?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: fix show super unknown flag output
[btrfs-progs-unstable/devel.git] / ctree.c
blob23399e23ff3c2db6101ae6d99239398374764350
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 btrfs_release_path(p);
52 kfree(p);
53 }
54
55 void btrfs_release_path(struct btrfs_path *p)
56 {
57 int i;
58 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
59 if (!p->nodes[i])
60 continue;
61 free_extent_buffer(p->nodes[i]);
62 }
63 memset(p, 0, sizeof(*p));
64 }
65
66 void add_root_to_dirty_list(struct btrfs_root *root)
67 {
68 if (root->track_dirty && list_empty(&root->dirty_list)) {
69 list_add(&root->dirty_list,
70 &root->fs_info->dirty_cowonly_roots);
71 }
72 }
73
74 int btrfs_copy_root(struct btrfs_trans_handle *trans,
75 struct btrfs_root *root,
76 struct extent_buffer *buf,
77 struct extent_buffer **cow_ret, u64 new_root_objectid)
78 {
79 struct extent_buffer *cow;
80 int ret = 0;
81 int level;
82 struct btrfs_root *new_root;
83 struct btrfs_disk_key disk_key;
84
85 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
86 if (!new_root)
87 return -ENOMEM;
88
89 memcpy(new_root, root, sizeof(*new_root));
90 new_root->root_key.objectid = new_root_objectid;
91
92 WARN_ON(root->ref_cows && trans->transid !=
93 root->fs_info->running_transaction->transid);
94 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
95
96 level = btrfs_header_level(buf);
97 if (level == 0)
98 btrfs_item_key(buf, &disk_key, 0);
99 else
100 btrfs_node_key(buf, &disk_key, 0);
101 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
102 new_root_objectid, &disk_key,
103 level, buf->start, 0);
104 if (IS_ERR(cow)) {
105 kfree(new_root);
106 return PTR_ERR(cow);
107 }
108
109 copy_extent_buffer(cow, buf, 0, 0, cow->len);
110 btrfs_set_header_bytenr(cow, cow->start);
111 btrfs_set_header_generation(cow, trans->transid);
112 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
113 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
114 BTRFS_HEADER_FLAG_RELOC);
115 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
116 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
117 else
118 btrfs_set_header_owner(cow, new_root_objectid);
119
120 write_extent_buffer(cow, root->fs_info->fsid,
121 btrfs_header_fsid(), BTRFS_FSID_SIZE);
122
123 WARN_ON(btrfs_header_generation(buf) > trans->transid);
124 ret = btrfs_inc_ref(trans, new_root, cow, 0);
125 kfree(new_root);
126
127 if (ret)
128 return ret;
129
130 btrfs_mark_buffer_dirty(cow);
131 *cow_ret = cow;
132 return 0;
133 }
134
135 /*
136 * check if the tree block can be shared by multiple trees
137 */
138 static int btrfs_block_can_be_shared(struct btrfs_root *root,
139 struct extent_buffer *buf)
140 {
141 /*
142 * Tree blocks not in refernece counted trees and tree roots
143 * are never shared. If a block was allocated after the last
144 * snapshot and the block was not allocated by tree relocation,
145 * we know the block is not shared.
146 */
147 if (root->ref_cows &&
148 buf != root->node && buf != root->commit_root &&
149 (btrfs_header_generation(buf) <=
150 btrfs_root_last_snapshot(&root->root_item) ||
151 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
152 return 1;
153 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
154 if (root->ref_cows &&
155 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
156 return 1;
157 #endif
158 return 0;
159 }
160
161 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
162 struct btrfs_root *root,
163 struct extent_buffer *buf,
164 struct extent_buffer *cow)
165 {
166 u64 refs;
167 u64 owner;
168 u64 flags;
169 u64 new_flags = 0;
170 int ret;
171
172 /*
173 * Backrefs update rules:
174 *
175 * Always use full backrefs for extent pointers in tree block
176 * allocated by tree relocation.
177 *
178 * If a shared tree block is no longer referenced by its owner
179 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
180 * use full backrefs for extent pointers in tree block.
181 *
182 * If a tree block is been relocating
183 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
184 * use full backrefs for extent pointers in tree block.
185 * The reason for this is some operations (such as drop tree)
186 * are only allowed for blocks use full backrefs.
187 */
188
189 if (btrfs_block_can_be_shared(root, buf)) {
190 ret = btrfs_lookup_extent_info(trans, root, buf->start,
191 btrfs_header_level(buf), 1,
192 &refs, &flags);
193 BUG_ON(ret);
194 BUG_ON(refs == 0);
195 } else {
196 refs = 1;
197 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
198 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
199 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
200 else
201 flags = 0;
202 }
203
204 owner = btrfs_header_owner(buf);
205 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
206 owner == BTRFS_TREE_RELOC_OBJECTID);
207
208 if (refs > 1) {
209 if ((owner == root->root_key.objectid ||
210 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
211 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
212 ret = btrfs_inc_ref(trans, root, buf, 1);
213 BUG_ON(ret);
214
215 if (root->root_key.objectid ==
216 BTRFS_TREE_RELOC_OBJECTID) {
217 ret = btrfs_dec_ref(trans, root, buf, 0);
218 BUG_ON(ret);
219 ret = btrfs_inc_ref(trans, root, cow, 1);
220 BUG_ON(ret);
221 }
222 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
223 } else {
224
225 if (root->root_key.objectid ==
226 BTRFS_TREE_RELOC_OBJECTID)
227 ret = btrfs_inc_ref(trans, root, cow, 1);
228 else
229 ret = btrfs_inc_ref(trans, root, cow, 0);
230 BUG_ON(ret);
231 }
232 if (new_flags != 0) {
233 ret = btrfs_set_block_flags(trans, root, buf->start,
234 btrfs_header_level(buf),
235 new_flags);
236 BUG_ON(ret);
237 }
238 } else {
239 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
240 if (root->root_key.objectid ==
241 BTRFS_TREE_RELOC_OBJECTID)
242 ret = btrfs_inc_ref(trans, root, cow, 1);
243 else
244 ret = btrfs_inc_ref(trans, root, cow, 0);
245 BUG_ON(ret);
246 ret = btrfs_dec_ref(trans, root, buf, 1);
247 BUG_ON(ret);
248 }
249 clean_tree_block(trans, root, buf);
250 }
251 return 0;
252 }
253
254 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
255 struct btrfs_root *root,
256 struct extent_buffer *buf,
257 struct extent_buffer *parent, int parent_slot,
258 struct extent_buffer **cow_ret,
259 u64 search_start, u64 empty_size)
260 {
261 struct extent_buffer *cow;
262 struct btrfs_disk_key disk_key;
263 int level;
264
265 WARN_ON(root->ref_cows && trans->transid !=
266 root->fs_info->running_transaction->transid);
267 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
268
269 level = btrfs_header_level(buf);
270
271 if (level == 0)
272 btrfs_item_key(buf, &disk_key, 0);
273 else
274 btrfs_node_key(buf, &disk_key, 0);
275
276 cow = btrfs_alloc_free_block(trans, root, buf->len,
277 root->root_key.objectid, &disk_key,
278 level, search_start, empty_size);
279 if (IS_ERR(cow))
280 return PTR_ERR(cow);
281
282 copy_extent_buffer(cow, buf, 0, 0, cow->len);
283 btrfs_set_header_bytenr(cow, cow->start);
284 btrfs_set_header_generation(cow, trans->transid);
285 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
286 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
287 BTRFS_HEADER_FLAG_RELOC);
288 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
289 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
290 else
291 btrfs_set_header_owner(cow, root->root_key.objectid);
292
293 write_extent_buffer(cow, root->fs_info->fsid,
294 btrfs_header_fsid(), BTRFS_FSID_SIZE);
295
296 WARN_ON(!(buf->flags & EXTENT_BAD_TRANSID) &&
297 btrfs_header_generation(buf) > trans->transid);
298
299 update_ref_for_cow(trans, root, buf, cow);
300
301 if (buf == root->node) {
302 root->node = cow;
303 extent_buffer_get(cow);
304
305 btrfs_free_extent(trans, root, buf->start, buf->len,
306 0, root->root_key.objectid, level, 0);
307 free_extent_buffer(buf);
308 add_root_to_dirty_list(root);
309 } else {
310 btrfs_set_node_blockptr(parent, parent_slot,
311 cow->start);
312 WARN_ON(trans->transid == 0);
313 btrfs_set_node_ptr_generation(parent, parent_slot,
314 trans->transid);
315 btrfs_mark_buffer_dirty(parent);
316 WARN_ON(btrfs_header_generation(parent) != trans->transid);
317
318 btrfs_free_extent(trans, root, buf->start, buf->len,
319 0, root->root_key.objectid, level, 1);
320 }
321 if (!list_empty(&buf->recow)) {
322 list_del_init(&buf->recow);
323 free_extent_buffer(buf);
324 }
325 free_extent_buffer(buf);
326 btrfs_mark_buffer_dirty(cow);
327 *cow_ret = cow;
328 return 0;
329 }
330
331 static inline int should_cow_block(struct btrfs_trans_handle *trans,
332 struct btrfs_root *root,
333 struct extent_buffer *buf)
334 {
335 if (btrfs_header_generation(buf) == trans->transid &&
336 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
337 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
338 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
339 return 0;
340 return 1;
341 }
342
343 int btrfs_cow_block(struct btrfs_trans_handle *trans,
344 struct btrfs_root *root, struct extent_buffer *buf,
345 struct extent_buffer *parent, int parent_slot,
346 struct extent_buffer **cow_ret)
347 {
348 u64 search_start;
349 int ret;
350 /*
351 if (trans->transaction != root->fs_info->running_transaction) {
352 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
353 root->fs_info->running_transaction->transid);
354 WARN_ON(1);
355 }
356 */
357 if (trans->transid != root->fs_info->generation) {
358 printk(KERN_CRIT "trans %llu running %llu\n",
359 (unsigned long long)trans->transid,
360 (unsigned long long)root->fs_info->generation);
361 WARN_ON(1);
362 }
363 if (!should_cow_block(trans, root, buf)) {
364 *cow_ret = buf;
365 return 0;
366 }
367
368 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
369 ret = __btrfs_cow_block(trans, root, buf, parent,
370 parent_slot, cow_ret, search_start, 0);
371 return ret;
372 }
373
374 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
375 {
376 if (k1->objectid > k2->objectid)
377 return 1;
378 if (k1->objectid < k2->objectid)
379 return -1;
380 if (k1->type > k2->type)
381 return 1;
382 if (k1->type < k2->type)
383 return -1;
384 if (k1->offset > k2->offset)
385 return 1;
386 if (k1->offset < k2->offset)
387 return -1;
388 return 0;
389 }
390
391 /*
392 * compare two keys in a memcmp fashion
393 */
394 static int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
395 {
396 struct btrfs_key k1;
397
398 btrfs_disk_key_to_cpu(&k1, disk);
399 return btrfs_comp_cpu_keys(&k1, k2);
400 }
401
402 /*
403 * The leaf data grows from end-to-front in the node.
