?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs progs v4.3
[btrfs-progs-unstable/devel.git] / ctree.c
blob46153e3567ff19b3183509137c480962e586e626
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 = kzalloc(sizeof(struct btrfs_path), GFP_NOFS);
46 return path;
47 }
48
49 void btrfs_free_path(struct btrfs_path *p)
50 {
51 if (!p)
52 return;
53 btrfs_release_path(p);
54 kfree(p);
55 }
56
57 void btrfs_release_path(struct btrfs_path *p)
58 {
59 int i;
60 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
61 if (!p->nodes[i])
62 continue;
63 free_extent_buffer(p->nodes[i]);
64 }
65 memset(p, 0, sizeof(*p));
66 }
67
68 void add_root_to_dirty_list(struct btrfs_root *root)
69 {
70 if (root->track_dirty && list_empty(&root->dirty_list)) {
71 list_add(&root->dirty_list,
72 &root->fs_info->dirty_cowonly_roots);
73 }
74 }
75
76 int btrfs_copy_root(struct btrfs_trans_handle *trans,
77 struct btrfs_root *root,
78 struct extent_buffer *buf,
79 struct extent_buffer **cow_ret, u64 new_root_objectid)
80 {
81 struct extent_buffer *cow;
82 int ret = 0;
83 int level;
84 struct btrfs_root *new_root;
85 struct btrfs_disk_key disk_key;
86
87 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
88 if (!new_root)
89 return -ENOMEM;
90
91 memcpy(new_root, root, sizeof(*new_root));
92 new_root->root_key.objectid = new_root_objectid;
93
94 WARN_ON(root->ref_cows && trans->transid !=
95 root->fs_info->running_transaction->transid);
96 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
97
98 level = btrfs_header_level(buf);
99 if (level == 0)
100 btrfs_item_key(buf, &disk_key, 0);
101 else
102 btrfs_node_key(buf, &disk_key, 0);
103 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
104 new_root_objectid, &disk_key,
105 level, buf->start, 0);
106 if (IS_ERR(cow)) {
107 kfree(new_root);
108 return PTR_ERR(cow);
109 }
110
111 copy_extent_buffer(cow, buf, 0, 0, cow->len);
112 btrfs_set_header_bytenr(cow, cow->start);
113 btrfs_set_header_generation(cow, trans->transid);
114 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
115 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
116 BTRFS_HEADER_FLAG_RELOC);
117 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
118 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
119 else
120 btrfs_set_header_owner(cow, new_root_objectid);
121
122 write_extent_buffer(cow, root->fs_info->fsid,
123 btrfs_header_fsid(), BTRFS_FSID_SIZE);
124
125 WARN_ON(btrfs_header_generation(buf) > trans->transid);
126 ret = btrfs_inc_ref(trans, new_root, cow, 0);
127 kfree(new_root);
128
129 if (ret)
130 return ret;
131
132 btrfs_mark_buffer_dirty(cow);
133 *cow_ret = cow;
134 return 0;
135 }
136
137 /*
138 * check if the tree block can be shared by multiple trees
139 */
140 static int btrfs_block_can_be_shared(struct btrfs_root *root,
141 struct extent_buffer *buf)
142 {
143 /*
144 * Tree blocks not in refernece counted trees and tree roots
145 * are never shared. If a block was allocated after the last
146 * snapshot and the block was not allocated by tree relocation,
147 * we know the block is not shared.
148 */
149 if (root->ref_cows &&
150 buf != root->node && buf != root->commit_root &&
151 (btrfs_header_generation(buf) <=
152 btrfs_root_last_snapshot(&root->root_item) ||
153 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
154 return 1;
155 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
156 if (root->ref_cows &&
157 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
158 return 1;
159 #endif
160 return 0;
161 }
162
163 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
164 struct btrfs_root *root,
165 struct extent_buffer *buf,
166 struct extent_buffer *cow)
167 {
168 u64 refs;
169 u64 owner;
170 u64 flags;
171 u64 new_flags = 0;
172 int ret;
173
174 /*
175 * Backrefs update rules:
176 *
177 * Always use full backrefs for extent pointers in tree block
178 * allocated by tree relocation.
179 *
180 * If a shared tree block is no longer referenced by its owner
181 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
182 * use full backrefs for extent pointers in tree block.
183 *
184 * If a tree block is been relocating
185 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
186 * use full backrefs for extent pointers in tree block.
187 * The reason for this is some operations (such as drop tree)
188 * are only allowed for blocks use full backrefs.
189 */
190
191 if (btrfs_block_can_be_shared(root, buf)) {
192 ret = btrfs_lookup_extent_info(trans, root, buf->start,
193 btrfs_header_level(buf), 1,
194 &refs, &flags);
195 BUG_ON(ret);
196 BUG_ON(refs == 0);
197 } else {
198 refs = 1;
199 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
200 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
201 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
202 else
203 flags = 0;
204 }
205
206 owner = btrfs_header_owner(buf);
207 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
208 owner == BTRFS_TREE_RELOC_OBJECTID);
209
210 if (refs > 1) {
211 if ((owner == root->root_key.objectid ||
212 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
213 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
214 ret = btrfs_inc_ref(trans, root, buf, 1);
215 BUG_ON(ret);
216
217 if (root->root_key.objectid ==
218 BTRFS_TREE_RELOC_OBJECTID) {
219 ret = btrfs_dec_ref(trans, root, buf, 0);
220 BUG_ON(ret);
221 ret = btrfs_inc_ref(trans, root, cow, 1);
222 BUG_ON(ret);
223 }
224 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
225 } else {
226
227 if (root->root_key.objectid ==
228 BTRFS_TREE_RELOC_OBJECTID)
229 ret = btrfs_inc_ref(trans, root, cow, 1);
230 else
231 ret = btrfs_inc_ref(trans, root, cow, 0);
232 BUG_ON(ret);
233 }
234 if (new_flags != 0) {
235 ret = btrfs_set_block_flags(trans, root, buf->start,
236 btrfs_header_level(buf),
237 new_flags);
238 BUG_ON(ret);
239 }
240 } else {
241 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
242 if (root->root_key.objectid ==
243 BTRFS_TREE_RELOC_OBJECTID)
244 ret = btrfs_inc_ref(trans, root, cow, 1);
245 else
246 ret = btrfs_inc_ref(trans, root, cow, 0);
247 BUG_ON(ret);
248 ret = btrfs_dec_ref(trans, root, buf, 1);
249 BUG_ON(ret);
250 }
251 clean_tree_block(trans, root, buf);
252 }
253 return 0;
254 }
255
256 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
257 struct btrfs_root *root,
258 struct extent_buffer *buf,
259 struct extent_buffer *parent, int parent_slot,
260 struct extent_buffer **cow_ret,
261 u64 search_start, u64 empty_size)
262 {
263 struct extent_buffer *cow;
264 struct btrfs_disk_key disk_key;
265 int level;
266
267 WARN_ON(root->ref_cows && trans->transid !=
268 root->fs_info->running_transaction->transid);
269 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
270
271 level = btrfs_header_level(buf);
272
273 if (level == 0)
274 btrfs_item_key(buf, &disk_key, 0);
275 else
276 btrfs_node_key(buf, &disk_key, 0);
277
278 cow = btrfs_alloc_free_block(trans, root, buf->len,
279 root->root_key.objectid, &disk_key,
280 level, search_start, empty_size);
281 if (IS_ERR(cow))
282 return PTR_ERR(cow);
283
284 copy_extent_buffer(cow, buf, 0, 0, cow->len);
285 btrfs_set_header_bytenr(cow, cow->start);
286 btrfs_set_header_generation(cow, trans->transid);
287 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
288 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
289 BTRFS_HEADER_FLAG_RELOC);
290 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
291 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
292 else
293 btrfs_set_header_owner(cow, root->root_key.objectid);
294
295 write_extent_buffer(cow, root->fs_info->fsid,
296 btrfs_header_fsid(), BTRFS_FSID_SIZE);
297
298 WARN_ON(!(buf->flags & EXTENT_BAD_TRANSID) &&
299 btrfs_header_generation(buf) > trans->transid);
300
301 update_ref_for_cow(trans, root, buf, cow);
302
303 if (buf == root->node) {
304 root->node = cow;
305 extent_buffer_get(cow);
306
307 btrfs_free_extent(trans, root, buf->start, buf->len,
308 0, root->root_key.objectid, level, 0);
309 free_extent_buffer(buf);
310 add_root_to_dirty_list(root);
311 } else {
312 btrfs_set_node_blockptr(parent, parent_slot,
313 cow->start);
314 WARN_ON(trans->transid == 0);
315 btrfs_set_node_ptr_generation(parent, parent_slot,
316 trans->transid);
317 btrfs_mark_buffer_dirty(parent);
318 WARN_ON(btrfs_header_generation(parent) != trans->transid);
319
320 btrfs_free_extent(trans, root, buf->start, buf->len,
321 0, root->root_key.objectid, level, 1);
322 }
323 if (!list_empty(&buf->recow)) {
324 list_del_init(&buf->recow);
325 free_extent_buffer(buf);
326 }
327 free_extent_buffer(buf);
328 btrfs_mark_buffer_dirty(cow);
329 *cow_ret = cow;
330 return 0;
331 }
332
333 static inline int should_cow_block(struct btrfs_trans_handle *trans,
334 struct btrfs_root *root,
335 struct extent_buffer *buf)
336 {
337 if (btrfs_header_generation(buf) == trans->transid &&
338 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
339 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
340 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
341 return 0;
342 return 1;
343 }
344
345 int btrfs_cow_block(struct btrfs_trans_handle *trans,
346 struct btrfs_root *root, struct extent_buffer *buf,
347 struct extent_buffer *parent, int parent_slot,
348 struct extent_buffer **cow_ret)
349 {
350 u64 search_start;
351 int ret;
352 /*
353 if (trans->transaction != root->fs_info->running_transaction) {
354 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
355 root->fs_info->running_transaction->transid);
356 WARN_ON(1);
357 }
358 */
359 if (trans->transid != root->fs_info->generation) {
360 printk(KERN_CRIT "trans %llu running %llu\n",
361 (unsigned long long)trans->transid,
362 (unsigned long long)root->fs_info->generation);
363 WARN_ON(1);
364 }
365 if (!should_cow_block(trans, root, buf)) {
366 *cow_ret = buf;
367 return 0;
368 }
369
370 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
371 ret = __btrfs_cow_block(trans, root, buf, parent,
372 parent_slot, cow_ret, search_start, 0);
373 return ret;
374 }
375
376 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
377 {
378 if (k1->objectid > k2->objectid)
379 return 1;
380 if (k1->objectid < k2->objectid)
381 return -1;
382 if (k1->type > k2->type)
383 return 1;
384 if (k1->type < k2->type)
385 return -1;
386 if (k1->offset > k2->offset)
387 return 1;
388 if (k1->offset < k2->offset)
389 return -1;
390 return 0;
391 }
392
393 /*
394 * compare two keys in a memcmp fashion
395 */
396 static int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
397 {
398 struct btrfs_key k1;
399
400 btrfs_disk_key_to_cpu(&k1, disk);
401 return btrfs_comp_cpu_keys(&k1, k2);
402 }
403
404 /*
405 * The leaf data grows from end-to-front in the node.
