?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: move utils code out of header
[btrfs-progs-unstable/devel.git] / ctree.c
blobe3d687fb03b3b3a620b668bc04401b31817bd07a
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 #include "internal.h"
24 #include "sizes.h"
25
26 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_path *path, int level);
28 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_key *ins_key,
30 struct btrfs_path *path, int data_size, int extend);
31 static int push_node_left(struct btrfs_trans_handle *trans,
32 struct btrfs_root *root, struct extent_buffer *dst,
33 struct extent_buffer *src, int empty);
34 static int balance_node_right(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root,
36 struct extent_buffer *dst_buf,
37 struct extent_buffer *src_buf);
38
39 inline void btrfs_init_path(struct btrfs_path *p)
40 {
41 memset(p, 0, sizeof(*p));
42 }
43
44 struct btrfs_path *btrfs_alloc_path(void)
45 {
46 struct btrfs_path *path;
47 path = kzalloc(sizeof(struct btrfs_path), GFP_NOFS);
48 return path;
49 }
50
51 void btrfs_free_path(struct btrfs_path *p)
52 {
53 if (!p)
54 return;
55 btrfs_release_path(p);
56 kfree(p);
57 }
58
59 void btrfs_release_path(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 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 btrfs_header_fsid(), BTRFS_FSID_SIZE);
126
127 WARN_ON(btrfs_header_generation(buf) > trans->transid);
128 ret = btrfs_inc_ref(trans, new_root, cow, 0);
129 kfree(new_root);
130
131 if (ret)
132 return ret;
133
134 btrfs_mark_buffer_dirty(cow);
135 *cow_ret = cow;
136 return 0;
137 }
138
139 /*
140 * check if the tree block can be shared by multiple trees
141 */
142 static int btrfs_block_can_be_shared(struct btrfs_root *root,
143 struct extent_buffer *buf)
144 {
145 /*
146 * Tree blocks not in reference counted trees and tree roots
147 * are never shared. If a block was allocated after the last
148 * snapshot and the block was not allocated by tree relocation,
149 * we know the block is not shared.
150 */
151 if (root->ref_cows &&
152 buf != root->node && buf != root->commit_root &&
153 (btrfs_header_generation(buf) <=
154 btrfs_root_last_snapshot(&root->root_item) ||
155 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
156 return 1;
157 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
158 if (root->ref_cows &&
159 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
160 return 1;
161 #endif
162 return 0;
163 }
164
165 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
166 struct btrfs_root *root,
167 struct extent_buffer *buf,
168 struct extent_buffer *cow)
169 {
170 u64 refs;
171 u64 owner;
172 u64 flags;
173 u64 new_flags = 0;
174 int ret;
175
176 /*
177 * Backrefs update rules:
178 *
179 * Always use full backrefs for extent pointers in tree block
180 * allocated by tree relocation.
181 *
182 * If a shared tree block is no longer referenced by its owner
183 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
184 * use full backrefs for extent pointers in tree block.
185 *
186 * If a tree block is been relocating
187 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
188 * use full backrefs for extent pointers in tree block.
189 * The reason for this is some operations (such as drop tree)
190 * are only allowed for blocks use full backrefs.
191 */
192
193 if (btrfs_block_can_be_shared(root, buf)) {
194 ret = btrfs_lookup_extent_info(trans, root, buf->start,
195 btrfs_header_level(buf), 1,
196 &refs, &flags);
197 BUG_ON(ret);
198 BUG_ON(refs == 0);
199 } else {
200 refs = 1;
201 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
202 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
203 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
204 else
205 flags = 0;
206 }
207
208 owner = btrfs_header_owner(buf);
209 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
210 owner == BTRFS_TREE_RELOC_OBJECTID);
211
212 if (refs > 1) {
213 if ((owner == root->root_key.objectid ||
214 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
215 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
216 ret = btrfs_inc_ref(trans, root, buf, 1);
217 BUG_ON(ret);
218
219 if (root->root_key.objectid ==
220 BTRFS_TREE_RELOC_OBJECTID) {
221 ret = btrfs_dec_ref(trans, root, buf, 0);
222 BUG_ON(ret);
223 ret = btrfs_inc_ref(trans, root, cow, 1);
224 BUG_ON(ret);
225 }
226 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
227 } else {
228
229 if (root->root_key.objectid ==
230 BTRFS_TREE_RELOC_OBJECTID)
231 ret = btrfs_inc_ref(trans, root, cow, 1);
232 else
233 ret = btrfs_inc_ref(trans, root, cow, 0);
234 BUG_ON(ret);
235 }
236 if (new_flags != 0) {
237 ret = btrfs_set_block_flags(trans, root, buf->start,
238 btrfs_header_level(buf),
239 new_flags);
240 BUG_ON(ret);
241 }
242 } else {
243 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
244 if (root->root_key.objectid ==
245 BTRFS_TREE_RELOC_OBJECTID)
246 ret = btrfs_inc_ref(trans, root, cow, 1);
247 else
248 ret = btrfs_inc_ref(trans, root, cow, 0);
249 BUG_ON(ret);
250 ret = btrfs_dec_ref(trans, root, buf, 1);
251 BUG_ON(ret);
252 }
253 clean_tree_block(trans, root, buf);
254 }
255 return 0;
256 }
257
258 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
259 struct btrfs_root *root,
260 struct extent_buffer *buf,
261 struct extent_buffer *parent, int parent_slot,
262 struct extent_buffer **cow_ret,
263 u64 search_start, u64 empty_size)
264 {
265 struct extent_buffer *cow;
266 struct btrfs_disk_key disk_key;
267 int level;
268
269 WARN_ON(root->ref_cows && trans->transid !=
270 root->fs_info->running_transaction->transid);
271 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
272
273 level = btrfs_header_level(buf);
274
275 if (level == 0)
276 btrfs_item_key(buf, &disk_key, 0);
277 else
278 btrfs_node_key(buf, &disk_key, 0);
279
280 cow = btrfs_alloc_free_block(trans, root, buf->len,
281 root->root_key.objectid, &disk_key,
282 level, search_start, empty_size);
283 if (IS_ERR(cow))
284 return PTR_ERR(cow);
285
286 copy_extent_buffer(cow, buf, 0, 0, cow->len);
287 btrfs_set_header_bytenr(cow, cow->start);
288 btrfs_set_header_generation(cow, trans->transid);
289 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
290 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
291 BTRFS_HEADER_FLAG_RELOC);
292 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
293 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
294 else
295 btrfs_set_header_owner(cow, root->root_key.objectid);
296
297 write_extent_buffer(cow, root->fs_info->fsid,
298 btrfs_header_fsid(), BTRFS_FSID_SIZE);
299
300 WARN_ON(!(buf->flags & EXTENT_BAD_TRANSID) &&
301 btrfs_header_generation(buf) > trans->transid);
302
303 update_ref_for_cow(trans, root, buf, cow);
304
305 if (buf == root->node) {
306 root->node = cow;
307 extent_buffer_get(cow);
308
309 btrfs_free_extent(trans, root, buf->start, buf->len,
310 0, root->root_key.objectid, level, 0);
311 free_extent_buffer(buf);
312 add_root_to_dirty_list(root);
313 } else {
314 btrfs_set_node_blockptr(parent, parent_slot,
315 cow->start);
316 WARN_ON(trans->transid == 0);
317 btrfs_set_node_ptr_generation(parent, parent_slot,
318 trans->transid);
319 btrfs_mark_buffer_dirty(parent);
320 WARN_ON(btrfs_header_generation(parent) != trans->transid);
321
322 btrfs_free_extent(trans, root, buf->start, buf->len,
323 0, root->root_key.objectid, level, 1);
324 }
325 if (!list_empty(&buf->recow)) {
326 list_del_init(&buf->recow);
327 free_extent_buffer(buf);
328 }
329 free_extent_buffer(buf);
330 btrfs_mark_buffer_dirty(cow);
331 *cow_ret = cow;
332 return 0;
333 }
334
335 static inline int should_cow_block(struct btrfs_trans_handle *trans,
336 struct btrfs_root *root,
337 struct extent_buffer *buf)
338 {
339 if (btrfs_header_generation(buf) == trans->transid &&
340 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
341 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
342 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
343 return 0;
344 return 1;
345 }
346
347 int btrfs_cow_block(struct btrfs_trans_handle *trans,
348 struct btrfs_root *root, struct extent_buffer *buf,
349 struct extent_buffer *parent, int parent_slot,
350 struct extent_buffer **cow_ret)
351 {
352 u64 search_start;
353 int ret;
354 /*
355 if (trans->transaction != root->fs_info->running_transaction) {
356 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
357 root->fs_info->running_transaction->transid);
358 WARN_ON(1);
359 }
360 */
361 if (trans->transid != root->fs_info->generation) {
362 printk(KERN_CRIT "trans %llu running %llu\n",
363 (unsigned long long)trans->transid,
364 (unsigned long long)root->fs_info->generation);
365 WARN_ON(1);
366 }
367 if (!should_cow_block(trans, root, buf)) {
368 *cow_ret = buf;
369 return 0;
370 }
371
372 search_start = buf->start & ~((u64)SZ_1G - 1);
373 ret = __btrfs_cow_block(trans, root, buf, parent,
374 parent_slot, cow_ret, search_start, 0);
375 return ret;
376 }
377
378 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
379 {
380 if (k1->objectid > k2->objectid)
381 return 1;
382 if (k1->objectid < k2->objectid)
383 return -1;
384 if (k1->type > k2->type)
385 return 1;
386 if (k1->type < k2->type)
387 return -1;
388 if (k1->offset > k2->offset)
389 return 1;
390 if (k1->offset < k2->offset)
391 return -1;
392 return 0;
393 }
394
395 /*
396 * compare two keys in a memcmp fashion
397 */
398 static int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
399 {
400 struct btrfs_key k1;
401
402 btrfs_disk_key_to_cpu(&k1, disk);
403 return btrfs_comp_cpu_keys(&k1, k2);
404 }
405
406 /*
407 * The leaf data grows from end-to-front in the node.