404 * this returns the address of the start of the last item,
405 * which is the stop of the leaf data stack
406 */
407 static inline unsigned int leaf_data_end(struct btrfs_root *root,
408 struct extent_buffer *leaf)
409 {
410 u32 nr = btrfs_header_nritems(leaf);
411 if (nr == 0)
412 return BTRFS_LEAF_DATA_SIZE(root);
413 return btrfs_item_offset_nr(leaf, nr - 1);
414 }
415
416 enum btrfs_tree_block_status
417 btrfs_check_node(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
418 struct extent_buffer *buf)
419 {
420 int i;
421 struct btrfs_key cpukey;
422 struct btrfs_disk_key key;
423 u32 nritems = btrfs_header_nritems(buf);
424 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
425
426 if (nritems == 0 || nritems > BTRFS_NODEPTRS_PER_BLOCK(root))
427 goto fail;
428
429 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
430 if (parent_key && parent_key->type) {
431 btrfs_node_key(buf, &key, 0);
432 if (memcmp(parent_key, &key, sizeof(key)))
433 goto fail;
434 }
435 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
436 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
437 btrfs_node_key(buf, &key, i);
438 btrfs_node_key_to_cpu(buf, &cpukey, i + 1);
439 if (btrfs_comp_keys(&key, &cpukey) >= 0)
440 goto fail;
441 }
442 return BTRFS_TREE_BLOCK_CLEAN;
443 fail:
444 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
445 if (parent_key)
446 btrfs_disk_key_to_cpu(&cpukey, parent_key);
447 else
448 btrfs_node_key_to_cpu(buf, &cpukey, 0);
449 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
450 buf->start, buf->len,
451 btrfs_header_level(buf));
452 }
453 return ret;
454 }
455
456 enum btrfs_tree_block_status
457 btrfs_check_leaf(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
458 struct extent_buffer *buf)
459 {
460 int i;
461 struct btrfs_key cpukey;
462 struct btrfs_disk_key key;
463 u32 nritems = btrfs_header_nritems(buf);
464 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
465
466 if (nritems * sizeof(struct btrfs_item) > buf->len) {
467 fprintf(stderr, "invalid number of items %llu\n",
468 (unsigned long long)buf->start);
469 goto fail;
470 }
471
472 if (btrfs_header_level(buf) != 0) {
473 ret = BTRFS_TREE_BLOCK_INVALID_LEVEL;
474 fprintf(stderr, "leaf is not a leaf %llu\n",
475 (unsigned long long)btrfs_header_bytenr(buf));
476 goto fail;
477 }
478 if (btrfs_leaf_free_space(root, buf) < 0) {
479 ret = BTRFS_TREE_BLOCK_INVALID_FREE_SPACE;
480 fprintf(stderr, "leaf free space incorrect %llu %d\n",
481 (unsigned long long)btrfs_header_bytenr(buf),
482 btrfs_leaf_free_space(root, buf));
483 goto fail;
484 }
485
486 if (nritems == 0)
487 return BTRFS_TREE_BLOCK_CLEAN;
488
489 btrfs_item_key(buf, &key, 0);
490 if (parent_key && parent_key->type &&
491 memcmp(parent_key, &key, sizeof(key))) {
492 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
493 fprintf(stderr, "leaf parent key incorrect %llu\n",
494 (unsigned long long)btrfs_header_bytenr(buf));
495 goto fail;
496 }
497 for (i = 0; nritems > 1 && i < nritems - 1; i++) {
498 btrfs_item_key(buf, &key, i);
499 btrfs_item_key_to_cpu(buf, &cpukey, i + 1);
500 if (btrfs_comp_keys(&key, &cpukey) >= 0) {
501 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
502 fprintf(stderr, "bad key ordering %d %d\n", i, i+1);
503 goto fail;
504 }
505 if (btrfs_item_offset_nr(buf, i) !=
506 btrfs_item_end_nr(buf, i + 1)) {
507 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
508 fprintf(stderr, "incorrect offsets %u %u\n",
509 btrfs_item_offset_nr(buf, i),
510 btrfs_item_end_nr(buf, i + 1));
511 goto fail;
512 }
513 if (i == 0 && btrfs_item_end_nr(buf, i) !=
514 BTRFS_LEAF_DATA_SIZE(root)) {
515 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
516 fprintf(stderr, "bad item end %u wanted %u\n",
517 btrfs_item_end_nr(buf, i),
518 (unsigned)BTRFS_LEAF_DATA_SIZE(root));
519 goto fail;
520 }
521 }
522 return BTRFS_TREE_BLOCK_CLEAN;
523 fail:
524 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
525 if (parent_key)
526 btrfs_disk_key_to_cpu(&cpukey, parent_key);
527 else
528 btrfs_item_key_to_cpu(buf, &cpukey, 0);
529
530 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
531 buf->start, buf->len, 0);
532 }
533 return ret;
534 }
535
536 static int noinline check_block(struct btrfs_root *root,
537 struct btrfs_path *path, int level)
538 {
539 struct btrfs_disk_key key;
540 struct btrfs_disk_key *key_ptr = NULL;
541 struct extent_buffer *parent;
542 enum btrfs_tree_block_status ret;
543
544 if (path->skip_check_block)
545 return 0;
546 if (path->nodes[level + 1]) {
547 parent = path->nodes[level + 1];
548 btrfs_node_key(parent, &key, path->slots[level + 1]);
549 key_ptr = &key;
550 }
551 if (level == 0)
552 ret = btrfs_check_leaf(root, key_ptr, path->nodes[0]);
553 else
554 ret = btrfs_check_node(root, key_ptr, path->nodes[level]);
555 if (ret == BTRFS_TREE_BLOCK_CLEAN)
556 return 0;
557 return -EIO;
558 }
559
560 /*
561 * search for key in the extent_buffer. The items start at offset p,
562 * and they are item_size apart. There are 'max' items in p.
563 *
564 * the slot in the array is returned via slot, and it points to
565 * the place where you would insert key if it is not found in
566 * the array.
567 *
568 * slot may point to max if the key is bigger than all of the keys
569 */
570 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
571 int item_size, struct btrfs_key *key,
572 int max, int *slot)
573 {
574 int low = 0;
575 int high = max;
576 int mid;
577 int ret;
578 unsigned long offset;
579 struct btrfs_disk_key *tmp;
580
581 while(low < high) {
582 mid = (low + high) / 2;
583 offset = p + mid * item_size;
584
585 tmp = (struct btrfs_disk_key *)(eb->data + offset);
586 ret = btrfs_comp_keys(tmp, key);
587
588 if (ret < 0)
589 low = mid + 1;
590 else if (ret > 0)
591 high = mid;
592 else {
593 *slot = mid;
594 return 0;
595 }
596 }
597 *slot = low;
598 return 1;
599 }
600
601 /*
602 * simple bin_search frontend that does the right thing for
603 * leaves vs nodes
604 */
605 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
606 int level, int *slot)
607 {
608 if (level == 0)
609 return generic_bin_search(eb,
610 offsetof(struct btrfs_leaf, items),
611 sizeof(struct btrfs_item),
612 key, btrfs_header_nritems(eb),
613 slot);
614 else
615 return generic_bin_search(eb,
616 offsetof(struct btrfs_node, ptrs),
617 sizeof(struct btrfs_key_ptr),
618 key, btrfs_header_nritems(eb),
619 slot);
620 }
621
622 struct extent_buffer *read_node_slot(struct btrfs_root *root,
623 struct extent_buffer *parent, int slot)
624 {
625 int level = btrfs_header_level(parent);
626 if (slot < 0)
627 return NULL;
628 if (slot >= btrfs_header_nritems(parent))
629 return NULL;
630
631 if (level == 0)
632 return NULL;
633
634 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
635 btrfs_level_size(root, level - 1),
636 btrfs_node_ptr_generation(parent, slot));
637 }
638
639 static int balance_level(struct btrfs_trans_handle *trans,
640 struct btrfs_root *root,
641 struct btrfs_path *path, int level)
642 {
643 struct extent_buffer *right = NULL;
644 struct extent_buffer *mid;
645 struct extent_buffer *left = NULL;
646 struct extent_buffer *parent = NULL;
647 int ret = 0;
648 int wret;
649 int pslot;
650 int orig_slot = path->slots[level];
651 u64 orig_ptr;
652
653 if (level == 0)
654 return 0;
655
656 mid = path->nodes[level];
657 WARN_ON(btrfs_header_generation(mid) != trans->transid);
658
659 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
660
661 if (level < BTRFS_MAX_LEVEL - 1) {
662 parent = path->nodes[level + 1];
663 pslot = path->slots[level + 1];
664 }
665
666 /*
667 * deal with the case where there is only one pointer in the root
668 * by promoting the node below to a root
669 */
670 if (!parent) {
671 struct extent_buffer *child;
672
673 if (btrfs_header_nritems(mid) != 1)
674 return 0;
675
676 /* promote the child to a root */
677 child = read_node_slot(root, mid, 0);
678 BUG_ON(!child);
679 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
680 BUG_ON(ret);
681
682 root->node = child;
683 add_root_to_dirty_list(root);
684 path->nodes[level] = NULL;
685 clean_tree_block(trans, root, mid);
686 wait_on_tree_block_writeback(root, mid);
687 /* once for the path */
688 free_extent_buffer(mid);
689
690 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
691 0, root->root_key.objectid,
692 level, 1);
693 /* once for the root ptr */
694 free_extent_buffer(mid);
695 return ret;
696 }
697 if (btrfs_header_nritems(mid) >
698 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
699 return 0;
700
701 left = read_node_slot(root, parent, pslot - 1);
702 if (left) {
703 wret = btrfs_cow_block(trans, root, left,
704 parent, pslot - 1, &left);
705 if (wret) {
706 ret = wret;
707 goto enospc;
708 }
709 }
710 right = read_node_slot(root, parent, pslot + 1);
711 if (right) {
712 wret = btrfs_cow_block(trans, root, right,
713 parent, pslot + 1, &right);
714 if (wret) {
715 ret = wret;
716 goto enospc;
717 }
718 }
719
720 /* first, try to make some room in the middle buffer */
721 if (left) {
722 orig_slot += btrfs_header_nritems(left);
723 wret = push_node_left(trans, root, left, mid, 1);
724 if (wret < 0)
725 ret = wret;
726 }
727
728 /*
729 * then try to empty the right most buffer into the middle
730 */
731 if (right) {
732 wret = push_node_left(trans, root, mid, right, 1);
733 if (wret < 0 && wret != -ENOSPC)
734 ret = wret;
735 if (btrfs_header_nritems(right) == 0) {
736 u64 bytenr = right->start;
737 u32 blocksize = right->len;
738
739 clean_tree_block(trans, root, right);
740 wait_on_tree_block_writeback(root, right);
741 free_extent_buffer(right);
742 right = NULL;
743 wret = btrfs_del_ptr(trans, root, path,
744 level + 1, pslot + 1);
745 if (wret)
746 ret = wret;
747 wret = btrfs_free_extent(trans, root, bytenr,
748 blocksize, 0,
749 root->root_key.objectid,
750 level, 0);
751 if (wret)
752 ret = wret;
753 } else {
754 struct btrfs_disk_key right_key;
755 btrfs_node_key(right, &right_key, 0);
756 btrfs_set_node_key(parent, &right_key, pslot + 1);
757 btrfs_mark_buffer_dirty(parent);
758 }
759 }
760 if (btrfs_header_nritems(mid) == 1) {
761 /*
762 * we're not allowed to leave a node with one item in the
763 * tree during a delete. A deletion from lower in the tree
764 * could try to delete the only pointer in this node.