406 * this returns the address of the start of the last item,
407 * which is the stop of the leaf data stack
408 */
409 static inline unsigned int leaf_data_end(struct btrfs_root *root,
410 struct extent_buffer *leaf)
411 {
412 u32 nr = btrfs_header_nritems(leaf);
413 if (nr == 0)
414 return BTRFS_LEAF_DATA_SIZE(root);
415 return btrfs_item_offset_nr(leaf, nr - 1);
416 }
417
418 enum btrfs_tree_block_status
419 btrfs_check_node(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
420 struct extent_buffer *buf)
421 {
422 int i;
423 struct btrfs_key cpukey;
424 struct btrfs_disk_key key;
425 u32 nritems = btrfs_header_nritems(buf);
426 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
427
428 if (nritems == 0 || nritems > BTRFS_NODEPTRS_PER_BLOCK(root))
429 goto fail;
430
431 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
432 if (parent_key && parent_key->type) {
433 btrfs_node_key(buf, &key, 0);
434 if (memcmp(parent_key, &key, sizeof(key)))
435 goto fail;
436 }
437 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
438 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
439 btrfs_node_key(buf, &key, i);
440 btrfs_node_key_to_cpu(buf, &cpukey, i + 1);
441 if (btrfs_comp_keys(&key, &cpukey) >= 0)
442 goto fail;
443 }
444 return BTRFS_TREE_BLOCK_CLEAN;
445 fail:
446 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
447 if (parent_key)
448 btrfs_disk_key_to_cpu(&cpukey, parent_key);
449 else
450 btrfs_node_key_to_cpu(buf, &cpukey, 0);
451 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
452 buf->start, buf->len,
453 btrfs_header_level(buf));
454 }
455 return ret;
456 }
457
458 enum btrfs_tree_block_status
459 btrfs_check_leaf(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
460 struct extent_buffer *buf)
461 {
462 int i;
463 struct btrfs_key cpukey;
464 struct btrfs_disk_key key;
465 u32 nritems = btrfs_header_nritems(buf);
466 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
467
468 if (nritems * sizeof(struct btrfs_item) > buf->len) {
469 fprintf(stderr, "invalid number of items %llu\n",
470 (unsigned long long)buf->start);
471 goto fail;
472 }
473
474 if (btrfs_header_level(buf) != 0) {
475 ret = BTRFS_TREE_BLOCK_INVALID_LEVEL;
476 fprintf(stderr, "leaf is not a leaf %llu\n",
477 (unsigned long long)btrfs_header_bytenr(buf));
478 goto fail;
479 }
480 if (btrfs_leaf_free_space(root, buf) < 0) {
481 ret = BTRFS_TREE_BLOCK_INVALID_FREE_SPACE;
482 fprintf(stderr, "leaf free space incorrect %llu %d\n",
483 (unsigned long long)btrfs_header_bytenr(buf),
484 btrfs_leaf_free_space(root, buf));
485 goto fail;
486 }
487
488 if (nritems == 0)
489 return BTRFS_TREE_BLOCK_CLEAN;
490
491 btrfs_item_key(buf, &key, 0);
492 if (parent_key && parent_key->type &&
493 memcmp(parent_key, &key, sizeof(key))) {
494 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
495 fprintf(stderr, "leaf parent key incorrect %llu\n",
496 (unsigned long long)btrfs_header_bytenr(buf));
497 goto fail;
498 }
499 for (i = 0; nritems > 1 && i < nritems - 1; i++) {
500 btrfs_item_key(buf, &key, i);
501 btrfs_item_key_to_cpu(buf, &cpukey, i + 1);
502 if (btrfs_comp_keys(&key, &cpukey) >= 0) {
503 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
504 fprintf(stderr, "bad key ordering %d %d\n", i, i+1);
505 goto fail;
506 }
507 if (btrfs_item_offset_nr(buf, i) !=
508 btrfs_item_end_nr(buf, i + 1)) {
509 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
510 fprintf(stderr, "incorrect offsets %u %u\n",
511 btrfs_item_offset_nr(buf, i),
512 btrfs_item_end_nr(buf, i + 1));
513 goto fail;
514 }
515 if (i == 0 && btrfs_item_end_nr(buf, i) !=
516 BTRFS_LEAF_DATA_SIZE(root)) {
517 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
518 fprintf(stderr, "bad item end %u wanted %u\n",
519 btrfs_item_end_nr(buf, i),
520 (unsigned)BTRFS_LEAF_DATA_SIZE(root));
521 goto fail;
522 }
523 }
524
525 for (i = 0; i < nritems; i++) {
526 if (btrfs_item_end_nr(buf, i) > BTRFS_LEAF_DATA_SIZE(root)) {
527 btrfs_item_key(buf, &key, 0);
528 btrfs_print_key(&key);
529 fflush(stdout);
530 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
531 fprintf(stderr, "slot end outside of leaf %llu > %llu\n",
532 (unsigned long long)btrfs_item_end_nr(buf, i),
533 (unsigned long long)BTRFS_LEAF_DATA_SIZE(root));
534 goto fail;
535 }
536 }
537
538 return BTRFS_TREE_BLOCK_CLEAN;
539 fail:
540 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
541 if (parent_key)
542 btrfs_disk_key_to_cpu(&cpukey, parent_key);
543 else
544 btrfs_item_key_to_cpu(buf, &cpukey, 0);
545
546 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
547 buf->start, buf->len, 0);
548 }
549 return ret;
550 }
551
552 static int noinline check_block(struct btrfs_root *root,
553 struct btrfs_path *path, int level)
554 {
555 struct btrfs_disk_key key;
556 struct btrfs_disk_key *key_ptr = NULL;
557 struct extent_buffer *parent;
558 enum btrfs_tree_block_status ret;
559
560 if (path->skip_check_block)
561 return 0;
562 if (path->nodes[level + 1]) {
563 parent = path->nodes[level + 1];
564 btrfs_node_key(parent, &key, path->slots[level + 1]);
565 key_ptr = &key;
566 }
567 if (level == 0)
568 ret = btrfs_check_leaf(root, key_ptr, path->nodes[0]);
569 else
570 ret = btrfs_check_node(root, key_ptr, path->nodes[level]);
571 if (ret == BTRFS_TREE_BLOCK_CLEAN)
572 return 0;
573 return -EIO;
574 }
575
576 /*
577 * search for key in the extent_buffer. The items start at offset p,
578 * and they are item_size apart. There are 'max' items in p.
579 *
580 * the slot in the array is returned via slot, and it points to
581 * the place where you would insert key if it is not found in
582 * the array.
583 *
584 * slot may point to max if the key is bigger than all of the keys
585 */
586 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
587 int item_size, struct btrfs_key *key,
588 int max, int *slot)
589 {
590 int low = 0;
591 int high = max;
592 int mid;
593 int ret;
594 unsigned long offset;
595 struct btrfs_disk_key *tmp;
596
597 while(low < high) {
598 mid = (low + high) / 2;
599 offset = p + mid * item_size;
600
601 tmp = (struct btrfs_disk_key *)(eb->data + offset);
602 ret = btrfs_comp_keys(tmp, key);
603
604 if (ret < 0)
605 low = mid + 1;
606 else if (ret > 0)
607 high = mid;
608 else {
609 *slot = mid;
610 return 0;
611 }
612 }
613 *slot = low;
614 return 1;
615 }
616
617 /*
618 * simple bin_search frontend that does the right thing for
619 * leaves vs nodes
620 */
621 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
622 int level, int *slot)
623 {
624 if (level == 0)
625 return generic_bin_search(eb,
626 offsetof(struct btrfs_leaf, items),
627 sizeof(struct btrfs_item),
628 key, btrfs_header_nritems(eb),
629 slot);
630 else
631 return generic_bin_search(eb,
632 offsetof(struct btrfs_node, ptrs),
633 sizeof(struct btrfs_key_ptr),
634 key, btrfs_header_nritems(eb),
635 slot);
636 }
637
638 struct extent_buffer *read_node_slot(struct btrfs_root *root,
639 struct extent_buffer *parent, int slot)
640 {
641 int level = btrfs_header_level(parent);
642 if (slot < 0)
643 return NULL;
644 if (slot >= btrfs_header_nritems(parent))
645 return NULL;
646
647 if (level == 0)
648 return NULL;
649
650 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
651 btrfs_level_size(root, level - 1),
652 btrfs_node_ptr_generation(parent, slot));
653 }
654
655 static int balance_level(struct btrfs_trans_handle *trans,
656 struct btrfs_root *root,
657 struct btrfs_path *path, int level)
658 {
659 struct extent_buffer *right = NULL;
660 struct extent_buffer *mid;
661 struct extent_buffer *left = NULL;
662 struct extent_buffer *parent = NULL;
663 int ret = 0;
664 int wret;
665 int pslot;
666 int orig_slot = path->slots[level];
667 u64 orig_ptr;
668
669 if (level == 0)
670 return 0;
671
672 mid = path->nodes[level];
673 WARN_ON(btrfs_header_generation(mid) != trans->transid);
674
675 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
676
677 if (level < BTRFS_MAX_LEVEL - 1) {
678 parent = path->nodes[level + 1];
679 pslot = path->slots[level + 1];
680 }
681
682 /*
683 * deal with the case where there is only one pointer in the root
684 * by promoting the node below to a root
685 */
686 if (!parent) {
687 struct extent_buffer *child;
688
689 if (btrfs_header_nritems(mid) != 1)
690 return 0;
691
692 /* promote the child to a root */
693 child = read_node_slot(root, mid, 0);
694 BUG_ON(!extent_buffer_uptodate(child));
695 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
696 BUG_ON(ret);
697
698 root->node = child;
699 add_root_to_dirty_list(root);
700 path->nodes[level] = NULL;
701 clean_tree_block(trans, root, mid);
702 wait_on_tree_block_writeback(root, mid);
703 /* once for the path */
704 free_extent_buffer(mid);
705
706 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
707 0, root->root_key.objectid,
708 level, 1);
709 /* once for the root ptr */
710 free_extent_buffer(mid);
711 return ret;
712 }
713 if (btrfs_header_nritems(mid) >
714 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
715 return 0;
716
717 left = read_node_slot(root, parent, pslot - 1);
718 if (extent_buffer_uptodate(left)) {
719 wret = btrfs_cow_block(trans, root, left,
720 parent, pslot - 1, &left);
721 if (wret) {
722 ret = wret;
723 goto enospc;
724 }
725 }
726 right = read_node_slot(root, parent, pslot + 1);
727 if (extent_buffer_uptodate(right)) {
728 wret = btrfs_cow_block(trans, root, right,
729 parent, pslot + 1, &right);
730 if (wret) {
731 ret = wret;
732 goto enospc;
733 }
734 }
735
736 /* first, try to make some room in the middle buffer */
737 if (left) {
738 orig_slot += btrfs_header_nritems(left);
739 wret = push_node_left(trans, root, left, mid, 1);
740 if (wret < 0)
741 ret = wret;
742 }
743
744 /*
745 * then try to empty the right most buffer into the middle
746 */
747 if (right) {
748 wret = push_node_left(trans, root, mid, right, 1);
749 if (wret < 0 && wret != -ENOSPC)
750 ret = wret;
751 if (btrfs_header_nritems(right) == 0) {
752 u64 bytenr = right->start;
753 u32 blocksize = right->len;
754
755 clean_tree_block(trans, root, right);
756 wait_on_tree_block_writeback(root, right);
757 free_extent_buffer(right);
758 right = NULL;
759 wret = btrfs_del_ptr(trans, root, path,
760 level + 1, pslot + 1);
761 if (wret)
762 ret = wret;
763 wret = btrfs_free_extent(trans, root, bytenr,
764 blocksize, 0,
765 root->root_key.objectid,
766 level, 0);
767 if (wret)
768 ret = wret;
769 } else {
770 struct btrfs_disk_key right_key;
771 btrfs_node_key(right, &right_key, 0);
772 btrfs_set_node_key(parent, &right_key, pslot + 1);
773 btrfs_mark_buffer_dirty(parent);
774 }
775 }
776 if (btrfs_header_nritems(mid) == 1) {
777 /*
778 * we're not allowed to leave a node with one item in the
779 * tree during a delete. A deletion from lower in the tree
780 * could try to delete the only pointer in this node.
