?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: btrfstune: rework option handling
[btrfs-progs-unstable/devel.git] / ctree.c
blob130c61fa036e12d50a02e008a4382c78a00b66bc
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18 #include "ctree.h"
19 #include "disk-io.h"
20 #include "transaction.h"
21 #include "print-tree.h"
22 #include "repair.h"
23
24 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
25 *root, struct btrfs_path *path, int level);
26 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_key *ins_key,
28 struct btrfs_path *path, int data_size, int extend);
29 static int push_node_left(struct btrfs_trans_handle *trans,
30 struct btrfs_root *root, struct extent_buffer *dst,
31 struct extent_buffer *src, int empty);
32 static int balance_node_right(struct btrfs_trans_handle *trans,
33 struct btrfs_root *root,
34 struct extent_buffer *dst_buf,
35 struct extent_buffer *src_buf);
36
37 inline void btrfs_init_path(struct btrfs_path *p)
38 {
39 memset(p, 0, sizeof(*p));
40 }
41
42 struct btrfs_path *btrfs_alloc_path(void)
43 {
44 struct btrfs_path *path;
45 path = kzalloc(sizeof(struct btrfs_path), GFP_NOFS);
46 return path;
47 }
48
49 void btrfs_free_path(struct btrfs_path *p)
50 {
51 if (!p)
52 return;
53 btrfs_release_path(p);
54 kfree(p);
55 }
56
57 void btrfs_release_path(struct btrfs_path *p)
58 {
59 int i;
60 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
61 if (!p->nodes[i])
62 continue;
63 free_extent_buffer(p->nodes[i]);
64 }
65 memset(p, 0, sizeof(*p));
66 }
67
68 void add_root_to_dirty_list(struct btrfs_root *root)
69 {
70 if (root->track_dirty && list_empty(&root->dirty_list)) {
71 list_add(&root->dirty_list,
72 &root->fs_info->dirty_cowonly_roots);
73 }
74 }
75
76 int btrfs_copy_root(struct btrfs_trans_handle *trans,
77 struct btrfs_root *root,
78 struct extent_buffer *buf,
79 struct extent_buffer **cow_ret, u64 new_root_objectid)
80 {
81 struct extent_buffer *cow;
82 int ret = 0;
83 int level;
84 struct btrfs_root *new_root;
85 struct btrfs_disk_key disk_key;
86
87 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
88 if (!new_root)
89 return -ENOMEM;
90
91 memcpy(new_root, root, sizeof(*new_root));
92 new_root->root_key.objectid = new_root_objectid;
93
94 WARN_ON(root->ref_cows && trans->transid !=
95 root->fs_info->running_transaction->transid);
96 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
97
98 level = btrfs_header_level(buf);
99 if (level == 0)
100 btrfs_item_key(buf, &disk_key, 0);
101 else
102 btrfs_node_key(buf, &disk_key, 0);
103 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
104 new_root_objectid, &disk_key,
105 level, buf->start, 0);
106 if (IS_ERR(cow)) {
107 kfree(new_root);
108 return PTR_ERR(cow);
109 }
110
111 copy_extent_buffer(cow, buf, 0, 0, cow->len);
112 btrfs_set_header_bytenr(cow, cow->start);
113 btrfs_set_header_generation(cow, trans->transid);
114 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
115 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
116 BTRFS_HEADER_FLAG_RELOC);
117 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
118 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
119 else
120 btrfs_set_header_owner(cow, new_root_objectid);
121
122 write_extent_buffer(cow, root->fs_info->fsid,
123 btrfs_header_fsid(), BTRFS_FSID_SIZE);
124
125 WARN_ON(btrfs_header_generation(buf) > trans->transid);
126 ret = btrfs_inc_ref(trans, new_root, cow, 0);
127 kfree(new_root);
128
129 if (ret)
130 return ret;
131
132 btrfs_mark_buffer_dirty(cow);
133 *cow_ret = cow;
134 return 0;
135 }
136
137 /*
138 * check if the tree block can be shared by multiple trees
139 */
140 static int btrfs_block_can_be_shared(struct btrfs_root *root,
141 struct extent_buffer *buf)
142 {
143 /*
144 * Tree blocks not in refernece counted trees and tree roots
145 * are never shared. If a block was allocated after the last
146 * snapshot and the block was not allocated by tree relocation,
147 * we know the block is not shared.
148 */
149 if (root->ref_cows &&
150 buf != root->node && buf != root->commit_root &&
151 (btrfs_header_generation(buf) <=
152 btrfs_root_last_snapshot(&root->root_item) ||
153 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
154 return 1;
155 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
156 if (root->ref_cows &&
157 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
158 return 1;
159 #endif
160 return 0;
161 }
162
163 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
164 struct btrfs_root *root,
165 struct extent_buffer *buf,
166 struct extent_buffer *cow)
167 {
168 u64 refs;
169 u64 owner;
170 u64 flags;
171 u64 new_flags = 0;
172 int ret;
173
174 /*
175 * Backrefs update rules:
176 *
177 * Always use full backrefs for extent pointers in tree block
178 * allocated by tree relocation.
179 *
180 * If a shared tree block is no longer referenced by its owner
181 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
182 * use full backrefs for extent pointers in tree block.
183 *
184 * If a tree block is been relocating
185 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
186 * use full backrefs for extent pointers in tree block.
187 * The reason for this is some operations (such as drop tree)
188 * are only allowed for blocks use full backrefs.
