point to the existing root
1480 *
1481 * returns zero on success or < 0 on failure.
1482 */
1483 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1484 struct btrfs_root *root,
1485 struct btrfs_path *path, int level)
1486 {
1487 u64 lower_gen;
1488 struct extent_buffer *lower;
1489 struct extent_buffer *c;
1490 struct extent_buffer *old;
1491 struct btrfs_disk_key lower_key;
1492
1493 BUG_ON(path->nodes[level]);
1494 BUG_ON(path->nodes[level-1] != root->node);
1495
1496 lower = path->nodes[level-1];
1497 if (level == 1)
1498 btrfs_item_key(lower, &lower_key, 0);
1499 else
1500 btrfs_node_key(lower, &lower_key, 0);
1501
1502 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1503 root->root_key.objectid, &lower_key,
1504 level, root->node->start, 0);
1505
1506 if (IS_ERR(c))
1507 return PTR_ERR(c);
1508
1509 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
1510 btrfs_set_header_nritems(c, 1);
1511 btrfs_set_header_level(c, level);
1512 btrfs_set_header_bytenr(c, c->start);
1513 btrfs_set_header_generation(c, trans->transid);
1514 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
1515 btrfs_set_header_owner(c, root->root_key.objectid);
1516
1517 write_extent_buffer(c, root->fs_info->fsid,
1518 (unsigned long)btrfs_header_fsid(c),
1519 BTRFS_FSID_SIZE);
1520
1521 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1522 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1523 BTRFS_UUID_SIZE);
1524
1525 btrfs_set_node_key(c, &lower_key, 0);
1526 btrfs_set_node_blockptr(c, 0, lower->start);
1527 lower_gen = btrfs_header_generation(lower);
1528 WARN_ON(lower_gen != trans->transid);
1529
1530 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1531
1532 btrfs_mark_buffer_dirty(c);
1533
1534 old = root->node;
1535 root->node = c;
1536
1537 /* the super has an extra ref to root->node */
1538 free_extent_buffer(old);
1539
1540 add_root_to_dirty_list(root);
1541 extent_buffer_get(c);
1542 path->nodes[level] = c;
1543 path->slots[level] = 0;
1544 return 0;
1545 }
1546
1547 /*
1548 * worker function to insert a single pointer in a node.
1549 * the node should have enough room for the pointer already
1550 *
1551 * slot and level indicate where you want the key to go, and
1552 * blocknr is the block the key points to.
1553 *
1554 * returns zero on success and < 0 on any error
1555 */
1556 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1557 *root, struct btrfs_path *path, struct btrfs_disk_key
1558 *key, u64 bytenr, int slot, int level)
1559 {
1560 struct extent_buffer *lower;
1561 int nritems;
1562
1563 BUG_ON(!path->nodes[level]);
1564 lower = path->nodes[level];
1565 nritems = btrfs_header_nritems(lower);
1566 if (slot > nritems)
1567 BUG();
1568 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1569 BUG();
1570 if (slot != nritems) {
1571 memmove_extent_buffer(lower,
1572 btrfs_node_key_ptr_offset(slot + 1),
1573 btrfs_node_key_ptr_offset(slot),
1574 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1575 }
1576 btrfs_set_node_key(lower, key, slot);
1577 btrfs_set_node_blockptr(lower, slot, bytenr);
1578 WARN_ON(trans->transid == 0);
1579 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1580 btrfs_set_header_nritems(lower, nritems + 1);
1581 btrfs_mark_buffer_dirty(lower);
1582 return 0;
1583 }
1584
1585 /*
1586 * split the node at the specified level in path in two.
1587 * The path is corrected to point to the appropriate node after the split
1588 *
1589 * Before splitting this tries to make some room in the node by pushing
1590 * left and right, if either one works, it returns right away.
1591 *
1592 * returns 0 on success and < 0 on failure
1593 */
1594 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1595 *root, struct btrfs_path *path, int level)
1596 {
1597 struct extent_buffer *c;
1598 struct extent_buffer *split;
1599 struct btrfs_disk_key disk_key;
1600 int mid;
1601 int ret;
1602 int wret;
1603 u32 c_nritems;
1604
1605 c = path->nodes[level];
1606 WARN_ON(btrfs_header_generation(c) != trans->transid);
1607 if (c == root->node) {
1608 /* trying to split the root, lets make a new one */
1609 ret = insert_new_root(trans, root, path, level + 1);
1610 if (ret)
1611 return ret;
1612 } else {
1613 ret = push_nodes_for_insert(trans, root, path, level);
1614 c = path->nodes[level];
1615 if (!ret && btrfs_header_nritems(c) <
1616 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1617 return 0;
1618 if (ret < 0)
1619 return ret;
1620 }
1621
1622 c_nritems = btrfs_header_nritems(c);
1623 mid = (c_nritems + 1) / 2;
1624 btrfs_node_key(c, &disk_key, mid);
1625
1626 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1627 root->root_key.objectid,
1628 &disk_key, level, c->start, 0);
1629 if (IS_ERR(split))
1630 return PTR_ERR(split);
1631
1632 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
1633 btrfs_set_header_level(split, btrfs_header_level(c));
1634 btrfs_set_header_bytenr(split, split->start);
1635 btrfs_set_header_generation(split, trans->transid);
1636 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
1637 btrfs_set_header_owner(split, root->root_key.objectid);
1638 write_extent_buffer(split, root->fs_info->fsid,
1639 (unsigned long)btrfs_header_fsid(split),
1640 BTRFS_FSID_SIZE);
1641 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1642 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1643 BTRFS_UUID_SIZE);
1644
1645
1646 copy_extent_buffer(split, c,
1647 btrfs_node_key_ptr_offset(0),
1648 btrfs_node_key_ptr_offset(mid),
1649 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1650 btrfs_set_header_nritems(split, c_nritems - mid);