408 * this returns the address of the start of the last item,
409 * which is the stop of the leaf data stack
410 */
411 static inline unsigned int leaf_data_end(struct btrfs_root *root,
412 struct extent_buffer *leaf)
413 {
414 u32 nr = btrfs_header_nritems(leaf);
415 if (nr == 0)
416 return BTRFS_LEAF_DATA_SIZE(root);
417 return btrfs_item_offset_nr(leaf, nr - 1);
418 }
419
420 enum btrfs_tree_block_status
421 btrfs_check_node(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
422 struct extent_buffer *buf)
423 {
424 int i;
425 struct btrfs_key cpukey;
426 struct btrfs_disk_key key;
427 u32 nritems = btrfs_header_nritems(buf);
428 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
429
430 if (nritems == 0 || nritems > BTRFS_NODEPTRS_PER_BLOCK(root))
431 goto fail;
432
433 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
434 if (parent_key && parent_key->type) {
435 btrfs_node_key(buf, &key, 0);
436 if (memcmp(parent_key, &key, sizeof(key)))
437 goto fail;
438 }
439 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
440 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
441 btrfs_node_key(buf, &key, i);
442 btrfs_node_key_to_cpu(buf, &cpukey, i + 1);
443 if (btrfs_comp_keys(&key, &cpukey) >= 0)
444 goto fail;
445 }
446 return BTRFS_TREE_BLOCK_CLEAN;
447 fail:
448 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
449 if (parent_key)
450 btrfs_disk_key_to_cpu(&cpukey, parent_key);
451 else
452 btrfs_node_key_to_cpu(buf, &cpukey, 0);
453 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
454 buf->start, buf->len,
455 btrfs_header_level(buf));
456 }
457 return ret;
458 }
459
460 enum btrfs_tree_block_status
461 btrfs_check_leaf(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
462 struct extent_buffer *buf)
463 {
464 int i;
465 struct btrfs_key cpukey;
466 struct btrfs_disk_key key;
467 u32 nritems = btrfs_header_nritems(buf);
468 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
469
470 if (nritems * sizeof(struct btrfs_item) > buf->len) {
471 fprintf(stderr, "invalid number of items %llu\n",
472 (unsigned long long)buf->start);
473 goto fail;
474 }
475
476 if (btrfs_header_level(buf) != 0) {
477 ret = BTRFS_TREE_BLOCK_INVALID_LEVEL;
478 fprintf(stderr, "leaf is not a leaf %llu\n",
479 (unsigned long long)btrfs_header_bytenr(buf));
480 goto fail;
481 }
482 if (btrfs_leaf_free_space(root, buf) < 0) {
483 ret = BTRFS_TREE_BLOCK_INVALID_FREE_SPACE;
484 fprintf(stderr, "leaf free space incorrect %llu %d\n",
485 (unsigned long long)btrfs_header_bytenr(buf),
486 btrfs_leaf_free_space(root, buf));
487 goto fail;
488 }
489
490 if (nritems == 0)
491 return BTRFS_TREE_BLOCK_CLEAN;
492
493 btrfs_item_key(buf, &key, 0);
494 if (parent_key && parent_key->type &&
495 memcmp(parent_key, &key, sizeof(key))) {
496 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
497 fprintf(stderr, "leaf parent key incorrect %llu\n",
498 (unsigned long long)btrfs_header_bytenr(buf));
499 goto fail;
500 }
501 for (i = 0; nritems > 1 && i < nritems - 1; i++) {
502 btrfs_item_key(buf, &key, i);
503 btrfs_item_key_to_cpu(buf, &cpukey, i + 1);
504 if (btrfs_comp_keys(&key, &cpukey) >= 0) {
505 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
506 fprintf(stderr, "bad key ordering %d %d\n", i, i+1);
507 goto fail;
508 }
509 if (btrfs_item_offset_nr(buf, i) !=
510 btrfs_item_end_nr(buf, i + 1)) {
511 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
512 fprintf(stderr, "incorrect offsets %u %u\n",
513 btrfs_item_offset_nr(buf, i),
514 btrfs_item_end_nr(buf, i + 1));
515 goto fail;
516 }
517 if (i == 0 && btrfs_item_end_nr(buf, i) !=
518 BTRFS_LEAF_DATA_SIZE(root)) {
519 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
520 fprintf(stderr, "bad item end %u wanted %u\n",
521 btrfs_item_end_nr(buf, i),
522 (unsigned)BTRFS_LEAF_DATA_SIZE(root));
523 goto fail;
524 }
525 }
526
527 for (i = 0; i < nritems; i++) {
528 if (btrfs_item_end_nr(buf, i) > BTRFS_LEAF_DATA_SIZE(root)) {
529 btrfs_item_key(buf, &key, 0);
530 btrfs_print_key(&key);
531 fflush(stdout);
532 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
533 fprintf(stderr, "slot end outside of leaf %llu > %llu\n",
534 (unsigned long long)btrfs_item_end_nr(buf, i),
535 (unsigned long long)BTRFS_LEAF_DATA_SIZE(root));
536 goto fail;
537 }
538 }
539
540 return BTRFS_TREE_BLOCK_CLEAN;
541 fail:
542 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
543 if (parent_key)
544 btrfs_disk_key_to_cpu(&cpukey, parent_key);
545 else
546 btrfs_item_key_to_cpu(buf, &cpukey, 0);
547
548 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
549 buf->start, buf->len, 0);
550 }
551 return ret;
552 }
553
554 static int noinline check_block(struct btrfs_root *root,
555 struct btrfs_path *path, int level)
556 {
557 struct btrfs_disk_key key;
558 struct btrfs_disk_key *key_ptr = NULL;
559 struct extent_buffer *parent;
560 enum btrfs_tree_block_status ret;
561
562 if (path->skip_check_block)
563 return 0;
564 if (path->nodes[level + 1]) {
565 parent = path->nodes[level + 1];
566 btrfs_node_key(parent, &key, path->slots[level + 1]);
567 key_ptr = &key;
568 }
569 if (level == 0)
570 ret = btrfs_check_leaf(root, key_ptr, path->nodes[0]);
571 else
572 ret = btrfs_check_node(root, key_ptr, path->nodes[level]);
573 if (ret == BTRFS_TREE_BLOCK_CLEAN)
574 return 0;
575 return -EIO;
576 }
577
578 /*
579 * search for key in the extent_buffer. The items start at offset p,
580 * and they are item_size apart. There are 'max' items in p.
581 *
582 * the slot in the array is returned via slot, and it points to
583 * the place where you would insert key if it is not found in
584 * the array.
585 *
586 * slot may point to max if the key is bigger than all of the keys
587 */
588 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
589 int item_size, struct btrfs_key *key,
590 int max, int *slot)
591 {
592 int low = 0;
593 int high = max;
594 int mid;
595 int ret;
596 unsigned long offset;
597 struct btrfs_disk_key *tmp;
598
599 while(low < high) {
600 mid = (low + high) / 2;
601 offset = p + mid * item_size;
602
603 tmp = (struct btrfs_disk_key *)(eb->data + offset);
604 ret = btrfs_comp_keys(tmp, key);
605
606 if (ret < 0)
607 low = mid + 1;
608 else if (ret > 0)
609 high = mid;
610 else {
611 *slot = mid;
612 return 0;
613 }
614 }
615 *slot = low;
616 return 1;
617 }
618
619 /*
620 * simple bin_search frontend that does the right thing for
621 * leaves vs nodes
622 */
623 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
624 int level, int *slot)
625 {
626 if (level == 0)
627 return generic_bin_search(eb,
628 offsetof(struct btrfs_leaf, items),
629 sizeof(struct btrfs_item),
630 key, btrfs_header_nritems(eb),
631 slot);
632 else
633 return generic_bin_search(eb,
634 offsetof(struct btrfs_node, ptrs),
635 sizeof(struct btrfs_key_ptr),
636 key, btrfs_header_nritems(eb),
637 slot);
638 }
639
640 struct extent_buffer *read_node_slot(struct btrfs_root *root,
641 struct extent_buffer *parent, int slot)
642 {
643 int level = btrfs_header_level(parent);
644 if (slot < 0)
645 return NULL;
646 if (slot >= btrfs_header_nritems(parent))
647 return NULL;
648
649 if (level == 0)
650 return NULL;
651
652 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
653 root->nodesize,
654 btrfs_node_ptr_generation(parent, slot));
655 }
656
657 static int balance_level(struct btrfs_trans_handle *trans,
658 struct btrfs_root *root,
659 struct btrfs_path *path, int level)
660 {
661 struct extent_buffer *right = NULL;
662 struct extent_buffer *mid;
663 struct extent_buffer *left = NULL;
664 struct extent_buffer *parent = NULL;
665 int ret = 0;
666 int wret;
667 int pslot;
668 int orig_slot = path->slots[level];
669 u64 orig_ptr;
670
671 if (level == 0)
672 return 0;
673
674 mid = path->nodes[level];
675 WARN_ON(btrfs_header_generation(mid) != trans->transid);
676
677 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
678
679 if (level < BTRFS_MAX_LEVEL - 1) {
680 parent = path->nodes[level + 1];
681 pslot = path->slots[level + 1];
682 }
683
684 /*
685 * deal with the case where there is only one pointer in the root
686 * by promoting the node below to a root
687 */
688 if (!parent) {
689 struct extent_buffer *child;
690
691 if (btrfs_header_nritems(mid) != 1)
692 return 0;
693
694 /* promote the child to a root */
695 child = read_node_slot(root, mid, 0);
696 BUG_ON(!extent_buffer_uptodate(child));
697 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
698 BUG_ON(ret);
699
700 root->node = child;
701 add_root_to_dirty_list(root);
702 path->nodes[level] = NULL;
703 clean_tree_block(trans, root, mid);
704 wait_on_tree_block_writeback(root, mid);
705 /* once for the path */
706 free_extent_buffer(mid);
707
708 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
709 0, root->root_key.objectid,
710 level, 1);
711 /* once for the root ptr */
712 free_extent_buffer(mid);
713 return ret;
714 }
715 if (btrfs_header_nritems(mid) >
716 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
717 return 0;
718
719 left = read_node_slot(root, parent, pslot - 1);
720 if (extent_buffer_uptodate(left)) {
721 wret = btrfs_cow_block(trans, root, left,
722 parent, pslot - 1, &left);
723 if (wret) {
724 ret = wret;
725 goto enospc;
726 }
727 }
728 right = read_node_slot(root, parent, pslot + 1);
729 if (extent_buffer_uptodate(right)) {
730 wret = btrfs_cow_block(trans, root, right,
731 parent, pslot + 1, &right);
732 if (wret) {
733 ret = wret;
734 goto enospc;
735 }
736 }
737
738 /* first, try to make some room in the middle buffer */
739 if (left) {
740 orig_slot += btrfs_header_nritems(left);
741 wret = push_node_left(trans, root, left, mid, 1);
742 if (wret < 0)
743 ret = wret;
744 }
745
746 /*
747 * then try to empty the right most buffer into the middle
748 */
749 if (right) {
750 wret = push_node_left(trans, root, mid, right, 1);
751 if (wret < 0 && wret != -ENOSPC)
752 ret = wret;
753 if (btrfs_header_nritems(right) == 0) {
754 u64 bytenr = right->start;
755 u32 blocksize = right->len;
756
757 clean_tree_block(trans, root, right);
758 wait_on_tree_block_writeback(root, right);
759 free_extent_buffer(right);
760 right = NULL;
761 wret = btrfs_del_ptr(trans, root, path,
762 level + 1, pslot + 1);
763 if (wret)
764 ret = wret;
765 wret = btrfs_free_extent(trans, root, bytenr,
766 blocksize, 0,
767 root->root_key.objectid,
768 level, 0);
769 if (wret)
770 ret = wret;
771 } else {
772 struct btrfs_disk_key right_key;
773 btrfs_node_key(right, &right_key, 0);
774 btrfs_set_node_key(parent, &right_key, pslot + 1);
775 btrfs_mark_buffer_dirty(parent);
776 }
777 }
778 if (btrfs_header_nritems(mid) == 1) {
779 /*
780 * we're not allowed to leave a node with one item in the
781 * tree during a delete. A deletion from lower in the tree
782 * could try to delete the only pointer in this node.
783 * So, pull some keys from the left.