189 */
190
191 if (btrfs_block_can_be_shared(root, buf)) {
192 ret = btrfs_lookup_extent_info(trans, root, buf->start,
193 btrfs_header_level(buf), 1,
194 &refs, &flags);
195 BUG_ON(ret);
196 BUG_ON(refs == 0);
197 } else {
198 refs = 1;
199 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
200 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
201 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
202 else
203 flags = 0;
204 }
205
206 owner = btrfs_header_owner(buf);
207 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
208 owner == BTRFS_TREE_RELOC_OBJECTID);
209
210 if (refs > 1) {
211 if ((owner == root->root_key.objectid ||
212 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
213 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
214 ret = btrfs_inc_ref(trans, root, buf, 1);
215 BUG_ON(ret);
216
217 if (root->root_key.objectid ==
218 BTRFS_TREE_RELOC_OBJECTID) {
219 ret = btrfs_dec_ref(trans, root, buf, 0);
220 BUG_ON(ret);
221 ret = btrfs_inc_ref(trans, root, cow, 1);
222 BUG_ON(ret);
223 }
224 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
225 } else {
226
227 if (root->root_key.objectid ==
228 BTRFS_TREE_RELOC_OBJECTID)
229 ret = btrfs_inc_ref(trans, root, cow, 1);
230 else
231 ret = btrfs_inc_ref(trans, root, cow, 0);
232 BUG_ON(ret);
233 }
234 if (new_flags != 0) {
235 ret = btrfs_set_block_flags(trans, root, buf->start,
236 btrfs_header_level(buf),
237 new_flags);
238 BUG_ON(ret);
239 }
240 } else {
241 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
242 if (root->root_key.objectid ==
243 BTRFS_TREE_RELOC_OBJECTID)
244 ret = btrfs_inc_ref(trans, root, cow, 1);
245 else
246 ret = btrfs_inc_ref(trans, root, cow, 0);
247 BUG_ON(ret);
248 ret = btrfs_dec_ref(trans, root, buf, 1);
249 BUG_ON(ret);
250 }
251 clean_tree_block(trans, root, buf);
252 }
253 return 0;
254 }
255
256 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
257 struct btrfs_root *root,
258 struct extent_buffer *buf,
259 struct extent_buffer *parent, int parent_slot,
260 struct extent_buffer **cow_ret,
261 u64 search_start, u64 empty_size)
262 {
263 struct extent_buffer *cow;
264 struct btrfs_disk_key disk_key;
265 int level;
266
267 WARN_ON(root->ref_cows && trans->transid !=
268 root->fs_info->running_transaction->transid);
269 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
270
271 level = btrfs_header_level(buf);
272
273 if (level == 0)
274 btrfs_item_key(buf, &disk_key, 0);
275 else
276 btrfs_node_key(buf, &disk_key, 0);
277
278 cow = btrfs_alloc_free_block(trans, root, buf->len,
279 root->root_key.objectid, &disk_key,
280 level, search_start, empty_size);
281 if (IS_ERR(cow))
282 return PTR_ERR(cow);
283
284 copy_extent_buffer(cow, buf, 0, 0, cow->len);
285 btrfs_set_header_bytenr(cow, cow->start);
286 btrfs_set_header_generation(cow, trans->transid);
287 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
288 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
289 BTRFS_HEADER_FLAG_RELOC);
290 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
291 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
292 else
293 btrfs_set_header_owner(cow, root->root_key.objectid);
294
295 write_extent_buffer(cow, root->fs_info->fsid,
296 btrfs_header_fsid(), BTRFS_FSID_SIZE);
297
298 WARN_ON(!(buf->flags & EXTENT_BAD_TRANSID) &&
299 btrfs_header_generation(buf) > trans->transid);
300
301 update_ref_for_cow(trans, root, buf, cow);
302
303 if (buf == root->node) {
304 root->node = cow;
305 extent_buffer_get(cow);
306
307 btrfs_free_extent(trans, root, buf->start, buf->len,
308 0, root->root_key.objectid, level, 0);
309 free_extent_buffer(buf);
310 add_root_to_dirty_list(root);
311 } else {
312 btrfs_set_node_blockptr(parent, parent_slot,
313 cow->start);
314 WARN_ON(trans->transid == 0);
315 btrfs_set_node_ptr_generation(parent, parent_slot,
316 trans->transid);
317 btrfs_mark_buffer_dirty(parent);
318 WARN_ON(btrfs_header_generation(parent) != trans->transid);
319
320 btrfs_free_extent(trans, root, buf->start, buf->len,
321 0, root->root_key.objectid, level, 1);
322 }
323 if (!list_empty(&buf->recow)) {
324 list_del_init(&buf->recow);
325 free_extent_buffer(buf);
326 }
327 free_extent_buffer(buf);
328 btrfs_mark_buffer_dirty(cow);
329 *cow_ret = cow;
330 return 0;
331 }
332
333 static inline int should_cow_block(struct btrfs_trans_handle *trans,
334 struct btrfs_root *root,
335 struct extent_buffer *buf)
336 {
337 if (btrfs_header_generation(buf) == trans->transid &&
338 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
339 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
340 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
341 return 0;
342 return 1;
343 }
344
345 int btrfs_cow_block(struct btrfs_trans_handle *trans,
346 struct btrfs_root *root, struct extent_buffer *buf,
347 struct extent_buffer *parent, int parent_slot,
348 struct extent_buffer **cow_ret)
349 {
350 u64 search_start;
351 int ret;
352 /*
353 if (trans->transaction != root->fs_info->running_transaction) {
354 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
355 root->fs_info->running_transaction->transid);
356 WARN_ON(1);
357 }
358 */
359 if (trans->transid != root->fs_info->generation) {
360 printk(KERN_CRIT "trans %llu running %llu\n",
361 (unsigned long long)trans->transid,
362 (unsigned long long)root->fs_info->generation);
363 WARN_ON(1);
364 }
365 if (!should_cow_block(trans, root, buf)) {
366 *cow_ret = buf;
367 return 0;
368 }
369
370 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
371 ret = __btrfs_cow_block(trans, root, buf, parent,
372 parent_slot, cow_ret, search_start, 0);
373 return ret;
374 }
375
376 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
377 {
378 if (k1->objectid > k2->objectid)
379 return 1;
380 if (k1->objectid < k2->objectid)
381 return -1;
382 if (k1->type > k2->type)
383 return 1;
384 if (k1->type < k2->type)
385 return -1;
386 if (k1->offset > k2->offset)
387 return 1;
388 if (k1->offset < k2->offset)
389 return -1;
390 return 0;
391 }
392
393 /*
394 * compare two keys in a memcmp fashion
395 */
396 static int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
397 {
398 struct btrfs_key k1;
399
400 btrfs_disk_key_to_cpu(&k1, disk);
401 return btrfs_comp_cpu_keys(&k1, k2);
402 }
403
404 /*
405 * The leaf data grows from end-to-front in the node.