1651 btrfs_set_header_nritems(c, mid);
1652 ret = 0;
1653
1654 btrfs_mark_buffer_dirty(c);
1655 btrfs_mark_buffer_dirty(split);
1656
1657 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1658 path->slots[level + 1] + 1,
1659 level + 1);
1660 if (wret)
1661 ret = wret;
1662
1663 if (path->slots[level] >= mid) {
1664 path->slots[level] -= mid;
1665 free_extent_buffer(c);
1666 path->nodes[level] = split;
1667 path->slots[level + 1] += 1;
1668 } else {
1669 free_extent_buffer(split);
1670 }
1671 return ret;
1672 }
1673
1674 /*
1675 * how many bytes are required to store the items in a leaf. start
1676 * and nr indicate which items in the leaf to check. This totals up the
1677 * space used both by the item structs and the item data
1678 */
1679 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1680 {
1681 int data_len;
1682 int nritems = btrfs_header_nritems(l);
1683 int end = min(nritems, start + nr) - 1;
1684
1685 if (!nr)
1686 return 0;
1687 data_len = btrfs_item_end_nr(l, start);
1688 data_len = data_len - btrfs_item_offset_nr(l, end);
1689 data_len += sizeof(struct btrfs_item) * nr;
1690 WARN_ON(data_len < 0);
1691 return data_len;
1692 }
1693
1694 /*
1695 * The space between the end of the leaf items and
1696 * the start of the leaf data. IOW, how much room
1697 * the leaf has left for both items and data
1698 */
1699 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1700 {
1701 int nritems = btrfs_header_nritems(leaf);
1702 int ret;
1703 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1704 if (ret < 0) {
1705 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1706 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1707 leaf_space_used(leaf, 0, nritems), nritems);
1708 }
1709 return ret;
1710 }
1711
1712 /*
1713 * push some data in the path leaf to the right, trying to free up at
1714 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1715 *
1716 * returns 1 if the push failed because the other node didn't have enough
1717 * room, 0 if everything worked out and < 0 if there were major errors.
1718 */
1719 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1720 *root, struct btrfs_path *path, int data_size,
1721 int empty)
1722 {
1723 struct extent_buffer *left = path->nodes[0];
1724 struct extent_buffer *right;
1725 struct extent_buffer *upper;
1726 struct btrfs_disk_key disk_key;
1727 int slot;
1728 u32 i;
1729 int free_space;
1730 int push_space = 0;
1731 int push_items = 0;
1732 struct btrfs_item *item;
1733 u32 left_nritems;
1734 u32 nr;
1735 u32 right_nritems;
1736 u32 data_end;
1737 u32 this_item_size;
1738 int ret;
1739
1740 slot = path->slots[1];
1741 if (!path->nodes[1]) {
1742 return 1;
1743 }
1744 upper = path->nodes[1];
1745 if (slot >= btrfs_header_nritems(upper) - 1)
1746 return 1;
1747
1748 right = read_node_slot(root, upper, slot + 1);
1749 free_space = btrfs_leaf_free_space(root, right);
1750 if (free_space < data_size) {
1751 free_extent_buffer(right);
1752 return 1;
1753 }
1754
1755 /* cow and double check */
1756 ret = btrfs_cow_block(trans, root, right, upper,
1757 slot + 1, &right);
1758 if (ret) {
1759 free_extent_buffer(right);
1760 return 1;
1761 }
1762 free_space = btrfs_leaf_free_space(root, right);
1763 if (free_space < data_size) {
1764 free_extent_buffer(right);
1765 return 1;
1766 }
1767
1768 left_nritems = btrfs_header_nritems(left);
1769 if (left_nritems == 0) {
1770 free_extent_buffer(right);
1771 return 1;
1772 }
1773
1774 if (empty)
1775 nr = 0;
1776 else
1777 nr = 1;
1778
1779 i = left_nritems - 1;
1780 while (i >= nr) {
1781 item = btrfs_item_nr(left, i);
1782
1783 if (path->slots[0] == i)
1784 push_space += data_size + sizeof(*item);
1785
1786 this_item_size = btrfs_item_size(left, item);
1787 if (this_item_size + sizeof(*item) + push_space > free_space)
1788 break;
1789 push_items++;
1790 push_space += this_item_size + sizeof(*item);
1791 if (i == 0)
1792 break;
1793 i--;
1794 }
1795
1796 if (push_items == 0) {
1797 free_extent_buffer(right);
1798 return 1;
1799 }
1800
1801 if (!empty && push_items == left_nritems)
1802 WARN_ON(1);
1803
1804 /* push left to right */
1805 right_nritems = btrfs_header_nritems(right);
1806
1807 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1808 push_space -= leaf_data_end(root, left);
1809
1810 /* make room in the right data area */
1811 data_end = leaf_data_end(root, right);
1812 memmove_extent_buffer(right,
1813 btrfs_leaf_data(right) + data_end - push_space,
1814 btrfs_leaf_data(right) + data_end,
1815 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1816
1817 /* copy from the left data area */
1818 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1819 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1820 btrfs_leaf_data(left) + leaf_data_end(root, left),
1821 push_space);
1822
1823 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1824 btrfs_item_nr_offset(0),
1825 right_nritems * sizeof(struct btrfs_item));
1826
1827 /* copy the items from left to right */
1828 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1829 btrfs_item_nr_offset(left_nritems - push_items),
1830 push_items * sizeof(struct btrfs_item));
1831
1832 /* update the item pointers */
1833 right_nritems += push_items;
1834 btrfs_set_header_nritems(right, right_nritems);
1835 push_space = BTRFS_LEAF_DATA_SIZE(root);
1836 for (i = 0; i < right_nritems; i++) {
1837 item = btrfs_item_nr(right, i);
1838 push_space -= btrfs_item_size(right, item);
1839 btrfs_set_item_offset(right, item, push_space);
1840 }
1841
1842 left_nritems -= push_items;