765 * So, pull some keys from the left.
766 * There has to be a left pointer at this point because
767 * otherwise we would have pulled some pointers from the
768 * right
769 */
770 BUG_ON(!left);
771 wret = balance_node_right(trans, root, mid, left);
772 if (wret < 0) {
773 ret = wret;
774 goto enospc;
775 }
776 if (wret == 1) {
777 wret = push_node_left(trans, root, left, mid, 1);
778 if (wret < 0)
779 ret = wret;
780 }
781 BUG_ON(wret == 1);
782 }
783 if (btrfs_header_nritems(mid) == 0) {
784 /* we've managed to empty the middle node, drop it */
785 u64 bytenr = mid->start;
786 u32 blocksize = mid->len;
787 clean_tree_block(trans, root, mid);
788 wait_on_tree_block_writeback(root, mid);
789 free_extent_buffer(mid);
790 mid = NULL;
791 wret = btrfs_del_ptr(trans, root, path, level + 1, pslot);
792 if (wret)
793 ret = wret;
794 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
795 0, root->root_key.objectid,
796 level, 0);
797 if (wret)
798 ret = wret;
799 } else {
800 /* update the parent key to reflect our changes */
801 struct btrfs_disk_key mid_key;
802 btrfs_node_key(mid, &mid_key, 0);
803 btrfs_set_node_key(parent, &mid_key, pslot);
804 btrfs_mark_buffer_dirty(parent);
805 }
806
807 /* update the path */
808 if (left) {
809 if (btrfs_header_nritems(left) > orig_slot) {
810 extent_buffer_get(left);
811 path->nodes[level] = left;
812 path->slots[level + 1] -= 1;
813 path->slots[level] = orig_slot;
814 if (mid)
815 free_extent_buffer(mid);
816 } else {
817 orig_slot -= btrfs_header_nritems(left);
818 path->slots[level] = orig_slot;
819 }
820 }
821 /* double check we haven't messed things up */
822 check_block(root, path, level);
823 if (orig_ptr !=
824 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
825 BUG();
826 enospc:
827 if (right)
828 free_extent_buffer(right);
829 if (left)
830 free_extent_buffer(left);
831 return ret;
832 }
833
834 /* returns zero if the push worked, non-zero otherwise */
835 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
836 struct btrfs_root *root,
837 struct btrfs_path *path, int level)
838 {
839 struct extent_buffer *right = NULL;
840 struct extent_buffer *mid;
841 struct extent_buffer *left = NULL;
842 struct extent_buffer *parent = NULL;
843 int ret = 0;
844 int wret;
845 int pslot;
846 int orig_slot = path->slots[level];
847
848 if (level == 0)
849 return 1;
850
851 mid = path->nodes[level];
852 WARN_ON(btrfs_header_generation(mid) != trans->transid);
853
854 if (level < BTRFS_MAX_LEVEL - 1) {
855 parent = path->nodes[level + 1];
856 pslot = path->slots[level + 1];
857 }
858
859 if (!parent)
860 return 1;
861
862 left = read_node_slot(root, parent, pslot - 1);
863
864 /* first, try to make some room in the middle buffer */
865 if (left) {
866 u32 left_nr;
867 left_nr = btrfs_header_nritems(left);
868 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
869 wret = 1;
870 } else {
871 ret = btrfs_cow_block(trans, root, left, parent,
872 pslot - 1, &left);
873 if (ret)
874 wret = 1;
875 else {
876 wret = push_node_left(trans, root,
877 left, mid, 0);
878 }
879 }
880 if (wret < 0)
881 ret = wret;
882 if (wret == 0) {
883 struct btrfs_disk_key disk_key;
884 orig_slot += left_nr;
885 btrfs_node_key(mid, &disk_key, 0);
886 btrfs_set_node_key(parent, &disk_key, pslot);
887 btrfs_mark_buffer_dirty(parent);
888 if (btrfs_header_nritems(left) > orig_slot) {
889 path->nodes[level] = left;
890 path->slots[level + 1] -= 1;
891 path->slots[level] = orig_slot;
892 free_extent_buffer(mid);
893 } else {
894 orig_slot -=
895 btrfs_header_nritems(left);
896 path->slots[level] = orig_slot;
897 free_extent_buffer(left);
898 }
899 return 0;
900 }
901 free_extent_buffer(left);
902 }
903 right= read_node_slot(root, parent, pslot + 1);
904
905 /*
906 * then try to empty the right most buffer into the middle
907 */
908 if (right) {
909 u32 right_nr;
910 right_nr = btrfs_header_nritems(right);
911 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
912 wret = 1;
913 } else {
914 ret = btrfs_cow_block(trans, root, right,
915 parent, pslot + 1,
916 &right);
917 if (ret)
918 wret = 1;
919 else {
920 wret = balance_node_right(trans, root,
921 right, mid);
922 }
923 }
924 if (wret < 0)
925 ret = wret;
926 if (wret == 0) {
927 struct btrfs_disk_key disk_key;
928
929 btrfs_node_key(right, &disk_key, 0);
930 btrfs_set_node_key(parent, &disk_key, pslot + 1);
931 btrfs_mark_buffer_dirty(parent);
932
933 if (btrfs_header_nritems(mid) <= orig_slot) {
934 path->nodes[level] = right;
935 path->slots[level + 1] += 1;
936 path->slots[level] = orig_slot -
937 btrfs_header_nritems(mid);
938 free_extent_buffer(mid);
939 } else {
940 free_extent_buffer(right);
941 }
942 return 0;
943 }
944 free_extent_buffer(right);
945 }
946 return 1;
947 }
948
949 /*
950 * readahead one full node of leaves
951 */
952 void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
953 int level, int slot, u64 objectid)
954 {
955 struct extent_buffer *node;
956 struct btrfs_disk_key disk_key;
957 u32 nritems;
958 u64 search;
959 u64 lowest_read;
960 u64 highest_read;
961 u64 nread = 0;
962 int direction = path->reada;
963 struct extent_buffer *eb;
964 u32 nr;
965 u32 blocksize;
966 u32 nscan = 0;
967
968 if (level != 1)
969 return;
970
971 if (!path->nodes[level])
972 return;
973
974 node = path->nodes[level];
975 search = btrfs_node_blockptr(node, slot);
976 blocksize = btrfs_level_size(root, level - 1);
977 eb = btrfs_find_tree_block(root, search, blocksize);
978 if (eb) {
979 free_extent_buffer(eb);
980 return;
981 }
982
983 highest_read = search;
984 lowest_read = search;
985
986 nritems = btrfs_header_nritems(node);
987 nr = slot;
988 while(1) {
989 if (direction < 0) {
990 if (nr == 0)
991 break;
992 nr--;
993 } else if (direction > 0) {
994 nr++;
995 if (nr >= nritems)
996 break;
997 }
998 if (path->reada < 0 && objectid) {
999 btrfs_node_key(node, &disk_key, nr);
1000 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1001 break;
1002 }
1003 search = btrfs_node_blockptr(node, nr);
1004 if ((search >= lowest_read && search <= highest_read) ||
1005 (search < lowest_read && lowest_read - search <= 32768) ||
1006 (search > highest_read && search - highest_read <= 32768)) {
1007 readahead_tree_block(root, search, blocksize,
1008 btrfs_node_ptr_generation(node, nr));
1009 nread += blocksize;
1010 }
1011 nscan++;
1012 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1013 break;
1014 if(nread > (1024 * 1024) || nscan > 128)
1015 break;
1016
1017 if (search < lowest_read)
1018 lowest_read = search;
1019 if (search > highest_read)
1020 highest_read = search;
1021 }
1022 }
1023
1024 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
1025 u64 iobjectid, u64 ioff, u8 key_type,
1026 struct btrfs_key *found_key)
1027 {
1028 int ret;
1029 struct btrfs_key key;
1030 struct extent_buffer *eb;
1031 struct btrfs_path *path;
1032
1033 key.type = key_type;
1034 key.objectid = iobjectid;
1035 key.offset = ioff;
1036
1037 if (found_path == NULL) {
1038 path = btrfs_alloc_path();
1039 if (!path)
1040 return -ENOMEM;
1041 } else
1042 path = found_path;
1043
1044 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1045 if ((ret < 0) || (found_key == NULL)) {
1046 if (path != found_path)
1047 btrfs_free_path(path);
1048 return ret;
1049 }
1050
1051 eb = path->nodes[0];
1052 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1053 ret = btrfs_next_leaf(fs_root, path);
1054 if (ret)
1055 return ret;
1056 eb = path->nodes[0];
1057 }
1058
1059 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1060 if (found_key->type != key.type ||
1061 found_key->objectid != key.objectid)
1062 return 1;
1063
1064 return 0;
1065 }
1066
1067 /*
1068 * look for key in the tree. path is filled in with nodes along the way
1069 * if key is found, we return zero and you can find the item in the leaf
1070 * level of the path (level 0)
1071 *
1072 * If the key isn't found, the path points to the slot where it should
1073 * be inserted, and 1 is returned. If there are other errors during the
1074 * search a negative error number is returned.
1075 *
1076 * if ins_len > 0, nodes and leaves will be split as we walk down the
1077 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1078 * possible)
1079 */
1080 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1081 *root, struct btrfs_key *key, struct btrfs_path *p, int
1082 ins_len, int cow)
1083 {
1084 struct extent_buffer *b;
1085 int slot;
1086 int ret;
1087 int level;
1088 int should_reada = p->reada;
1089 u8 lowest_level = 0;
1090
1091 lowest_level = p->lowest_level;
1092 WARN_ON(lowest_level && ins_len > 0);
1093 WARN_ON(p->nodes[0] != NULL);
1094 /*
1095 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1096 */
1097 again:
1098 b = root->node;
1099 extent_buffer_get(b);
1100 while (b) {
1101 level = btrfs_header_level(b);
1102 if (cow) {
1103 int wret;
1104 wret = btrfs_cow_block(trans, root, b,
1105 p->nodes[level + 1],
1106 p->slots[level + 1],
1107 &b);
1108 if (wret) {
1109 free_extent_buffer(b);
1110 return wret;
1111 }
1112 }
1113 BUG_ON(!cow && ins_len);
1114 if (level != btrfs_header_level(b))
1115 WARN_ON(1);
1116 level = btrfs_header_level(b);
1117 p->nodes[level] = b;
1118 ret = check_block(root, p, level);
1119 if (ret)
1120 return -1;
1121 ret = bin_search(b, key, level, &slot);
1122 if (level != 0) {
1123 if (ret && slot > 0)
1124 slot -= 1;
1125 p->slots[level] = slot;
1126 if ((p->search_for_split || ins_len > 0) &&
1127 btrfs_header_nritems(b) >=
1128 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1129 int sret = split_node(trans, root, p, level);
1130 BUG_ON(sret > 0);
1131 if (sret)
1132 return sret;
1133 b = p->nodes[level];
1134 slot = p->slots[level];
1135 } else if (ins_len < 0) {
1136 int sret = balance_level(trans, root, p,
1137 level);
1138 if (sret)
1139 return sret;
1140 b = p->nodes[level];
1141 if (!b) {
1142 btrfs_release_path(p);
1143 goto again;
1144 }
1145 slot = p->slots[level];
1146 BUG_ON(btrfs_header_nritems(b) == 1);
1147 }
1148 /* this is only true while dropping a snapshot */
1149 if (level == lowest_level)
1150 break;
1151
1152 if (should_reada)
1153 reada_for_search(root, p, level, slot,
1154 key->objectid);
1155
1156 b = read_node_slot(root, b, slot);
1157 if (!extent_buffer_uptodate(b))
1158 return -EIO;
1159 } else {
1160 p->slots[level] = slot;
1161 if (ins_len > 0 &&
1162 ins_len > btrfs_leaf_free_space(root, b)) {
1163 int sret = split_leaf(trans, root, key,
1164 p, ins_len, ret == 0);
1165 BUG_ON(sret > 0);
1166 if (sret)
1167 return sret;
1168 }
1169 return ret;
1170 }
1171 }
1172 return 1;
1173 }
1174
1175 /*
1176 * adjust the pointers going up the tree, starting at level
1177 * making sure the right key of each node is points to 'key'.