406 * this returns the address of the start of the last item,
407 * which is the stop of the leaf data stack
408 */
409 static inline unsigned int leaf_data_end(struct btrfs_root *root,
410 struct extent_buffer *leaf)
411 {
412 u32 nr = btrfs_header_nritems(leaf);
413 if (nr == 0)
414 return BTRFS_LEAF_DATA_SIZE(root);
415 return btrfs_item_offset_nr(leaf, nr - 1);
416 }
417
418 enum btrfs_tree_block_status
419 btrfs_check_node(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
420 struct extent_buffer *buf)
421 {
422 int i;
423 struct btrfs_key cpukey;
424 struct btrfs_disk_key key;
425 u32 nritems = btrfs_header_nritems(buf);
426 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
427
428 if (nritems == 0 || nritems > BTRFS_NODEPTRS_PER_BLOCK(root))
429 goto fail;
430
431 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
432 if (parent_key && parent_key->type) {
433 btrfs_node_key(buf, &key, 0);
434 if (memcmp(parent_key, &key, sizeof(key)))
435 goto fail;
436 }
437 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
438 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
439 btrfs_node_key(buf, &key, i);
440 btrfs_node_key_to_cpu(buf, &cpukey, i + 1);
441 if (btrfs_comp_keys(&key, &cpukey) >= 0)
442 goto fail;
443 }
444 return BTRFS_TREE_BLOCK_CLEAN;
445 fail:
446 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
447 if (parent_key)
448 btrfs_disk_key_to_cpu(&cpukey, parent_key);
449 else
450 btrfs_node_key_to_cpu(buf, &cpukey, 0);
451 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
452 buf->start, buf->len,
453 btrfs_header_level(buf));
454 }
455 return ret;
456 }
457
458 enum btrfs_tree_block_status
459 btrfs_check_leaf(struct btrfs_root *root, struct btrfs_disk_key *parent_key,
460 struct extent_buffer *buf)
461 {
462 int i;
463 struct btrfs_key cpukey;
464 struct btrfs_disk_key key;
465 u32 nritems = btrfs_header_nritems(buf);
466 enum btrfs_tree_block_status ret = BTRFS_TREE_BLOCK_INVALID_NRITEMS;
467
468 if (nritems * sizeof(struct btrfs_item) > buf->len) {
469 fprintf(stderr, "invalid number of items %llu\n",
470 (unsigned long long)buf->start);
471 goto fail;
472 }
473
474 if (btrfs_header_level(buf) != 0) {
475 ret = BTRFS_TREE_BLOCK_INVALID_LEVEL;
476 fprintf(stderr, "leaf is not a leaf %llu\n",
477 (unsigned long long)btrfs_header_bytenr(buf));
478 goto fail;
479 }
480 if (btrfs_leaf_free_space(root, buf) < 0) {
481 ret = BTRFS_TREE_BLOCK_INVALID_FREE_SPACE;
482 fprintf(stderr, "leaf free space incorrect %llu %d\n",
483 (unsigned long long)btrfs_header_bytenr(buf),
484 btrfs_leaf_free_space(root, buf));
485 goto fail;
486 }
487
488 if (nritems == 0)
489 return BTRFS_TREE_BLOCK_CLEAN;
490
491 btrfs_item_key(buf, &key, 0);
492 if (parent_key && parent_key->type &&
493 memcmp(parent_key, &key, sizeof(key))) {
494 ret = BTRFS_TREE_BLOCK_INVALID_PARENT_KEY;
495 fprintf(stderr, "leaf parent key incorrect %llu\n",
496 (unsigned long long)btrfs_header_bytenr(buf));
497 goto fail;
498 }
499 for (i = 0; nritems > 1 && i < nritems - 1; i++) {
500 btrfs_item_key(buf, &key, i);
501 btrfs_item_key_to_cpu(buf, &cpukey, i + 1);
502 if (btrfs_comp_keys(&key, &cpukey) >= 0) {
503 ret = BTRFS_TREE_BLOCK_BAD_KEY_ORDER;
504 fprintf(stderr, "bad key ordering %d %d\n", i, i+1);
505 goto fail;
506 }
507 if (btrfs_item_offset_nr(buf, i) !=
508 btrfs_item_end_nr(buf, i + 1)) {
509 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
510 fprintf(stderr, "incorrect offsets %u %u\n",
511 btrfs_item_offset_nr(buf, i),
512 btrfs_item_end_nr(buf, i + 1));
513 goto fail;
514 }
515 if (i == 0 && btrfs_item_end_nr(buf, i) !=
516 BTRFS_LEAF_DATA_SIZE(root)) {
517 ret = BTRFS_TREE_BLOCK_INVALID_OFFSETS;
518 fprintf(stderr, "bad item end %u wanted %u\n",
519 btrfs_item_end_nr(buf, i),
520 (unsigned)BTRFS_LEAF_DATA_SIZE(root));
521 goto fail;
522 }
523 }
524 return BTRFS_TREE_BLOCK_CLEAN;
525 fail:
526 if (btrfs_header_owner(buf) == BTRFS_EXTENT_TREE_OBJECTID) {
527 if (parent_key)
528 btrfs_disk_key_to_cpu(&cpukey, parent_key);
529 else
530 btrfs_item_key_to_cpu(buf, &cpukey, 0);
531
532 btrfs_add_corrupt_extent_record(root->fs_info, &cpukey,
533 buf->start, buf->len, 0);
534 }
535 return ret;
536 }
537
538 static int noinline check_block(struct btrfs_root *root,
539 struct btrfs_path *path, int level)
540 {
541 struct btrfs_disk_key key;
542 struct btrfs_disk_key *key_ptr = NULL;
543 struct extent_buffer *parent;
544 enum btrfs_tree_block_status ret;
545
546 if (path->skip_check_block)
547 return 0;
548 if (path->nodes[level + 1]) {
549 parent = path->nodes[level + 1];
550 btrfs_node_key(parent, &key, path->slots[level + 1]);
551 key_ptr = &key;
552 }
553 if (level == 0)
554 ret = btrfs_check_leaf(root, key_ptr, path->nodes[0]);
555 else
556 ret = btrfs_check_node(root, key_ptr, path->nodes[level]);
557 if (ret == BTRFS_TREE_BLOCK_CLEAN)
558 return 0;
559 return -EIO;
560 }
561
562 /*
563 * search for key in the extent_buffer. The items start at offset p,
564 * and they are item_size apart. There are 'max' items in p.
565 *
566 * the slot in the array is returned via slot, and it points to
567 * the place where you would insert key if it is not found in
568 * the array.