784 * There has to be a left pointer at this point because
785 * otherwise we would have pulled some pointers from the
786 * right
787 */
788 BUG_ON(!left);
789 wret = balance_node_right(trans, root, mid, left);
790 if (wret < 0) {
791 ret = wret;
792 goto enospc;
793 }
794 if (wret == 1) {
795 wret = push_node_left(trans, root, left, mid, 1);
796 if (wret < 0)
797 ret = wret;
798 }
799 BUG_ON(wret == 1);
800 }
801 if (btrfs_header_nritems(mid) == 0) {
802 /* we've managed to empty the middle node, drop it */
803 u64 bytenr = mid->start;
804 u32 blocksize = mid->len;
805 clean_tree_block(trans, root, mid);
806 wait_on_tree_block_writeback(root, mid);
807 free_extent_buffer(mid);
808 mid = NULL;
809 wret = btrfs_del_ptr(trans, root, path, level + 1, pslot);
810 if (wret)
811 ret = wret;
812 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
813 0, root->root_key.objectid,
814 level, 0);
815 if (wret)
816 ret = wret;
817 } else {
818 /* update the parent key to reflect our changes */
819 struct btrfs_disk_key mid_key;
820 btrfs_node_key(mid, &mid_key, 0);
821 btrfs_set_node_key(parent, &mid_key, pslot);
822 btrfs_mark_buffer_dirty(parent);
823 }
824
825 /* update the path */
826 if (left) {
827 if (btrfs_header_nritems(left) > orig_slot) {
828 extent_buffer_get(left);
829 path->nodes[level] = left;
830 path->slots[level + 1] -= 1;
831 path->slots[level] = orig_slot;
832 if (mid)
833 free_extent_buffer(mid);
834 } else {
835 orig_slot -= btrfs_header_nritems(left);
836 path->slots[level] = orig_slot;
837 }
838 }
839 /* double check we haven't messed things up */
840 check_block(root, path, level);
841 if (orig_ptr !=
842 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
843 BUG();
844 enospc:
845 if (right)
846 free_extent_buffer(right);
847 if (left)
848 free_extent_buffer(left);
849 return ret;
850 }
851
852 /* returns zero if the push worked, non-zero otherwise */
853 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
854 struct btrfs_root *root,
855 struct btrfs_path *path, int level)
856 {
857 struct extent_buffer *right = NULL;
858 struct extent_buffer *mid;
859 struct extent_buffer *left = NULL;
860 struct extent_buffer *parent = NULL;
861 int ret = 0;
862 int wret;
863 int pslot;
864 int orig_slot = path->slots[level];
865
866 if (level == 0)
867 return 1;
868
869 mid = path->nodes[level];
870 WARN_ON(btrfs_header_generation(mid) != trans->transid);
871
872 if (level < BTRFS_MAX_LEVEL - 1) {
873 parent = path->nodes[level + 1];
874 pslot = path->slots[level + 1];
875 }
876
877 if (!parent)
878 return 1;
879
880 left = read_node_slot(root, parent, pslot - 1);
881
882 /* first, try to make some room in the middle buffer */
883 if (extent_buffer_uptodate(left)) {
884 u32 left_nr;
885 left_nr = btrfs_header_nritems(left);
886 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
887 wret = 1;
888 } else {
889 ret = btrfs_cow_block(trans, root, left, parent,
890 pslot - 1, &left);
891 if (ret)
892 wret = 1;
893 else {
894 wret = push_node_left(trans, root,
895 left, mid, 0);
896 }
897 }
898 if (wret < 0)
899 ret = wret;
900 if (wret == 0) {
901 struct btrfs_disk_key disk_key;
902 orig_slot += left_nr;
903 btrfs_node_key(mid, &disk_key, 0);
904 btrfs_set_node_key(parent, &disk_key, pslot);
905 btrfs_mark_buffer_dirty(parent);
906 if (btrfs_header_nritems(left) > orig_slot) {
907 path->nodes[level] = left;
908 path->slots[level + 1] -= 1;
909 path->slots[level] = orig_slot;
910 free_extent_buffer(mid);
911 } else {
912 orig_slot -=
913 btrfs_header_nritems(left);
914 path->slots[level] = orig_slot;
915 free_extent_buffer(left);
916 }
917 return 0;
918 }
919 free_extent_buffer(left);
920 }
921 right= read_node_slot(root, parent, pslot + 1);
922
923 /*
924 * then try to empty the right most buffer into the middle
925 */
926 if (extent_buffer_uptodate(right)) {
927 u32 right_nr;
928 right_nr = btrfs_header_nritems(right);
929 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
930 wret = 1;
931 } else {
932 ret = btrfs_cow_block(trans, root, right,
933 parent, pslot + 1,
934 &right);
935 if (ret)
936 wret = 1;
937 else {
938 wret = balance_node_right(trans, root,
939 right, mid);
940 }
941 }
942 if (wret < 0)
943 ret = wret;
944 if (wret == 0) {
945 struct btrfs_disk_key disk_key;
946
947 btrfs_node_key(right, &disk_key, 0);
948 btrfs_set_node_key(parent, &disk_key, pslot + 1);
949 btrfs_mark_buffer_dirty(parent);
950
951 if (btrfs_header_nritems(mid) <= orig_slot) {
952 path->nodes[level] = right;
953 path->slots[level + 1] += 1;
954 path->slots[level] = orig_slot -
955 btrfs_header_nritems(mid);
956 free_extent_buffer(mid);
957 } else {
958 free_extent_buffer(right);
959 }
960 return 0;
961 }
962 free_extent_buffer(right);
963 }
964 return 1;
965 }
966
967 /*
968 * readahead one full node of leaves
969 */
970 void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
971 int level, int slot, u64 objectid)
972 {
973 struct extent_buffer *node;
974 struct btrfs_disk_key disk_key;
975 u32 nritems;
976 u64 search;
977 u64 lowest_read;
978 u64 highest_read;
979 u64 nread = 0;
980 int direction = path->reada;
981 struct extent_buffer *eb;
982 u32 nr;
983 u32 blocksize;
984 u32 nscan = 0;
985
986 if (level != 1)
987 return;
988
989 if (!path->nodes[level])
990 return;
991
992 node = path->nodes[level];
993 search = btrfs_node_blockptr(node, slot);
994 blocksize = root->nodesize;
995 eb = btrfs_find_tree_block(root, search, blocksize);
996 if (eb) {
997 free_extent_buffer(eb);
998 return;
999 }
1000
1001 highest_read = search;
1002 lowest_read = search;
1003
1004 nritems = btrfs_header_nritems(node);
1005 nr = slot;
1006 while(1) {
1007 if (direction < 0) {
1008 if (nr == 0)
1009 break;
1010 nr--;
1011 } else if (direction > 0) {
1012 nr++;
1013 if (nr >= nritems)
1014 break;
1015 }
1016 if (path->reada < 0 && objectid) {
1017 btrfs_node_key(node, &disk_key, nr);
1018 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1019 break;
1020 }
1021 search = btrfs_node_blockptr(node, nr);
1022 if ((search >= lowest_read && search <= highest_read) ||
1023 (search < lowest_read && lowest_read - search <= 32768) ||
1024 (search > highest_read && search - highest_read <= 32768)) {
1025 readahead_tree_block(root, search, blocksize,
1026 btrfs_node_ptr_generation(node, nr));
1027 nread += blocksize;
1028 }
1029 nscan++;
1030 if (path->reada < 2 && (nread > SZ_256K || nscan > 32))
1031 break;
1032 if(nread > SZ_1M || nscan > 128)
1033 break;
1034
1035 if (search < lowest_read)
1036 lowest_read = search;
1037 if (search > highest_read)
1038 highest_read = search;
1039 }
1040 }
1041
1042 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
1043 u64 iobjectid, u64 ioff, u8 key_type,
1044 struct btrfs_key *found_key)
1045 {
1046 int ret;
1047 struct btrfs_key key;
1048 struct extent_buffer *eb;
1049 struct btrfs_path *path;
1050
1051 key.type = key_type;
1052 key.objectid = iobjectid;
1053 key.offset = ioff;
1054
1055 if (found_path == NULL) {
1056 path = btrfs_alloc_path();
1057 if (!path)
1058 return -ENOMEM;
1059 } else
1060 path = found_path;
1061
1062 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1063 if ((ret < 0) || (found_key == NULL))
1064 goto out;
1065
1066 eb = path->nodes[0];
1067 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1068 ret = btrfs_next_leaf(fs_root, path);
1069 if (ret)
1070 goto out;
1071 eb = path->nodes[0];
1072 }
1073
1074 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1075 if (found_key->type != key.type ||
1076 found_key->objectid != key.objectid) {
1077 ret = 1;
1078 goto out;
1079 }
1080
1081 out:
1082 if (path != found_path)
1083 btrfs_free_path(path);
1084 return ret;
1085 }
1086
1087 /*
1088 * look for key in the tree. path is filled in with nodes along the way
1089 * if key is found, we return zero and you can find the item in the leaf
1090 * level of the path (level 0)
1091 *
1092 * If the key isn't found, the path points to the slot where it should
1093 * be inserted, and 1 is returned. If there are other errors during the
1094 * search a negative error number is returned.
1095 *
1096 * if ins_len > 0, nodes and leaves will be split as we walk down the
1097 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1098 * possible)
1099 */
1100 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1101 *root, struct btrfs_key *key, struct btrfs_path *p, int
1102 ins_len, int cow)
1103 {
1104 struct extent_buffer *b;
1105 int slot;
1106 int ret;
1107 int level;
1108 int should_reada = p->reada;
1109 u8 lowest_level = 0;
1110
1111 lowest_level = p->lowest_level;
1112 WARN_ON(lowest_level && ins_len > 0);
1113 WARN_ON(p->nodes[0] != NULL);
1114 /*
1115 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1116 */
1117 again:
1118 b = root->node;
1119 extent_buffer_get(b);
1120 while (b) {
1121 level = btrfs_header_level(b);
1122 if (cow) {
1123 int wret;
1124 wret = btrfs_cow_block(trans, root, b,
1125 p->nodes[level + 1],
1126 p->slots[level + 1],
1127 &b);
1128 if (wret) {
1129 free_extent_buffer(b);
1130 return wret;
1131 }
1132 }
1133 BUG_ON(!cow && ins_len);
1134 if (level != btrfs_header_level(b))
1135 WARN_ON(1);
1136 level = btrfs_header_level(b);
1137 p->nodes[level] = b;
1138 ret = check_block(root, p, level);
1139 if (ret)
1140 return -1;
1141 ret = bin_search(b, key, level, &slot);
1142 if (level != 0) {
1143 if (ret && slot > 0)
1144 slot -= 1;
1145 p->slots[level] = slot;
1146 if ((p->search_for_split || ins_len > 0) &&
1147 btrfs_header_nritems(b) >=
1148 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1149 int sret = split_node(trans, root, p, level);
1150 BUG_ON(sret > 0);
1151 if (sret)
1152 return sret;
1153 b = p->nodes[level];
1154 slot = p->slots[level];
1155 } else if (ins_len < 0) {
1156 int sret = balance_level(trans, root, p,
1157 level);
1158 if (sret)
1159 return sret;
1160 b = p->nodes[level];
1161 if (!b) {
1162 btrfs_release_path(p);
1163 goto again;
1164 }
1165 slot = p->slots[level];
1166 BUG_ON(btrfs_header_nritems(b) == 1);
1167 }
1168 /* this is only true while dropping a snapshot */
1169 if (level == lowest_level)
1170 break;
1171
1172 if (should_reada)
1173 reada_for_search(root, p, level, slot,
1174 key->objectid);
1175
1176 b = read_node_slot(root, b, slot);
1177 if (!extent_buffer_uptodate(b))
1178 return -EIO;
1179 } else {
1180 p->slots[level] = slot;
1181 if (ins_len > 0 &&
1182 ins_len > btrfs_leaf_free_space(root, b)) {
1183 int sret = split_leaf(trans, root, key,
1184 p, ins_len, ret == 0);
1185 BUG_ON(sret > 0);
1186 if (sret)
1187 return sret;
1188 }
1189 return ret;
1190 }
1191 }
1192 return 1;
1193 }
1194
1195 /*
1196 * adjust the pointers going up the tree, starting at level
1197 * making sure the right key of each node is points to 'key'.