1843 btrfs_set_header_nritems(left, left_nritems);
1844
1845 if (left_nritems)
1846 btrfs_mark_buffer_dirty(left);
1847 btrfs_mark_buffer_dirty(right);
1848
1849 btrfs_item_key(right, &disk_key, 0);
1850 btrfs_set_node_key(upper, &disk_key, slot + 1);
1851 btrfs_mark_buffer_dirty(upper);
1852
1853 /* then fixup the leaf pointer in the path */
1854 if (path->slots[0] >= left_nritems) {
1855 path->slots[0] -= left_nritems;
1856 free_extent_buffer(path->nodes[0]);
1857 path->nodes[0] = right;
1858 path->slots[1] += 1;
1859 } else {
1860 free_extent_buffer(right);
1861 }
1862 return 0;
1863 }
1864 /*
1865 * push some data in the path leaf to the left, trying to free up at
1866 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1867 */
1868 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1869 *root, struct btrfs_path *path, int data_size,
1870 int empty)
1871 {
1872 struct btrfs_disk_key disk_key;
1873 struct extent_buffer *right = path->nodes[0];
1874 struct extent_buffer *left;
1875 int slot;
1876 int i;
1877 int free_space;
1878 int push_space = 0;
1879 int push_items = 0;
1880 struct btrfs_item *item;
1881 u32 old_left_nritems;
1882 u32 right_nritems;
1883 u32 nr;
1884 int ret = 0;
1885 int wret;
1886 u32 this_item_size;
1887 u32 old_left_item_size;
1888
1889 slot = path->slots[1];
1890 if (slot == 0)
1891 return 1;
1892 if (!path->nodes[1])
1893 return 1;
1894
1895 right_nritems = btrfs_header_nritems(right);
1896 if (right_nritems == 0) {
1897 return 1;
1898 }
1899
1900 left = read_node_slot(root, path->nodes[1], slot - 1);
1901 free_space = btrfs_leaf_free_space(root, left);
1902 if (free_space < data_size) {
1903 free_extent_buffer(left);
1904 return 1;
1905 }
1906
1907 /* cow and double check */
1908 ret = btrfs_cow_block(trans, root, left,
1909 path->nodes[1], slot - 1, &left);
1910 if (ret) {
1911 /* we hit -ENOSPC, but it isn't fatal here */
1912 free_extent_buffer(left);
1913 return 1;
1914 }
1915
1916 free_space = btrfs_leaf_free_space(root, left);
1917 if (free_space < data_size) {
1918 free_extent_buffer(left);
1919 return 1;
1920 }
1921
1922 if (empty)
1923 nr = right_nritems;
1924 else
1925 nr = right_nritems - 1;
1926
1927 for (i = 0; i < nr; i++) {
1928 item = btrfs_item_nr(right, i);
1929
1930 if (path->slots[0] == i)
1931 push_space += data_size + sizeof(*item);
1932
1933 this_item_size = btrfs_item_size(right, item);
1934 if (this_item_size + sizeof(*item) + push_space > free_space)
1935 break;
1936
1937 push_items++;
1938 push_space += this_item_size + sizeof(*item);
1939 }
1940
1941 if (push_items == 0) {
1942 free_extent_buffer(left);
1943 return 1;
1944 }
1945 if (!empty && push_items == btrfs_header_nritems(right))
1946 WARN_ON(1);
1947
1948 /* push data from right to left */
1949 copy_extent_buffer(left, right,
1950 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1951 btrfs_item_nr_offset(0),
1952 push_items * sizeof(struct btrfs_item));
1953
1954 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1955 btrfs_item_offset_nr(right, push_items -1);
1956
1957 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1958 leaf_data_end(root, left) - push_space,
1959 btrfs_leaf_data(right) +
1960 btrfs_item_offset_nr(right, push_items - 1),
1961 push_space);
1962 old_left_nritems = btrfs_header_nritems(left);
1963 BUG_ON(old_left_nritems < 0);
1964
1965 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1966 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1967 u32 ioff;
1968
1969 item = btrfs_item_nr(left, i);
1970 ioff = btrfs_item_offset(left, item);
1971 btrfs_set_item_offset(left, item,
1972 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1973 }
1974 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1975
1976 /* fixup right node */
1977 if (push_items > right_nritems) {
1978 printk("push items %d nr %u\n", push_items, right_nritems);
1979 WARN_ON(1);
1980 }
1981
1982 if (push_items < right_nritems) {
1983 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1984 leaf_data_end(root, right);
1985 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1986 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1987 btrfs_leaf_data(right) +
1988 leaf_data_end(root, right), push_space);
1989
1990 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1991 btrfs_item_nr_offset(push_items),
1992 (btrfs_header_nritems(right) - push_items) *
1993 sizeof(struct btrfs_item));
1994 }
1995 right_nritems -= push_items;
1996 btrfs_set_header_nritems(right, right_nritems);
1997 push_space = BTRFS_LEAF_DATA_SIZE(root);
1998 for (i = 0; i < right_nritems; i++) {
1999 item = btrfs_item_nr(right, i);
2000 push_space = push_space - btrfs_item_size(right, item);
2001 btrfs_set_item_offset(right, item, push_space);
2002 }
2003
2004 btrfs_mark_buffer_dirty(left);
2005 if (right_nritems)
2006 btrfs_mark_buffer_dirty(right);
2007
2008 btrfs_item_key(right, &disk_key, 0);
2009 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2010 if (wret)
2011 ret = wret;
2012
2013 /* then fixup the leaf pointer in the path */
2014 if (path->slots[0] < push_items) {
2015 path->slots[0] += old_left_nritems;
2016 free_extent_buffer(path->nodes[0]);
2017 path->nodes[0] = left;
2018 path->slots[1] -= 1;
2019 } else {
2020 free_extent_buffer(left);
2021 path->slots[0] -= push_items;
2022 }
2023 BUG_ON(path->slots[0] < 0);
2024 return ret;
2025 }
2026
2027 /*
2028 * split the path's leaf in two, making sure there is at least data_size
2029 * available for the resulting leaf level of the path.
2030 *
2031 * returns 0 if all went well and < 0 on failure.