569 *
570 * slot may point to max if the key is bigger than all of the keys
571 */
572 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
573 int item_size, struct btrfs_key *key,
574 int max, int *slot)
575 {
576 int low = 0;
577 int high = max;
578 int mid;
579 int ret;
580 unsigned long offset;
581 struct btrfs_disk_key *tmp;
582
583 while(low < high) {
584 mid = (low + high) / 2;
585 offset = p + mid * item_size;
586
587 tmp = (struct btrfs_disk_key *)(eb->data + offset);
588 ret = btrfs_comp_keys(tmp, key);
589
590 if (ret < 0)
591 low = mid + 1;
592 else if (ret > 0)
593 high = mid;
594 else {
595 *slot = mid;
596 return 0;
597 }
598 }
599 *slot = low;
600 return 1;
601 }
602
603 /*
604 * simple bin_search frontend that does the right thing for
605 * leaves vs nodes
606 */
607 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
608 int level, int *slot)
609 {
610 if (level == 0)
611 return generic_bin_search(eb,
612 offsetof(struct btrfs_leaf, items),
613 sizeof(struct btrfs_item),
614 key, btrfs_header_nritems(eb),
615 slot);
616 else
617 return generic_bin_search(eb,
618 offsetof(struct btrfs_node, ptrs),
619 sizeof(struct btrfs_key_ptr),
620 key, btrfs_header_nritems(eb),
621 slot);
622 }
623
624 struct extent_buffer *read_node_slot(struct btrfs_root *root,
625 struct extent_buffer *parent, int slot)
626 {
627 int level = btrfs_header_level(parent);
628 if (slot < 0)
629 return NULL;
630 if (slot >= btrfs_header_nritems(parent))
631 return NULL;
632
633 if (level == 0)
634 return NULL;
635
636 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
637 btrfs_level_size(root, level - 1),
638 btrfs_node_ptr_generation(parent, slot));
639 }
640
641 static int balance_level(struct btrfs_trans_handle *trans,
642 struct btrfs_root *root,
643 struct btrfs_path *path, int level)
644 {
645 struct extent_buffer *right = NULL;
646 struct extent_buffer *mid;
647 struct extent_buffer *left = NULL;
648 struct extent_buffer *parent = NULL;
649 int ret = 0;
650 int wret;
651 int pslot;
652 int orig_slot = path->slots[level];
653 u64 orig_ptr;
654
655 if (level == 0)
656 return 0;
657
658 mid = path->nodes[level];
659 WARN_ON(btrfs_header_generation(mid) != trans->transid);
660
661 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
662
663 if (level < BTRFS_MAX_LEVEL - 1) {
664 parent = path->nodes[level + 1];
665 pslot = path->slots[level + 1];
666 }
667
668 /*
669 * deal with the case where there is only one pointer in the root
670 * by promoting the node below to a root
671 */
672 if (!parent) {
673 struct extent_buffer *child;
674
675 if (btrfs_header_nritems(mid) != 1)
676 return 0;
677
678 /* promote the child to a root */
679 child = read_node_slot(root, mid, 0);
680 BUG_ON(!extent_buffer_uptodate(child));
681 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
682 BUG_ON(ret);
683
684 root->node = child;
685 add_root_to_dirty_list(root);
686 path->nodes[level] = NULL;
687 clean_tree_block(trans, root, mid);
688 wait_on_tree_block_writeback(root, mid);
689 /* once for the path */
690 free_extent_buffer(mid);
691
692 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
693 0, root->root_key.objectid,
694 level, 1);
695 /* once for the root ptr */
696 free_extent_buffer(mid);
697 return ret;
698 }
699 if (btrfs_header_nritems(mid) >
700 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
701 return 0;
702
703 left = read_node_slot(root, parent, pslot - 1);
704 if (extent_buffer_uptodate(left)) {
705 wret = btrfs_cow_block(trans, root, left,
706 parent, pslot - 1, &left);
707 if (wret) {
708 ret = wret;
709 goto enospc;
710 }
711 }
712 right = read_node_slot(root, parent, pslot + 1);
713 if (extent_buffer_uptodate(right)) {
714 wret = btrfs_cow_block(trans, root, right,
715 parent, pslot + 1, &right);
716 if (wret) {
717 ret = wret;
718 goto enospc;
719 }
720 }
721
722 /* first, try to make some room in the middle buffer */
723 if (left) {
724 orig_slot += btrfs_header_nritems(left);
725 wret = push_node_left(trans, root, left, mid, 1);
726 if (wret < 0)
727 ret = wret;
728 }
729
730 /*
731 * then try to empty the right most buffer into the middle
732 */
733 if (right) {
734 wret = push_node_left(trans, root, mid, right, 1);
735 if (wret < 0 && wret != -ENOSPC)
736 ret = wret;
737 if (btrfs_header_nritems(right) == 0) {
738 u64 bytenr = right->start;
739 u32 blocksize = right->len;
740
741 clean_tree_block(trans, root, right);
742 wait_on_tree_block_writeback(root, right);
743 free_extent_buffer(right);
744 right = NULL;
745 wret = btrfs_del_ptr(trans, root, path,
746 level + 1, pslot + 1);
747 if (wret)
748 ret = wret;
749 wret = btrfs_free_extent(trans, root, bytenr,
750 blocksize, 0,
751 root->root_key.objectid,
752 level, 0);
753 if (wret)
754 ret = wret;
755 } else {
756 struct btrfs_disk_key right_key;
757 btrfs_node_key(right, &right_key, 0);
758 btrfs_set_node_key(parent, &right_key, pslot + 1);
759 btrfs_mark_buffer_dirty(parent);
760 }
761 }
762 if (btrfs_header_nritems(mid) == 1) {
763 /*
764 * we're not allowed to leave a node with one item in the
765 * tree during a delete. A deletion from lower in the tree
766 * could try to delete the only pointer in this node.
767 * So, pull some keys from the left.
768 * There has to be a left pointer at this point because
769 * otherwise we would have pulled some pointers from the
770 * right
771 */
772 BUG_ON(!left);
773 wret = balance_node_right(trans, root, mid, left);
774 if (wret < 0) {
775 ret = wret;
776 goto enospc;
777 }
778 if (wret == 1) {
779 wret = push_node_left(trans, root, left, mid, 1);
780 if (wret < 0)
781 ret = wret;
782 }
783 BUG_ON(wret == 1);
784 }
785 if (btrfs_header_nritems(mid) == 0) {
786 /* we've managed to empty the middle node, drop it */
787 u64 bytenr = mid->start;
788 u32 blocksize = mid->len;
789 clean_tree_block(trans, root, mid);
790 wait_on_tree_block_writeback(root, mid);
791 free_extent_buffer(mid);
792 mid = NULL;
793 wret = btrfs_del_ptr(trans, root, path, level + 1, pslot);
794 if (wret)
795 ret = wret;
796 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
797 0, root->root_key.objectid,
798 level, 0);
799 if (wret)
800 ret = wret;
801 } else {
802 /* update the parent key to reflect our changes */
803 struct btrfs_disk_key mid_key;
804 btrfs_node_key(mid, &mid_key, 0);
805 btrfs_set_node_key(parent, &mid_key, pslot);
806 btrfs_mark_buffer_dirty(parent);
807 }
808
809 /* update the path */
810 if (left) {
811 if (btrfs_header_nritems(left) > orig_slot) {
812 extent_buffer_get(left);
813 path->nodes[level] = left;
814 path->slots[level + 1] -= 1;
815 path->slots[level] = orig_slot;
816 if (mid)
817 free_extent_buffer(mid);