1178 * This is used after shifting pointers to the left, so it stops
1179 * fixing up pointers when a given leaf/node is not in slot 0 of the
1180 * higher levels
1181 */
1182 void btrfs_fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
1183 struct btrfs_disk_key *key, int level)
1184 {
1185 int i;
1186 struct extent_buffer *t;
1187
1188 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1189 int tslot = path->slots[i];
1190 if (!path->nodes[i])
1191 break;
1192 t = path->nodes[i];
1193 btrfs_set_node_key(t, key, tslot);
1194 btrfs_mark_buffer_dirty(path->nodes[i]);
1195 if (tslot != 0)
1196 break;
1197 }
1198 }
1199
1200 /*
1201 * update item key.
1202 *
1203 * This function isn't completely safe. It's the caller's responsibility
1204 * that the new key won't break the order
1205 */
1206 int btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
1207 struct btrfs_key *new_key)
1208 {
1209 struct btrfs_disk_key disk_key;
1210 struct extent_buffer *eb;
1211 int slot;
1212
1213 eb = path->nodes[0];
1214 slot = path->slots[0];
1215 if (slot > 0) {
1216 btrfs_item_key(eb, &disk_key, slot - 1);
1217 if (btrfs_comp_keys(&disk_key, new_key) >= 0)
1218 return -1;
1219 }
1220 if (slot < btrfs_header_nritems(eb) - 1) {
1221 btrfs_item_key(eb, &disk_key, slot + 1);
1222 if (btrfs_comp_keys(&disk_key, new_key) <= 0)
1223 return -1;
1224 }
1225
1226 btrfs_cpu_key_to_disk(&disk_key, new_key);
1227 btrfs_set_item_key(eb, &disk_key, slot);
1228 btrfs_mark_buffer_dirty(eb);
1229 if (slot == 0)
1230 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1231 return 0;
1232 }
1233
1234 /*
1235 * update an item key without the safety checks. This is meant to be called by
1236 * fsck only.
1237 */
1238 void btrfs_set_item_key_unsafe(struct btrfs_root *root,
1239 struct btrfs_path *path,
1240 struct btrfs_key *new_key)
1241 {
1242 struct btrfs_disk_key disk_key;
1243 struct extent_buffer *eb;
1244 int slot;
1245
1246 eb = path->nodes[0];
1247 slot = path->slots[0];
1248
1249 btrfs_cpu_key_to_disk(&disk_key, new_key);
1250 btrfs_set_item_key(eb, &disk_key, slot);
1251 btrfs_mark_buffer_dirty(eb);
1252 if (slot == 0)
1253 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1254 }
1255
1256 /*
1257 * try to push data from one node into the next node left in the
1258 * tree.
1259 *
1260 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1261 * error, and > 0 if there was no room in the left hand block.
1262 */
1263 static int push_node_left(struct btrfs_trans_handle *trans,
1264 struct btrfs_root *root, struct extent_buffer *dst,
1265 struct extent_buffer *src, int empty)
1266 {
1267 int push_items = 0;
1268 int src_nritems;
1269 int dst_nritems;
1270 int ret = 0;
1271
1272 src_nritems = btrfs_header_nritems(src);
1273 dst_nritems = btrfs_header_nritems(dst);
1274 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1275 WARN_ON(btrfs_header_generation(src) != trans->transid);
1276 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1277
1278 if (!empty && src_nritems <= 8)
1279 return 1;
1280
1281 if (push_items <= 0) {
1282 return 1;
1283 }
1284
1285 if (empty) {
1286 push_items = min(src_nritems, push_items);
1287 if (push_items < src_nritems) {
1288 /* leave at least 8 pointers in the node if
1289 * we aren't going to empty it
1290 */
1291 if (src_nritems - push_items < 8) {
1292 if (push_items <= 8)
1293 return 1;
1294 push_items -= 8;
1295 }
1296 }
1297 } else
1298 push_items = min(src_nritems - 8, push_items);
1299
1300 copy_extent_buffer(dst, src,
1301 btrfs_node_key_ptr_offset(dst_nritems),
1302 btrfs_node_key_ptr_offset(0),
1303 push_items * sizeof(struct btrfs_key_ptr));
1304
1305 if (push_items < src_nritems) {
1306 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1307 btrfs_node_key_ptr_offset(push_items),
1308 (src_nritems - push_items) *
1309 sizeof(struct btrfs_key_ptr));
1310 }
1311 btrfs_set_header_nritems(src, src_nritems - push_items);
1312 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1313 btrfs_mark_buffer_dirty(src);
1314 btrfs_mark_buffer_dirty(dst);
1315
1316 return ret;
1317 }
1318
1319 /*
1320 * try to push data from one node into the next node right in the
1321 * tree.
1322 *
1323 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1324 * error, and > 0 if there was no room in the right hand block.
1325 *
1326 * this will only push up to 1/2 the contents of the left node over
1327 */
1328 static int balance_node_right(struct btrfs_trans_handle *trans,
1329 struct btrfs_root *root,
1330 struct extent_buffer *dst,
1331 struct extent_buffer *src)
1332 {
1333 int push_items = 0;
1334 int max_push;
1335 int src_nritems;
1336 int dst_nritems;
1337 int ret = 0;
1338
1339 WARN_ON(btrfs_header_generation(src) != trans->transid);
1340 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1341
1342 src_nritems = btrfs_header_nritems(src);
1343 dst_nritems = btrfs_header_nritems(dst);
1344 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1345 if (push_items <= 0) {
1346 return 1;
1347 }
1348
1349 if (src_nritems < 4) {
1350 return 1;
1351 }
1352
1353 max_push = src_nritems / 2 + 1;
1354 /* don't try to empty the node */
1355 if (max_push >= src_nritems) {
1356 return 1;
1357 }
1358
1359 if (max_push < push_items)
1360 push_items = max_push;
1361
1362 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1363 btrfs_node_key_ptr_offset(0),
1364 (dst_nritems) *
1365 sizeof(struct btrfs_key_ptr));
1366
1367 copy_extent_buffer(dst, src,
1368 btrfs_node_key_ptr_offset(0),
1369 btrfs_node_key_ptr_offset(src_nritems - push_items),
1370 push_items * sizeof(struct btrfs_key_ptr));
1371
1372 btrfs_set_header_nritems(src, src_nritems - push_items);
1373 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1374
1375 btrfs_mark_buffer_dirty(src);
1376 btrfs_mark_buffer_dirty(dst);
1377
1378 return ret;
1379 }
1380
1381 /*
1382 * helper function to insert a new root level in the tree.
1383 * A new node is allocated, and a single item is inserted to
1384 * point to the existing root
1385 *
1386 * returns zero on success or < 0 on failure.
1387 */
1388 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1389 struct btrfs_root *root,
1390 struct btrfs_path *path, int level)
1391 {
1392 u64 lower_gen;
1393 struct extent_buffer *lower;
1394 struct extent_buffer *c;
1395 struct extent_buffer *old;
1396 struct btrfs_disk_key lower_key;
1397
1398 BUG_ON(path->nodes[level]);
1399 BUG_ON(path->nodes[level-1] != root->node);
1400
1401 lower = path->nodes[level-1];
1402 if (level == 1)
1403 btrfs_item_key(lower, &lower_key, 0);
1404 else
1405 btrfs_node_key(lower, &lower_key, 0);
1406
1407 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1408 root->root_key.objectid, &lower_key,
1409 level, root->node->start, 0);
1410
1411 if (IS_ERR(c))
1412 return PTR_ERR(c);
1413
1414 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
1415 btrfs_set_header_nritems(c, 1);
1416 btrfs_set_header_level(c, level);
1417 btrfs_set_header_bytenr(c, c->start);
1418 btrfs_set_header_generation(c, trans->transid);
1419 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
1420 btrfs_set_header_owner(c, root->root_key.objectid);
1421
1422 write_extent_buffer(c, root->fs_info->fsid,
1423 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1424
1425 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1426 btrfs_header_chunk_tree_uuid(c),
1427 BTRFS_UUID_SIZE);
1428
1429 btrfs_set_node_key(c, &lower_key, 0);
1430 btrfs_set_node_blockptr(c, 0, lower->start);
1431 lower_gen = btrfs_header_generation(lower);
1432 WARN_ON(lower_gen != trans->transid);
1433
1434 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1435
1436 btrfs_mark_buffer_dirty(c);
1437
1438 old = root->node;
1439 root->node = c;
1440
1441 /* the super has an extra ref to root->node */
1442 free_extent_buffer(old);
1443
1444 add_root_to_dirty_list(root);
1445 extent_buffer_get(c);
1446 path->nodes[level] = c;
1447 path->slots[level] = 0;
1448 return 0;
1449 }
1450
1451 /*
1452 * worker function to insert a single pointer in a node.
1453 * the node should have enough room for the pointer already
1454 *
1455 * slot and level indicate where you want the key to go, and
1456 * blocknr is the block the key points to.
1457 *
1458 * returns zero on success and < 0 on any error
1459 */
1460 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1461 *root, struct btrfs_path *path, struct btrfs_disk_key
1462 *key, u64 bytenr, int slot, int level)
1463 {
1464 struct extent_buffer *lower;
1465 int nritems;
1466
1467 BUG_ON(!path->nodes[level]);
1468 lower = path->nodes[level];
1469 nritems = btrfs_header_nritems(lower);
1470 if (slot > nritems)
1471 BUG();
1472 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1473 BUG();
1474 if (slot != nritems) {
1475 memmove_extent_buffer(lower,
1476 btrfs_node_key_ptr_offset(slot + 1),
1477 btrfs_node_key_ptr_offset(slot),
1478 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1479 }
1480 btrfs_set_node_key(lower, key, slot);
1481 btrfs_set_node_blockptr(lower, slot, bytenr);
1482 WARN_ON(trans->transid == 0);
1483 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1484 btrfs_set_header_nritems(lower, nritems + 1);
1485 btrfs_mark_buffer_dirty(lower);
1486 return 0;
1487 }
1488
1489 /*
1490 * split the node at the specified level in path in two.