2032 */
2033 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
2034 struct btrfs_root *root,
2035 struct btrfs_path *path,
2036 struct extent_buffer *l,
2037 struct extent_buffer *right,
2038 int slot, int mid, int nritems)
2039 {
2040 int data_copy_size;
2041 int rt_data_off;
2042 int i;
2043 int ret = 0;
2044 int wret;
2045 struct btrfs_disk_key disk_key;
2046
2047 nritems = nritems - mid;
2048 btrfs_set_header_nritems(right, nritems);
2049 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2050
2051 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2052 btrfs_item_nr_offset(mid),
2053 nritems * sizeof(struct btrfs_item));
2054
2055 copy_extent_buffer(right, l,
2056 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2057 data_copy_size, btrfs_leaf_data(l) +
2058 leaf_data_end(root, l), data_copy_size);
2059
2060 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2061 btrfs_item_end_nr(l, mid);
2062
2063 for (i = 0; i < nritems; i++) {
2064 struct btrfs_item *item = btrfs_item_nr(right, i);
2065 u32 ioff = btrfs_item_offset(right, item);
2066 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2067 }
2068
2069 btrfs_set_header_nritems(l, mid);
2070 ret = 0;
2071 btrfs_item_key(right, &disk_key, 0);
2072 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2073 path->slots[1] + 1, 1);
2074 if (wret)
2075 ret = wret;
2076
2077 btrfs_mark_buffer_dirty(right);
2078 btrfs_mark_buffer_dirty(l);
2079 BUG_ON(path->slots[0] != slot);
2080
2081 if (mid <= slot) {
2082 free_extent_buffer(path->nodes[0]);
2083 path->nodes[0] = right;
2084 path->slots[0] -= mid;
2085 path->slots[1] += 1;
2086 } else {
2087 free_extent_buffer(right);
2088 }
2089
2090 BUG_ON(path->slots[0] < 0);
2091
2092 return ret;
2093 }
2094
2095 /*
2096 * split the path's leaf in two, making sure there is at least data_size
2097 * available for the resulting leaf level of the path.
2098 *
2099 * returns 0 if all went well and < 0 on failure.
2100 */
2101 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2102 struct btrfs_root *root,
2103 struct btrfs_key *ins_key,
2104 struct btrfs_path *path, int data_size,
2105 int extend)
2106 {
2107 struct btrfs_disk_key disk_key;
2108 struct extent_buffer *l;
2109 u32 nritems;
2110 int mid;
2111 int slot;
2112 struct extent_buffer *right;
2113 int ret = 0;
2114 int wret;
2115 int split;
2116 int num_doubles = 0;
2117
2118 /* first try to make some room by pushing left and right */
2119 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2120 wret = push_leaf_right(trans, root, path, data_size, 0);
2121 if (wret < 0)
2122 return wret;
2123 if (wret) {
2124 wret = push_leaf_left(trans, root, path, data_size, 0);
2125 if (wret < 0)
2126 return wret;
2127 }
2128 l = path->nodes[0];
2129
2130 /* did the pushes work? */
2131 if (btrfs_leaf_free_space(root, l) >= data_size)
2132 return 0;
2133 }
2134
2135 if (!path->nodes[1]) {
2136 ret = insert_new_root(trans, root, path, 1);
2137 if (ret)
2138 return ret;
2139 }
2140 again:
2141 split = 1;
2142 l = path->nodes[0];
2143 slot = path->slots[0];
2144 nritems = btrfs_header_nritems(l);
2145 mid = (nritems + 1) / 2;
2146
2147 if (mid <= slot) {
2148 if (nritems == 1 ||
2149 leaf_space_used(l, mid, nritems - mid) + data_size >
2150 BTRFS_LEAF_DATA_SIZE(root)) {
2151 if (slot >= nritems) {
2152 split = 0;
2153 } else {
2154 mid = slot;
2155 if (mid != nritems &&
2156 leaf_space_used(l, mid, nritems - mid) +
2157 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2158 split = 2;
2159 }
2160 }
2161 }
2162 } else {
2163 if (leaf_space_used(l, 0, mid) + data_size >
2164 BTRFS_LEAF_DATA_SIZE(root)) {
2165 if (!extend && data_size && slot == 0) {
2166 split = 0;
2167 } else if ((extend || !data_size) && slot == 0) {
2168 mid = 1;
2169 } else {
2170 mid = slot;
2171 if (mid != nritems &&
2172 leaf_space_used(l, mid, nritems - mid) +
2173 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2174 split = 2 ;
2175 }
2176 }
2177 }
2178 }
2179
2180 if (split == 0)
2181 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2182 else
2183 btrfs_item_key(l, &disk_key, mid);
2184
2185 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2186 root->root_key.objectid,
2187 &disk_key, 0, l->start, 0);
2188 if (IS_ERR(right)) {
2189 BUG_ON(1);
2190 return PTR_ERR(right);
2191 }
2192
2193 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2194 btrfs_set_header_bytenr(right, right->start);
2195 btrfs_set_header_generation(right, trans->transid);
2196 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2197 btrfs_set_header_owner(right, root->root_key.objectid);
2198 btrfs_set_header_level(right, 0);
2199 write_extent_buffer(right, root->fs_info->fsid,
2200 (unsigned long)btrfs_header_fsid(right),
2201 BTRFS_FSID_SIZE);
2202
2203 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2204 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2205 BTRFS_UUID_SIZE);
2206
2207 if (split == 0) {
2208 if (mid <= slot) {
2209 btrfs_set_header_nritems(right, 0);
2210 wret = insert_ptr(trans, root, path,
2211 &disk_key, right->start,
2212 path->slots[1] + 1, 1);
2213 if (wret)
2214 ret = wret;
2215
2216 free_extent_buffer(path->nodes[0]);
2217 path->nodes[0] = right;
2218 path->slots[0] = 0;
2219 path->slots[1] += 1;
2220 } else {
2221 btrfs_set_header_nritems(right, 0);
2222 wret = insert_ptr(trans, root, path,
2223 &disk_key,
2224 right->start,
2225 path->slots[1], 1);
2226 if (wret)
2227 ret = wret;
2228 free_extent_buffer(path->nodes[0]);
2229 path->nodes[0] = right;
2230 path->slots[0] = 0;
2231 if (path->slots[1] == 0) {
2232 wret = fixup_low_keys(trans, root,
2233 path, &disk_key, 1);
2234 if (wret)
2235 ret = wret;
2236 }
2237 }
2238 btrfs_mark_buffer_dirty(right);
2239 return ret;
2240 }
2241
2242 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2243 BUG_ON(ret);
2244
2245 if (split == 2) {
2246 BUG_ON(num_doubles != 0);
2247 num_doubles++;
2248 goto again;
2249 }
2250
2251 return ret;
2252 }
2253
2254 /*
2255 * This function splits a single item into two items,
2256 * giving 'new_key' to the new item and splitting the
2257 * old one at split_offset (from the start of the item).