818 } else {
819 orig_slot -= btrfs_header_nritems(left);
820 path->slots[level] = orig_slot;
821 }
822 }
823 /* double check we haven't messed things up */
824 check_block(root, path, level);
825 if (orig_ptr !=
826 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
827 BUG();
828 enospc:
829 if (right)
830 free_extent_buffer(right);
831 if (left)
832 free_extent_buffer(left);
833 return ret;
834 }
835
836 /* returns zero if the push worked, non-zero otherwise */
837 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
838 struct btrfs_root *root,
839 struct btrfs_path *path, int level)
840 {
841 struct extent_buffer *right = NULL;
842 struct extent_buffer *mid;
843 struct extent_buffer *left = NULL;
844 struct extent_buffer *parent = NULL;
845 int ret = 0;
846 int wret;
847 int pslot;
848 int orig_slot = path->slots[level];
849
850 if (level == 0)
851 return 1;
852
853 mid = path->nodes[level];
854 WARN_ON(btrfs_header_generation(mid) != trans->transid);
855
856 if (level < BTRFS_MAX_LEVEL - 1) {
857 parent = path->nodes[level + 1];
858 pslot = path->slots[level + 1];
859 }
860
861 if (!parent)
862 return 1;
863
864 left = read_node_slot(root, parent, pslot - 1);
865
866 /* first, try to make some room in the middle buffer */
867 if (extent_buffer_uptodate(left)) {
868 u32 left_nr;
869 left_nr = btrfs_header_nritems(left);
870 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
871 wret = 1;
872 } else {
873 ret = btrfs_cow_block(trans, root, left, parent,
874 pslot - 1, &left);
875 if (ret)
876 wret = 1;
877 else {
878 wret = push_node_left(trans, root,
879 left, mid, 0);
880 }
881 }
882 if (wret < 0)
883 ret = wret;
884 if (wret == 0) {
885 struct btrfs_disk_key disk_key;
886 orig_slot += left_nr;
887 btrfs_node_key(mid, &disk_key, 0);
888 btrfs_set_node_key(parent, &disk_key, pslot);
889 btrfs_mark_buffer_dirty(parent);
890 if (btrfs_header_nritems(left) > orig_slot) {
891 path->nodes[level] = left;
892 path->slots[level + 1] -= 1;
893 path->slots[level] = orig_slot;
894 free_extent_buffer(mid);
895 } else {
896 orig_slot -=
897 btrfs_header_nritems(left);
898 path->slots[level] = orig_slot;
899 free_extent_buffer(left);
900 }
901 return 0;
902 }
903 free_extent_buffer(left);
904 }
905 right= read_node_slot(root, parent, pslot + 1);
906
907 /*
908 * then try to empty the right most buffer into the middle
909 */
910 if (extent_buffer_uptodate(right)) {
911 u32 right_nr;
912 right_nr = btrfs_header_nritems(right);
913 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
914 wret = 1;
915 } else {
916 ret = btrfs_cow_block(trans, root, right,
917 parent, pslot + 1,
918 &right);
919 if (ret)
920 wret = 1;
921 else {
922 wret = balance_node_right(trans, root,
923 right, mid);
924 }
925 }
926 if (wret < 0)
927 ret = wret;
928 if (wret == 0) {
929 struct btrfs_disk_key disk_key;
930
931 btrfs_node_key(right, &disk_key, 0);
932 btrfs_set_node_key(parent, &disk_key, pslot + 1);
933 btrfs_mark_buffer_dirty(parent);
934
935 if (btrfs_header_nritems(mid) <= orig_slot) {
936 path->nodes[level] = right;
937 path->slots[level + 1] += 1;
938 path->slots[level] = orig_slot -
939 btrfs_header_nritems(mid);
940 free_extent_buffer(mid);
941 } else {
942 free_extent_buffer(right);
943 }
944 return 0;
945 }
946 free_extent_buffer(right);
947 }
948 return 1;
949 }
950
951 /*
952 * readahead one full node of leaves
953 */
954 void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
955 int level, int slot, u64 objectid)
956 {
957 struct extent_buffer *node;
958 struct btrfs_disk_key disk_key;
959 u32 nritems;
960 u64 search;
961 u64 lowest_read;
962 u64 highest_read;
963 u64 nread = 0;
964 int direction = path->reada;
965 struct extent_buffer *eb;
966 u32 nr;
967 u32 blocksize;
968 u32 nscan = 0;
969
970 if (level != 1)
971 return;
972
973 if (!path->nodes[level])
974 return;
975
976 node = path->nodes[level];
977 search = btrfs_node_blockptr(node, slot);
978 blocksize = btrfs_level_size(root, level - 1);
979 eb = btrfs_find_tree_block(root, search, blocksize);
980 if (eb) {
981 free_extent_buffer(eb);
982 return;
983 }
984
985 highest_read = search;
986 lowest_read = search;
987
988 nritems = btrfs_header_nritems(node);
989 nr = slot;
990 while(1) {
991 if (direction < 0) {
992 if (nr == 0)
993 break;
994 nr--;
995 } else if (direction > 0) {
996 nr++;
997 if (nr >= nritems)
998 break;
999 }
1000 if (path->reada < 0 && objectid) {
1001 btrfs_node_key(node, &disk_key, nr);
1002 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1003 break;
1004 }
1005 search = btrfs_node_blockptr(node, nr);
1006 if ((search >= lowest_read && search <= highest_read) ||
1007 (search < lowest_read && lowest_read - search <= 32768) ||
1008 (search > highest_read && search - highest_read <= 32768)) {
1009 readahead_tree_block(root, search, blocksize,
1010 btrfs_node_ptr_generation(node, nr));
1011 nread += blocksize;
1012 }
1013 nscan++;
1014 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1015 break;
1016 if(nread > (1024 * 1024) || nscan > 128)
1017 break;
1018
1019 if (search < lowest_read)
1020 lowest_read = search;
1021 if (search > highest_read)
1022 highest_read = search;
1023 }
1024 }
1025
1026 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
1027 u64 iobjectid, u64 ioff, u8 key_type,
1028 struct btrfs_key *found_key)
1029 {
1030 int ret;
1031 struct btrfs_key key;
1032 struct extent_buffer *eb;
1033 struct btrfs_path *path;
1034
1035 key.type = key_type;
1036 key.objectid = iobjectid;
1037 key.offset = ioff;
1038
1039 if (found_path == NULL) {
1040 path = btrfs_alloc_path();
1041 if (!path)
1042 return -ENOMEM;
1043 } else
1044 path = found_path;
1045
1046 ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1047 if ((ret < 0) || (found_key == NULL)) {
1048 if (path != found_path)
1049 btrfs_free_path(path);
1050 return ret;
1051 }
1052
1053 eb = path->nodes[0];
1054 if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1055 ret = btrfs_next_leaf(fs_root, path);
1056 if (ret)
1057 return ret;
1058 eb = path->nodes[0];
1059 }
1060
1061 btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1062 if (found_key->type != key.type ||
1063 found_key->objectid != key.objectid)
1064 return 1;
1065
1066 return 0;
1067 }
1068
1069 /*
1070 * look for key in the tree. path is filled in with nodes along the way
1071 * if key is found, we return zero and you can find the item in the leaf
1072 * level of the path (level 0)
1073 *
1074 * If the key isn't found, the path points to the slot where it should
1075 * be inserted, and 1 is returned. If there are other errors during the
1076 * search a negative error number is returned.