781 * So, pull some keys from the left.
782 * There has to be a left pointer at this point because
783 * otherwise we would have pulled some pointers from the
784 * right
785 */
786 BUG_ON(!left);
787 wret = balance_node_right(trans, root, mid, left);
788 if (wret < 0) {
789 ret = wret;
790 goto enospc;
791 }
792 if (wret == 1) {
793 wret = push_node_left(trans, root, left, mid, 1);
794 if (wret < 0)
795 ret = wret;
796 }
797 BUG_ON(wret == 1);
798 }
799 if (btrfs_header_nritems(mid) == 0) {
800 /* we've managed to empty the middle node, drop it */
801 u64 bytenr = mid->start;
802 u32 blocksize = mid->len;
803 clean_tree_block(trans, root, mid);
804 wait_on_tree_block_writeback(root, mid);
805 free_extent_buffer(mid);
806 mid = NULL;
807 wret = btrfs_del_ptr(trans, root, path, level + 1, pslot);
808 if (wret)
809 ret = wret;
810 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
811 0, root->root_key.objectid,
812 level, 0);
813 if (wret)
814 ret = wret;
815 } else {
816 /* update the parent key to reflect our changes */
817 struct btrfs_disk_key mid_key;
818 btrfs_node_key(mid, &mid_key, 0);
819 btrfs_set_node_key(parent, &mid_key, pslot);
820 btrfs_mark_buffer_dirty(parent);
821 }
822
823 /* update the path */
824 if (left) {
825 if (btrfs_header_nritems(left) > orig_slot) {
826 extent_buffer_get(left);
827 path->nodes[level] = left;
828 path->slots[level + 1] -= 1;
829 path->slots[level] = orig_slot;
830 if (mid)
831 free_extent_buffer(mid);
832 } else {
833 orig_slot -= btrfs_header_nritems(left);
834 path->slots[level] = orig_slot;
835 }
836 }
837 /* double check we haven't messed things up */
838 check_block(root, path, level);
839 if (orig_ptr !=
840 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
841 BUG();
842 enospc:
843 if (right)
844 free_extent_buffer(right);
845 if (left)
846 free_extent_buffer(left);
847 return ret;
848 }
849
850 /* returns zero if the push worked, non-zero otherwise */
851 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
852 struct btrfs_root *root,
853 struct btrfs_path *path, int level)
854 {
855 struct extent_buffer *right = NULL;
856 struct extent_buffer *mid;
857 struct extent_buffer *left = NULL;
858 struct extent_buffer *parent = NULL;
859 int ret = 0;
860 int wret;
861 int pslot;
862 int orig_slot = path->slots[level];
863
864 if (level == 0)
865 return 1;
866
867 mid = path->nodes[level];
868 WARN_ON(btrfs_header_generation(mid) != trans->transid);
869
870 if (level < BTRFS_MAX_LEVEL - 1) {
871 parent = path->nodes[level + 1];
872 pslot = path->slots[level + 1];
873 }
874
875 if (!parent)
876 return 1;
877
878 left = read_node_slot(root, parent, pslot - 1);
879
880 /* first, try to make some room in the middle buffer */
881 if (extent_buffer_uptodate(left)) {
882 u32 left_nr;
883 left_nr = btrfs_header_nritems(left);
884 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
885 wret = 1;
886 } else {
887 ret = btrfs_cow_block(trans, root, left, parent,
888 pslot - 1, &left);
889 if (ret)
890 wret = 1;
891 else {
892 wret = push_node_left(trans, root,
893 left, mid, 0);
894 }
895 }
896 if (wret < 0)
897 ret = wret;
898 if (wret == 0) {
899 struct btrfs_disk_key disk_key;
900 orig_slot += left_nr;
901 btrfs_node_key(mid, &disk_key, 0);
902 btrfs_set_node_key(parent, &disk_key, pslot);
903 btrfs_mark_buffer_dirty(parent);
904 if (btrfs_header_nritems(left) > orig_slot) {
905 path->nodes[level] = left;
906 path->slots[level + 1] -= 1;
907 path->slots[level] = orig_slot;
908 free_extent_buffer(mid);
909 } else {
910 orig_slot -=
911 btrfs_header_nritems(left);
912 path->slots[level] = orig_slot;
913 free_extent_buffer(left);
914 }
915 return 0;
916 }
917 free_extent_buffer(left);
918 }
919 right= read_node_slot(root, parent, pslot + 1);
920
921 /*
922 * then try to empty the right most buffer into the middle
923 */
924 if (extent_buffer_uptodate(right)) {
925 u32 right_nr;
926 right_nr = btrfs_header_nritems(right);
927 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
928 wret = 1;
929 } else {
930 ret = btrfs_cow_block(trans, root, right,
931 parent, pslot + 1,
932 &right);
933 if (ret)
934 wret = 1;
935 else {
936 wret = balance_node_right(trans, root,
937 right, mid);
938 }
939 }
940 if (wret < 0)
941 ret = wret;
942 if (wret == 0) {
943 struct btrfs_disk_key disk_key;
944
945 btrfs_node_key(right, &disk_key, 0);
946 btrfs_set_node_key(parent, &disk_key, pslot + 1);
947 btrfs_mark_buffer_dirty(parent);
948
949 if (btrfs_header_nritems(mid) <= orig_slot) {
950 path->nodes[level] = right;
951 path->slots[level + 1] += 1;
952 path->slots[level] = orig_slot -
953 btrfs_header_nritems(mid);
954 free_extent_buffer(mid);
955 } else {
956 free_extent_buffer(right);
957 }
958 return 0;
959 }
960 free_extent_buffer(right);
961 }
962 return 1;
963 }
964
965 /*
966 * readahead one full node of leaves
967 */
968 void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
969 int level, int slot, u64 objectid)
970 {
971 struct extent_buffer *node;
972 struct btrfs_disk_key disk_key;
973 u32 nritems;
974 u64 search;
975 u64 lowest_read;
976 u64 highest_read;
977 u64 nread = 0;
978 int direction = path->reada;
979 struct extent_buffer *eb;
980 u32 nr;
981 u32 blocksize;
982 u32 nscan = 0;
983
984 if (level != 1)
985 return;
986
987 if (!path->nodes[level])
988 return;
989
990 node = path->nodes[level];
991 search = btrfs_node_blockptr(node, slot);
992 blocksize = btrfs_level_size(root, level - 1);
993 eb = btrfs_find_tree_block(root, search, blocksize);
994 if (eb) {
995 free_extent_buffer(eb);
996 return;
997 }
998
999 highest_read = search;
1000 lowest_read = search;
1001
1002 nritems = btrfs_header_nritems(node);
1003 nr = slot;
1004 while(1) {
1005 if (direction < 0) {
1006 if (nr == 0)
1007 break;
1008 nr--;
1009 } else if (direction > 0) {
1010 nr++;
1011 if (nr >= nritems)
1012 break;
1013 }
1014 if (path->reada < 0 && objectid) {
1015 btrfs_node_key(node, &disk_key, nr);
1016 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1017 break;
1018 }
1019 search = btrfs_node_blockptr(node, nr);
1020 if ((search >= lowest_read && search <= highest_read) ||
1021 (search < lowest_read && lowest_read - search <= 32768) ||
1022 (search > highest_read && search - highest_read <= 32768)) {
1023 readahead_tree_block(root, search, blocksize,
1024 btrfs_node_ptr_generation(node, nr));
1025 nread += blocksize;
1026 }
1027 nscan++;
1028 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1029 break;
1030 if(nread > (1024 * 1024) || nscan > 128)
1031 break;
1032
1033 if (search < lowest_read)
1034 lowest_read = search;
1035 if (search > highest_read)
1036 highest_read = search;
1037 }
1038 }
1039
1040 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
1041 u64 iobjectid, u64 ioff, u8 key_type,
1042 struct btrfs_key *found_key)
1043 {
1044 int ret;
1045 struct btrfs_key key;
1046 struct extent_buffer *eb;
1047 struct btrfs_path *path;
1048
1049 key.type = key_type;
1050 key.objectid = iobjectid;
1051 key.offset = ioff;
1052
1053 if (found_path == NULL) {
1054 path = btrfs_alloc_path();
1055 if (!path)
1056 return -ENOMEM;
1057 } else
1058 path = found_path;
1059
1060 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1061 if ((ret < 0) || (found_key == NULL))
1062 goto out;
1063
1064 eb = path->nodes[0];
1065 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1066 ret = btrfs_next_leaf(fs_root, path);
1067 if (ret)
1068 goto out;
1069 eb = path->nodes[0];
1070 }
1071
1072 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1073 if (found_key->type != key.type ||
1074 found_key->objectid != key.objectid) {
1075 ret = 1;
1076 goto out;
1077 }
1078
1079 out:
1080 if (path != found_path)
1081 btrfs_free_path(path);
1082 return ret;
1083 }
1084
1085 /*
1086 * look for key in the tree. path is filled in with nodes along the way
1087 * if key is found, we return zero and you can find the item in the leaf
1088 * level of the path (level 0)
1089 *
1090 * If the key isn't found, the path points to the slot where it should
1091 * be inserted, and 1 is returned. If there are other errors during the
1092 * search a negative error number is returned.