2258 *
2259 * The path may be released by this operation. After
2260 * the split, the path is pointing to the old item. The
2261 * new item is going to be in the same node as the old one.
2262 *
2263 * Note, the item being split must be smaller enough to live alone on
2264 * a tree block with room for one extra struct btrfs_item
2265 *
2266 * This allows us to split the item in place, keeping a lock on the
2267 * leaf the entire time.
2268 */
2269 int btrfs_split_item(struct btrfs_trans_handle *trans,
2270 struct btrfs_root *root,
2271 struct btrfs_path *path,
2272 struct btrfs_key *new_key,
2273 unsigned long split_offset)
2274 {
2275 u32 item_size;
2276 struct extent_buffer *leaf;
2277 struct btrfs_key orig_key;
2278 struct btrfs_item *item;
2279 struct btrfs_item *new_item;
2280 int ret = 0;
2281 int slot;
2282 u32 nritems;
2283 u32 orig_offset;
2284 struct btrfs_disk_key disk_key;
2285 char *buf;
2286
2287 leaf = path->nodes[0];
2288 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2289 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2290 goto split;
2291
2292 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2293 btrfs_release_path(root, path);
2294
2295 path->search_for_split = 1;
2296
2297 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2298 path->search_for_split = 0;
2299
2300 /* if our item isn't there or got smaller, return now */
2301 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2302 path->slots[0])) {
2303 return -EAGAIN;
2304 }
2305
2306 ret = split_leaf(trans, root, &orig_key, path, 0, 0);
2307 BUG_ON(ret);
2308
2309 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2310 leaf = path->nodes[0];
2311
2312 split:
2313 item = btrfs_item_nr(leaf, path->slots[0]);
2314 orig_offset = btrfs_item_offset(leaf, item);
2315 item_size = btrfs_item_size(leaf, item);
2316
2317
2318 buf = kmalloc(item_size, GFP_NOFS);
2319 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2320 path->slots[0]), item_size);
2321 slot = path->slots[0] + 1;
2322 leaf = path->nodes[0];
2323
2324 nritems = btrfs_header_nritems(leaf);
2325
2326 if (slot != nritems) {
2327 /* shift the items */
2328 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2329 btrfs_item_nr_offset(slot),
2330 (nritems - slot) * sizeof(struct btrfs_item));
2331
2332 }
2333
2334 btrfs_cpu_key_to_disk(&disk_key, new_key);
2335 btrfs_set_item_key(leaf, &disk_key, slot);
2336
2337 new_item = btrfs_item_nr(leaf, slot);
2338
2339 btrfs_set_item_offset(leaf, new_item, orig_offset);
2340 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
2341
2342 btrfs_set_item_offset(leaf, item,
2343 orig_offset + item_size - split_offset);
2344 btrfs_set_item_size(leaf, item, split_offset);
2345
2346 btrfs_set_header_nritems(leaf, nritems + 1);
2347
2348 /* write the data for the start of the original item */
2349 write_extent_buffer(leaf, buf,
2350 btrfs_item_ptr_offset(leaf, path->slots[0]),
2351 split_offset);
2352
2353 /* write the data for the new item */
2354 write_extent_buffer(leaf, buf + split_offset,
2355 btrfs_item_ptr_offset(leaf, slot),
2356 item_size - split_offset);
2357 btrfs_mark_buffer_dirty(leaf);
2358
2359 ret = 0;
2360 if (btrfs_leaf_free_space(root, leaf) < 0) {
2361 btrfs_print_leaf(root, leaf);
2362 BUG();
2363 }
2364 kfree(buf);
2365 return ret;
2366 }
2367
2368 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2369 struct btrfs_root *root,
2370 struct btrfs_path *path,
2371 u32 new_size, int from_end)
2372 {
2373 int ret = 0;
2374 int slot;
2375 struct extent_buffer *leaf;
2376 struct btrfs_item *item;
2377 u32 nritems;
2378 unsigned int data_end;
2379 unsigned int old_data_start;
2380 unsigned int old_size;
2381 unsigned int size_diff;
2382 int i;
2383
2384 leaf = path->nodes[0];
2385 slot = path->slots[0];
2386
2387 old_size = btrfs_item_size_nr(leaf, slot);
2388 if (old_size == new_size)
2389 return 0;
2390
2391 nritems = btrfs_header_nritems(leaf);
2392 data_end = leaf_data_end(root, leaf);
2393
2394 old_data_start = btrfs_item_offset_nr(leaf, slot);
2395
2396 size_diff = old_size - new_size;
2397
2398 BUG_ON(slot < 0);
2399 BUG_ON(slot >= nritems);
2400
2401 /*
2402 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2403 */
2404 /* first correct the data pointers */
2405 for (i = slot; i < nritems; i++) {
2406 u32 ioff;
2407 item = btrfs_item_nr(leaf, i);
2408 ioff = btrfs_item_offset(leaf, item);
2409 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2410 }
2411
2412 /* shift the data */
2413 if (from_end) {
2414 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2415 data_end + size_diff, btrfs_leaf_data(leaf) +
2416 data_end, old_data_start + new_size - data_end);
2417 } else {
2418 struct btrfs_disk_key disk_key;
2419 u64 offset;
2420
2421 btrfs_item_key(leaf, &disk_key, slot);
2422
2423 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2424 unsigned long ptr;
2425 struct btrfs_file_extent_item *fi;
2426
2427 fi = btrfs_item_ptr(leaf, slot,
2428 struct btrfs_file_extent_item);
2429 fi = (struct btrfs_file_extent_item *)(
2430 (unsigned long)fi - size_diff);
2431
2432 if (btrfs_file_extent_type(leaf, fi) ==
2433 BTRFS_FILE_EXTENT_INLINE) {
2434 ptr = btrfs_item_ptr_offset(leaf, slot);
2435 memmove_extent_buffer(leaf, ptr,
2436 (unsigned long)fi,
2437 offsetof(struct btrfs_file_extent_item,
2438 disk_bytenr));
2439 }
2440 }
2441
2442 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2443 data_end + size_diff, btrfs_leaf_data(leaf) +