1077 *
1078 * if ins_len > 0, nodes and leaves will be split as we walk down the
1079 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1080 * possible)
1081 */
1082 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1083 *root, struct btrfs_key *key, struct btrfs_path *p, int
1084 ins_len, int cow)
1085 {
1086 struct extent_buffer *b;
1087 int slot;
1088 int ret;
1089 int level;
1090 int should_reada = p->reada;
1091 u8 lowest_level = 0;
1092
1093 lowest_level = p->lowest_level;
1094 WARN_ON(lowest_level && ins_len > 0);
1095 WARN_ON(p->nodes[0] != NULL);
1096 /*
1097 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1098 */
1099 again:
1100 b = root->node;
1101 extent_buffer_get(b);
1102 while (b) {
1103 level = btrfs_header_level(b);
1104 if (cow) {
1105 int wret;
1106 wret = btrfs_cow_block(trans, root, b,
1107 p->nodes[level + 1],
1108 p->slots[level + 1],
1109 &b);
1110 if (wret) {
1111 free_extent_buffer(b);
1112 return wret;
1113 }
1114 }
1115 BUG_ON(!cow && ins_len);
1116 if (level != btrfs_header_level(b))
1117 WARN_ON(1);
1118 level = btrfs_header_level(b);
1119 p->nodes[level] = b;
1120 ret = check_block(root, p, level);
1121 if (ret)
1122 return -1;
1123 ret = bin_search(b, key, level, &slot);
1124 if (level != 0) {
1125 if (ret && slot > 0)
1126 slot -= 1;
1127 p->slots[level] = slot;
1128 if ((p->search_for_split || ins_len > 0) &&
1129 btrfs_header_nritems(b) >=
1130 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1131 int sret = split_node(trans, root, p, level);
1132 BUG_ON(sret > 0);
1133 if (sret)
1134 return sret;
1135 b = p->nodes[level];
1136 slot = p->slots[level];
1137 } else if (ins_len < 0) {
1138 int sret = balance_level(trans, root, p,
1139 level);
1140 if (sret)
1141 return sret;
1142 b = p->nodes[level];
1143 if (!b) {
1144 btrfs_release_path(p);
1145 goto again;
1146 }
1147 slot = p->slots[level];
1148 BUG_ON(btrfs_header_nritems(b) == 1);
1149 }
1150 /* this is only true while dropping a snapshot */
1151 if (level == lowest_level)
1152 break;
1153
1154 if (should_reada)
1155 reada_for_search(root, p, level, slot,
1156 key->objectid);
1157
1158 b = read_node_slot(root, b, slot);
1159 if (!extent_buffer_uptodate(b))
1160 return -EIO;
1161 } else {
1162 p->slots[level] = slot;
1163 if (ins_len > 0 &&
1164 ins_len > btrfs_leaf_free_space(root, b)) {
1165 int sret = split_leaf(trans, root, key,
1166 p, ins_len, ret == 0);
1167 BUG_ON(sret > 0);
1168 if (sret)
1169 return sret;
1170 }
1171 return ret;
1172 }
1173 }
1174 return 1;
1175 }
1176
1177 /*
1178 * adjust the pointers going up the tree, starting at level
1179 * making sure the right key of each node is points to 'key'.
1180 * This is used after shifting pointers to the left, so it stops
1181 * fixing up pointers when a given leaf/node is not in slot 0 of the
1182 * higher levels
1183 */
1184 void btrfs_fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
1185 struct btrfs_disk_key *key, int level)
1186 {
1187 int i;
1188 struct extent_buffer *t;
1189
1190 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1191 int tslot = path->slots[i];
1192 if (!path->nodes[i])
1193 break;
1194 t = path->nodes[i];
1195 btrfs_set_node_key(t, key, tslot);
1196 btrfs_mark_buffer_dirty(path->nodes[i]);
1197 if (tslot != 0)
1198 break;
1199 }
1200 }
1201
1202 /*
1203 * update item key.
1204 *
1205 * This function isn't completely safe. It's the caller's responsibility
1206 * that the new key won't break the order
1207 */
1208 int btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
1209 struct btrfs_key *new_key)
1210 {
1211 struct btrfs_disk_key disk_key;
1212 struct extent_buffer *eb;
1213 int slot;
1214
1215 eb = path->nodes[0];
1216 slot = path->slots[0];
1217 if (slot > 0) {
1218 btrfs_item_key(eb, &disk_key, slot - 1);
1219 if (btrfs_comp_keys(&disk_key, new_key) >= 0)
1220 return -1;
1221 }
1222 if (slot < btrfs_header_nritems(eb) - 1) {
1223 btrfs_item_key(eb, &disk_key, slot + 1);
1224 if (btrfs_comp_keys(&disk_key, new_key) <= 0)
1225 return -1;
1226 }
1227
1228 btrfs_cpu_key_to_disk(&disk_key, new_key);
1229 btrfs_set_item_key(eb, &disk_key, slot);
1230 btrfs_mark_buffer_dirty(eb);
1231 if (slot == 0)
1232 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1233 return 0;
1234 }
1235
1236 /*
1237 * update an item key without the safety checks. This is meant to be called by
1238 * fsck only.
1239 */
1240 void btrfs_set_item_key_unsafe(struct btrfs_root *root,
1241 struct btrfs_path *path,
1242 struct btrfs_key *new_key)
1243 {
1244 struct btrfs_disk_key disk_key;
1245 struct extent_buffer *eb;
1246 int slot;
1247
1248 eb = path->nodes[0];
1249 slot = path->slots[0];
1250
1251 btrfs_cpu_key_to_disk(&disk_key, new_key);
1252 btrfs_set_item_key(eb, &disk_key, slot);
1253 btrfs_mark_buffer_dirty(eb);
1254 if (slot == 0)
1255 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1256 }
1257
1258 /*
1259 * try to push data from one node into the next node left in the
1260 * tree.
1261 *
1262 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1263 * error, and > 0 if there was no room in the left hand block.
1264 */
1265 static int push_node_left(struct btrfs_trans_handle *trans,
1266 struct btrfs_root *root, struct extent_buffer *dst,
1267 struct extent_buffer *src, int empty)
1268 {
1269 int push_items = 0;
1270 int src_nritems;
1271 int dst_nritems;
1272 int ret = 0;
1273
1274 src_nritems = btrfs_header_nritems(src);
1275 dst_nritems = btrfs_header_nritems(dst);
1276 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1277 WARN_ON(btrfs_header_generation(src) != trans->transid);
1278 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1279
1280 if (!empty && src_nritems <= 8)
1281 return 1;
1282
1283 if (push_items <= 0) {
1284 return 1;
1285 }
1286
1287 if (empty) {
1288 push_items = min(src_nritems, push_items);
1289 if (push_items < src_nritems) {
1290 /* leave at least 8 pointers in the node if
1291 * we aren't going to empty it
1292 */
1293 if (src_nritems - push_items < 8) {
1294 if (push_items <= 8)
1295 return 1;
1296 push_items -= 8;
1297 }
1298 }
1299 } else
1300 push_items = min(src_nritems - 8, push_items);
1301
1302 copy_extent_buffer(dst, src,
1303 btrfs_node_key_ptr_offset(dst_nritems),
1304 btrfs_node_key_ptr_offset(0),
1305 push_items * sizeof(struct btrfs_key_ptr));
1306
1307 if (push_items < src_nritems) {
1308 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1309 btrfs_node_key_ptr_offset(push_items),
1310 (src_nritems - push_items) *
1311 sizeof(struct btrfs_key_ptr));
1312 }
1313 btrfs_set_header_nritems(src, src_nritems - push_items);
1314 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1315 btrfs_mark_buffer_dirty(src);
1316 btrfs_mark_buffer_dirty(dst);
1317
1318 return ret;
1319 }
1320
1321 /*
1322 * try to push data from one node into the next node right in the
1323 * tree.