1093 *
1094 * if ins_len > 0, nodes and leaves will be split as we walk down the
1095 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1096 * possible)
1097 */
1098 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1099 *root, struct btrfs_key *key, struct btrfs_path *p, int
1100 ins_len, int cow)
1101 {
1102 struct extent_buffer *b;
1103 int slot;
1104 int ret;
1105 int level;
1106 int should_reada = p->reada;
1107 u8 lowest_level = 0;
1108
1109 lowest_level = p->lowest_level;
1110 WARN_ON(lowest_level && ins_len > 0);
1111 WARN_ON(p->nodes[0] != NULL);
1112 /*
1113 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1114 */
1115 again:
1116 b = root->node;
1117 extent_buffer_get(b);
1118 while (b) {
1119 level = btrfs_header_level(b);
1120 if (cow) {
1121 int wret;
1122 wret = btrfs_cow_block(trans, root, b,
1123 p->nodes[level + 1],
1124 p->slots[level + 1],
1125 &b);
1126 if (wret) {
1127 free_extent_buffer(b);
1128 return wret;
1129 }
1130 }
1131 BUG_ON(!cow && ins_len);
1132 if (level != btrfs_header_level(b))
1133 WARN_ON(1);
1134 level = btrfs_header_level(b);
1135 p->nodes[level] = b;
1136 ret = check_block(root, p, level);
1137 if (ret)
1138 return -1;
1139 ret = bin_search(b, key, level, &slot);
1140 if (level != 0) {
1141 if (ret && slot > 0)
1142 slot -= 1;
1143 p->slots[level] = slot;
1144 if ((p->search_for_split || ins_len > 0) &&
1145 btrfs_header_nritems(b) >=
1146 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1147 int sret = split_node(trans, root, p, level);
1148 BUG_ON(sret > 0);
1149 if (sret)
1150 return sret;
1151 b = p->nodes[level];
1152 slot = p->slots[level];
1153 } else if (ins_len < 0) {
1154 int sret = balance_level(trans, root, p,
1155 level);
1156 if (sret)
1157 return sret;
1158 b = p->nodes[level];
1159 if (!b) {
1160 btrfs_release_path(p);
1161 goto again;
1162 }
1163 slot = p->slots[level];
1164 BUG_ON(btrfs_header_nritems(b) == 1);
1165 }
1166 /* this is only true while dropping a snapshot */
1167 if (level == lowest_level)
1168 break;
1169
1170 if (should_reada)
1171 reada_for_search(root, p, level, slot,
1172 key->objectid);
1173
1174 b = read_node_slot(root, b, slot);
1175 if (!extent_buffer_uptodate(b))
1176 return -EIO;
1177 } else {
1178 p->slots[level] = slot;
1179 if (ins_len > 0 &&
1180 ins_len > btrfs_leaf_free_space(root, b)) {
1181 int sret = split_leaf(trans, root, key,
1182 p, ins_len, ret == 0);
1183 BUG_ON(sret > 0);
1184 if (sret)
1185 return sret;
1186 }
1187 return ret;
1188 }
1189 }
1190 return 1;
1191 }
1192
1193 /*
1194 * adjust the pointers going up the tree, starting at level
1195 * making sure the right key of each node is points to 'key'.
1196 * This is used after shifting pointers to the left, so it stops
1197 * fixing up pointers when a given leaf/node is not in slot 0 of the
1198 * higher levels
1199 */
1200 void btrfs_fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
1201 struct btrfs_disk_key *key, int level)
1202 {
1203 int i;
1204 struct extent_buffer *t;
1205
1206 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1207 int tslot = path->slots[i];
1208 if (!path->nodes[i])
1209 break;
1210 t = path->nodes[i];
1211 btrfs_set_node_key(t, key, tslot);
1212 btrfs_mark_buffer_dirty(path->nodes[i]);
1213 if (tslot != 0)
1214 break;
1215 }
1216 }
1217
1218 /*
1219 * update item key.
1220 *
1221 * This function isn't completely safe. It's the caller's responsibility
1222 * that the new key won't break the order
1223 */
1224 int btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
1225 struct btrfs_key *new_key)
1226 {
1227 struct btrfs_disk_key disk_key;
1228 struct extent_buffer *eb;
1229 int slot;
1230
1231 eb = path->nodes[0];
1232 slot = path->slots[0];
1233 if (slot > 0) {
1234 btrfs_item_key(eb, &disk_key, slot - 1);
1235 if (btrfs_comp_keys(&disk_key, new_key) >= 0)
1236 return -1;
1237 }
1238 if (slot < btrfs_header_nritems(eb) - 1) {
1239 btrfs_item_key(eb, &disk_key, slot + 1);
1240 if (btrfs_comp_keys(&disk_key, new_key) <= 0)
1241 return -1;
1242 }
1243
1244 btrfs_cpu_key_to_disk(&disk_key, new_key);
1245 btrfs_set_item_key(eb, &disk_key, slot);
1246 btrfs_mark_buffer_dirty(eb);
1247 if (slot == 0)
1248 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1249 return 0;
1250 }
1251
1252 /*
1253 * update an item key without the safety checks. This is meant to be called by
1254 * fsck only.
1255 */
1256 void btrfs_set_item_key_unsafe(struct btrfs_root *root,
1257 struct btrfs_path *path,
1258 struct btrfs_key *new_key)
1259 {
1260 struct btrfs_disk_key disk_key;
1261 struct extent_buffer *eb;
1262 int slot;
1263
1264 eb = path->nodes[0];
1265 slot = path->slots[0];
1266
1267 btrfs_cpu_key_to_disk(&disk_key, new_key);
1268 btrfs_set_item_key(eb, &disk_key, slot);
1269 btrfs_mark_buffer_dirty(eb);
1270 if (slot == 0)
1271 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1272 }
1273
1274 /*
1275 * try to push data from one node into the next node left in the
1276 * tree.
1277 *
1278 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1279 * error, and > 0 if there was no room in the left hand block.
1280 */
1281 static int push_node_left(struct btrfs_trans_handle *trans,
1282 struct btrfs_root *root, struct extent_buffer *dst,
1283 struct extent_buffer *src, int empty)
1284 {
1285 int push_items = 0;
1286 int src_nritems;
1287 int dst_nritems;
1288 int ret = 0;
1289
1290 src_nritems = btrfs_header_nritems(src);
1291 dst_nritems = btrfs_header_nritems(dst);
1292 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1293 WARN_ON(btrfs_header_generation(src) != trans->transid);
1294 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1295
1296 if (!empty && src_nritems <= 8)
1297 return 1;
1298
1299 if (push_items <= 0) {
1300 return 1;
1301 }
1302
1303 if (empty) {
1304 push_items = min(src_nritems, push_items);
1305 if (push_items < src_nritems) {
1306 /* leave at least 8 pointers in the node if
1307 * we aren't going to empty it
1308 */
1309 if (src_nritems - push_items < 8) {
1310 if (push_items <= 8)
1311 return 1;
1312 push_items -= 8;
1313 }
1314 }
1315 } else
1316 push_items = min(src_nritems - 8, push_items);
1317
1318 copy_extent_buffer(dst, src,
1319 btrfs_node_key_ptr_offset(dst_nritems),
1320 btrfs_node_key_ptr_offset(0),
1321 push_items * sizeof(struct btrfs_key_ptr));
1322
1323 if (push_items < src_nritems) {
1324 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1325 btrfs_node_key_ptr_offset(push_items),
1326 (src_nritems - push_items) *
1327 sizeof(struct btrfs_key_ptr));
1328 }
1329 btrfs_set_header_nritems(src, src_nritems - push_items);
1330 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1331 btrfs_mark_buffer_dirty(src);
1332 btrfs_mark_buffer_dirty(dst);
1333
1334 return ret;
1335 }
1336
1337 /*
1338 * try to push data from one node into the next node right in the
1339 * tree.
1340 *
1341 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1342 * error, and > 0 if there was no room in the right hand block.
1343 *
1344 * this will only push up to 1/2 the contents of the left node over
1345 */
1346 static int balance_node_right(struct btrfs_trans_handle *trans,
1347 struct btrfs_root *root,
1348 struct extent_buffer *dst,
1349 struct extent_buffer *src)
1350 {
1351 int push_items = 0;
1352 int max_push;
1353 int src_nritems;
1354 int dst_nritems;
1355 int ret = 0;
1356
1357 WARN_ON(btrfs_header_generation(src) != trans->transid);
1358 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1359
1360 src_nritems = btrfs_header_nritems(src);
1361 dst_nritems = btrfs_header_nritems(dst);
1362 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1363 if (push_items <= 0) {
1364 return 1;
1365 }
1366
1367 if (src_nritems < 4) {
1368 return 1;
1369 }
1370
1371 max_push = src_nritems / 2 + 1;
1372 /* don't try to empty the node */
1373 if (max_push >= src_nritems) {
1374 return 1;
1375 }
1376
1377 if (max_push < push_items)
1378 push_items = max_push;
1379
1380 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1381 btrfs_node_key_ptr_offset(0),
1382 (dst_nritems) *
1383 sizeof(struct btrfs_key_ptr));
1384
1385 copy_extent_buffer(dst, src,
1386 btrfs_node_key_ptr_offset(0),
1387 btrfs_node_key_ptr_offset(src_nritems - push_items),
1388 push_items * sizeof(struct btrfs_key_ptr));
1389
1390 btrfs_set_header_nritems(src, src_nritems - push_items);
1391 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1392
1393 btrfs_mark_buffer_dirty(src);
1394 btrfs_mark_buffer_dirty(dst);
1395
1396 return ret;
1397 }
1398
1399 /*
1400 * helper function to insert a new root level in the tree.
1401 * A new node is allocated, and a single item is inserted to
1402 * point to the existing root
1403 *
1404 * returns zero on success or < 0 on failure.
1405 */
1406 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1407 struct btrfs_root *root,
1408 struct btrfs_path *path, int level)
1409 {
1410 u64 lower_gen;
1411 struct extent_buffer *lower;
1412 struct extent_buffer *c;
1413 struct extent_buffer *old;
1414 struct btrfs_disk_key lower_key;
1415
1416 BUG_ON(path->nodes[level]);
1417 BUG_ON(path->nodes[level-1] != root->node);
1418
1419 lower = path->nodes[level-1];
1420 if (level == 1)
1421 btrfs_item_key(lower, &lower_key, 0);
1422 else
1423 btrfs_node_key(lower, &lower_key, 0);
1424
1425 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1426 root->root_key.objectid, &lower_key,
1427 level, root->node->start, 0);
1428
1429 if (IS_ERR(c))
1430 return PTR_ERR(c);
1431
1432 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
1433 btrfs_set_header_nritems(c, 1);
1434 btrfs_set_header_level(c, level);
1435 btrfs_set_header_bytenr(c, c->start);
1436 btrfs_set_header_generation(c, trans->transid);
1437 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
1438 btrfs_set_header_owner(c, root->root_key.objectid);
1439
1440 write_extent_buffer(c, root->fs_info->fsid,
1441 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1442
1443 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1444 btrfs_header_chunk_tree_uuid(c),
1445 BTRFS_UUID_SIZE);
1446
1447 btrfs_set_node_key(c, &lower_key, 0);
1448 btrfs_set_node_blockptr(c, 0, lower->start);
1449 lower_gen = btrfs_header_generation(lower);
1450 WARN_ON(lower_gen != trans->transid);
1451
1452 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1453
1454 btrfs_mark_buffer_dirty(c);
1455
1456 old = root->node;
1457 root->node = c;
1458
1459 /* the super has an extra ref to root->node */
1460 free_extent_buffer(old);
1461
1462 add_root_to_dirty_list(root);
1463 extent_buffer_get(c);
1464 path->nodes[level] = c;
1465 path->slots[level] = 0;
1466 return 0;
1467 }
1468
1469 /*
1470 * worker function to insert a single pointer in a node.
1471 * the node should have enough room for the pointer already
1472 *
1473 * slot and level indicate where you want the key to go, and
1474 * blocknr is the block the key points to.
1475 *
1476 * returns zero on success and < 0 on any error
1477 */
1478 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1479 *root, struct btrfs_path *path, struct btrfs_disk_key
1480 *key, u64 bytenr, int slot, int level)
1481 {
1482 struct extent_buffer *lower;
1483 int nritems;
1484
1485 BUG_ON(!path->nodes[level]);
1486 lower = path->nodes[level];
1487 nritems = btrfs_header_nritems(lower);
1488 if (slot > nritems)
1489 BUG();
1490 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1491 BUG();
1492 if (slot != nritems) {
1493 memmove_extent_buffer(lower,
1494 btrfs_node_key_ptr_offset(slot + 1),
1495 btrfs_node_key_ptr_offset(slot),
1496 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1497 }
1498 btrfs_set_node_key(lower, key, slot);
1499 btrfs_set_node_blockptr(lower, slot, bytenr);
1500 WARN_ON(trans->transid == 0);
1501 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1502 btrfs_set_header_nritems(lower, nritems + 1);
1503 btrfs_mark_buffer_dirty(lower);
1504 return 0;
1505 }
1506
1507 /*
1508 * split the node at the specified level in path in two.