1198 * This is used after shifting pointers to the left, so it stops
1199 * fixing up pointers when a given leaf/node is not in slot 0 of the
1200 * higher levels
1201 */
1202 void btrfs_fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
1203 struct btrfs_disk_key *key, int level)
1204 {
1205 int i;
1206 struct extent_buffer *t;
1207
1208 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1209 int tslot = path->slots[i];
1210 if (!path->nodes[i])
1211 break;
1212 t = path->nodes[i];
1213 btrfs_set_node_key(t, key, tslot);
1214 btrfs_mark_buffer_dirty(path->nodes[i]);
1215 if (tslot != 0)
1216 break;
1217 }
1218 }
1219
1220 /*
1221 * update item key.
1222 *
1223 * This function isn't completely safe. It's the caller's responsibility
1224 * that the new key won't break the order
1225 */
1226 int btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
1227 struct btrfs_key *new_key)
1228 {
1229 struct btrfs_disk_key disk_key;
1230 struct extent_buffer *eb;
1231 int slot;
1232
1233 eb = path->nodes[0];
1234 slot = path->slots[0];
1235 if (slot > 0) {
1236 btrfs_item_key(eb, &disk_key, slot - 1);
1237 if (btrfs_comp_keys(&disk_key, new_key) >= 0)
1238 return -1;
1239 }
1240 if (slot < btrfs_header_nritems(eb) - 1) {
1241 btrfs_item_key(eb, &disk_key, slot + 1);
1242 if (btrfs_comp_keys(&disk_key, new_key) <= 0)
1243 return -1;
1244 }
1245
1246 btrfs_cpu_key_to_disk(&disk_key, new_key);
1247 btrfs_set_item_key(eb, &disk_key, slot);
1248 btrfs_mark_buffer_dirty(eb);
1249 if (slot == 0)
1250 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1251 return 0;
1252 }
1253
1254 /*
1255 * update an item key without the safety checks. This is meant to be called by
1256 * fsck only.
1257 */
1258 void btrfs_set_item_key_unsafe(struct btrfs_root *root,
1259 struct btrfs_path *path,
1260 struct btrfs_key *new_key)
1261 {
1262 struct btrfs_disk_key disk_key;
1263 struct extent_buffer *eb;
1264 int slot;
1265
1266 eb = path->nodes[0];
1267 slot = path->slots[0];
1268
1269 btrfs_cpu_key_to_disk(&disk_key, new_key);
1270 btrfs_set_item_key(eb, &disk_key, slot);
1271 btrfs_mark_buffer_dirty(eb);
1272 if (slot == 0)
1273 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1274 }
1275
1276 /*
1277 * try to push data from one node into the next node left in the
1278 * tree.
1279 *
1280 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1281 * error, and > 0 if there was no room in the left hand block.
1282 */
1283 static int push_node_left(struct btrfs_trans_handle *trans,
1284 struct btrfs_root *root, struct extent_buffer *dst,
1285 struct extent_buffer *src, int empty)
1286 {
1287 int push_items = 0;
1288 int src_nritems;
1289 int dst_nritems;
1290 int ret = 0;
1291
1292 src_nritems = btrfs_header_nritems(src);
1293 dst_nritems = btrfs_header_nritems(dst);
1294 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1295 WARN_ON(btrfs_header_generation(src) != trans->transid);
1296 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1297
1298 if (!empty && src_nritems <= 8)
1299 return 1;
1300
1301 if (push_items <= 0) {
1302 return 1;
1303 }
1304
1305 if (empty) {
1306 push_items = min(src_nritems, push_items);
1307 if (push_items < src_nritems) {
1308 /* leave at least 8 pointers in the node if
1309 * we aren't going to empty it
1310 */
1311 if (src_nritems - push_items < 8) {
1312 if (push_items <= 8)
1313 return 1;
1314 push_items -= 8;
1315 }
1316 }
1317 } else
1318 push_items = min(src_nritems - 8, push_items);
1319
1320 copy_extent_buffer(dst, src,
1321 btrfs_node_key_ptr_offset(dst_nritems),
1322 btrfs_node_key_ptr_offset(0),
1323 push_items * sizeof(struct btrfs_key_ptr));
1324
1325 if (push_items < src_nritems) {
1326 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1327 btrfs_node_key_ptr_offset(push_items),
1328 (src_nritems - push_items) *
1329 sizeof(struct btrfs_key_ptr));
1330 }
1331 btrfs_set_header_nritems(src, src_nritems - push_items);
1332 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1333 btrfs_mark_buffer_dirty(src);
1334 btrfs_mark_buffer_dirty(dst);
1335
1336 return ret;
1337 }
1338
1339 /*
1340 * try to push data from one node into the next node right in the
1341 * tree.
1342 *
1343 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1344 * error, and > 0 if there was no room in the right hand block.
1345 *
1346 * this will only push up to 1/2 the contents of the left node over
1347 */
1348 static int balance_node_right(struct btrfs_trans_handle *trans,
1349 struct btrfs_root *root,
1350 struct extent_buffer *dst,
1351 struct extent_buffer *src)
1352 {
1353 int push_items = 0;
1354 int max_push;
1355 int src_nritems;
1356 int dst_nritems;
1357 int ret = 0;
1358
1359 WARN_ON(btrfs_header_generation(src) != trans->transid);
1360 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1361
1362 src_nritems = btrfs_header_nritems(src);
1363 dst_nritems = btrfs_header_nritems(dst);
1364 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1365 if (push_items <= 0) {
1366 return 1;
1367 }
1368
1369 if (src_nritems < 4) {
1370 return 1;
1371 }
1372
1373 max_push = src_nritems / 2 + 1;
1374 /* don't try to empty the node */
1375 if (max_push >= src_nritems) {
1376 return 1;
1377 }
1378
1379 if (max_push < push_items)
1380 push_items = max_push;
1381
1382 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1383 btrfs_node_key_ptr_offset(0),
1384 (dst_nritems) *
1385 sizeof(struct btrfs_key_ptr));
1386
1387 copy_extent_buffer(dst, src,
1388 btrfs_node_key_ptr_offset(0),
1389 btrfs_node_key_ptr_offset(src_nritems - push_items),
1390 push_items * sizeof(struct btrfs_key_ptr));
1391
1392 btrfs_set_header_nritems(src, src_nritems - push_items);
1393 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1394
1395 btrfs_mark_buffer_dirty(src);
1396 btrfs_mark_buffer_dirty(dst);
1397
1398 return ret;
1399 }
1400
1401 /*
1402 * helper function to insert a new root level in the tree.
1403 * A new node is allocated, and a single item is inserted to
1404 * point to the existing root
1405 *
1406 * returns zero on success or < 0 on failure.
1407 */
1408 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1409 struct btrfs_root *root,
1410 struct btrfs_path *path, int level)
1411 {
1412 u64 lower_gen;
1413 struct extent_buffer *lower;
1414 struct extent_buffer *c;
1415 struct extent_buffer *old;
1416 struct btrfs_disk_key lower_key;
1417
1418 BUG_ON(path->nodes[level]);
1419 BUG_ON(path->nodes[level-1] != root->node);
1420
1421 lower = path->nodes[level-1];
1422 if (level == 1)
1423 btrfs_item_key(lower, &lower_key, 0);
1424 else
1425 btrfs_node_key(lower, &lower_key, 0);
1426
1427 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1428 root->root_key.objectid, &lower_key,
1429 level, root->node->start, 0);
1430
1431 if (IS_ERR(c))
1432 return PTR_ERR(c);
1433
1434 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
1435 btrfs_set_header_nritems(c, 1);
1436 btrfs_set_header_level(c, level);
1437 btrfs_set_header_bytenr(c, c->start);
1438 btrfs_set_header_generation(c, trans->transid);
1439 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
1440 btrfs_set_header_owner(c, root->root_key.objectid);
1441
1442 write_extent_buffer(c, root->fs_info->fsid,
1443 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1444
1445 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1446 btrfs_header_chunk_tree_uuid(c),
1447 BTRFS_UUID_SIZE);
1448
1449 btrfs_set_node_key(c, &lower_key, 0);
1450 btrfs_set_node_blockptr(c, 0, lower->start);
1451 lower_gen = btrfs_header_generation(lower);
1452 WARN_ON(lower_gen != trans->transid);
1453
1454 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1455
1456 btrfs_mark_buffer_dirty(c);
1457
1458 old = root->node;
1459 root->node = c;
1460
1461 /* the super has an extra ref to root->node */
1462 free_extent_buffer(old);
1463
1464 add_root_to_dirty_list(root);
1465 extent_buffer_get(c);
1466 path->nodes[level] = c;
1467 path->slots[level] = 0;
1468 return 0;
1469 }
1470
1471 /*
1472 * worker function to insert a single pointer in a node.
1473 * the node should have enough room for the pointer already
1474 *
1475 * slot and level indicate where you want the key to go, and
1476 * blocknr is the block the key points to.
1477 *
1478 * returns zero on success and < 0 on any error
1479 */
1480 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1481 *root, struct btrfs_path *path, struct btrfs_disk_key
1482 *key, u64 bytenr, int slot, int level)
1483 {
1484 struct extent_buffer *lower;
1485 int nritems;
1486
1487 BUG_ON(!path->nodes[level]);
1488 lower = path->nodes[level];
1489 nritems = btrfs_header_nritems(lower);
1490 if (slot > nritems)
1491 BUG();
1492 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1493 BUG();
1494 if (slot != nritems) {
1495 memmove_extent_buffer(lower,
1496 btrfs_node_key_ptr_offset(slot + 1),
1497 btrfs_node_key_ptr_offset(slot),
1498 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1499 }
1500 btrfs_set_node_key(lower, key, slot);
1501 btrfs_set_node_blockptr(lower, slot, bytenr);
1502 WARN_ON(trans->transid == 0);
1503 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1504 btrfs_set_header_nritems(lower, nritems + 1);
1505 btrfs_mark_buffer_dirty(lower);
1506 return 0;
1507 }
1508
1509 /*
1510 * split the node at the specified level in path in two.
1511 * The path is corrected to point to the appropriate node after the split
1512 *
1513 * Before splitting this tries to make some room in the node by pushing
1514 * left and right, if either one works, it returns right away.