1491 * The path is corrected to point to the appropriate node after the split
1492 *
1493 * Before splitting this tries to make some room in the node by pushing
1494 * left and right, if either one works, it returns right away.
1495 *
1496 * returns 0 on success and < 0 on failure
1497 */
1498 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1499 *root, struct btrfs_path *path, int level)
1500 {
1501 struct extent_buffer *c;
1502 struct extent_buffer *split;
1503 struct btrfs_disk_key disk_key;
1504 int mid;
1505 int ret;
1506 int wret;
1507 u32 c_nritems;
1508
1509 c = path->nodes[level];
1510 WARN_ON(btrfs_header_generation(c) != trans->transid);
1511 if (c == root->node) {
1512 /* trying to split the root, lets make a new one */
1513 ret = insert_new_root(trans, root, path, level + 1);
1514 if (ret)
1515 return ret;
1516 } else {
1517 ret = push_nodes_for_insert(trans, root, path, level);
1518 c = path->nodes[level];
1519 if (!ret && btrfs_header_nritems(c) <
1520 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1521 return 0;
1522 if (ret < 0)
1523 return ret;
1524 }
1525
1526 c_nritems = btrfs_header_nritems(c);
1527 mid = (c_nritems + 1) / 2;
1528 btrfs_node_key(c, &disk_key, mid);
1529
1530 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1531 root->root_key.objectid,
1532 &disk_key, level, c->start, 0);
1533 if (IS_ERR(split))
1534 return PTR_ERR(split);
1535
1536 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
1537 btrfs_set_header_level(split, btrfs_header_level(c));
1538 btrfs_set_header_bytenr(split, split->start);
1539 btrfs_set_header_generation(split, trans->transid);
1540 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
1541 btrfs_set_header_owner(split, root->root_key.objectid);
1542 write_extent_buffer(split, root->fs_info->fsid,
1543 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1544 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1545 btrfs_header_chunk_tree_uuid(split),
1546 BTRFS_UUID_SIZE);
1547
1548
1549 copy_extent_buffer(split, c,
1550 btrfs_node_key_ptr_offset(0),
1551 btrfs_node_key_ptr_offset(mid),
1552 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1553 btrfs_set_header_nritems(split, c_nritems - mid);
1554 btrfs_set_header_nritems(c, mid);
1555 ret = 0;
1556
1557 btrfs_mark_buffer_dirty(c);
1558 btrfs_mark_buffer_dirty(split);
1559
1560 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1561 path->slots[level + 1] + 1,
1562 level + 1);
1563 if (wret)
1564 ret = wret;
1565
1566 if (path->slots[level] >= mid) {
1567 path->slots[level] -= mid;
1568 free_extent_buffer(c);
1569 path->nodes[level] = split;
1570 path->slots[level + 1] += 1;
1571 } else {
1572 free_extent_buffer(split);
1573 }
1574 return ret;
1575 }
1576
1577 /*
1578 * how many bytes are required to store the items in a leaf. start
1579 * and nr indicate which items in the leaf to check. This totals up the
1580 * space used both by the item structs and the item data
1581 */
1582 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1583 {
1584 int data_len;
1585 int nritems = btrfs_header_nritems(l);
1586 int end = min(nritems, start + nr) - 1;
1587
1588 if (!nr)
1589 return 0;
1590 data_len = btrfs_item_end_nr(l, start);
1591 data_len = data_len - btrfs_item_offset_nr(l, end);
1592 data_len += sizeof(struct btrfs_item) * nr;
1593 WARN_ON(data_len < 0);
1594 return data_len;
1595 }
1596
1597 /*
1598 * The space between the end of the leaf items and
1599 * the start of the leaf data. IOW, how much room
1600 * the leaf has left for both items and data
1601 */
1602 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1603 {
1604 int nritems = btrfs_header_nritems(leaf);
1605 int ret;
1606 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1607 if (ret < 0) {
1608 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1609 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1610 leaf_space_used(leaf, 0, nritems), nritems);
1611 }
1612 return ret;
1613 }
1614
1615 /*
1616 * push some data in the path leaf to the right, trying to free up at
1617 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1618 *
1619 * returns 1 if the push failed because the other node didn't have enough
1620 * room, 0 if everything worked out and < 0 if there were major errors.
1621 */
1622 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1623 *root, struct btrfs_path *path, int data_size,
1624 int empty)
1625 {
1626 struct extent_buffer *left = path->nodes[0];
1627 struct extent_buffer *right;
1628 struct extent_buffer *upper;
1629 struct btrfs_disk_key disk_key;
1630 int slot;
1631 u32 i;
1632 int free_space;
1633 int push_space = 0;
1634 int push_items = 0;
1635 struct btrfs_item *item;
1636 u32 left_nritems;
1637 u32 nr;
1638 u32 right_nritems;
1639 u32 data_end;
1640 u32 this_item_size;
1641 int ret;
1642
1643 slot = path->slots[1];
1644 if (!path->nodes[1]) {
1645 return 1;
1646 }
1647 upper = path->nodes[1];
1648 if (slot >= btrfs_header_nritems(upper) - 1)
1649 return 1;
1650
1651 right = read_node_slot(root, upper, slot + 1);
1652 free_space = btrfs_leaf_free_space(root, right);
1653 if (free_space < data_size) {
1654 free_extent_buffer(right);
1655 return 1;
1656 }
1657
1658 /* cow and double check */
1659 ret = btrfs_cow_block(trans, root, right, upper,
1660 slot + 1, &right);
1661 if (ret) {
1662 free_extent_buffer(right);
1663 return 1;
1664 }
1665 free_space = btrfs_leaf_free_space(root, right);
1666 if (free_space < data_size) {
1667 free_extent_buffer(right);
1668 return 1;
1669 }
1670
1671 left_nritems = btrfs_header_nritems(left);
1672 if (left_nritems == 0) {
1673 free_extent_buffer(right);
1674 return 1;
1675 }
1676
1677 if (empty)
1678 nr = 0;
1679 else
1680 nr = 1;
1681
1682 i = left_nritems - 1;
1683 while (i >= nr) {
1684 item = btrfs_item_nr(i);
1685
1686 if (path->slots[0] == i)
1687 push_space += data_size + sizeof(*item);
1688
1689 this_item_size = btrfs_item_size(left, item);
1690 if (this_item_size + sizeof(*item) + push_space > free_space)
1691 break;
1692 push_items++;
1693 push_space += this_item_size + sizeof(*item);
1694 if (i == 0)
1695 break;
1696 i--;
1697 }
1698
1699 if (push_items == 0) {
1700 free_extent_buffer(right);
1701 return 1;
1702 }
1703
1704 if (!empty && push_items == left_nritems)
1705 WARN_ON(1);
1706
1707 /* push left to right */
1708 right_nritems = btrfs_header_nritems(right);
1709
1710 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1711 push_space -= leaf_data_end(root, left);
1712
1713 /* make room in the right data area */
1714 data_end = leaf_data_end(root, right);
1715 memmove_extent_buffer(right,
1716 btrfs_leaf_data(right) + data_end - push_space,
1717 btrfs_leaf_data(right) + data_end,
1718 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1719
1720 /* copy from the left data area */
1721 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1722 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1723 btrfs_leaf_data(left) + leaf_data_end(root, left),
1724 push_space);
1725
1726 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1727 btrfs_item_nr_offset(0),
1728 right_nritems * sizeof(struct btrfs_item));
1729
1730 /* copy the items from left to right */
1731 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1732 btrfs_item_nr_offset(left_nritems - push_items),
1733 push_items * sizeof(struct btrfs_item));
1734
1735 /* update the item pointers */
1736 right_nritems += push_items;
1737 btrfs_set_header_nritems(right, right_nritems);
1738 push_space = BTRFS_LEAF_DATA_SIZE(root);
1739 for (i = 0; i < right_nritems; i++) {
1740 item = btrfs_item_nr(i);
1741 push_space -= btrfs_item_size(right, item);
1742 btrfs_set_item_offset(right, item, push_space);
1743 }
1744
1745 left_nritems -= push_items;
1746 btrfs_set_header_nritems(left, left_nritems);
1747
1748 if (left_nritems)
1749 btrfs_mark_buffer_dirty(left);
1750 btrfs_mark_buffer_dirty(right);
1751
1752 btrfs_item_key(right, &disk_key, 0);
1753 btrfs_set_node_key(upper, &disk_key, slot + 1);
1754 btrfs_mark_buffer_dirty(upper);
1755
1756 /* then fixup the leaf pointer in the path */
1757 if (path->slots[0] >= left_nritems) {
1758 path->slots[0] -= left_nritems;
1759 free_extent_buffer(path->nodes[0]);
1760 path->nodes[0] = right;
1761 path->slots[1] += 1;
1762 } else {
1763 free_extent_buffer(right);
1764 }
1765 return 0;
1766 }
1767 /*
1768 * push some data in the path leaf to the left, trying to free up at
1769 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1770 */
1771 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1772 *root, struct btrfs_path *path, int data_size,
1773 int empty)
1774 {
1775 struct btrfs_disk_key disk_key;
1776 struct extent_buffer *right = path->nodes[0];
1777 struct extent_buffer *left;
1778 int slot;
1779 int i;
1780 int free_space;
1781 int push_space = 0;
1782 int push_items = 0;
1783 struct btrfs_item *item;
1784 u32 old_left_nritems;
1785 u32 right_nritems;
1786 u32 nr;
1787 int ret = 0;
1788 u32 this_item_size;
1789 u32 old_left_item_size;
1790
1791 slot = path->slots[1];
1792 if (slot == 0)
1793 return 1;
1794 if (!path->nodes[1])
1795 return 1;
1796
1797 right_nritems = btrfs_header_nritems(right);
1798 if (right_nritems == 0) {
1799 return 1;
1800 }
1801
1802 left = read_node_slot(root, path->nodes[1], slot - 1);
1803 free_space = btrfs_leaf_free_space(root, left);
1804 if (free_space < data_size) {
1805 free_extent_buffer(left);
1806 return 1;
1807 }
1808
1809 /* cow and double check */
1810 ret = btrfs_cow_block(trans, root, left,
1811 path->nodes[1], slot - 1, &left);
1812 if (ret) {
1813 /* we hit -ENOSPC, but it isn't fatal here */
1814 free_extent_buffer(left);
1815 return 1;
1816 }
1817
1818 free_space = btrfs_leaf_free_space(root, left);
1819 if (free_space < data_size) {
1820 free_extent_buffer(left);
1821 return 1;
1822 }
1823
1824 if (empty)
1825 nr = right_nritems;
1826 else
1827 nr = right_nritems - 1;
1828
1829 for (i = 0; i < nr; i++) {
1830 item = btrfs_item_nr(i);
1831
1832 if (path->slots[0] == i)
1833 push_space += data_size + sizeof(*item);
1834
1835 this_item_size = btrfs_item_size(right, item);
1836 if (this_item_size + sizeof(*item) + push_space > free_space)
1837 break;
1838
1839 push_items++;
1840 push_space += this_item_size + sizeof(*item);
1841 }
1842
1843 if (push_items == 0) {
1844 free_extent_buffer(left);
1845 return 1;
1846 }
1847 if (!empty && push_items == btrfs_header_nritems(right))
1848 WARN_ON(1);
1849
1850 /* push data from right to left */
1851 copy_extent_buffer(left, right,
1852 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1853 btrfs_item_nr_offset(0),
1854 push_items * sizeof(struct btrfs_item));
1855
1856 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1857 btrfs_item_offset_nr(right, push_items -1);
1858
1859 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1860 leaf_data_end(root, left) - push_space,
1861 btrfs_leaf_data(right) +
1862 btrfs_item_offset_nr(right, push_items - 1),
1863 push_space);
1864 old_left_nritems = btrfs_header_nritems(left);
1865 BUG_ON(old_left_nritems == 0);
1866
1867 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1868 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1869 u32 ioff;
1870
1871 item = btrfs_item_nr(i);
1872 ioff = btrfs_item_offset(left, item);
1873 btrfs_set_item_offset(left, item,
1874 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1875 }
1876 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1877
1878 /* fixup right node */
1879 if (push_items > right_nritems) {
1880 printk("push items %d nr %u\n", push_items, right_nritems);
1881 WARN_ON(1);
1882 }
1883
1884 if (push_items < right_nritems) {
1885 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1886 leaf_data_end(root, right);
1887 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1888 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1889 btrfs_leaf_data(right) +
1890 leaf_data_end(root, right), push_space);
1891
1892 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1893 btrfs_item_nr_offset(push_items),
1894 (btrfs_header_nritems(right) - push_items) *
1895 sizeof(struct btrfs_item));
1896 }
1897 right_nritems -= push_items;
1898 btrfs_set_header_nritems(right, right_nritems);
1899 push_space = BTRFS_LEAF_DATA_SIZE(root);
1900 for (i = 0; i < right_nritems; i++) {
1901 item = btrfs_item_nr(i);
1902 push_space = push_space - btrfs_item_size(right, item);
1903 btrfs_set_item_offset(right, item, push_space);
1904 }
1905
1906 btrfs_mark_buffer_dirty(left);
1907 if (right_nritems)
1908 btrfs_mark_buffer_dirty(right);
1909
1910 btrfs_item_key(right, &disk_key, 0);
1911 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1912
1913 /* then fixup the leaf pointer in the path */
1914 if (path->slots[0] < push_items) {
1915 path->slots[0] += old_left_nritems;
1916 free_extent_buffer(path->nodes[0]);
1917 path->nodes[0] = left;
1918 path->slots[1] -= 1;
1919 } else {
1920 free_extent_buffer(left);
1921 path->slots[0] -= push_items;
1922 }
1923 BUG_ON(path->slots[0] < 0);
1924 return ret;
1925 }
1926
1927 /*
1928 * split the path's leaf in two, making sure there is at least data_size
1929 * available for the resulting leaf level of the path.