1509 * The path is corrected to point to the appropriate node after the split
1510 *
1511 * Before splitting this tries to make some room in the node by pushing
1512 * left and right, if either one works, it returns right away.
1513 *
1514 * returns 0 on success and < 0 on failure
1515 */
1516 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1517 *root, struct btrfs_path *path, int level)
1518 {
1519 struct extent_buffer *c;
1520 struct extent_buffer *split;
1521 struct btrfs_disk_key disk_key;
1522 int mid;
1523 int ret;
1524 int wret;
1525 u32 c_nritems;
1526
1527 c = path->nodes[level];
1528 WARN_ON(btrfs_header_generation(c) != trans->transid);
1529 if (c == root->node) {
1530 /* trying to split the root, lets make a new one */
1531 ret = insert_new_root(trans, root, path, level + 1);
1532 if (ret)
1533 return ret;
1534 } else {
1535 ret = push_nodes_for_insert(trans, root, path, level);
1536 c = path->nodes[level];
1537 if (!ret && btrfs_header_nritems(c) <
1538 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1539 return 0;
1540 if (ret < 0)
1541 return ret;
1542 }
1543
1544 c_nritems = btrfs_header_nritems(c);
1545 mid = (c_nritems + 1) / 2;
1546 btrfs_node_key(c, &disk_key, mid);
1547
1548 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1549 root->root_key.objectid,
1550 &disk_key, level, c->start, 0);
1551 if (IS_ERR(split))
1552 return PTR_ERR(split);
1553
1554 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
1555 btrfs_set_header_level(split, btrfs_header_level(c));
1556 btrfs_set_header_bytenr(split, split->start);
1557 btrfs_set_header_generation(split, trans->transid);
1558 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
1559 btrfs_set_header_owner(split, root->root_key.objectid);
1560 write_extent_buffer(split, root->fs_info->fsid,
1561 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1562 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1563 btrfs_header_chunk_tree_uuid(split),
1564 BTRFS_UUID_SIZE);
1565
1566
1567 copy_extent_buffer(split, c,
1568 btrfs_node_key_ptr_offset(0),
1569 btrfs_node_key_ptr_offset(mid),
1570 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1571 btrfs_set_header_nritems(split, c_nritems - mid);
1572 btrfs_set_header_nritems(c, mid);
1573 ret = 0;
1574
1575 btrfs_mark_buffer_dirty(c);
1576 btrfs_mark_buffer_dirty(split);
1577
1578 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1579 path->slots[level + 1] + 1,
1580 level + 1);
1581 if (wret)
1582 ret = wret;
1583
1584 if (path->slots[level] >= mid) {
1585 path->slots[level] -= mid;
1586 free_extent_buffer(c);
1587 path->nodes[level] = split;
1588 path->slots[level + 1] += 1;
1589 } else {
1590 free_extent_buffer(split);
1591 }
1592 return ret;
1593 }
1594
1595 /*
1596 * how many bytes are required to store the items in a leaf. start
1597 * and nr indicate which items in the leaf to check. This totals up the
1598 * space used both by the item structs and the item data
1599 */
1600 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1601 {
1602 int data_len;
1603 int nritems = btrfs_header_nritems(l);
1604 int end = min(nritems, start + nr) - 1;
1605
1606 if (!nr)
1607 return 0;
1608 data_len = btrfs_item_end_nr(l, start);
1609 data_len = data_len - btrfs_item_offset_nr(l, end);
1610 data_len += sizeof(struct btrfs_item) * nr;
1611 WARN_ON(data_len < 0);
1612 return data_len;
1613 }
1614
1615 /*
1616 * The space between the end of the leaf items and
1617 * the start of the leaf data. IOW, how much room
1618 * the leaf has left for both items and data
1619 */
1620 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1621 {
1622 u32 nodesize = (root ? BTRFS_LEAF_DATA_SIZE(root) : leaf->len);
1623 int nritems = btrfs_header_nritems(leaf);
1624 int ret;
1625 ret = nodesize - leaf_space_used(leaf, 0, nritems);
1626 if (ret < 0) {
1627 printk("leaf free space ret %d, leaf data size %u, used %d nritems %d\n",
1628 ret, nodesize, leaf_space_used(leaf, 0, nritems),
1629 nritems);
1630 }
1631 return ret;
1632 }
1633
1634 /*
1635 * push some data in the path leaf to the right, trying to free up at
1636 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1637 *
1638 * returns 1 if the push failed because the other node didn't have enough
1639 * room, 0 if everything worked out and < 0 if there were major errors.
1640 */
1641 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1642 *root, struct btrfs_path *path, int data_size,
1643 int empty)
1644 {
1645 struct extent_buffer *left = path->nodes[0];
1646 struct extent_buffer *right;
1647 struct extent_buffer *upper;
1648 struct btrfs_disk_key disk_key;
1649 int slot;
1650 u32 i;
1651 int free_space;
1652 int push_space = 0;
1653 int push_items = 0;
1654 struct btrfs_item *item;
1655 u32 left_nritems;
1656 u32 nr;
1657 u32 right_nritems;
1658 u32 data_end;
1659 u32 this_item_size;
1660 int ret;
1661
1662 slot = path->slots[1];
1663 if (!path->nodes[1]) {
1664 return 1;
1665 }
1666 upper = path->nodes[1];
1667 if (slot >= btrfs_header_nritems(upper) - 1)
1668 return 1;
1669
1670 right = read_node_slot(root, upper, slot + 1);
1671 if (!extent_buffer_uptodate(right)) {
1672 if (IS_ERR(right))
1673 return PTR_ERR(right);
1674 return -EIO;
1675 }
1676 free_space = btrfs_leaf_free_space(root, right);
1677 if (free_space < data_size) {
1678 free_extent_buffer(right);
1679 return 1;
1680 }
1681
1682 /* cow and double check */
1683 ret = btrfs_cow_block(trans, root, right, upper,
1684 slot + 1, &right);
1685 if (ret) {
1686 free_extent_buffer(right);
1687 return 1;
1688 }
1689 free_space = btrfs_leaf_free_space(root, right);
1690 if (free_space < data_size) {
1691 free_extent_buffer(right);
1692 return 1;
1693 }
1694
1695 left_nritems = btrfs_header_nritems(left);
1696 if (left_nritems == 0) {
1697 free_extent_buffer(right);
1698 return 1;
1699 }
1700
1701 if (empty)
1702 nr = 0;
1703 else
1704 nr = 1;
1705
1706 i = left_nritems - 1;
1707 while (i >= nr) {
1708 item = btrfs_item_nr(i);
1709
1710 if (path->slots[0] == i)
1711 push_space += data_size + sizeof(*item);
1712
1713 this_item_size = btrfs_item_size(left, item);
1714 if (this_item_size + sizeof(*item) + push_space > free_space)
1715 break;
1716 push_items++;
1717 push_space += this_item_size + sizeof(*item);
1718 if (i == 0)
1719 break;
1720 i--;
1721 }
1722
1723 if (push_items == 0) {
1724 free_extent_buffer(right);
1725 return 1;
1726 }
1727
1728 if (!empty && push_items == left_nritems)
1729 WARN_ON(1);
1730
1731 /* push left to right */
1732 right_nritems = btrfs_header_nritems(right);
1733
1734 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1735 push_space -= leaf_data_end(root, left);
1736
1737 /* make room in the right data area */
1738 data_end = leaf_data_end(root, right);
1739 memmove_extent_buffer(right,
1740 btrfs_leaf_data(right) + data_end - push_space,
1741 btrfs_leaf_data(right) + data_end,
1742 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1743
1744 /* copy from the left data area */
1745 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1746 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1747 btrfs_leaf_data(left) + leaf_data_end(root, left),
1748 push_space);
1749
1750 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1751 btrfs_item_nr_offset(0),
1752 right_nritems * sizeof(struct btrfs_item));
1753
1754 /* copy the items from left to right */
1755 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1756 btrfs_item_nr_offset(left_nritems - push_items),
1757 push_items * sizeof(struct btrfs_item));
1758
1759 /* update the item pointers */
1760 right_nritems += push_items;
1761 btrfs_set_header_nritems(right, right_nritems);
1762 push_space = BTRFS_LEAF_DATA_SIZE(root);
1763 for (i = 0; i < right_nritems; i++) {
1764 item = btrfs_item_nr(i);
1765 push_space -= btrfs_item_size(right, item);
1766 btrfs_set_item_offset(right, item, push_space);
1767 }
1768
1769 left_nritems -= push_items;
1770 btrfs_set_header_nritems(left, left_nritems);
1771
1772 if (left_nritems)
1773 btrfs_mark_buffer_dirty(left);
1774 btrfs_mark_buffer_dirty(right);
1775
1776 btrfs_item_key(right, &disk_key, 0);
1777 btrfs_set_node_key(upper, &disk_key, slot + 1);
1778 btrfs_mark_buffer_dirty(upper);
1779
1780 /* then fixup the leaf pointer in the path */
1781 if (path->slots[0] >= left_nritems) {
1782 path->slots[0] -= left_nritems;
1783 free_extent_buffer(path->nodes[0]);
1784 path->nodes[0] = right;
1785 path->slots[1] += 1;
1786 } else {
1787 free_extent_buffer(right);
1788 }
1789 return 0;
1790 }
1791 /*
1792 * push some data in the path leaf to the left, trying to free up at
1793 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1794 */
1795 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1796 *root, struct btrfs_path *path, int data_size,
1797 int empty)
1798 {
1799 struct btrfs_disk_key disk_key;
1800 struct extent_buffer *right = path->nodes[0];
1801 struct extent_buffer *left;
1802 int slot;
1803 int i;
1804 int free_space;
1805 int push_space = 0;
1806 int push_items = 0;
1807 struct btrfs_item *item;
1808 u32 old_left_nritems;
1809 u32 right_nritems;
1810 u32 nr;
1811 int ret = 0;
1812 u32 this_item_size;
1813 u32 old_left_item_size;
1814
1815 slot = path->slots[1];
1816 if (slot == 0)
1817 return 1;
1818 if (!path->nodes[1])
1819 return 1;
1820
1821 right_nritems = btrfs_header_nritems(right);
1822 if (right_nritems == 0) {
1823 return 1;
1824 }
1825
1826 left = read_node_slot(root, path->nodes[1], slot - 1);
1827 free_space = btrfs_leaf_free_space(root, left);
1828 if (free_space < data_size) {
1829 free_extent_buffer(left);
1830 return 1;
1831 }
1832
1833 /* cow and double check */
1834 ret = btrfs_cow_block(trans, root, left,
1835 path->nodes[1], slot - 1, &left);
1836 if (ret) {
1837 /* we hit -ENOSPC, but it isn't fatal here */
1838 free_extent_buffer(left);
1839 return 1;
1840 }
1841
1842 free_space = btrfs_leaf_free_space(root, left);
1843 if (free_space < data_size) {
1844 free_extent_buffer(left);
1845 return 1;
1846 }
1847
1848 if (empty)
1849 nr = right_nritems;
1850 else
1851 nr = right_nritems - 1;
1852
1853 for (i = 0; i < nr; i++) {
1854 item = btrfs_item_nr(i);
1855
1856 if (path->slots[0] == i)
1857 push_space += data_size + sizeof(*item);
1858
1859 this_item_size = btrfs_item_size(right, item);
1860 if (this_item_size + sizeof(*item) + push_space > free_space)
1861 break;
1862
1863 push_items++;
1864 push_space += this_item_size + sizeof(*item);
1865 }
1866
1867 if (push_items == 0) {
1868 free_extent_buffer(left);
1869 return 1;
1870 }
1871 if (!empty && push_items == btrfs_header_nritems(right))
1872 WARN_ON(1);
1873
1874 /* push data from right to left */
1875 copy_extent_buffer(left, right,
1876 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1877 btrfs_item_nr_offset(0),
1878 push_items * sizeof(struct btrfs_item));
1879
1880 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1881 btrfs_item_offset_nr(right, push_items -1);
1882
1883 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1884 leaf_data_end(root, left) - push_space,
1885 btrfs_leaf_data(right) +
1886 btrfs_item_offset_nr(right, push_items - 1),
1887 push_space);
1888 old_left_nritems = btrfs_header_nritems(left);
1889 BUG_ON(old_left_nritems == 0);
1890
1891 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1892 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1893 u32 ioff;
1894
1895 item = btrfs_item_nr(i);
1896 ioff = btrfs_item_offset(left, item);
1897 btrfs_set_item_offset(left, item,
1898 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1899 }
1900 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1901
1902 /* fixup right node */
1903 if (push_items > right_nritems) {
1904 printk("push items %d nr %u\n", push_items, right_nritems);
1905 WARN_ON(1);
1906 }
1907
1908 if (push_items < right_nritems) {
1909 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1910 leaf_data_end(root, right);
1911 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1912 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1913 btrfs_leaf_data(right) +
1914 leaf_data_end(root, right), push_space);
1915
1916 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1917 btrfs_item_nr_offset(push_items),
1918 (btrfs_header_nritems(right) - push_items) *
1919 sizeof(struct btrfs_item));
1920 }
1921 right_nritems -= push_items;
1922 btrfs_set_header_nritems(right, right_nritems);
1923 push_space = BTRFS_LEAF_DATA_SIZE(root);
1924 for (i = 0; i < right_nritems; i++) {
1925 item = btrfs_item_nr(i);
1926 push_space = push_space - btrfs_item_size(right, item);
1927 btrfs_set_item_offset(right, item, push_space);
1928 }
1929
1930 btrfs_mark_buffer_dirty(left);
1931 if (right_nritems)
1932 btrfs_mark_buffer_dirty(right);
1933
1934 btrfs_item_key(right, &disk_key, 0);
1935 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1936
1937 /* then fixup the leaf pointer in the path */
1938 if (path->slots[0] < push_items) {
1939 path->slots[0] += old_left_nritems;
1940 free_extent_buffer(path->nodes[0]);
1941 path->nodes[0] = left;
1942 path->slots[1] -= 1;
1943 } else {
1944 free_extent_buffer(left);
1945 path->slots[0] -= push_items;
1946 }
1947 BUG_ON(path->slots[0] < 0);
1948 return ret;
1949 }
1950
1951 /*
1952 * split the path's leaf in two, making sure there is at least data_size
1953 * available for the resulting leaf level of the path.