1515 *
1516 * returns 0 on success and < 0 on failure
1517 */
1518 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1519 *root, struct btrfs_path *path, int level)
1520 {
1521 struct extent_buffer *c;
1522 struct extent_buffer *split;
1523 struct btrfs_disk_key disk_key;
1524 int mid;
1525 int ret;
1526 int wret;
1527 u32 c_nritems;
1528
1529 c = path->nodes[level];
1530 WARN_ON(btrfs_header_generation(c) != trans->transid);
1531 if (c == root->node) {
1532 /* trying to split the root, lets make a new one */
1533 ret = insert_new_root(trans, root, path, level + 1);
1534 if (ret)
1535 return ret;
1536 } else {
1537 ret = push_nodes_for_insert(trans, root, path, level);
1538 c = path->nodes[level];
1539 if (!ret && btrfs_header_nritems(c) <
1540 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1541 return 0;
1542 if (ret < 0)
1543 return ret;
1544 }
1545
1546 c_nritems = btrfs_header_nritems(c);
1547 mid = (c_nritems + 1) / 2;
1548 btrfs_node_key(c, &disk_key, mid);
1549
1550 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1551 root->root_key.objectid,
1552 &disk_key, level, c->start, 0);
1553 if (IS_ERR(split))
1554 return PTR_ERR(split);
1555
1556 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
1557 btrfs_set_header_level(split, btrfs_header_level(c));
1558 btrfs_set_header_bytenr(split, split->start);
1559 btrfs_set_header_generation(split, trans->transid);
1560 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
1561 btrfs_set_header_owner(split, root->root_key.objectid);
1562 write_extent_buffer(split, root->fs_info->fsid,
1563 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1564 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1565 btrfs_header_chunk_tree_uuid(split),
1566 BTRFS_UUID_SIZE);
1567
1568
1569 copy_extent_buffer(split, c,
1570 btrfs_node_key_ptr_offset(0),
1571 btrfs_node_key_ptr_offset(mid),
1572 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1573 btrfs_set_header_nritems(split, c_nritems - mid);
1574 btrfs_set_header_nritems(c, mid);
1575 ret = 0;
1576
1577 btrfs_mark_buffer_dirty(c);
1578 btrfs_mark_buffer_dirty(split);
1579
1580 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1581 path->slots[level + 1] + 1,
1582 level + 1);
1583 if (wret)
1584 ret = wret;
1585
1586 if (path->slots[level] >= mid) {
1587 path->slots[level] -= mid;
1588 free_extent_buffer(c);
1589 path->nodes[level] = split;
1590 path->slots[level + 1] += 1;
1591 } else {
1592 free_extent_buffer(split);
1593 }
1594 return ret;
1595 }
1596
1597 /*
1598 * how many bytes are required to store the items in a leaf. start
1599 * and nr indicate which items in the leaf to check. This totals up the
1600 * space used both by the item structs and the item data
1601 */
1602 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1603 {
1604 int data_len;
1605 int nritems = btrfs_header_nritems(l);
1606 int end = min(nritems, start + nr) - 1;
1607
1608 if (!nr)
1609 return 0;
1610 data_len = btrfs_item_end_nr(l, start);
1611 data_len = data_len - btrfs_item_offset_nr(l, end);
1612 data_len += sizeof(struct btrfs_item) * nr;
1613 WARN_ON(data_len < 0);
1614 return data_len;
1615 }
1616
1617 /*
1618 * The space between the end of the leaf items and
1619 * the start of the leaf data. IOW, how much room
1620 * the leaf has left for both items and data
1621 */
1622 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1623 {
1624 u32 nodesize = (root ? BTRFS_LEAF_DATA_SIZE(root) : leaf->len);
1625 int nritems = btrfs_header_nritems(leaf);
1626 int ret;
1627 ret = nodesize - leaf_space_used(leaf, 0, nritems);
1628 if (ret < 0) {
1629 printk("leaf free space ret %d, leaf data size %u, used %d nritems %d\n",
1630 ret, nodesize, leaf_space_used(leaf, 0, nritems),
1631 nritems);
1632 }
1633 return ret;
1634 }
1635
1636 /*
1637 * push some data in the path leaf to the right, trying to free up at
1638 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1639 *
1640 * returns 1 if the push failed because the other node didn't have enough
1641 * room, 0 if everything worked out and < 0 if there were major errors.
1642 */
1643 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1644 *root, struct btrfs_path *path, int data_size,
1645 int empty)
1646 {
1647 struct extent_buffer *left = path->nodes[0];
1648 struct extent_buffer *right;
1649 struct extent_buffer *upper;
1650 struct btrfs_disk_key disk_key;
1651 int slot;
1652 u32 i;
1653 int free_space;
1654 int push_space = 0;
1655 int push_items = 0;
1656 struct btrfs_item *item;
1657 u32 left_nritems;
1658 u32 nr;
1659 u32 right_nritems;
1660 u32 data_end;
1661 u32 this_item_size;
1662 int ret;
1663
1664 slot = path->slots[1];
1665 if (!path->nodes[1]) {
1666 return 1;
1667 }
1668 upper = path->nodes[1];
1669 if (slot >= btrfs_header_nritems(upper) - 1)
1670 return 1;
1671
1672 right = read_node_slot(root, upper, slot + 1);
1673 if (!extent_buffer_uptodate(right)) {
1674 if (IS_ERR(right))
1675 return PTR_ERR(right);
1676 return -EIO;
1677 }
1678 free_space = btrfs_leaf_free_space(root, right);
1679 if (free_space < data_size) {
1680 free_extent_buffer(right);
1681 return 1;
1682 }
1683
1684 /* cow and double check */
1685 ret = btrfs_cow_block(trans, root, right, upper,
1686 slot + 1, &right);
1687 if (ret) {
1688 free_extent_buffer(right);
1689 return 1;
1690 }
1691 free_space = btrfs_leaf_free_space(root, right);
1692 if (free_space < data_size) {
1693 free_extent_buffer(right);
1694 return 1;
1695 }
1696
1697 left_nritems = btrfs_header_nritems(left);
1698 if (left_nritems == 0) {
1699 free_extent_buffer(right);
1700 return 1;
1701 }
1702
1703 if (empty)
1704 nr = 0;
1705 else
1706 nr = 1;
1707
1708 i = left_nritems - 1;
1709 while (i >= nr) {
1710 item = btrfs_item_nr(i);
1711
1712 if (path->slots[0] == i)
1713 push_space += data_size + sizeof(*item);
1714
1715 this_item_size = btrfs_item_size(left, item);
1716 if (this_item_size + sizeof(*item) + push_space > free_space)
1717 break;
1718 push_items++;
1719 push_space += this_item_size + sizeof(*item);
1720 if (i == 0)
1721 break;
1722 i--;
1723 }
1724
1725 if (push_items == 0) {
1726 free_extent_buffer(right);
1727 return 1;
1728 }
1729
1730 if (!empty && push_items == left_nritems)
1731 WARN_ON(1);
1732
1733 /* push left to right */
1734 right_nritems = btrfs_header_nritems(right);
1735
1736 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1737 push_space -= leaf_data_end(root, left);
1738
1739 /* make room in the right data area */
1740 data_end = leaf_data_end(root, right);
1741 memmove_extent_buffer(right,
1742 btrfs_leaf_data(right) + data_end - push_space,
1743 btrfs_leaf_data(right) + data_end,
1744 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1745
1746 /* copy from the left data area */
1747 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1748 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1749 btrfs_leaf_data(left) + leaf_data_end(root, left),
1750 push_space);
1751
1752 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1753 btrfs_item_nr_offset(0),
1754 right_nritems * sizeof(struct btrfs_item));
1755
1756 /* copy the items from left to right */
1757 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1758 btrfs_item_nr_offset(left_nritems - push_items),
1759 push_items * sizeof(struct btrfs_item));
1760
1761 /* update the item pointers */
1762 right_nritems += push_items;
1763 btrfs_set_header_nritems(right, right_nritems);
1764 push_space = BTRFS_LEAF_DATA_SIZE(root);
1765 for (i = 0; i < right_nritems; i++) {
1766 item = btrfs_item_nr(i);
1767 push_space -= btrfs_item_size(right, item);
1768 btrfs_set_item_offset(right, item, push_space);
1769 }
1770
1771 left_nritems -= push_items;
1772 btrfs_set_header_nritems(left, left_nritems);
1773
1774 if (left_nritems)
1775 btrfs_mark_buffer_dirty(left);
1776 btrfs_mark_buffer_dirty(right);
1777
1778 btrfs_item_key(right, &disk_key, 0);
1779 btrfs_set_node_key(upper, &disk_key, slot + 1);
1780 btrfs_mark_buffer_dirty(upper);
1781
1782 /* then fixup the leaf pointer in the path */
1783 if (path->slots[0] >= left_nritems) {
1784 path->slots[0] -= left_nritems;
1785 free_extent_buffer(path->nodes[0]);
1786 path->nodes[0] = right;
1787 path->slots[1] += 1;
1788 } else {
1789 free_extent_buffer(right);
1790 }
1791 return 0;
1792 }
1793 /*
1794 * push some data in the path leaf to the left, trying to free up at
1795 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1796 */
1797 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1798 *root, struct btrfs_path *path, int data_size,
1799 int empty)
1800 {
1801 struct btrfs_disk_key disk_key;
1802 struct extent_buffer *right = path->nodes[0];
1803 struct extent_buffer *left;
1804 int slot;
1805 int i;
1806 int free_space;
1807 int push_space = 0;
1808 int push_items = 0;
1809 struct btrfs_item *item;
1810 u32 old_left_nritems;
1811 u32 right_nritems;
1812 u32 nr;
1813 int ret = 0;
1814 u32 this_item_size;
1815 u32 old_left_item_size;
1816
1817 slot = path->slots[1];
1818 if (slot == 0)
1819 return 1;
1820 if (!path->nodes[1])
1821 return 1;
1822
1823 right_nritems = btrfs_header_nritems(right);
1824 if (right_nritems == 0) {
1825 return 1;
1826 }
1827
1828 left = read_node_slot(root, path->nodes[1], slot - 1);
1829 free_space = btrfs_leaf_free_space(root, left);
1830 if (free_space < data_size) {
1831 free_extent_buffer(left);
1832 return 1;
1833 }
1834
1835 /* cow and double check */
1836 ret = btrfs_cow_block(trans, root, left,
1837 path->nodes[1], slot - 1, &left);
1838 if (ret) {
1839 /* we hit -ENOSPC, but it isn't fatal here */
1840 free_extent_buffer(left);
1841 return 1;
1842 }
1843
1844 free_space = btrfs_leaf_free_space(root, left);
1845 if (free_space < data_size) {
1846 free_extent_buffer(left);
1847 return 1;
1848 }
1849
1850 if (empty)
1851 nr = right_nritems;
1852 else
1853 nr = right_nritems - 1;
1854
1855 for (i = 0; i < nr; i++) {
1856 item = btrfs_item_nr(i);
1857
1858 if (path->slots[0] == i)
1859 push_space += data_size + sizeof(*item);
1860
1861 this_item_size = btrfs_item_size(right, item);
1862 if (this_item_size + sizeof(*item) + push_space > free_space)
1863 break;
1864
1865 push_items++;
1866 push_space += this_item_size + sizeof(*item);
1867 }
1868
1869 if (push_items == 0) {
1870 free_extent_buffer(left);
1871 return 1;
1872 }
1873 if (!empty && push_items == btrfs_header_nritems(right))
1874 WARN_ON(1);
1875
1876 /* push data from right to left */
1877 copy_extent_buffer(left, right,
1878 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1879 btrfs_item_nr_offset(0),
1880 push_items * sizeof(struct btrfs_item));
1881
1882 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1883 btrfs_item_offset_nr(right, push_items -1);
1884
1885 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1886 leaf_data_end(root, left) - push_space,
1887 btrfs_leaf_data(right) +
1888 btrfs_item_offset_nr(right, push_items - 1),
1889 push_space);
1890 old_left_nritems = btrfs_header_nritems(left);
1891 BUG_ON(old_left_nritems == 0);
1892
1893 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1894 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1895 u32 ioff;
1896
1897 item = btrfs_item_nr(i);
1898 ioff = btrfs_item_offset(left, item);
1899 btrfs_set_item_offset(left, item,
1900 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1901 }
1902 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1903
1904 /* fixup right node */
1905 if (push_items > right_nritems) {
1906 printk("push items %d nr %u\n", push_items, right_nritems);
1907 WARN_ON(1);
1908 }
1909
1910 if (push_items < right_nritems) {
1911 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1912 leaf_data_end(root, right);
1913 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1914 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1915 btrfs_leaf_data(right) +
1916 leaf_data_end(root, right), push_space);
1917
1918 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1919 btrfs_item_nr_offset(push_items),
1920 (btrfs_header_nritems(right) - push_items) *
1921 sizeof(struct btrfs_item));
1922 }
1923 right_nritems -= push_items;
1924 btrfs_set_header_nritems(right, right_nritems);
1925 push_space = BTRFS_LEAF_DATA_SIZE(root);
1926 for (i = 0; i < right_nritems; i++) {
1927 item = btrfs_item_nr(i);
1928 push_space = push_space - btrfs_item_size(right, item);
1929 btrfs_set_item_offset(right, item, push_space);
1930 }
1931
1932 btrfs_mark_buffer_dirty(left);
1933 if (right_nritems)
1934 btrfs_mark_buffer_dirty(right);
1935
1936 btrfs_item_key(right, &disk_key, 0);
1937 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1938
1939 /* then fixup the leaf pointer in the path */
1940 if (path->slots[0] < push_items) {
1941 path->slots[0] += old_left_nritems;
1942 free_extent_buffer(path->nodes[0]);
1943 path->nodes[0] = left;
1944 path->slots[1] -= 1;
1945 } else {
1946 free_extent_buffer(left);
1947 path->slots[0] -= push_items;
1948 }
1949 BUG_ON(path->slots[0] < 0);
1950 return ret;
1951 }
1952
1953 /*
1954 * split the path's leaf in two, making sure there is at least data_size
1955 * available for the resulting leaf level of the path.