2444 data_end, old_data_start - data_end);
2445
2446 offset = btrfs_disk_key_offset(&disk_key);
2447 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2448 btrfs_set_item_key(leaf, &disk_key, slot);
2449 if (slot == 0)
2450 fixup_low_keys(trans, root, path, &disk_key, 1);
2451 }
2452
2453 item = btrfs_item_nr(leaf, slot);
2454 btrfs_set_item_size(leaf, item, new_size);
2455 btrfs_mark_buffer_dirty(leaf);
2456
2457 ret = 0;
2458 if (btrfs_leaf_free_space(root, leaf) < 0) {
2459 btrfs_print_leaf(root, leaf);
2460 BUG();
2461 }
2462 return ret;
2463 }
2464
2465 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2466 struct btrfs_root *root, struct btrfs_path *path,
2467 u32 data_size)
2468 {
2469 int ret = 0;
2470 int slot;
2471 struct extent_buffer *leaf;
2472 struct btrfs_item *item;
2473 u32 nritems;
2474 unsigned int data_end;
2475 unsigned int old_data;
2476 unsigned int old_size;
2477 int i;
2478
2479 leaf = path->nodes[0];
2480
2481 nritems = btrfs_header_nritems(leaf);
2482 data_end = leaf_data_end(root, leaf);
2483
2484 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2485 btrfs_print_leaf(root, leaf);
2486 BUG();
2487 }
2488 slot = path->slots[0];
2489 old_data = btrfs_item_end_nr(leaf, slot);
2490
2491 BUG_ON(slot < 0);
2492 if (slot >= nritems) {
2493 btrfs_print_leaf(root, leaf);
2494 printk("slot %d too large, nritems %d\n", slot, nritems);
2495 BUG_ON(1);
2496 }
2497
2498 /*
2499 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2500 */
2501 /* first correct the data pointers */
2502 for (i = slot; i < nritems; i++) {
2503 u32 ioff;
2504 item = btrfs_item_nr(leaf, i);
2505 ioff = btrfs_item_offset(leaf, item);
2506 btrfs_set_item_offset(leaf, item, ioff - data_size);
2507 }
2508
2509 /* shift the data */
2510 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2511 data_end - data_size, btrfs_leaf_data(leaf) +
2512 data_end, old_data - data_end);
2513
2514 data_end = old_data;
2515 old_size = btrfs_item_size_nr(leaf, slot);
2516 item = btrfs_item_nr(leaf, slot);
2517 btrfs_set_item_size(leaf, item, old_size + data_size);
2518 btrfs_mark_buffer_dirty(leaf);
2519
2520 ret = 0;
2521 if (btrfs_leaf_free_space(root, leaf) < 0) {
2522 btrfs_print_leaf(root, leaf);
2523 BUG();
2524 }
2525 return ret;
2526 }
2527
2528 /*
2529 * Given a key and some data, insert an item into the tree.
2530 * This does all the path init required, making room in the tree if needed.
2531 */
2532 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2533 struct btrfs_root *root,
2534 struct btrfs_path *path,
2535 struct btrfs_key *cpu_key, u32 *data_size,
2536 int nr)
2537 {
2538 struct extent_buffer *leaf;
2539 struct btrfs_item *item;
2540 int ret = 0;
2541 int slot;
2542 int i;
2543 u32 nritems;
2544 u32 total_size = 0;
2545 u32 total_data = 0;
2546 unsigned int data_end;
2547 struct btrfs_disk_key disk_key;
2548
2549 for (i = 0; i < nr; i++) {
2550 total_data += data_size[i];
2551 }
2552
2553 /* create a root if there isn't one */
2554 if (!root->node)
2555 BUG();
2556
2557 total_size = total_data + nr * sizeof(struct btrfs_item);
2558 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2559 if (ret == 0) {
2560 return -EEXIST;
2561 }
2562 if (ret < 0)
2563 goto out;
2564
2565 leaf = path->nodes[0];
2566
2567 nritems = btrfs_header_nritems(leaf);
2568 data_end = leaf_data_end(root, leaf);
2569
2570 if (btrfs_leaf_free_space(root, leaf) < total_size) {
2571 btrfs_print_leaf(root, leaf);
2572 printk("not enough freespace need %u have %d\n",
2573 total_size, btrfs_leaf_free_space(root, leaf));
2574 BUG();
2575 }
2576
2577 slot = path->slots[0];
2578 BUG_ON(slot < 0);
2579
2580 if (slot != nritems) {
2581 int i;
2582 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2583
2584 if (old_data < data_end) {
2585 btrfs_print_leaf(root, leaf);
2586 printk("slot %d old_data %d data_end %d\n",
2587 slot, old_data, data_end);
2588 BUG_ON(1);
2589 }
2590 /*
2591 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2592 */
2593 /* first correct the data pointers */
2594 for (i = slot; i < nritems; i++) {
2595 u32 ioff;
2596
2597 item = btrfs_item_nr(leaf, i);
2598 ioff = btrfs_item_offset(leaf, item);
2599 btrfs_set_item_offset(leaf, item, ioff - total_data);
2600 }
2601
2602 /* shift the items */
2603 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2604 btrfs_item_nr_offset(slot),
2605 (nritems - slot) * sizeof(struct btrfs_item));
2606
2607 /* shift the data */
2608 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2609 data_end - total_data, btrfs_leaf_data(leaf) +
2610 data_end, old_data - data_end);
2611 data_end = old_data;
2612 }
2613
2614 /* setup the item for the new data */
2615 for (i = 0; i < nr; i++) {
2616 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2617 btrfs_set_item_key(leaf, &disk_key, slot + i);
2618 item = btrfs_item_nr(leaf, slot + i);
2619 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2620 data_end -= data_size[i];
2621 btrfs_set_item_size(leaf, item, data_size[i]);
2622 }
2623 btrfs_set_header_nritems(leaf, nritems + nr);
2624 btrfs_mark_buffer_dirty(leaf);
2625
2626 ret = 0;
2627 if (slot == 0) {
2628 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2629 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2630 }
2631
2632 if (btrfs_leaf_free_space(root, leaf) < 0) {
2633 btrfs_print_leaf(root, leaf);
2634 BUG();
2635 }
2636
2637 out:
2638 return ret;
2639 }
2640
2641 /*
2642 * Given a key and some data, insert an item into the tree.