1954 *
1955 * returns 0 if all went well and < 0 on failure.
1956 */
1957 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
1958 struct btrfs_root *root,
1959 struct btrfs_path *path,
1960 struct extent_buffer *l,
1961 struct extent_buffer *right,
1962 int slot, int mid, int nritems)
1963 {
1964 int data_copy_size;
1965 int rt_data_off;
1966 int i;
1967 int ret = 0;
1968 int wret;
1969 struct btrfs_disk_key disk_key;
1970
1971 nritems = nritems - mid;
1972 btrfs_set_header_nritems(right, nritems);
1973 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
1974
1975 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
1976 btrfs_item_nr_offset(mid),
1977 nritems * sizeof(struct btrfs_item));
1978
1979 copy_extent_buffer(right, l,
1980 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
1981 data_copy_size, btrfs_leaf_data(l) +
1982 leaf_data_end(root, l), data_copy_size);
1983
1984 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
1985 btrfs_item_end_nr(l, mid);
1986
1987 for (i = 0; i < nritems; i++) {
1988 struct btrfs_item *item = btrfs_item_nr(i);
1989 u32 ioff = btrfs_item_offset(right, item);
1990 btrfs_set_item_offset(right, item, ioff + rt_data_off);
1991 }
1992
1993 btrfs_set_header_nritems(l, mid);
1994 ret = 0;
1995 btrfs_item_key(right, &disk_key, 0);
1996 wret = insert_ptr(trans, root, path, &disk_key, right->start,
1997 path->slots[1] + 1, 1);
1998 if (wret)
1999 ret = wret;
2000
2001 btrfs_mark_buffer_dirty(right);
2002 btrfs_mark_buffer_dirty(l);
2003 BUG_ON(path->slots[0] != slot);
2004
2005 if (mid <= slot) {
2006 free_extent_buffer(path->nodes[0]);
2007 path->nodes[0] = right;
2008 path->slots[0] -= mid;
2009 path->slots[1] += 1;
2010 } else {
2011 free_extent_buffer(right);
2012 }
2013
2014 BUG_ON(path->slots[0] < 0);
2015
2016 return ret;
2017 }
2018
2019 /*
2020 * split the path's leaf in two, making sure there is at least data_size
2021 * available for the resulting leaf level of the path.
2022 *
2023 * returns 0 if all went well and < 0 on failure.
2024 */
2025 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2026 struct btrfs_root *root,
2027 struct btrfs_key *ins_key,
2028 struct btrfs_path *path, int data_size,
2029 int extend)
2030 {
2031 struct btrfs_disk_key disk_key;
2032 struct extent_buffer *l;
2033 u32 nritems;
2034 int mid;
2035 int slot;
2036 struct extent_buffer *right;
2037 int ret = 0;
2038 int wret;
2039 int split;
2040 int num_doubles = 0;
2041
2042 l = path->nodes[0];
2043 slot = path->slots[0];
2044 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2045 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2046 return -EOVERFLOW;
2047
2048 /* first try to make some room by pushing left and right */
2049 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2050 wret = push_leaf_right(trans, root, path, data_size, 0);
2051 if (wret < 0)
2052 return wret;
2053 if (wret) {
2054 wret = push_leaf_left(trans, root, path, data_size, 0);
2055 if (wret < 0)
2056 return wret;
2057 }
2058 l = path->nodes[0];
2059
2060 /* did the pushes work? */
2061 if (btrfs_leaf_free_space(root, l) >= data_size)
2062 return 0;
2063 }
2064
2065 if (!path->nodes[1]) {
2066 ret = insert_new_root(trans, root, path, 1);
2067 if (ret)
2068 return ret;
2069 }
2070 again:
2071 split = 1;
2072 l = path->nodes[0];
2073 slot = path->slots[0];
2074 nritems = btrfs_header_nritems(l);
2075 mid = (nritems + 1) / 2;
2076
2077 if (mid <= slot) {
2078 if (nritems == 1 ||
2079 leaf_space_used(l, mid, nritems - mid) + data_size >
2080 BTRFS_LEAF_DATA_SIZE(root)) {
2081 if (slot >= nritems) {
2082 split = 0;
2083 } else {
2084 mid = slot;
2085 if (mid != nritems &&
2086 leaf_space_used(l, mid, nritems - mid) +
2087 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2088 split = 2;
2089 }
2090 }
2091 }
2092 } else {
2093 if (leaf_space_used(l, 0, mid) + data_size >
2094 BTRFS_LEAF_DATA_SIZE(root)) {
2095 if (!extend && data_size && slot == 0) {
2096 split = 0;
2097 } else if ((extend || !data_size) && slot == 0) {
2098 mid = 1;
2099 } else {
2100 mid = slot;
2101 if (mid != nritems &&
2102 leaf_space_used(l, mid, nritems - mid) +
2103 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2104 split = 2 ;
2105 }
2106 }
2107 }
2108 }
2109
2110 if (split == 0)
2111 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2112 else
2113 btrfs_item_key(l, &disk_key, mid);
2114
2115 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2116 root->root_key.objectid,
2117 &disk_key, 0, l->start, 0);
2118 if (IS_ERR(right)) {
2119 BUG_ON(1);
2120 return PTR_ERR(right);
2121 }
2122
2123 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2124 btrfs_set_header_bytenr(right, right->start);
2125 btrfs_set_header_generation(right, trans->transid);
2126 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2127 btrfs_set_header_owner(right, root->root_key.objectid);
2128 btrfs_set_header_level(right, 0);
2129 write_extent_buffer(right, root->fs_info->fsid,
2130 btrfs_header_fsid(), BTRFS_FSID_SIZE);
2131
2132 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2133 btrfs_header_chunk_tree_uuid(right),
2134 BTRFS_UUID_SIZE);
2135
2136 if (split == 0) {
2137 if (mid <= slot) {
2138 btrfs_set_header_nritems(right, 0);
2139 wret = insert_ptr(trans, root, path,
2140 &disk_key, right->start,
2141 path->slots[1] + 1, 1);
2142 if (wret)
2143 ret = wret;
2144
2145 free_extent_buffer(path->nodes[0]);
2146 path->nodes[0] = right;
2147 path->slots[0] = 0;
2148 path->slots[1] += 1;
2149 } else {
2150 btrfs_set_header_nritems(right, 0);
2151 wret = insert_ptr(trans, root, path,
2152 &disk_key,
2153 right->start,
2154 path->slots[1], 1);
2155 if (wret)
2156 ret = wret;
2157 free_extent_buffer(path->nodes[0]);
2158 path->nodes[0] = right;
2159 path->slots[0] = 0;
2160 if (path->slots[1] == 0) {
2161 btrfs_fixup_low_keys(root, path,
2162 &disk_key, 1);
2163 }
2164 }
2165 btrfs_mark_buffer_dirty(right);
2166 return ret;
2167 }
2168
2169 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2170 BUG_ON(ret);
2171
2172 if (split == 2) {
2173 BUG_ON(num_doubles != 0);
2174 num_doubles++;
2175 goto again;
2176 }
2177
2178 return ret;
2179 }
2180
2181 /*
2182 * This function splits a single item into two items,
2183 * giving 'new_key' to the new item and splitting the
2184 * old one at split_offset (from the start of the item).
2185 *
2186 * The path may be released by this operation. After
2187 * the split, the path is pointing to the old item. The
2188 * new item is going to be in the same node as the old one.
2189 *
2190 * Note, the item being split must be smaller enough to live alone on
2191 * a tree block with room for one extra struct btrfs_item
2192 *
2193 * This allows us to split the item in place, keeping a lock on the
2194 * leaf the entire time.