1956 *
1957 * returns 0 if all went well and < 0 on failure.
1958 */
1959 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
1960 struct btrfs_root *root,
1961 struct btrfs_path *path,
1962 struct extent_buffer *l,
1963 struct extent_buffer *right,
1964 int slot, int mid, int nritems)
1965 {
1966 int data_copy_size;
1967 int rt_data_off;
1968 int i;
1969 int ret = 0;
1970 int wret;
1971 struct btrfs_disk_key disk_key;
1972
1973 nritems = nritems - mid;
1974 btrfs_set_header_nritems(right, nritems);
1975 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
1976
1977 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
1978 btrfs_item_nr_offset(mid),
1979 nritems * sizeof(struct btrfs_item));
1980
1981 copy_extent_buffer(right, l,
1982 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
1983 data_copy_size, btrfs_leaf_data(l) +
1984 leaf_data_end(root, l), data_copy_size);
1985
1986 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
1987 btrfs_item_end_nr(l, mid);
1988
1989 for (i = 0; i < nritems; i++) {
1990 struct btrfs_item *item = btrfs_item_nr(i);
1991 u32 ioff = btrfs_item_offset(right, item);
1992 btrfs_set_item_offset(right, item, ioff + rt_data_off);
1993 }
1994
1995 btrfs_set_header_nritems(l, mid);
1996 ret = 0;
1997 btrfs_item_key(right, &disk_key, 0);
1998 wret = insert_ptr(trans, root, path, &disk_key, right->start,
1999 path->slots[1] + 1, 1);
2000 if (wret)
2001 ret = wret;
2002
2003 btrfs_mark_buffer_dirty(right);
2004 btrfs_mark_buffer_dirty(l);
2005 BUG_ON(path->slots[0] != slot);
2006
2007 if (mid <= slot) {
2008 free_extent_buffer(path->nodes[0]);
2009 path->nodes[0] = right;
2010 path->slots[0] -= mid;
2011 path->slots[1] += 1;
2012 } else {
2013 free_extent_buffer(right);
2014 }
2015
2016 BUG_ON(path->slots[0] < 0);
2017
2018 return ret;
2019 }
2020
2021 /*
2022 * split the path's leaf in two, making sure there is at least data_size
2023 * available for the resulting leaf level of the path.
2024 *
2025 * returns 0 if all went well and < 0 on failure.
2026 */
2027 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2028 struct btrfs_root *root,
2029 struct btrfs_key *ins_key,
2030 struct btrfs_path *path, int data_size,
2031 int extend)
2032 {
2033 struct btrfs_disk_key disk_key;
2034 struct extent_buffer *l;
2035 u32 nritems;
2036 int mid;
2037 int slot;
2038 struct extent_buffer *right;
2039 int ret = 0;
2040 int wret;
2041 int split;
2042 int num_doubles = 0;
2043
2044 l = path->nodes[0];
2045 slot = path->slots[0];
2046 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2047 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2048 return -EOVERFLOW;
2049
2050 /* first try to make some room by pushing left and right */
2051 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2052 wret = push_leaf_right(trans, root, path, data_size, 0);
2053 if (wret < 0)
2054 return wret;
2055 if (wret) {
2056 wret = push_leaf_left(trans, root, path, data_size, 0);
2057 if (wret < 0)
2058 return wret;
2059 }
2060 l = path->nodes[0];
2061
2062 /* did the pushes work? */
2063 if (btrfs_leaf_free_space(root, l) >= data_size)
2064 return 0;
2065 }
2066
2067 if (!path->nodes[1]) {
2068 ret = insert_new_root(trans, root, path, 1);
2069 if (ret)
2070 return ret;
2071 }
2072 again:
2073 split = 1;
2074 l = path->nodes[0];
2075 slot = path->slots[0];
2076 nritems = btrfs_header_nritems(l);
2077 mid = (nritems + 1) / 2;
2078
2079 if (mid <= slot) {
2080 if (nritems == 1 ||
2081 leaf_space_used(l, mid, nritems - mid) + data_size >
2082 BTRFS_LEAF_DATA_SIZE(root)) {
2083 if (slot >= nritems) {
2084 split = 0;
2085 } else {
2086 mid = slot;
2087 if (mid != nritems &&
2088 leaf_space_used(l, mid, nritems - mid) +
2089 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2090 split = 2;
2091 }
2092 }
2093 }
2094 } else {
2095 if (leaf_space_used(l, 0, mid) + data_size >
2096 BTRFS_LEAF_DATA_SIZE(root)) {
2097 if (!extend && data_size && slot == 0) {
2098 split = 0;
2099 } else if ((extend || !data_size) && slot == 0) {
2100 mid = 1;
2101 } else {
2102 mid = slot;
2103 if (mid != nritems &&
2104 leaf_space_used(l, mid, nritems - mid) +
2105 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2106 split = 2 ;
2107 }
2108 }
2109 }
2110 }
2111
2112 if (split == 0)
2113 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2114 else
2115 btrfs_item_key(l, &disk_key, mid);
2116
2117 right = btrfs_alloc_free_block(trans, root, root->nodesize,
2118 root->root_key.objectid,
2119 &disk_key, 0, l->start, 0);
2120 if (IS_ERR(right)) {
2121 BUG_ON(1);
2122 return PTR_ERR(right);
2123 }
2124
2125 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2126 btrfs_set_header_bytenr(right, right->start);
2127 btrfs_set_header_generation(right, trans->transid);
2128 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2129 btrfs_set_header_owner(right, root->root_key.objectid);
2130 btrfs_set_header_level(right, 0);
2131 write_extent_buffer(right, root->fs_info->fsid,
2132 btrfs_header_fsid(), BTRFS_FSID_SIZE);
2133
2134 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2135 btrfs_header_chunk_tree_uuid(right),
2136 BTRFS_UUID_SIZE);
2137
2138 if (split == 0) {
2139 if (mid <= slot) {
2140 btrfs_set_header_nritems(right, 0);
2141 wret = insert_ptr(trans, root, path,
2142 &disk_key, right->start,
2143 path->slots[1] + 1, 1);
2144 if (wret)
2145 ret = wret;
2146
2147 free_extent_buffer(path->nodes[0]);
2148 path->nodes[0] = right;
2149 path->slots[0] = 0;
2150 path->slots[1] += 1;
2151 } else {
2152 btrfs_set_header_nritems(right, 0);
2153 wret = insert_ptr(trans, root, path,
2154 &disk_key,
2155 right->start,
2156 path->slots[1], 1);
2157 if (wret)
2158 ret = wret;
2159 free_extent_buffer(path->nodes[0]);
2160 path->nodes[0] = right;
2161 path->slots[0] = 0;
2162 if (path->slots[1] == 0) {
2163 btrfs_fixup_low_keys(root, path,
2164 &disk_key, 1);
2165 }
2166 }
2167 btrfs_mark_buffer_dirty(right);
2168 return ret;
2169 }
2170
2171 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2172 BUG_ON(ret);
2173
2174 if (split == 2) {
2175 BUG_ON(num_doubles != 0);
2176 num_doubles++;
2177 goto again;
2178 }
2179
2180 return ret;
2181 }
2182
2183 /*
2184 * This function splits a single item into two items,
2185 * giving 'new_key' to the new item and splitting the
2186 * old one at split_offset (from the start of the item).
2187 *
2188 * The path may be released by this operation. After
2189 * the split, the path is pointing to the old item. The
2190 * new item is going to be in the same node as the old one.
2191 *
2192 * Note, the item being split must be smaller enough to live alone on
2193 * a tree block with room for one extra struct btrfs_item
2194 *
2195 * This allows us to split the item in place, keeping a lock on the
2196 * leaf the entire time.