1324 *
1325 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1326 * error, and > 0 if there was no room in the right hand block.
1327 *
1328 * this will only push up to 1/2 the contents of the left node over
1329 */
1330 static int balance_node_right(struct btrfs_trans_handle *trans,
1331 struct btrfs_root *root,
1332 struct extent_buffer *dst,
1333 struct extent_buffer *src)
1334 {
1335 int push_items = 0;
1336 int max_push;
1337 int src_nritems;
1338 int dst_nritems;
1339 int ret = 0;
1340
1341 WARN_ON(btrfs_header_generation(src) != trans->transid);
1342 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1343
1344 src_nritems = btrfs_header_nritems(src);
1345 dst_nritems = btrfs_header_nritems(dst);
1346 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1347 if (push_items <= 0) {
1348 return 1;
1349 }
1350
1351 if (src_nritems < 4) {
1352 return 1;
1353 }
1354
1355 max_push = src_nritems / 2 + 1;
1356 /* don't try to empty the node */
1357 if (max_push >= src_nritems) {
1358 return 1;
1359 }
1360
1361 if (max_push < push_items)
1362 push_items = max_push;
1363
1364 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1365 btrfs_node_key_ptr_offset(0),
1366 (dst_nritems) *
1367 sizeof(struct btrfs_key_ptr));
1368
1369 copy_extent_buffer(dst, src,
1370 btrfs_node_key_ptr_offset(0),
1371 btrfs_node_key_ptr_offset(src_nritems - push_items),
1372 push_items * sizeof(struct btrfs_key_ptr));
1373
1374 btrfs_set_header_nritems(src, src_nritems - push_items);
1375 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1376
1377 btrfs_mark_buffer_dirty(src);
1378 btrfs_mark_buffer_dirty(dst);
1379
1380 return ret;
1381 }
1382
1383 /*
1384 * helper function to insert a new root level in the tree.
1385 * A new node is allocated, and a single item is inserted to
1386 * point to the existing root
1387 *
1388 * returns zero on success or < 0 on failure.
1389 */
1390 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1391 struct btrfs_root *root,
1392 struct btrfs_path *path, int level)
1393 {
1394 u64 lower_gen;
1395 struct extent_buffer *lower;
1396 struct extent_buffer *c;
1397 struct extent_buffer *old;
1398 struct btrfs_disk_key lower_key;
1399
1400 BUG_ON(path->nodes[level]);
1401 BUG_ON(path->nodes[level-1] != root->node);
1402
1403 lower = path->nodes[level-1];
1404 if (level == 1)
1405 btrfs_item_key(lower, &lower_key, 0);
1406 else
1407 btrfs_node_key(lower, &lower_key, 0);
1408
1409 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1410 root->root_key.objectid, &lower_key,
1411 level, root->node->start, 0);
1412
1413 if (IS_ERR(c))
1414 return PTR_ERR(c);
1415
1416 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
1417 btrfs_set_header_nritems(c, 1);
1418 btrfs_set_header_level(c, level);
1419 btrfs_set_header_bytenr(c, c->start);
1420 btrfs_set_header_generation(c, trans->transid);
1421 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
1422 btrfs_set_header_owner(c, root->root_key.objectid);
1423
1424 write_extent_buffer(c, root->fs_info->fsid,
1425 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1426
1427 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1428 btrfs_header_chunk_tree_uuid(c),
1429 BTRFS_UUID_SIZE);
1430
1431 btrfs_set_node_key(c, &lower_key, 0);
1432 btrfs_set_node_blockptr(c, 0, lower->start);
1433 lower_gen = btrfs_header_generation(lower);
1434 WARN_ON(lower_gen != trans->transid);
1435
1436 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1437
1438 btrfs_mark_buffer_dirty(c);
1439
1440 old = root->node;
1441 root->node = c;
1442
1443 /* the super has an extra ref to root->node */
1444 free_extent_buffer(old);
1445
1446 add_root_to_dirty_list(root);
1447 extent_buffer_get(c);
1448 path->nodes[level] = c;
1449 path->slots[level] = 0;
1450 return 0;
1451 }
1452
1453 /*
1454 * worker function to insert a single pointer in a node.
1455 * the node should have enough room for the pointer already
1456 *
1457 * slot and level indicate where you want the key to go, and
1458 * blocknr is the block the key points to.
1459 *
1460 * returns zero on success and < 0 on any error
1461 */
1462 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1463 *root, struct btrfs_path *path, struct btrfs_disk_key
1464 *key, u64 bytenr, int slot, int level)
1465 {
1466 struct extent_buffer *lower;
1467 int nritems;
1468
1469 BUG_ON(!path->nodes[level]);
1470 lower = path->nodes[level];
1471 nritems = btrfs_header_nritems(lower);
1472 if (slot > nritems)
1473 BUG();
1474 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1475 BUG();
1476 if (slot != nritems) {
1477 memmove_extent_buffer(lower,
1478 btrfs_node_key_ptr_offset(slot + 1),
1479 btrfs_node_key_ptr_offset(slot),
1480 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1481 }
1482 btrfs_set_node_key(lower, key, slot);
1483 btrfs_set_node_blockptr(lower, slot, bytenr);
1484 WARN_ON(trans->transid == 0);
1485 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1486 btrfs_set_header_nritems(lower, nritems + 1);
1487 btrfs_mark_buffer_dirty(lower);
1488 return 0;
1489 }
1490
1491 /*
1492 * split the node at the specified level in path in two.
1493 * The path is corrected to point to the appropriate node after the split
1494 *
1495 * Before splitting this tries to make some room in the node by pushing
1496 * left and right, if either one works, it returns right away.