2195 */
2196 int btrfs_split_item(struct btrfs_trans_handle *trans,
2197 struct btrfs_root *root,
2198 struct btrfs_path *path,
2199 struct btrfs_key *new_key,
2200 unsigned long split_offset)
2201 {
2202 u32 item_size;
2203 struct extent_buffer *leaf;
2204 struct btrfs_key orig_key;
2205 struct btrfs_item *item;
2206 struct btrfs_item *new_item;
2207 int ret = 0;
2208 int slot;
2209 u32 nritems;
2210 u32 orig_offset;
2211 struct btrfs_disk_key disk_key;
2212 char *buf;
2213
2214 leaf = path->nodes[0];
2215 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2216 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2217 goto split;
2218
2219 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2220 btrfs_release_path(path);
2221
2222 path->search_for_split = 1;
2223
2224 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2225 path->search_for_split = 0;
2226
2227 /* if our item isn't there or got smaller, return now */
2228 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2229 path->slots[0])) {
2230 return -EAGAIN;
2231 }
2232
2233 ret = split_leaf(trans, root, &orig_key, path, 0, 0);
2234 BUG_ON(ret);
2235
2236 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2237 leaf = path->nodes[0];
2238
2239 split:
2240 item = btrfs_item_nr(path->slots[0]);
2241 orig_offset = btrfs_item_offset(leaf, item);
2242 item_size = btrfs_item_size(leaf, item);
2243
2244
2245 buf = kmalloc(item_size, GFP_NOFS);
2246 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2247 path->slots[0]), item_size);
2248 slot = path->slots[0] + 1;
2249 leaf = path->nodes[0];
2250
2251 nritems = btrfs_header_nritems(leaf);
2252
2253 if (slot != nritems) {
2254 /* shift the items */
2255 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2256 btrfs_item_nr_offset(slot),
2257 (nritems - slot) * sizeof(struct btrfs_item));
2258
2259 }
2260
2261 btrfs_cpu_key_to_disk(&disk_key, new_key);
2262 btrfs_set_item_key(leaf, &disk_key, slot);
2263
2264 new_item = btrfs_item_nr(slot);
2265
2266 btrfs_set_item_offset(leaf, new_item, orig_offset);
2267 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
2268
2269 btrfs_set_item_offset(leaf, item,
2270 orig_offset + item_size - split_offset);
2271 btrfs_set_item_size(leaf, item, split_offset);
2272
2273 btrfs_set_header_nritems(leaf, nritems + 1);
2274
2275 /* write the data for the start of the original item */
2276 write_extent_buffer(leaf, buf,
2277 btrfs_item_ptr_offset(leaf, path->slots[0]),
2278 split_offset);
2279
2280 /* write the data for the new item */
2281 write_extent_buffer(leaf, buf + split_offset,
2282 btrfs_item_ptr_offset(leaf, slot),
2283 item_size - split_offset);
2284 btrfs_mark_buffer_dirty(leaf);
2285
2286 ret = 0;
2287 if (btrfs_leaf_free_space(root, leaf) < 0) {
2288 btrfs_print_leaf(root, leaf);
2289 BUG();
2290 }
2291 kfree(buf);
2292 return ret;
2293 }
2294
2295 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2296 struct btrfs_root *root,
2297 struct btrfs_path *path,
2298 u32 new_size, int from_end)
2299 {
2300 int ret = 0;
2301 int slot;
2302 struct extent_buffer *leaf;
2303 struct btrfs_item *item;
2304 u32 nritems;
2305 unsigned int data_end;
2306 unsigned int old_data_start;
2307 unsigned int old_size;
2308 unsigned int size_diff;
2309 int i;
2310
2311 leaf = path->nodes[0];
2312 slot = path->slots[0];
2313
2314 old_size = btrfs_item_size_nr(leaf, slot);
2315 if (old_size == new_size)
2316 return 0;
2317
2318 nritems = btrfs_header_nritems(leaf);
2319 data_end = leaf_data_end(root, leaf);
2320
2321 old_data_start = btrfs_item_offset_nr(leaf, slot);
2322
2323 size_diff = old_size - new_size;
2324
2325 BUG_ON(slot < 0);
2326 BUG_ON(slot >= nritems);
2327
2328 /*
2329 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2330 */
2331 /* first correct the data pointers */
2332 for (i = slot; i < nritems; i++) {
2333 u32 ioff;
2334 item = btrfs_item_nr(i);
2335 ioff = btrfs_item_offset(leaf, item);
2336 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2337 }
2338
2339 /* shift the data */
2340 if (from_end) {
2341 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2342 data_end + size_diff, btrfs_leaf_data(leaf) +
2343 data_end, old_data_start + new_size - data_end);
2344 } else {
2345 struct btrfs_disk_key disk_key;
2346 u64 offset;
2347
2348 btrfs_item_key(leaf, &disk_key, slot);
2349
2350 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2351 unsigned long ptr;
2352 struct btrfs_file_extent_item *fi;
2353
2354 fi = btrfs_item_ptr(leaf, slot,
2355 struct btrfs_file_extent_item);
2356 fi = (struct btrfs_file_extent_item *)(
2357 (unsigned long)fi - size_diff);
2358
2359 if (btrfs_file_extent_type(leaf, fi) ==
2360 BTRFS_FILE_EXTENT_INLINE) {
2361 ptr = btrfs_item_ptr_offset(leaf, slot);
2362 memmove_extent_buffer(leaf, ptr,
2363 (unsigned long)fi,
2364 offsetof(struct btrfs_file_extent_item,
2365 disk_bytenr));
2366 }
2367 }
2368
2369 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2370 data_end + size_diff, btrfs_leaf_data(leaf) +
2371 data_end, old_data_start - data_end);
2372
2373 offset = btrfs_disk_key_offset(&disk_key);
2374 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2375 btrfs_set_item_key(leaf, &disk_key, slot);
2376 if (slot == 0)
2377 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2378 }
2379
2380 item = btrfs_item_nr(slot);
2381 btrfs_set_item_size(leaf, item, new_size);
2382 btrfs_mark_buffer_dirty(leaf);
2383
2384 ret = 0;
2385 if (btrfs_leaf_free_space(root, leaf) < 0) {
2386 btrfs_print_leaf(root, leaf);
2387 BUG();
2388 }
2389 return ret;
2390 }
2391
2392 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2393 struct btrfs_root *root, struct btrfs_path *path,
2394 u32 data_size)
2395 {
2396 int ret = 0;
2397 int slot;
2398 struct extent_buffer *leaf;
2399 struct btrfs_item *item;
2400 u32 nritems;
2401 unsigned int data_end;
2402 unsigned int old_data;
2403 unsigned int old_size;
2404 int i;
2405
2406 leaf = path->nodes[0];
2407
2408 nritems = btrfs_header_nritems(leaf);
2409 data_end = leaf_data_end(root, leaf);
2410
2411 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2412 btrfs_print_leaf(root, leaf);
2413 BUG();
2414 }
2415 slot = path->slots[0];
2416 old_data = btrfs_item_end_nr(leaf, slot);
2417
2418 BUG_ON(slot < 0);
2419 if (slot >= nritems) {
2420 btrfs_print_leaf(root, leaf);
2421 printk("slot %d too large, nritems %d\n", slot, nritems);
2422 BUG_ON(1);
2423 }
2424
2425 /*
2426 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2427 */
2428 /* first correct the data pointers */
2429 for (i = slot; i < nritems; i++) {
2430 u32 ioff;
2431 item = btrfs_item_nr(i);
2432 ioff = btrfs_item_offset(leaf, item);
2433 btrfs_set_item_offset(leaf, item, ioff - data_size);
2434 }
2435
2436 /* shift the data */
2437 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2438 data_end - data_size, btrfs_leaf_data(leaf) +
2439 data_end, old_data - data_end);
2440
2441 data_end = old_data;
2442 old_size = btrfs_item_size_nr(leaf, slot);
2443 item = btrfs_item_nr(slot);
2444 btrfs_set_item_size(leaf, item, old_size + data_size);
2445 btrfs_mark_buffer_dirty(leaf);
2446
2447 ret = 0;
2448 if (btrfs_leaf_free_space(root, leaf) < 0) {
2449 btrfs_print_leaf(root, leaf);
2450 BUG();
2451 }
2452 return ret;
2453 }
2454
2455 /*
2456 * Given a key and some data, insert an item into the tree.
2457 * This does all the path init required, making room in the tree if needed.
2458 */
2459 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2460 struct btrfs_root *root,
2461 struct btrfs_path *path,
2462 struct btrfs_key *cpu_key, u32 *data_size,
2463 int nr)
2464 {
2465 struct extent_buffer *leaf;
2466 struct btrfs_item *item;
2467 int ret = 0;
2468 int slot;
2469 int i;
2470 u32 nritems;
2471 u32 total_size = 0;
2472 u32 total_data = 0;
2473 unsigned int data_end;
2474 struct btrfs_disk_key disk_key;
2475
2476 for (i = 0; i < nr; i++) {
2477 total_data += data_size[i];
2478 }
2479
2480 /* create a root if there isn't one */
2481 if (!root->node)
2482 BUG();
2483
2484 total_size = total_data + nr * sizeof(struct btrfs_item);
2485 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2486 if (ret == 0) {
2487 return -EEXIST;
2488 }
2489 if (ret < 0)
2490 goto out;
2491
2492 leaf = path->nodes[0];
2493
2494 nritems = btrfs_header_nritems(leaf);
2495 data_end = leaf_data_end(root, leaf);
2496
2497 if (btrfs_leaf_free_space(root, leaf) < total_size) {
2498 btrfs_print_leaf(root, leaf);
2499 printk("not enough freespace need %u have %d\n",
2500 total_size, btrfs_leaf_free_space(root, leaf));
2501 BUG();
2502 }
2503
2504 slot = path->slots[0];
2505 BUG_ON(slot < 0);
2506
2507 if (slot != nritems) {
2508 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2509
2510 if (old_data < data_end) {
2511 btrfs_print_leaf(root, leaf);
2512 printk("slot %d old_data %d data_end %d\n",
2513 slot, old_data, data_end);
2514 BUG_ON(1);
2515 }
2516 /*
2517 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2518 */
2519 /* first correct the data pointers */
2520 for (i = slot; i < nritems; i++) {
2521 u32 ioff;
2522
2523 item = btrfs_item_nr(i);
2524 ioff = btrfs_item_offset(leaf, item);
2525 btrfs_set_item_offset(leaf, item, ioff - total_data);
2526 }
2527
2528 /* shift the items */
2529 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2530 btrfs_item_nr_offset(slot),
2531 (nritems - slot) * sizeof(struct btrfs_item));
2532
2533 /* shift the data */
2534 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2535 data_end - total_data, btrfs_leaf_data(leaf) +
2536 data_end, old_data - data_end);
2537 data_end = old_data;
2538 }
2539
2540 /* setup the item for the new data */
2541 for (i = 0; i < nr; i++) {
2542 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2543 btrfs_set_item_key(leaf, &disk_key, slot + i);
2544 item = btrfs_item_nr(slot + i);
2545 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2546 data_end -= data_size[i];
2547 btrfs_set_item_size(leaf, item, data_size[i]);
2548 }
2549 btrfs_set_header_nritems(leaf, nritems + nr);
2550 btrfs_mark_buffer_dirty(leaf);
2551
2552 ret = 0;
2553 if (slot == 0) {
2554 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2555 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2556 }
2557
2558 if (btrfs_leaf_free_space(root, leaf) < 0) {
2559 btrfs_print_leaf(root, leaf);
2560 BUG();
2561 }
2562
2563 out:
2564 return ret;
2565 }
2566
2567 /*
2568 * Given a key and some data, insert an item into the tree.
2569 * This does all the path init required, making room in the tree if needed.