1930 *
1931 * returns 0 if all went well and < 0 on failure.
1932 */
1933 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root,
1935 struct btrfs_path *path,
1936 struct extent_buffer *l,
1937 struct extent_buffer *right,
1938 int slot, int mid, int nritems)
1939 {
1940 int data_copy_size;
1941 int rt_data_off;
1942 int i;
1943 int ret = 0;
1944 int wret;
1945 struct btrfs_disk_key disk_key;
1946
1947 nritems = nritems - mid;
1948 btrfs_set_header_nritems(right, nritems);
1949 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
1950
1951 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
1952 btrfs_item_nr_offset(mid),
1953 nritems * sizeof(struct btrfs_item));
1954
1955 copy_extent_buffer(right, l,
1956 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
1957 data_copy_size, btrfs_leaf_data(l) +
1958 leaf_data_end(root, l), data_copy_size);
1959
1960 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
1961 btrfs_item_end_nr(l, mid);
1962
1963 for (i = 0; i < nritems; i++) {
1964 struct btrfs_item *item = btrfs_item_nr(i);
1965 u32 ioff = btrfs_item_offset(right, item);
1966 btrfs_set_item_offset(right, item, ioff + rt_data_off);
1967 }
1968
1969 btrfs_set_header_nritems(l, mid);
1970 ret = 0;
1971 btrfs_item_key(right, &disk_key, 0);
1972 wret = insert_ptr(trans, root, path, &disk_key, right->start,
1973 path->slots[1] + 1, 1);
1974 if (wret)
1975 ret = wret;
1976
1977 btrfs_mark_buffer_dirty(right);
1978 btrfs_mark_buffer_dirty(l);
1979 BUG_ON(path->slots[0] != slot);
1980
1981 if (mid <= slot) {
1982 free_extent_buffer(path->nodes[0]);
1983 path->nodes[0] = right;
1984 path->slots[0] -= mid;
1985 path->slots[1] += 1;
1986 } else {
1987 free_extent_buffer(right);
1988 }
1989
1990 BUG_ON(path->slots[0] < 0);
1991
1992 return ret;
1993 }
1994
1995 /*
1996 * split the path's leaf in two, making sure there is at least data_size
1997 * available for the resulting leaf level of the path.
1998 *
1999 * returns 0 if all went well and < 0 on failure.
2000 */
2001 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2002 struct btrfs_root *root,
2003 struct btrfs_key *ins_key,
2004 struct btrfs_path *path, int data_size,
2005 int extend)
2006 {
2007 struct btrfs_disk_key disk_key;
2008 struct extent_buffer *l;
2009 u32 nritems;
2010 int mid;
2011 int slot;
2012 struct extent_buffer *right;
2013 int ret = 0;
2014 int wret;
2015 int split;
2016 int num_doubles = 0;
2017
2018 /* first try to make some room by pushing left and right */
2019 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2020 wret = push_leaf_right(trans, root, path, data_size, 0);
2021 if (wret < 0)
2022 return wret;
2023 if (wret) {
2024 wret = push_leaf_left(trans, root, path, data_size, 0);
2025 if (wret < 0)
2026 return wret;
2027 }
2028 l = path->nodes[0];
2029
2030 /* did the pushes work? */
2031 if (btrfs_leaf_free_space(root, l) >= data_size)
2032 return 0;
2033 }
2034
2035 if (!path->nodes[1]) {
2036 ret = insert_new_root(trans, root, path, 1);
2037 if (ret)
2038 return ret;
2039 }
2040 again:
2041 split = 1;
2042 l = path->nodes[0];
2043 slot = path->slots[0];
2044 nritems = btrfs_header_nritems(l);
2045 mid = (nritems + 1) / 2;
2046
2047 if (mid <= slot) {
2048 if (nritems == 1 ||
2049 leaf_space_used(l, mid, nritems - mid) + data_size >
2050 BTRFS_LEAF_DATA_SIZE(root)) {
2051 if (slot >= nritems) {
2052 split = 0;
2053 } else {
2054 mid = slot;
2055 if (mid != nritems &&
2056 leaf_space_used(l, mid, nritems - mid) +
2057 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2058 split = 2;
2059 }
2060 }
2061 }
2062 } else {
2063 if (leaf_space_used(l, 0, mid) + data_size >
2064 BTRFS_LEAF_DATA_SIZE(root)) {
2065 if (!extend && data_size && slot == 0) {
2066 split = 0;
2067 } else if ((extend || !data_size) && slot == 0) {
2068 mid = 1;
2069 } else {
2070 mid = slot;
2071 if (mid != nritems &&
2072 leaf_space_used(l, mid, nritems - mid) +
2073 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2074 split = 2 ;
2075 }
2076 }
2077 }
2078 }
2079
2080 if (split == 0)
2081 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2082 else
2083 btrfs_item_key(l, &disk_key, mid);
2084
2085 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2086 root->root_key.objectid,
2087 &disk_key, 0, l->start, 0);
2088 if (IS_ERR(right)) {
2089 BUG_ON(1);
2090 return PTR_ERR(right);
2091 }
2092
2093 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2094 btrfs_set_header_bytenr(right, right->start);
2095 btrfs_set_header_generation(right, trans->transid);
2096 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2097 btrfs_set_header_owner(right, root->root_key.objectid);
2098 btrfs_set_header_level(right, 0);
2099 write_extent_buffer(right, root->fs_info->fsid,
2100 btrfs_header_fsid(), BTRFS_FSID_SIZE);
2101
2102 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2103 btrfs_header_chunk_tree_uuid(right),
2104 BTRFS_UUID_SIZE);
2105
2106 if (split == 0) {
2107 if (mid <= slot) {
2108 btrfs_set_header_nritems(right, 0);
2109 wret = insert_ptr(trans, root, path,
2110 &disk_key, right->start,
2111 path->slots[1] + 1, 1);
2112 if (wret)
2113 ret = wret;
2114
2115 free_extent_buffer(path->nodes[0]);
2116 path->nodes[0] = right;
2117 path->slots[0] = 0;
2118 path->slots[1] += 1;
2119 } else {
2120 btrfs_set_header_nritems(right, 0);
2121 wret = insert_ptr(trans, root, path,
2122 &disk_key,
2123 right->start,
2124 path->slots[1], 1);
2125 if (wret)
2126 ret = wret;
2127 free_extent_buffer(path->nodes[0]);
2128 path->nodes[0] = right;
2129 path->slots[0] = 0;
2130 if (path->slots[1] == 0) {
2131 btrfs_fixup_low_keys(root, path,
2132 &disk_key, 1);
2133 }
2134 }
2135 btrfs_mark_buffer_dirty(right);
2136 return ret;
2137 }
2138
2139 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2140 BUG_ON(ret);
2141
2142 if (split == 2) {
2143 BUG_ON(num_doubles != 0);
2144 num_doubles++;
2145 goto again;
2146 }
2147
2148 return ret;
2149 }
2150
2151 /*
2152 * This function splits a single item into two items,
2153 * giving 'new_key' to the new item and splitting the
2154 * old one at split_offset (from the start of the item).
2155 *
2156 * The path may be released by this operation. After
2157 * the split, the path is pointing to the old item. The
2158 * new item is going to be in the same node as the old one.
2159 *
2160 * Note, the item being split must be smaller enough to live alone on
2161 * a tree block with room for one extra struct btrfs_item
2162 *
2163 * This allows us to split the item in place, keeping a lock on the
2164 * leaf the entire time.