2643 * This does all the path init required, making room in the tree if needed.
2644 */
2645 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2646 *root, struct btrfs_key *cpu_key, void *data, u32
2647 data_size)
2648 {
2649 int ret = 0;
2650 struct btrfs_path *path;
2651 struct extent_buffer *leaf;
2652 unsigned long ptr;
2653
2654 path = btrfs_alloc_path();
2655 BUG_ON(!path);
2656 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2657 if (!ret) {
2658 leaf = path->nodes[0];
2659 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2660 write_extent_buffer(leaf, data, ptr, data_size);
2661 btrfs_mark_buffer_dirty(leaf);
2662 }
2663 btrfs_free_path(path);
2664 return ret;
2665 }
2666
2667 /*
2668 * delete the pointer from a given node.
2669 *
2670 * If the delete empties a node, the node is removed from the tree,
2671 * continuing all the way the root if required. The root is converted into
2672 * a leaf if all the nodes are emptied.
2673 */
2674 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2675 struct btrfs_path *path, int level, int slot)
2676 {
2677 struct extent_buffer *parent = path->nodes[level];
2678 u32 nritems;
2679 int ret = 0;
2680 int wret;
2681
2682 nritems = btrfs_header_nritems(parent);
2683 if (slot != nritems -1) {
2684 memmove_extent_buffer(parent,
2685 btrfs_node_key_ptr_offset(slot),
2686 btrfs_node_key_ptr_offset(slot + 1),
2687 sizeof(struct btrfs_key_ptr) *
2688 (nritems - slot - 1));
2689 }
2690 nritems--;
2691 btrfs_set_header_nritems(parent, nritems);
2692 if (nritems == 0 && parent == root->node) {
2693 BUG_ON(btrfs_header_level(root->node) != 1);
2694 /* just turn the root into a leaf and break */
2695 btrfs_set_header_level(root->node, 0);
2696 } else if (slot == 0) {
2697 struct btrfs_disk_key disk_key;
2698
2699 btrfs_node_key(parent, &disk_key, 0);
2700 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2701 if (wret)
2702 ret = wret;
2703 }
2704 btrfs_mark_buffer_dirty(parent);
2705 return ret;
2706 }
2707
2708 /*
2709 * a helper function to delete the leaf pointed to by path->slots[1] and
2710 * path->nodes[1].
2711 *
2712 * This deletes the pointer in path->nodes[1] and frees the leaf
2713 * block extent. zero is returned if it all worked out, < 0 otherwise.
2714 *
2715 * The path must have already been setup for deleting the leaf, including
2716 * all the proper balancing. path->nodes[1] must be locked.
2717 */
2718 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
2719 struct btrfs_root *root,
2720 struct btrfs_path *path,
2721 struct extent_buffer *leaf)
2722 {
2723 int ret;
2724
2725 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
2726 ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]);
2727 if (ret)
2728 return ret;
2729
2730 ret = btrfs_free_extent(trans, root, leaf->start, leaf->len,
2731 0, root->root_key.objectid, 0, 0);
2732 return ret;
2733 }
2734
2735 /*
2736 * delete the item at the leaf level in path. If that empties
2737 * the leaf, remove it from the tree
2738 */
2739 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2740 struct btrfs_path *path, int slot, int nr)
2741 {
2742 struct extent_buffer *leaf;
2743 struct btrfs_item *item;
2744 int last_off;
2745 int dsize = 0;
2746 int ret = 0;
2747 int wret;
2748 int i;
2749 u32 nritems;
2750
2751 leaf = path->nodes[0];
2752 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2753
2754 for (i = 0; i < nr; i++)
2755 dsize += btrfs_item_size_nr(leaf, slot + i);
2756
2757 nritems = btrfs_header_nritems(leaf);
2758
2759 if (slot + nr != nritems) {
2760 int i;
2761 int data_end = leaf_data_end(root, leaf);
2762
2763 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2764 data_end + dsize,
2765 btrfs_leaf_data(leaf) + data_end,
2766 last_off - data_end);
2767
2768 for (i = slot + nr; i < nritems; i++) {
2769 u32 ioff;
2770
2771 item = btrfs_item_nr(leaf, i);
2772 ioff = btrfs_item_offset(leaf, item);
2773 btrfs_set_item_offset(leaf, item, ioff + dsize);
2774 }
2775
2776 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2777 btrfs_item_nr_offset(slot + nr),
2778 sizeof(struct btrfs_item) *
2779 (nritems - slot - nr));
2780 }
2781 btrfs_set_header_nritems(leaf, nritems - nr);
2782 nritems -= nr;
2783
2784 /* delete the leaf if we've emptied it */
2785 if (nritems == 0) {
2786 if (leaf == root->node) {
2787 btrfs_set_header_level(leaf, 0);
2788 } else {
2789 clean_tree_block(trans, root, leaf);
2790 wait_on_tree_block_writeback(root, leaf);
2791
2792 wret = btrfs_del_leaf(trans, root, path, leaf);
2793 BUG_ON(ret);
2794 if (wret)
2795 ret = wret;
2796 }
2797 } else {
2798 int used = leaf_space_used(leaf, 0, nritems);
2799 if (slot == 0) {
2800 struct btrfs_disk_key disk_key;
2801
2802 btrfs_item_key(leaf, &disk_key, 0);
2803 wret = fixup_low_keys(trans, root, path,
2804 &disk_key, 1);
2805 if (wret)
2806 ret = wret;
2807 }
2808
2809 /* delete the leaf if it is mostly empty */
2810 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2811 /* push_leaf_left fixes the path.