2570 */
2571 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2572 *root, struct btrfs_key *cpu_key, void *data, u32
2573 data_size)
2574 {
2575 int ret = 0;
2576 struct btrfs_path *path;
2577 struct extent_buffer *leaf;
2578 unsigned long ptr;
2579
2580 path = btrfs_alloc_path();
2581 BUG_ON(!path);
2582 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2583 if (!ret) {
2584 leaf = path->nodes[0];
2585 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2586 write_extent_buffer(leaf, data, ptr, data_size);
2587 btrfs_mark_buffer_dirty(leaf);
2588 }
2589 btrfs_free_path(path);
2590 return ret;
2591 }
2592
2593 /*
2594 * delete the pointer from a given node.
2595 *
2596 * If the delete empties a node, the node is removed from the tree,
2597 * continuing all the way the root if required. The root is converted into
2598 * a leaf if all the nodes are emptied.
2599 */
2600 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2601 struct btrfs_path *path, int level, int slot)
2602 {
2603 struct extent_buffer *parent = path->nodes[level];
2604 u32 nritems;
2605 int ret = 0;
2606
2607 nritems = btrfs_header_nritems(parent);
2608 if (slot != nritems -1) {
2609 memmove_extent_buffer(parent,
2610 btrfs_node_key_ptr_offset(slot),
2611 btrfs_node_key_ptr_offset(slot + 1),
2612 sizeof(struct btrfs_key_ptr) *
2613 (nritems - slot - 1));
2614 }
2615 nritems--;
2616 btrfs_set_header_nritems(parent, nritems);
2617 if (nritems == 0 && parent == root->node) {
2618 BUG_ON(btrfs_header_level(root->node) != 1);
2619 /* just turn the root into a leaf and break */
2620 btrfs_set_header_level(root->node, 0);
2621 } else if (slot == 0) {
2622 struct btrfs_disk_key disk_key;
2623
2624 btrfs_node_key(parent, &disk_key, 0);
2625 btrfs_fixup_low_keys(root, path, &disk_key, level + 1);
2626 }
2627 btrfs_mark_buffer_dirty(parent);
2628 return ret;
2629 }
2630
2631 /*
2632 * a helper function to delete the leaf pointed to by path->slots[1] and
2633 * path->nodes[1].
2634 *
2635 * This deletes the pointer in path->nodes[1] and frees the leaf
2636 * block extent. zero is returned if it all worked out, < 0 otherwise.
2637 *
2638 * The path must have already been setup for deleting the leaf, including
2639 * all the proper balancing. path->nodes[1] must be locked.
2640 */
2641 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
2642 struct btrfs_root *root,
2643 struct btrfs_path *path,
2644 struct extent_buffer *leaf)
2645 {
2646 int ret;
2647
2648 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
2649 ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]);
2650 if (ret)
2651 return ret;
2652
2653 ret = btrfs_free_extent(trans, root, leaf->start, leaf->len,
2654 0, root->root_key.objectid, 0, 0);
2655 return ret;
2656 }
2657
2658 /*
2659 * delete the item at the leaf level in path. If that empties
2660 * the leaf, remove it from the tree
2661 */
2662 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2663 struct btrfs_path *path, int slot, int nr)
2664 {
2665 struct extent_buffer *leaf;
2666 struct btrfs_item *item;
2667 int last_off;
2668 int dsize = 0;
2669 int ret = 0;
2670 int wret;
2671 int i;
2672 u32 nritems;
2673
2674 leaf = path->nodes[0];
2675 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2676
2677 for (i = 0; i < nr; i++)
2678 dsize += btrfs_item_size_nr(leaf, slot + i);
2679
2680 nritems = btrfs_header_nritems(leaf);
2681
2682 if (slot + nr != nritems) {
2683 int data_end = leaf_data_end(root, leaf);
2684
2685 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2686 data_end + dsize,
2687 btrfs_leaf_data(leaf) + data_end,
2688 last_off - data_end);
2689
2690 for (i = slot + nr; i < nritems; i++) {
2691 u32 ioff;
2692
2693 item = btrfs_item_nr(i);
2694 ioff = btrfs_item_offset(leaf, item);
2695 btrfs_set_item_offset(leaf, item, ioff + dsize);
2696 }
2697
2698 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2699 btrfs_item_nr_offset(slot + nr),
2700 sizeof(struct btrfs_item) *
2701 (nritems - slot - nr));
2702 }
2703 btrfs_set_header_nritems(leaf, nritems - nr);
2704 nritems -= nr;
2705
2706 /* delete the leaf if we've emptied it */
2707 if (nritems == 0) {
2708 if (leaf == root->node) {
2709 btrfs_set_header_level(leaf, 0);
2710 } else {
2711 clean_tree_block(trans, root, leaf);
2712 wait_on_tree_block_writeback(root, leaf);
2713
2714 wret = btrfs_del_leaf(trans, root, path, leaf);
2715 BUG_ON(ret);
2716 if (wret)
2717 ret = wret;
2718 }
2719 } else {
2720 int used = leaf_space_used(leaf, 0, nritems);
2721 if (slot == 0) {
2722 struct btrfs_disk_key disk_key;
2723
2724 btrfs_item_key(leaf, &disk_key, 0);
2725 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2726 }
2727
2728 /* delete the leaf if it is mostly empty */
2729 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2730 /* push_leaf_left fixes the path.
2731 * make sure the path still points to our leaf
2732 * for possible call to del_ptr below
2733 */
2734 slot = path->slots[1];
2735 extent_buffer_get(leaf);
2736
2737 wret = push_leaf_left(trans, root, path, 1, 1);
2738 if (wret < 0 && wret != -ENOSPC)
2739 ret = wret;
2740
2741 if (path->nodes[0] == leaf &&
2742 btrfs_header_nritems(leaf)) {
2743 wret = push_leaf_right(trans, root, path, 1, 1);
2744 if (wret < 0 && wret != -ENOSPC)
2745 ret = wret;
2746 }
2747
2748 if (btrfs_header_nritems(leaf) == 0) {
2749 clean_tree_block(trans, root, leaf);
2750 wait_on_tree_block_writeback(root, leaf);
2751
2752 path->slots[1] = slot;
2753 ret = btrfs_del_leaf(trans, root, path, leaf);
2754 BUG_ON(ret);
2755 free_extent_buffer(leaf);
2756
2757 } else {
2758 btrfs_mark_buffer_dirty(leaf);
2759 free_extent_buffer(leaf);
2760 }
2761 } else {
2762 btrfs_mark_buffer_dirty(leaf);
2763 }
2764 }
2765 return ret;
2766 }
2767
2768 /*
2769 * walk up the tree as far as required to find the previous leaf.
2770 * returns 0 if it found something or 1 if there are no lesser leaves.
2771 * returns < 0 on io errors.
2772 */
2773 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2774 {
2775 int slot;
2776 int level = 1;
2777 struct extent_buffer *c;
2778 struct extent_buffer *next = NULL;
2779
2780 while(level < BTRFS_MAX_LEVEL) {
2781 if (!path->nodes[level])
2782 return 1;
2783
2784 slot = path->slots[level];
2785 c = path->nodes[level];
2786 if (slot == 0) {
2787 level++;
2788 if (level == BTRFS_MAX_LEVEL)
2789 return 1;
2790 continue;
2791 }
2792 slot--;
2793
2794 next = read_node_slot(root, c, slot);
2795 if (!extent_buffer_uptodate(next)) {
2796 if (IS_ERR(next))
2797 return PTR_ERR(next);
2798 return -EIO;
2799 }
2800 break;
2801 }
2802 path->slots[level] = slot;
2803 while(1) {
2804 level--;
2805 c = path->nodes[level];
2806 free_extent_buffer(c);
2807 slot = btrfs_header_nritems(next);
2808 if (slot != 0)
2809 slot--;
2810 path->nodes[level] = next;
2811 path->slots[level] = slot;
2812 if (!level)
2813 break;
2814 next = read_node_slot(root, next, slot);
2815 if (!extent_buffer_uptodate(next)) {
2816 if (IS_ERR(next))
2817 return PTR_ERR(next);
2818 return -EIO;
2819 }
2820 }
2821 return 0;
2822 }
2823
2824 /*
2825 * walk up the tree as far as required to find the next leaf.
2826 * returns 0 if it found something or 1 if there are no greater leaves.
2827 * returns < 0 on io errors.
2828 */
2829 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2830 {
2831 int slot;
2832 int level = 1;
2833 struct extent_buffer *c;
2834 struct extent_buffer *next = NULL;
2835
2836 while(level < BTRFS_MAX_LEVEL) {
2837 if (!path->nodes[level])
2838 return 1;
2839
2840 slot = path->slots[level] + 1;
2841 c = path->nodes[level];
2842 if (slot >= btrfs_header_nritems(c)) {
2843 level++;
2844 if (level == BTRFS_MAX_LEVEL)
2845 return 1;
2846 continue;
2847 }
2848
2849 if (path->reada)
2850 reada_for_search(root, path, level, slot, 0);
2851
2852 next = read_node_slot(root, c, slot);
2853 if (!extent_buffer_uptodate(next))
2854 return -EIO;
2855 break;
2856 }
2857 path->slots[level] = slot;
2858 while(1) {
2859 level--;
2860 c = path->nodes[level];
2861 free_extent_buffer(c);
2862 path->nodes[level] = next;
2863 path->slots[level] = 0;
2864 if (!level)
2865 break;
2866 if (path->reada)
2867 reada_for_search(root, path, level, 0, 0);
2868 next = read_node_slot(root, next, 0);
2869 if (!extent_buffer_uptodate(next))
2870 return -EIO;
2871 }
2872 return 0;
2873 }
2874
2875 int btrfs_previous_item(struct btrfs_root *root,
2876 struct btrfs_path *path, u64 min_objectid,
2877 int type)
2878 {
2879 struct btrfs_key found_key;
2880 struct extent_buffer *leaf;
2881 int ret;
2882
2883 while(1) {
2884 if (path->slots[0] == 0) {
2885 ret = btrfs_prev_leaf(root, path);
2886 if (ret != 0)
2887 return ret;
2888 } else {
2889 path->slots[0]--;
2890 }
2891 leaf = path->nodes[0];
2892 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2893 if (found_key.type == type)
2894 return 0;
2895 }
2896 return 1;
2897 }
2898
2899 /*
2900 * search in extent tree to find a previous Metadata/Data extent item with
2901 * min objecitd.
2902 *
2903 * returns 0 if something is found, 1 if nothing was found and < 0 on error
2904 */
2905 int btrfs_previous_extent_item(struct btrfs_root *root,
2906 struct btrfs_path *path, u64 min_objectid)
2907 {
2908 struct btrfs_key found_key;
2909 struct extent_buffer *leaf;
2910 u32 nritems;
2911 int ret;
2912
2913 while (1) {
2914 if (path->slots[0] == 0) {
2915 ret = btrfs_prev_leaf(root, path);
2916 if (ret != 0)
2917 return ret;
2918 } else {
2919 path->slots[0]--;
2920 }
2921 leaf = path->nodes[0];
2922 nritems = btrfs_header_nritems(leaf);
2923 if (nritems == 0)
2924 return 1;
2925 if (path->slots[0] == nritems)
2926 path->slots[0]--;
2927
2928 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2929 if (found_key.objectid < min_objectid)
2930 break;
2931 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
2932 found_key.type == BTRFS_METADATA_ITEM_KEY)
2933 return 0;
2934 if (found_key.objectid == min_objectid &&
2935 found_key.type < BTRFS_EXTENT_ITEM_KEY)
2936 break;
2937 }
2938 return 1;
2939 }
(deprecated) Btrfs progs developments and patch integration