2197 */
2198 int btrfs_split_item(struct btrfs_trans_handle *trans,
2199 struct btrfs_root *root,
2200 struct btrfs_path *path,
2201 struct btrfs_key *new_key,
2202 unsigned long split_offset)
2203 {
2204 u32 item_size;
2205 struct extent_buffer *leaf;
2206 struct btrfs_key orig_key;
2207 struct btrfs_item *item;
2208 struct btrfs_item *new_item;
2209 int ret = 0;
2210 int slot;
2211 u32 nritems;
2212 u32 orig_offset;
2213 struct btrfs_disk_key disk_key;
2214 char *buf;
2215
2216 leaf = path->nodes[0];
2217 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2218 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2219 goto split;
2220
2221 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2222 btrfs_release_path(path);
2223
2224 path->search_for_split = 1;
2225
2226 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2227 path->search_for_split = 0;
2228
2229 /* if our item isn't there or got smaller, return now */
2230 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2231 path->slots[0])) {
2232 return -EAGAIN;
2233 }
2234
2235 ret = split_leaf(trans, root, &orig_key, path, 0, 0);
2236 BUG_ON(ret);
2237
2238 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2239 leaf = path->nodes[0];
2240
2241 split:
2242 item = btrfs_item_nr(path->slots[0]);
2243 orig_offset = btrfs_item_offset(leaf, item);
2244 item_size = btrfs_item_size(leaf, item);
2245
2246
2247 buf = kmalloc(item_size, GFP_NOFS);
2248 BUG_ON(!buf);
2249 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2250 path->slots[0]), item_size);
2251 slot = path->slots[0] + 1;
2252 leaf = path->nodes[0];
2253
2254 nritems = btrfs_header_nritems(leaf);
2255
2256 if (slot != nritems) {
2257 /* shift the items */
2258 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2259 btrfs_item_nr_offset(slot),
2260 (nritems - slot) * sizeof(struct btrfs_item));
2261
2262 }
2263
2264 btrfs_cpu_key_to_disk(&disk_key, new_key);
2265 btrfs_set_item_key(leaf, &disk_key, slot);
2266
2267 new_item = btrfs_item_nr(slot);
2268
2269 btrfs_set_item_offset(leaf, new_item, orig_offset);
2270 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
2271
2272 btrfs_set_item_offset(leaf, item,
2273 orig_offset + item_size - split_offset);
2274 btrfs_set_item_size(leaf, item, split_offset);
2275
2276 btrfs_set_header_nritems(leaf, nritems + 1);
2277
2278 /* write the data for the start of the original item */
2279 write_extent_buffer(leaf, buf,
2280 btrfs_item_ptr_offset(leaf, path->slots[0]),
2281 split_offset);
2282
2283 /* write the data for the new item */
2284 write_extent_buffer(leaf, buf + split_offset,
2285 btrfs_item_ptr_offset(leaf, slot),
2286 item_size - split_offset);
2287 btrfs_mark_buffer_dirty(leaf);
2288
2289 ret = 0;
2290 if (btrfs_leaf_free_space(root, leaf) < 0) {
2291 btrfs_print_leaf(root, leaf);
2292 BUG();
2293 }
2294 kfree(buf);
2295 return ret;
2296 }
2297
2298 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2299 struct btrfs_root *root,
2300 struct btrfs_path *path,
2301 u32 new_size, int from_end)
2302 {
2303 int ret = 0;
2304 int slot;
2305 struct extent_buffer *leaf;
2306 struct btrfs_item *item;
2307 u32 nritems;
2308 unsigned int data_end;
2309 unsigned int old_data_start;
2310 unsigned int old_size;
2311 unsigned int size_diff;
2312 int i;
2313
2314 leaf = path->nodes[0];
2315 slot = path->slots[0];
2316
2317 old_size = btrfs_item_size_nr(leaf, slot);
2318 if (old_size == new_size)
2319 return 0;
2320
2321 nritems = btrfs_header_nritems(leaf);
2322 data_end = leaf_data_end(root, leaf);
2323
2324 old_data_start = btrfs_item_offset_nr(leaf, slot);
2325
2326 size_diff = old_size - new_size;
2327
2328 BUG_ON(slot < 0);
2329 BUG_ON(slot >= nritems);
2330
2331 /*
2332 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2333 */
2334 /* first correct the data pointers */
2335 for (i = slot; i < nritems; i++) {
2336 u32 ioff;
2337 item = btrfs_item_nr(i);
2338 ioff = btrfs_item_offset(leaf, item);
2339 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2340 }
2341
2342 /* shift the data */
2343 if (from_end) {
2344 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2345 data_end + size_diff, btrfs_leaf_data(leaf) +
2346 data_end, old_data_start + new_size - data_end);
2347 } else {
2348 struct btrfs_disk_key disk_key;
2349 u64 offset;
2350
2351 btrfs_item_key(leaf, &disk_key, slot);
2352
2353 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2354 unsigned long ptr;
2355 struct btrfs_file_extent_item *fi;
2356
2357 fi = btrfs_item_ptr(leaf, slot,
2358 struct btrfs_file_extent_item);
2359 fi = (struct btrfs_file_extent_item *)(
2360 (unsigned long)fi - size_diff);
2361
2362 if (btrfs_file_extent_type(leaf, fi) ==
2363 BTRFS_FILE_EXTENT_INLINE) {
2364 ptr = btrfs_item_ptr_offset(leaf, slot);
2365 memmove_extent_buffer(leaf, ptr,
2366 (unsigned long)fi,
2367 offsetof(struct btrfs_file_extent_item,
2368 disk_bytenr));
2369 }
2370 }
2371
2372 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2373 data_end + size_diff, btrfs_leaf_data(leaf) +
2374 data_end, old_data_start - data_end);
2375
2376 offset = btrfs_disk_key_offset(&disk_key);
2377 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2378 btrfs_set_item_key(leaf, &disk_key, slot);
2379 if (slot == 0)
2380 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2381 }
2382
2383 item = btrfs_item_nr(slot);
2384 btrfs_set_item_size(leaf, item, new_size);
2385 btrfs_mark_buffer_dirty(leaf);
2386
2387 ret = 0;
2388 if (btrfs_leaf_free_space(root, leaf) < 0) {
2389 btrfs_print_leaf(root, leaf);
2390 BUG();
2391 }
2392 return ret;
2393 }
2394
2395 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2396 struct btrfs_root *root, struct btrfs_path *path,
2397 u32 data_size)
2398 {
2399 int ret = 0;
2400 int slot;
2401 struct extent_buffer *leaf;
2402 struct btrfs_item *item;
2403 u32 nritems;
2404 unsigned int data_end;
2405 unsigned int old_data;
2406 unsigned int old_size;
2407 int i;
2408
2409 leaf = path->nodes[0];
2410
2411 nritems = btrfs_header_nritems(leaf);
2412 data_end = leaf_data_end(root, leaf);
2413
2414 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2415 btrfs_print_leaf(root, leaf);
2416 BUG();
2417 }
2418 slot = path->slots[0];
2419 old_data = btrfs_item_end_nr(leaf, slot);
2420
2421 BUG_ON(slot < 0);
2422 if (slot >= nritems) {
2423 btrfs_print_leaf(root, leaf);
2424 printk("slot %d too large, nritems %d\n", slot, nritems);
2425 BUG_ON(1);
2426 }
2427
2428 /*
2429 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2430 */
2431 /* first correct the data pointers */
2432 for (i = slot; i < nritems; i++) {
2433 u32 ioff;
2434 item = btrfs_item_nr(i);
2435 ioff = btrfs_item_offset(leaf, item);
2436 btrfs_set_item_offset(leaf, item, ioff - data_size);
2437 }
2438
2439 /* shift the data */
2440 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2441 data_end - data_size, btrfs_leaf_data(leaf) +
2442 data_end, old_data - data_end);
2443
2444 data_end = old_data;
2445 old_size = btrfs_item_size_nr(leaf, slot);
2446 item = btrfs_item_nr(slot);
2447 btrfs_set_item_size(leaf, item, old_size + data_size);
2448 btrfs_mark_buffer_dirty(leaf);
2449
2450 ret = 0;
2451 if (btrfs_leaf_free_space(root, leaf) < 0) {
2452 btrfs_print_leaf(root, leaf);
2453 BUG();
2454 }
2455 return ret;
2456 }
2457
2458 /*
2459 * Given a key and some data, insert an item into the tree.
2460 * This does all the path init required, making room in the tree if needed.
2461 */
2462 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2463 struct btrfs_root *root,
2464 struct btrfs_path *path,
2465 struct btrfs_key *cpu_key, u32 *data_size,
2466 int nr)
2467 {
2468 struct extent_buffer *leaf;
2469 struct btrfs_item *item;
2470 int ret = 0;
2471 int slot;
2472 int i;
2473 u32 nritems;
2474 u32 total_size = 0;
2475 u32 total_data = 0;
2476 unsigned int data_end;
2477 struct btrfs_disk_key disk_key;
2478
2479 for (i = 0; i < nr; i++) {
2480 total_data += data_size[i];
2481 }
2482
2483 /* create a root if there isn't one */
2484 if (!root->node)
2485 BUG();
2486
2487 total_size = total_data + nr * sizeof(struct btrfs_item);
2488 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2489 if (ret == 0) {
2490 return -EEXIST;
2491 }
2492 if (ret < 0)
2493 goto out;
2494
2495 leaf = path->nodes[0];
2496
2497 nritems = btrfs_header_nritems(leaf);
2498 data_end = leaf_data_end(root, leaf);
2499
2500 if (btrfs_leaf_free_space(root, leaf) < total_size) {
2501 btrfs_print_leaf(root, leaf);
2502 printk("not enough freespace need %u have %d\n",
2503 total_size, btrfs_leaf_free_space(root, leaf));
2504 BUG();
2505 }
2506
2507 slot = path->slots[0];
2508 BUG_ON(slot < 0);
2509
2510 if (slot != nritems) {
2511 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2512
2513 if (old_data < data_end) {
2514 btrfs_print_leaf(root, leaf);
2515 printk("slot %d old_data %d data_end %d\n",
2516 slot, old_data, data_end);
2517 BUG_ON(1);
2518 }
2519 /*
2520 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2521 */
2522 /* first correct the data pointers */
2523 for (i = slot; i < nritems; i++) {
2524 u32 ioff;
2525
2526 item = btrfs_item_nr(i);
2527 ioff = btrfs_item_offset(leaf, item);
2528 btrfs_set_item_offset(leaf, item, ioff - total_data);
2529 }
2530
2531 /* shift the items */
2532 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2533 btrfs_item_nr_offset(slot),
2534 (nritems - slot) * sizeof(struct btrfs_item));
2535
2536 /* shift the data */
2537 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2538 data_end - total_data, btrfs_leaf_data(leaf) +
2539 data_end, old_data - data_end);
2540 data_end = old_data;
2541 }
2542
2543 /* setup the item for the new data */
2544 for (i = 0; i < nr; i++) {
2545 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2546 btrfs_set_item_key(leaf, &disk_key, slot + i);
2547 item = btrfs_item_nr(slot + i);
2548 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2549 data_end -= data_size[i];
2550 btrfs_set_item_size(leaf, item, data_size[i]);
2551 }
2552 btrfs_set_header_nritems(leaf, nritems + nr);
2553 btrfs_mark_buffer_dirty(leaf);
2554
2555 ret = 0;
2556 if (slot == 0) {
2557 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2558 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2559 }
2560
2561 if (btrfs_leaf_free_space(root, leaf) < 0) {
2562 btrfs_print_leaf(root, leaf);
2563 BUG();
2564 }
2565
2566 out:
2567 return ret;
2568 }
2569
2570 /*
2571 * Given a key and some data, insert an item into the tree.
2572 * This does all the path init required, making room in the tree if needed.
2573 */
2574 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2575 *root, struct btrfs_key *cpu_key, void *data, u32
2576 data_size)
2577 {
2578 int ret = 0;
2579 struct btrfs_path *path;
2580 struct extent_buffer *leaf;
2581 unsigned long ptr;
2582
2583 path = btrfs_alloc_path();
2584 if (!path)
2585 return -ENOMEM;
2586
2587 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2588 if (!ret) {
2589 leaf = path->nodes[0];
2590 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2591 write_extent_buffer(leaf, data, ptr, data_size);
2592 btrfs_mark_buffer_dirty(leaf);
2593 }
2594 btrfs_free_path(path);
2595 return ret;
2596 }
2597
2598 /*
2599 * delete the pointer from a given node.
2600 *
2601 * If the delete empties a node, the node is removed from the tree,
2602 * continuing all the way the root if required. The root is converted into
2603 * a leaf if all the nodes are emptied.