2812 * make sure the path still points to our leaf
2813 * for possible call to del_ptr below
2814 */
2815 slot = path->slots[1];
2816 extent_buffer_get(leaf);
2817
2818 wret = push_leaf_left(trans, root, path, 1, 1);
2819 if (wret < 0 && wret != -ENOSPC)
2820 ret = wret;
2821
2822 if (path->nodes[0] == leaf &&
2823 btrfs_header_nritems(leaf)) {
2824 wret = push_leaf_right(trans, root, path, 1, 1);
2825 if (wret < 0 && wret != -ENOSPC)
2826 ret = wret;
2827 }
2828
2829 if (btrfs_header_nritems(leaf) == 0) {
2830 clean_tree_block(trans, root, leaf);
2831 wait_on_tree_block_writeback(root, leaf);
2832
2833 path->slots[1] = slot;
2834 ret = btrfs_del_leaf(trans, root, path, leaf);
2835 BUG_ON(ret);
2836 free_extent_buffer(leaf);
2837
2838 } else {
2839 btrfs_mark_buffer_dirty(leaf);
2840 free_extent_buffer(leaf);
2841 }
2842 } else {
2843 btrfs_mark_buffer_dirty(leaf);
2844 }
2845 }
2846 return ret;
2847 }
2848
2849 /*
2850 * walk up the tree as far as required to find the previous leaf.
2851 * returns 0 if it found something or 1 if there are no lesser leaves.
2852 * returns < 0 on io errors.
2853 */
2854 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2855 {
2856 int slot;
2857 int level = 1;
2858 struct extent_buffer *c;
2859 struct extent_buffer *next = NULL;
2860
2861 while(level < BTRFS_MAX_LEVEL) {
2862 if (!path->nodes[level])
2863 return 1;
2864
2865 slot = path->slots[level];
2866 c = path->nodes[level];
2867 if (slot == 0) {
2868 level++;
2869 if (level == BTRFS_MAX_LEVEL)
2870 return 1;
2871 continue;
2872 }
2873 slot--;
2874
2875 if (next)
2876 free_extent_buffer(next);
2877
2878 next = read_node_slot(root, c, slot);
2879 break;
2880 }
2881 path->slots[level] = slot;
2882 while(1) {
2883 level--;
2884 c = path->nodes[level];
2885 free_extent_buffer(c);
2886 slot = btrfs_header_nritems(next);
2887 if (slot != 0)
2888 slot--;
2889 path->nodes[level] = next;
2890 path->slots[level] = slot;
2891 if (!level)
2892 break;
2893 next = read_node_slot(root, next, slot);
2894 }
2895 return 0;
2896 }
2897
2898 /*
2899 * walk up the tree as far as required to find the next leaf.
2900 * returns 0 if it found something or 1 if there are no greater leaves.
2901 * returns < 0 on io errors.
2902 */
2903 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2904 {
2905 int slot;
2906 int level = 1;
2907 struct extent_buffer *c;
2908 struct extent_buffer *next = NULL;
2909
2910 while(level < BTRFS_MAX_LEVEL) {
2911 if (!path->nodes[level])
2912 return 1;
2913
2914 slot = path->slots[level] + 1;
2915 c = path->nodes[level];
2916 if (slot >= btrfs_header_nritems(c)) {
2917 level++;
2918 if (level == BTRFS_MAX_LEVEL)
2919 return 1;
2920 continue;
2921 }
2922
2923 if (next)
2924 free_extent_buffer(next);
2925
2926 if (path->reada)
2927 reada_for_search(root, path, level, slot, 0);
2928
2929 next = read_node_slot(root, c, slot);
2930 if (!next)
2931 return -EIO;
2932 break;
2933 }
2934 path->slots[level] = slot;
2935 while(1) {
2936 level--;
2937 c = path->nodes[level];
2938 free_extent_buffer(c);
2939 path->nodes[level] = next;
2940 path->slots[level] = 0;
2941 if (!level)
2942 break;
2943 if (path->reada)
2944 reada_for_search(root, path, level, 0, 0);
2945 next = read_node_slot(root, next, 0);
2946 if (!next)
2947 return -EIO;
2948 }
2949 return 0;
2950 }
2951
2952 int btrfs_previous_item(struct btrfs_root *root,
2953 struct btrfs_path *path, u64 min_objectid,
2954 int type)
2955 {
2956 struct btrfs_key found_key;
2957 struct extent_buffer *leaf;
2958 int ret;
2959
2960 while(1) {
2961 if (path->slots[0] == 0) {
2962 ret = btrfs_prev_leaf(root, path);
2963 if (ret != 0)
2964 return ret;
2965 } else {
2966 path->slots[0]--;
2967 }
2968 leaf = path->nodes[0];
2969 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2970 if (found_key.type == type)
2971 return 0;
2972 }
2973 return 1;
2974 }
2975
(deprecated) Btrfs progs developments and patch integration