2165 */
2166 int btrfs_split_item(struct btrfs_trans_handle *trans,
2167 struct btrfs_root *root,
2168 struct btrfs_path *path,
2169 struct btrfs_key *new_key,
2170 unsigned long split_offset)
2171 {
2172 u32 item_size;
2173 struct extent_buffer *leaf;
2174 struct btrfs_key orig_key;
2175 struct btrfs_item *item;
2176 struct btrfs_item *new_item;
2177 int ret = 0;
2178 int slot;
2179 u32 nritems;
2180 u32 orig_offset;
2181 struct btrfs_disk_key disk_key;
2182 char *buf;
2183
2184 leaf = path->nodes[0];
2185 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2186 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2187 goto split;
2188
2189 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2190 btrfs_release_path(path);
2191
2192 path->search_for_split = 1;
2193
2194 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2195 path->search_for_split = 0;
2196
2197 /* if our item isn't there or got smaller, return now */
2198 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2199 path->slots[0])) {
2200 return -EAGAIN;
2201 }
2202
2203 ret = split_leaf(trans, root, &orig_key, path, 0, 0);
2204 BUG_ON(ret);
2205
2206 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2207 leaf = path->nodes[0];
2208
2209 split:
2210 item = btrfs_item_nr(path->slots[0]);
2211 orig_offset = btrfs_item_offset(leaf, item);
2212 item_size = btrfs_item_size(leaf, item);
2213
2214
2215 buf = kmalloc(item_size, GFP_NOFS);
2216 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2217 path->slots[0]), item_size);
2218 slot = path->slots[0] + 1;
2219 leaf = path->nodes[0];
2220
2221 nritems = btrfs_header_nritems(leaf);
2222
2223 if (slot != nritems) {
2224 /* shift the items */
2225 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2226 btrfs_item_nr_offset(slot),
2227 (nritems - slot) * sizeof(struct btrfs_item));
2228
2229 }
2230
2231 btrfs_cpu_key_to_disk(&disk_key, new_key);
2232 btrfs_set_item_key(leaf, &disk_key, slot);
2233
2234 new_item = btrfs_item_nr(slot);
2235
2236 btrfs_set_item_offset(leaf, new_item, orig_offset);
2237 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
2238
2239 btrfs_set_item_offset(leaf, item,
2240 orig_offset + item_size - split_offset);
2241 btrfs_set_item_size(leaf, item, split_offset);
2242
2243 btrfs_set_header_nritems(leaf, nritems + 1);
2244
2245 /* write the data for the start of the original item */
2246 write_extent_buffer(leaf, buf,
2247 btrfs_item_ptr_offset(leaf, path->slots[0]),
2248 split_offset);
2249
2250 /* write the data for the new item */
2251 write_extent_buffer(leaf, buf + split_offset,
2252 btrfs_item_ptr_offset(leaf, slot),
2253 item_size - split_offset);
2254 btrfs_mark_buffer_dirty(leaf);
2255
2256 ret = 0;
2257 if (btrfs_leaf_free_space(root, leaf) < 0) {
2258 btrfs_print_leaf(root, leaf);
2259 BUG();
2260 }
2261 kfree(buf);
2262 return ret;
2263 }
2264
2265 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2266 struct btrfs_root *root,
2267 struct btrfs_path *path,
2268 u32 new_size, int from_end)
2269 {
2270 int ret = 0;
2271 int slot;
2272 struct extent_buffer *leaf;
2273 struct btrfs_item *item;
2274 u32 nritems;
2275 unsigned int data_end;
2276 unsigned int old_data_start;
2277 unsigned int old_size;
2278 unsigned int size_diff;
2279 int i;
2280
2281 leaf = path->nodes[0];
2282 slot = path->slots[0];
2283
2284 old_size = btrfs_item_size_nr(leaf, slot);
2285 if (old_size == new_size)
2286 return 0;
2287
2288 nritems = btrfs_header_nritems(leaf);
2289 data_end = leaf_data_end(root, leaf);
2290
2291 old_data_start = btrfs_item_offset_nr(leaf, slot);
2292
2293 size_diff = old_size - new_size;
2294
2295 BUG_ON(slot < 0);
2296 BUG_ON(slot >= nritems);
2297
2298 /*
2299 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2300 */
2301 /* first correct the data pointers */
2302 for (i = slot; i < nritems; i++) {
2303 u32 ioff;
2304 item = btrfs_item_nr(i);
2305 ioff = btrfs_item_offset(leaf, item);
2306 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2307 }
2308
2309 /* shift the data */
2310 if (from_end) {
2311 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2312 data_end + size_diff, btrfs_leaf_data(leaf) +
2313 data_end, old_data_start + new_size - data_end);
2314 } else {
2315 struct btrfs_disk_key disk_key;
2316 u64 offset;
2317
2318 btrfs_item_key(leaf, &disk_key, slot);
2319
2320 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2321 unsigned long ptr;
2322 struct btrfs_file_extent_item *fi;
2323
2324 fi = btrfs_item_ptr(leaf, slot,
2325 struct btrfs_file_extent_item);
2326 fi = (struct btrfs_file_extent_item *)(
2327 (unsigned long)fi - size_diff);
2328
2329 if (btrfs_file_extent_type(leaf, fi) ==
2330 BTRFS_FILE_EXTENT_INLINE) {
2331 ptr = btrfs_item_ptr_offset(leaf, slot);
2332 memmove_extent_buffer(leaf, ptr,
2333 (unsigned long)fi,
2334 offsetof(struct btrfs_file_extent_item,
2335 disk_bytenr));
2336 }
2337 }
2338
2339 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2340 data_end + size_diff, btrfs_leaf_data(leaf) +
2341 data_end, old_data_start - data_end);
2342
2343 offset = btrfs_disk_key_offset(&disk_key);
2344 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2345 btrfs_set_item_key(leaf, &disk_key, slot);
2346 if (slot == 0)
2347 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2348 }
2349
2350 item = btrfs_item_nr(slot);
2351 btrfs_set_item_size(leaf, item, new_size);
2352 btrfs_mark_buffer_dirty(leaf);
2353
2354 ret = 0;
2355 if (btrfs_leaf_free_space(root, leaf) < 0) {
2356 btrfs_print_leaf(root, leaf);
2357 BUG();
2358 }
2359 return ret;
2360 }
2361
2362 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2363 struct btrfs_root *root, struct btrfs_path *path,
2364 u32 data_size)
2365 {
2366 int ret = 0;
2367 int slot;
2368 struct extent_buffer *leaf;
2369 struct btrfs_item *item;
2370 u32 nritems;
2371 unsigned int data_end;
2372 unsigned int old_data;
2373 unsigned int old_size;
2374 int i;
2375
2376 leaf = path->nodes[0];
2377
2378 nritems = btrfs_header_nritems(leaf);
2379 data_end = leaf_data_end(root, leaf);
2380
2381 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2382 btrfs_print_leaf(root, leaf);
2383 BUG();
2384 }
2385 slot = path->slots[0];
2386 old_data = btrfs_item_end_nr(leaf, slot);
2387
2388 BUG_ON(slot < 0);
2389 if (slot >= nritems) {
2390 btrfs_print_leaf(root, leaf);
2391 printk("slot %d too large, nritems %d\n", slot, nritems);
2392 BUG_ON(1);
2393 }
2394
2395 /*
2396 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2397 */
2398 /* first correct the data pointers */
2399 for (i = slot; i < nritems; i++) {
2400 u32 ioff;
2401 item = btrfs_item_nr(i);
2402 ioff = btrfs_item_offset(leaf, item);
2403 btrfs_set_item_offset(leaf, item, ioff - data_size);
2404 }
2405
2406 /* shift the data */
2407 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2408 data_end - data_size, btrfs_leaf_data(leaf) +
2409 data_end, old_data - data_end);
2410
2411 data_end = old_data;
2412 old_size = btrfs_item_size_nr(leaf, slot);
2413 item = btrfs_item_nr(slot);
2414 btrfs_set_item_size(leaf, item, old_size + data_size);
2415 btrfs_mark_buffer_dirty(leaf);
2416
2417 ret = 0;
2418 if (btrfs_leaf_free_space(root, leaf) < 0) {
2419 btrfs_print_leaf(root, leaf);
2420 BUG();
2421 }
2422 return ret;
2423 }
2424
2425 /*
2426 * Given a key and some data, insert an item into the tree.
2427 * This does all the path init required, making room in the tree if needed.
2428 */
2429 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2430 struct btrfs_root *root,
2431 struct btrfs_path *path,
2432 struct btrfs_key *cpu_key, u32 *data_size,
2433 int nr)
2434 {
2435 struct extent_buffer *leaf;
2436 struct btrfs_item *item;
2437 int ret = 0;
2438 int slot;
2439 int i;
2440 u32 nritems;
2441 u32 total_size = 0;
2442 u32 total_data = 0;
2443 unsigned int data_end;
2444 struct btrfs_disk_key disk_key;
2445
2446 for (i = 0; i < nr; i++) {
2447 total_data += data_size[i];
2448 }
2449
2450 /* create a root if there isn't one */
2451 if (!root->node)
2452 BUG();
2453
2454 total_size = total_data + nr * sizeof(struct btrfs_item);
2455 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2456 if (ret == 0) {
2457 return -EEXIST;
2458 }
2459 if (ret < 0)
2460 goto out;
2461
2462 leaf = path->nodes[0];
2463
2464 nritems = btrfs_header_nritems(leaf);
2465 data_end = leaf_data_end(root, leaf);
2466
2467 if (btrfs_leaf_free_space(root, leaf) < total_size) {
2468 btrfs_print_leaf(root, leaf);
2469 printk("not enough freespace need %u have %d\n",
2470 total_size, btrfs_leaf_free_space(root, leaf));
2471 BUG();
2472 }
2473
2474 slot = path->slots[0];
2475 BUG_ON(slot < 0);
2476
2477 if (slot != nritems) {
2478 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2479
2480 if (old_data < data_end) {
2481 btrfs_print_leaf(root, leaf);
2482 printk("slot %d old_data %d data_end %d\n",
2483 slot, old_data, data_end);
2484 BUG_ON(1);
2485 }
2486 /*
2487 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2488 */
2489 /* first correct the data pointers */
2490 for (i = slot; i < nritems; i++) {
2491 u32 ioff;
2492
2493 item = btrfs_item_nr(i);
2494 ioff = btrfs_item_offset(leaf, item);
2495 btrfs_set_item_offset(leaf, item, ioff - total_data);
2496 }
2497
2498 /* shift the items */
2499 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2500 btrfs_item_nr_offset(slot),
2501 (nritems - slot) * sizeof(struct btrfs_item));
2502
2503 /* shift the data */
2504 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2505 data_end - total_data, btrfs_leaf_data(leaf) +
2506 data_end, old_data - data_end);
2507 data_end = old_data;
2508 }
2509
2510 /* setup the item for the new data */
2511 for (i = 0; i < nr; i++) {
2512 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2513 btrfs_set_item_key(leaf, &disk_key, slot + i);
2514 item = btrfs_item_nr(slot + i);
2515 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2516 data_end -= data_size[i];
2517 btrfs_set_item_size(leaf, item, data_size[i]);
2518 }
2519 btrfs_set_header_nritems(leaf, nritems + nr);
2520 btrfs_mark_buffer_dirty(leaf);
2521
2522 ret = 0;
2523 if (slot == 0) {
2524 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2525 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2526 }
2527
2528 if (btrfs_leaf_free_space(root, leaf) < 0) {
2529 btrfs_print_leaf(root, leaf);
2530 BUG();
2531 }
2532
2533 out:
2534 return ret;
2535 }
2536
2537 /*
2538 * Given a key and some data, insert an item into the tree.
2539 * This does all the path init required, making room in the tree if needed.