2604 */
2605 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2606 struct btrfs_path *path, int level, int slot)
2607 {
2608 struct extent_buffer *parent = path->nodes[level];
2609 u32 nritems;
2610 int ret = 0;
2611
2612 nritems = btrfs_header_nritems(parent);
2613 if (slot != nritems -1) {
2614 memmove_extent_buffer(parent,
2615 btrfs_node_key_ptr_offset(slot),
2616 btrfs_node_key_ptr_offset(slot + 1),
2617 sizeof(struct btrfs_key_ptr) *
2618 (nritems - slot - 1));
2619 }
2620 nritems--;
2621 btrfs_set_header_nritems(parent, nritems);
2622 if (nritems == 0 && parent == root->node) {
2623 BUG_ON(btrfs_header_level(root->node) != 1);
2624 /* just turn the root into a leaf and break */
2625 btrfs_set_header_level(root->node, 0);
2626 } else if (slot == 0) {
2627 struct btrfs_disk_key disk_key;
2628
2629 btrfs_node_key(parent, &disk_key, 0);
2630 btrfs_fixup_low_keys(root, path, &disk_key, level + 1);
2631 }
2632 btrfs_mark_buffer_dirty(parent);
2633 return ret;
2634 }
2635
2636 /*
2637 * a helper function to delete the leaf pointed to by path->slots[1] and
2638 * path->nodes[1].
2639 *
2640 * This deletes the pointer in path->nodes[1] and frees the leaf
2641 * block extent. zero is returned if it all worked out, < 0 otherwise.
2642 *
2643 * The path must have already been setup for deleting the leaf, including
2644 * all the proper balancing. path->nodes[1] must be locked.
2645 */
2646 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
2647 struct btrfs_root *root,
2648 struct btrfs_path *path,
2649 struct extent_buffer *leaf)
2650 {
2651 int ret;
2652
2653 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
2654 ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]);
2655 if (ret)
2656 return ret;
2657
2658 ret = btrfs_free_extent(trans, root, leaf->start, leaf->len,
2659 0, root->root_key.objectid, 0, 0);
2660 return ret;
2661 }
2662
2663 /*
2664 * delete the item at the leaf level in path. If that empties
2665 * the leaf, remove it from the tree
2666 */
2667 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2668 struct btrfs_path *path, int slot, int nr)
2669 {
2670 struct extent_buffer *leaf;
2671 struct btrfs_item *item;
2672 int last_off;
2673 int dsize = 0;
2674 int ret = 0;
2675 int wret;
2676 int i;
2677 u32 nritems;
2678
2679 leaf = path->nodes[0];
2680 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2681
2682 for (i = 0; i < nr; i++)
2683 dsize += btrfs_item_size_nr(leaf, slot + i);
2684
2685 nritems = btrfs_header_nritems(leaf);
2686
2687 if (slot + nr != nritems) {
2688 int data_end = leaf_data_end(root, leaf);
2689
2690 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2691 data_end + dsize,
2692 btrfs_leaf_data(leaf) + data_end,
2693 last_off - data_end);
2694
2695 for (i = slot + nr; i < nritems; i++) {
2696 u32 ioff;
2697
2698 item = btrfs_item_nr(i);
2699 ioff = btrfs_item_offset(leaf, item);
2700 btrfs_set_item_offset(leaf, item, ioff + dsize);
2701 }
2702
2703 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2704 btrfs_item_nr_offset(slot + nr),
2705 sizeof(struct btrfs_item) *
2706 (nritems - slot - nr));
2707 }
2708 btrfs_set_header_nritems(leaf, nritems - nr);
2709 nritems -= nr;
2710
2711 /* delete the leaf if we've emptied it */
2712 if (nritems == 0) {
2713 if (leaf == root->node) {
2714 btrfs_set_header_level(leaf, 0);
2715 } else {
2716 clean_tree_block(trans, root, leaf);
2717 wait_on_tree_block_writeback(root, leaf);
2718
2719 wret = btrfs_del_leaf(trans, root, path, leaf);
2720 BUG_ON(ret);
2721 if (wret)
2722 ret = wret;
2723 }
2724 } else {
2725 int used = leaf_space_used(leaf, 0, nritems);
2726 if (slot == 0) {
2727 struct btrfs_disk_key disk_key;
2728
2729 btrfs_item_key(leaf, &disk_key, 0);
2730 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2731 }
2732
2733 /* delete the leaf if it is mostly empty */
2734 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2735 /* push_leaf_left fixes the path.
2736 * make sure the path still points to our leaf
2737 * for possible call to del_ptr below
2738 */
2739 slot = path->slots[1];
2740 extent_buffer_get(leaf);
2741
2742 wret = push_leaf_left(trans, root, path, 1, 1);
2743 if (wret < 0 && wret != -ENOSPC)
2744 ret = wret;
2745
2746 if (path->nodes[0] == leaf &&
2747 btrfs_header_nritems(leaf)) {
2748 wret = push_leaf_right(trans, root, path, 1, 1);
2749 if (wret < 0 && wret != -ENOSPC)
2750 ret = wret;
2751 }
2752
2753 if (btrfs_header_nritems(leaf) == 0) {
2754 clean_tree_block(trans, root, leaf);
2755 wait_on_tree_block_writeback(root, leaf);
2756
2757 path->slots[1] = slot;
2758 ret = btrfs_del_leaf(trans, root, path, leaf);
2759 BUG_ON(ret);
2760 free_extent_buffer(leaf);
2761
2762 } else {
2763 btrfs_mark_buffer_dirty(leaf);
2764 free_extent_buffer(leaf);
2765 }
2766 } else {
2767 btrfs_mark_buffer_dirty(leaf);
2768 }
2769 }
2770 return ret;
2771 }
2772
2773 /*
2774 * walk up the tree as far as required to find the previous leaf.
2775 * returns 0 if it found something or 1 if there are no lesser leaves.
2776 * returns < 0 on io errors.
2777 */
2778 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2779 {
2780 int slot;
2781 int level = 1;
2782 struct extent_buffer *c;
2783 struct extent_buffer *next = NULL;
2784
2785 while(level < BTRFS_MAX_LEVEL) {
2786 if (!path->nodes[level])
2787 return 1;
2788
2789 slot = path->slots[level];
2790 c = path->nodes[level];
2791 if (slot == 0) {
2792 level++;
2793 if (level == BTRFS_MAX_LEVEL)
2794 return 1;
2795 continue;
2796 }
2797 slot--;
2798
2799 next = read_node_slot(root, c, slot);
2800 if (!extent_buffer_uptodate(next)) {
2801 if (IS_ERR(next))
2802 return PTR_ERR(next);
2803 return -EIO;
2804 }
2805 break;
2806 }
2807 path->slots[level] = slot;
2808 while(1) {
2809 level--;
2810 c = path->nodes[level];
2811 free_extent_buffer(c);
2812 slot = btrfs_header_nritems(next);
2813 if (slot != 0)
2814 slot--;
2815 path->nodes[level] = next;
2816 path->slots[level] = slot;
2817 if (!level)
2818 break;
2819 next = read_node_slot(root, next, slot);
2820 if (!extent_buffer_uptodate(next)) {
2821 if (IS_ERR(next))
2822 return PTR_ERR(next);
2823 return -EIO;
2824 }
2825 }
2826 return 0;
2827 }
2828
2829 /*
2830 * walk up the tree as far as required to find the next leaf.
2831 * returns 0 if it found something or 1 if there are no greater leaves.
2832 * returns < 0 on io errors.
2833 */
2834 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2835 {
2836 int slot;
2837 int level = 1;
2838 struct extent_buffer *c;
2839 struct extent_buffer *next = NULL;
2840
2841 while(level < BTRFS_MAX_LEVEL) {
2842 if (!path->nodes[level])
2843 return 1;
2844
2845 slot = path->slots[level] + 1;
2846 c = path->nodes[level];
2847 if (slot >= btrfs_header_nritems(c)) {
2848 level++;
2849 if (level == BTRFS_MAX_LEVEL)
2850 return 1;
2851 continue;
2852 }
2853
2854 if (path->reada)
2855 reada_for_search(root, path, level, slot, 0);
2856
2857 next = read_node_slot(root, c, slot);
2858 if (!extent_buffer_uptodate(next))
2859 return -EIO;
2860 break;
2861 }
2862 path->slots[level] = slot;
2863 while(1) {
2864 level--;
2865 c = path->nodes[level];
2866 free_extent_buffer(c);
2867 path->nodes[level] = next;
2868 path->slots[level] = 0;
2869 if (!level)
2870 break;
2871 if (path->reada)
2872 reada_for_search(root, path, level, 0, 0);
2873 next = read_node_slot(root, next, 0);
2874 if (!extent_buffer_uptodate(next))
2875 return -EIO;
2876 }
2877 return 0;
2878 }
2879
2880 int btrfs_previous_item(struct btrfs_root *root,
2881 struct btrfs_path *path, u64 min_objectid,
2882 int type)
2883 {
2884 struct btrfs_key found_key;
2885 struct extent_buffer *leaf;
2886 u32 nritems;
2887 int ret;
2888
2889 while(1) {
2890 if (path->slots[0] == 0) {
2891 ret = btrfs_prev_leaf(root, path);
2892 if (ret != 0)
2893 return ret;
2894 } else {
2895 path->slots[0]--;
2896 }
2897 leaf = path->nodes[0];
2898 nritems = btrfs_header_nritems(leaf);
2899 if (nritems == 0)
2900 return 1;
2901 if (path->slots[0] == nritems)
2902 path->slots[0]--;
2903
2904 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2905 if (found_key.objectid < min_objectid)
2906 break;
2907 if (found_key.type == type)
2908 return 0;
2909 if (found_key.objectid == min_objectid &&
2910 found_key.type < type)
2911 break;
2912 }
2913 return 1;
2914 }
2915
2916 /*
2917 * search in extent tree to find a previous Metadata/Data extent item with
2918 * min objecitd.
2919 *
2920 * returns 0 if something is found, 1 if nothing was found and < 0 on error
2921 */
2922 int btrfs_previous_extent_item(struct btrfs_root *root,
2923 struct btrfs_path *path, u64 min_objectid)
2924 {
2925 struct btrfs_key found_key;
2926 struct extent_buffer *leaf;
2927 u32 nritems;
2928 int ret;
2929
2930 while (1) {
2931 if (path->slots[0] == 0) {
2932 ret = btrfs_prev_leaf(root, path);
2933 if (ret != 0)
2934 return ret;
2935 } else {
2936 path->slots[0]--;
2937 }
2938 leaf = path->nodes[0];
2939 nritems = btrfs_header_nritems(leaf);
2940 if (nritems == 0)
2941 return 1;
2942 if (path->slots[0] == nritems)
2943 path->slots[0]--;
2944
2945 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2946 if (found_key.objectid < min_objectid)
2947 break;
2948 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
2949 found_key.type == BTRFS_METADATA_ITEM_KEY)
2950 return 0;
2951 if (found_key.objectid == min_objectid &&
2952 found_key.type < BTRFS_EXTENT_ITEM_KEY)
2953 break;
2954 }
2955 return 1;
2956 }
2957
2958 /*
2959 * Search in extent tree to found next meta/data extent
2960 * Caller needs to check for no-hole or skinny metadata features.
2961 */
2962 int btrfs_next_extent_item(struct btrfs_root *root,
2963 struct btrfs_path *path, u64 max_objectid)
2964 {
2965 struct btrfs_key found_key;
2966 int ret;
2967
2968 while (1) {
2969 ret = btrfs_next_item(root, path);
2970 if (ret)
2971 return ret;
2972 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2973 path->slots[0]);
2974 if (found_key.objectid > max_objectid)
2975 return 1;
2976 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
2977 found_key.type == BTRFS_METADATA_ITEM_KEY)
2978 return 0;
2979 }
2980 }
(deprecated) Btrfs progs developments and patch integration