2540 */
2541 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2542 *root, struct btrfs_key *cpu_key, void *data, u32
2543 data_size)
2544 {
2545 int ret = 0;
2546 struct btrfs_path *path;
2547 struct extent_buffer *leaf;
2548 unsigned long ptr;
2549
2550 path = btrfs_alloc_path();
2551 BUG_ON(!path);
2552 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2553 if (!ret) {
2554 leaf = path->nodes[0];
2555 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2556 write_extent_buffer(leaf, data, ptr, data_size);
2557 btrfs_mark_buffer_dirty(leaf);
2558 }
2559 btrfs_free_path(path);
2560 return ret;
2561 }
2562
2563 /*
2564 * delete the pointer from a given node.
2565 *
2566 * If the delete empties a node, the node is removed from the tree,
2567 * continuing all the way the root if required. The root is converted into
2568 * a leaf if all the nodes are emptied.
2569 */
2570 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2571 struct btrfs_path *path, int level, int slot)
2572 {
2573 struct extent_buffer *parent = path->nodes[level];
2574 u32 nritems;
2575 int ret = 0;
2576
2577 nritems = btrfs_header_nritems(parent);
2578 if (slot != nritems -1) {
2579 memmove_extent_buffer(parent,
2580 btrfs_node_key_ptr_offset(slot),
2581 btrfs_node_key_ptr_offset(slot + 1),
2582 sizeof(struct btrfs_key_ptr) *
2583 (nritems - slot - 1));
2584 }
2585 nritems--;
2586 btrfs_set_header_nritems(parent, nritems);
2587 if (nritems == 0 && parent == root->node) {
2588 BUG_ON(btrfs_header_level(root->node) != 1);
2589 /* just turn the root into a leaf and break */
2590 btrfs_set_header_level(root->node, 0);
2591 } else if (slot == 0) {
2592 struct btrfs_disk_key disk_key;
2593
2594 btrfs_node_key(parent, &disk_key, 0);
2595 btrfs_fixup_low_keys(root, path, &disk_key, level + 1);
2596 }
2597 btrfs_mark_buffer_dirty(parent);
2598 return ret;
2599 }
2600
2601 /*
2602 * a helper function to delete the leaf pointed to by path->slots[1] and
2603 * path->nodes[1].
2604 *
2605 * This deletes the pointer in path->nodes[1] and frees the leaf
2606 * block extent. zero is returned if it all worked out, < 0 otherwise.
2607 *
2608 * The path must have already been setup for deleting the leaf, including
2609 * all the proper balancing. path->nodes[1] must be locked.
2610 */
2611 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
2612 struct btrfs_root *root,
2613 struct btrfs_path *path,
2614 struct extent_buffer *leaf)
2615 {
2616 int ret;
2617
2618 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
2619 ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]);
2620 if (ret)
2621 return ret;
2622
2623 ret = btrfs_free_extent(trans, root, leaf->start, leaf->len,
2624 0, root->root_key.objectid, 0, 0);
2625 return ret;
2626 }
2627
2628 /*
2629 * delete the item at the leaf level in path. If that empties
2630 * the leaf, remove it from the tree
2631 */
2632 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2633 struct btrfs_path *path, int slot, int nr)
2634 {
2635 struct extent_buffer *leaf;
2636 struct btrfs_item *item;
2637 int last_off;
2638 int dsize = 0;
2639 int ret = 0;
2640 int wret;
2641 int i;
2642 u32 nritems;
2643
2644 leaf = path->nodes[0];
2645 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2646
2647 for (i = 0; i < nr; i++)
2648 dsize += btrfs_item_size_nr(leaf, slot + i);
2649
2650 nritems = btrfs_header_nritems(leaf);
2651
2652 if (slot + nr != nritems) {
2653 int data_end = leaf_data_end(root, leaf);
2654
2655 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2656 data_end + dsize,
2657 btrfs_leaf_data(leaf) + data_end,
2658 last_off - data_end);
2659
2660 for (i = slot + nr; i < nritems; i++) {
2661 u32 ioff;
2662
2663 item = btrfs_item_nr(i);
2664 ioff = btrfs_item_offset(leaf, item);
2665 btrfs_set_item_offset(leaf, item, ioff + dsize);
2666 }
2667
2668 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2669 btrfs_item_nr_offset(slot + nr),
2670 sizeof(struct btrfs_item) *
2671 (nritems - slot - nr));
2672 }
2673 btrfs_set_header_nritems(leaf, nritems - nr);
2674 nritems -= nr;
2675
2676 /* delete the leaf if we've emptied it */
2677 if (nritems == 0) {
2678 if (leaf == root->node) {
2679 btrfs_set_header_level(leaf, 0);
2680 } else {
2681 clean_tree_block(trans, root, leaf);
2682 wait_on_tree_block_writeback(root, leaf);
2683
2684 wret = btrfs_del_leaf(trans, root, path, leaf);
2685 BUG_ON(ret);
2686 if (wret)
2687 ret = wret;
2688 }
2689 } else {
2690 int used = leaf_space_used(leaf, 0, nritems);
2691 if (slot == 0) {
2692 struct btrfs_disk_key disk_key;
2693
2694 btrfs_item_key(leaf, &disk_key, 0);
2695 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2696 }
2697
2698 /* delete the leaf if it is mostly empty */
2699 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2700 /* push_leaf_left fixes the path.
2701 * make sure the path still points to our leaf
2702 * for possible call to del_ptr below
2703 */
2704 slot = path->slots[1];
2705 extent_buffer_get(leaf);
2706
2707 wret = push_leaf_left(trans, root, path, 1, 1);
2708 if (wret < 0 && wret != -ENOSPC)
2709 ret = wret;
2710
2711 if (path->nodes[0] == leaf &&
2712 btrfs_header_nritems(leaf)) {
2713 wret = push_leaf_right(trans, root, path, 1, 1);
2714 if (wret < 0 && wret != -ENOSPC)
2715 ret = wret;
2716 }
2717
2718 if (btrfs_header_nritems(leaf) == 0) {
2719 clean_tree_block(trans, root, leaf);
2720 wait_on_tree_block_writeback(root, leaf);
2721
2722 path->slots[1] = slot;
2723 ret = btrfs_del_leaf(trans, root, path, leaf);
2724 BUG_ON(ret);
2725 free_extent_buffer(leaf);
2726
2727 } else {
2728 btrfs_mark_buffer_dirty(leaf);
2729 free_extent_buffer(leaf);
2730 }
2731 } else {
2732 btrfs_mark_buffer_dirty(leaf);
2733 }
2734 }
2735 return ret;
2736 }
2737
2738 /*
2739 * walk up the tree as far as required to find the previous leaf.
2740 * returns 0 if it found something or 1 if there are no lesser leaves.
2741 * returns < 0 on io errors.
2742 */
2743 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2744 {
2745 int slot;
2746 int level = 1;
2747 struct extent_buffer *c;
2748 struct extent_buffer *next = NULL;
2749
2750 while(level < BTRFS_MAX_LEVEL) {
2751 if (!path->nodes[level])
2752 return 1;
2753
2754 slot = path->slots[level];
2755 c = path->nodes[level];
2756 if (slot == 0) {
2757 level++;
2758 if (level == BTRFS_MAX_LEVEL)
2759 return 1;
2760 continue;
2761 }
2762 slot--;
2763
2764 next = read_node_slot(root, c, slot);
2765 break;
2766 }
2767 path->slots[level] = slot;
2768 while(1) {
2769 level--;
2770 c = path->nodes[level];
2771 free_extent_buffer(c);
2772 slot = btrfs_header_nritems(next);
2773 if (slot != 0)
2774 slot--;
2775 path->nodes[level] = next;
2776 path->slots[level] = slot;
2777 if (!level)
2778 break;
2779 next = read_node_slot(root, next, slot);
2780 }
2781 return 0;
2782 }
2783
2784 /*
2785 * walk up the tree as far as required to find the next leaf.
2786 * returns 0 if it found something or 1 if there are no greater leaves.
2787 * returns < 0 on io errors.
2788 */
2789 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2790 {
2791 int slot;
2792 int level = 1;
2793 struct extent_buffer *c;
2794 struct extent_buffer *next = NULL;
2795
2796 while(level < BTRFS_MAX_LEVEL) {
2797 if (!path->nodes[level])
2798 return 1;
2799
2800 slot = path->slots[level] + 1;
2801 c = path->nodes[level];
2802 if (slot >= btrfs_header_nritems(c)) {
2803 level++;
2804 if (level == BTRFS_MAX_LEVEL)
2805 return 1;
2806 continue;
2807 }
2808
2809 if (path->reada)
2810 reada_for_search(root, path, level, slot, 0);
2811
2812 next = read_node_slot(root, c, slot);
2813 if (!next)
2814 return -EIO;
2815 break;
2816 }
2817 path->slots[level] = slot;
2818 while(1) {
2819 level--;
2820 c = path->nodes[level];
2821 free_extent_buffer(c);
2822 path->nodes[level] = next;
2823 path->slots[level] = 0;
2824 if (!level)
2825 break;
2826 if (path->reada)
2827 reada_for_search(root, path, level, 0, 0);
2828 next = read_node_slot(root, next, 0);
2829 if (!next)
2830 return -EIO;
2831 }
2832 return 0;
2833 }
2834
2835 int btrfs_previous_item(struct btrfs_root *root,
2836 struct btrfs_path *path, u64 min_objectid,
2837 int type)
2838 {
2839 struct btrfs_key found_key;
2840 struct extent_buffer *leaf;
2841 int ret;
2842
2843 while(1) {
2844 if (path->slots[0] == 0) {
2845 ret = btrfs_prev_leaf(root, path);
2846 if (ret != 0)
2847 return ret;
2848 } else {
2849 path->slots[0]--;
2850 }
2851 leaf = path->nodes[0];
2852 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2853 if (found_key.type == type)
2854 return 0;
2855 }
2856 return 1;
2857 }
2858
2859 /*
2860 * search in extent tree to find a previous Metadata/Data extent item with
2861 * min objecitd.
2862 *
2863 * returns 0 if something is found, 1 if nothing was found and < 0 on error
2864 */
2865 int btrfs_previous_extent_item(struct btrfs_root *root,
2866 struct btrfs_path *path, u64 min_objectid)
2867 {
2868 struct btrfs_key found_key;
2869 struct extent_buffer *leaf;
2870 u32 nritems;
2871 int ret;
2872
2873 while (1) {
2874 if (path->slots[0] == 0) {
2875 ret = btrfs_prev_leaf(root, path);
2876 if (ret != 0)
2877 return ret;
2878 } else {
2879 path->slots[0]--;
2880 }
2881 leaf = path->nodes[0];
2882 nritems = btrfs_header_nritems(leaf);
2883 if (nritems == 0)
2884 return 1;
2885 if (path->slots[0] == nritems)
2886 path->slots[0]--;
2887
2888 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2889 if (found_key.objectid < min_objectid)
2890 break;
2891 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
2892 found_key.type == BTRFS_METADATA_ITEM_KEY)
2893 return 0;
2894 if (found_key.objectid == min_objectid &&
2895 found_key.type < BTRFS_EXTENT_ITEM_KEY)
2896 break;
2897 }
2898 return 1;
2899 }
(deprecated) Btrfs progs developments and patch integration