1497 *
1498 * returns 0 on success and < 0 on failure
1499 */
1500 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1501 *root, struct btrfs_path *path, int level)
1502 {
1503 struct extent_buffer *c;
1504 struct extent_buffer *split;
1505 struct btrfs_disk_key disk_key;
1506 int mid;
1507 int ret;
1508 int wret;
1509 u32 c_nritems;
1510
1511 c = path->nodes[level];
1512 WARN_ON(btrfs_header_generation(c) != trans->transid);
1513 if (c == root->node) {
1514 /* trying to split the root, lets make a new one */
1515 ret = insert_new_root(trans, root, path, level + 1);
1516 if (ret)
1517 return ret;
1518 } else {
1519 ret = push_nodes_for_insert(trans, root, path, level);
1520 c = path->nodes[level];
1521 if (!ret && btrfs_header_nritems(c) <
1522 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1523 return 0;
1524 if (ret < 0)
1525 return ret;
1526 }
1527
1528 c_nritems = btrfs_header_nritems(c);
1529 mid = (c_nritems + 1) / 2;
1530 btrfs_node_key(c, &disk_key, mid);
1531
1532 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1533 root->root_key.objectid,
1534 &disk_key, level, c->start, 0);
1535 if (IS_ERR(split))
1536 return PTR_ERR(split);
1537
1538 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
1539 btrfs_set_header_level(split, btrfs_header_level(c));
1540 btrfs_set_header_bytenr(split, split->start);
1541 btrfs_set_header_generation(split, trans->transid);
1542 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
1543 btrfs_set_header_owner(split, root->root_key.objectid);
1544 write_extent_buffer(split, root->fs_info->fsid,
1545 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1546 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1547 btrfs_header_chunk_tree_uuid(split),
1548 BTRFS_UUID_SIZE);
1549
1550
1551 copy_extent_buffer(split, c,
1552 btrfs_node_key_ptr_offset(0),
1553 btrfs_node_key_ptr_offset(mid),
1554 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1555 btrfs_set_header_nritems(split, c_nritems - mid);
1556 btrfs_set_header_nritems(c, mid);
1557 ret = 0;
1558
1559 btrfs_mark_buffer_dirty(c);
1560 btrfs_mark_buffer_dirty(split);
1561
1562 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1563 path->slots[level + 1] + 1,
1564 level + 1);
1565 if (wret)
1566 ret = wret;
1567
1568 if (path->slots[level] >= mid) {
1569 path->slots[level] -= mid;
1570 free_extent_buffer(c);
1571 path->nodes[level] = split;
1572 path->slots[level + 1] += 1;
1573 } else {
1574 free_extent_buffer(split);
1575 }
1576 return ret;
1577 }
1578
1579 /*
1580 * how many bytes are required to store the items in a leaf. start
1581 * and nr indicate which items in the leaf to check. This totals up the
1582 * space used both by the item structs and the item data
1583 */
1584 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1585 {
1586 int data_len;
1587 int nritems = btrfs_header_nritems(l);
1588 int end = min(nritems, start + nr) - 1;
1589
1590 if (!nr)
1591 return 0;
1592 data_len = btrfs_item_end_nr(l, start);
1593 data_len = data_len - btrfs_item_offset_nr(l, end);
1594 data_len += sizeof(struct btrfs_item) * nr;
1595 WARN_ON(data_len < 0);
1596 return data_len;
1597 }
1598
1599 /*
1600 * The space between the end of the leaf items and
1601 * the start of the leaf data. IOW, how much room
1602 * the leaf has left for both items and data
1603 */
1604 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1605 {
1606 int nritems = btrfs_header_nritems(leaf);
1607 int ret;
1608 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1609 if (ret < 0) {
1610 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1611 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1612 leaf_space_used(leaf, 0, nritems), nritems);
1613 }
1614 return ret;
1615 }
1616
1617 /*
1618 * push some data in the path leaf to the right, trying to free up at
1619 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1620 *
1621 * returns 1 if the push failed because the other node didn't have enough
1622 * room, 0 if everything worked out and < 0 if there were major errors.
1623 */
1624 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1625 *root, struct btrfs_path *path, int data_size,
1626 int empty)
1627 {
1628 struct extent_buffer *left = path->nodes[0];
1629 struct extent_buffer *right;
1630 struct extent_buffer *upper;
1631 struct btrfs_disk_key disk_key;
1632 int slot;
1633 u32 i;
1634 int free_space;
1635 int push_space = 0;
1636 int push_items = 0;
1637 struct btrfs_item *item;
1638 u32 left_nritems;
1639 u32 nr;
1640 u32 right_nritems;
1641 u32 data_end;
1642 u32 this_item_size;
1643 int ret;
1644
1645 slot = path->slots[1];
1646 if (!path->nodes[1]) {
1647 return 1;
1648 }
1649 upper = path->nodes[1];
1650 if (slot >= btrfs_header_nritems(upper) - 1)
1651 return 1;
1652
1653 right = read_node_slot(root, upper, slot + 1);
1654 if (!extent_buffer_uptodate(right)) {
1655 if (IS_ERR(right))
1656 return PTR_ERR(right);
1657 return -EIO;
1658 }
1659 free_space = btrfs_leaf_free_space(root, right);
1660 if (free_space < data_size) {
1661 free_extent_buffer(right);
1662 return 1;
1663 }
1664
1665 /* cow and double check */
1666 ret = btrfs_cow_block(trans, root, right, upper,
1667 slot + 1, &right);
1668 if (ret) {
1669 free_extent_buffer(right);
1670 return 1;
1671 }
1672 free_space = btrfs_leaf_free_space(root, right);
1673 if (free_space < data_size) {
1674 free_extent_buffer(right);
1675 return 1;
1676 }
1677
1678 left_nritems = btrfs_header_nritems(left);
1679 if (left_nritems == 0) {
1680 free_extent_buffer(right);
1681 return 1;
1682 }
1683
1684 if (empty)
1685 nr = 0;
1686 else
1687 nr = 1;
1688
1689 i = left_nritems - 1;
1690 while (i >= nr) {
1691 item = btrfs_item_nr(i);
1692
1693 if (path->slots[0] == i)
1694 push_space += data_size + sizeof(*item);
1695
1696 this_item_size = btrfs_item_size(left, item);
1697 if (this_item_size + sizeof(*item) + push_space > free_space)
1698 break;
1699 push_items++;
1700 push_space += this_item_size + sizeof(*item);
1701 if (i == 0)
1702 break;
1703 i--;
1704 }
1705
1706 if (push_items == 0) {
1707 free_extent_buffer(right);
1708 return 1;
1709 }
1710
1711 if (!empty && push_items == left_nritems)
1712 WARN_ON(1);
1713
1714 /* push left to right */
1715 right_nritems = btrfs_header_nritems(right);
1716
1717 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1718 push_space -= leaf_data_end(root, left);
1719
1720 /* make room in the right data area */
1721 data_end = leaf_data_end(root, right);
1722 memmove_extent_buffer(right,
1723 btrfs_leaf_data(right) + data_end - push_space,
1724 btrfs_leaf_data(right) + data_end,
1725 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1726
1727 /* copy from the left data area */
1728 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1729 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1730 btrfs_leaf_data(left) + leaf_data_end(root, left),
1731 push_space);
1732
1733 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1734 btrfs_item_nr_offset(0),
1735 right_nritems * sizeof(struct btrfs_item));
1736
1737 /* copy the items from left to right */
1738 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1739 btrfs_item_nr_offset(left_nritems - push_items),
1740 push_items * sizeof(struct btrfs_item));
1741
1742 /* update the item pointers */
1743 right_nritems += push_items;
1744 btrfs_set_header_nritems(right, right_nritems);
1745 push_space = BTRFS_LEAF_DATA_SIZE(root);
1746 for (i = 0; i < right_nritems; i++) {
1747 item = btrfs_item_nr(i);
1748 push_space -= btrfs_item_size(right, item);
1749 btrfs_set_item_offset(right, item, push_space);
1750 }
1751
1752 left_nritems -= push_items;
1753 btrfs_set_header_nritems(left, left_nritems);
1754
1755 if (left_nritems)
1756 btrfs_mark_buffer_dirty(left);
1757 btrfs_mark_buffer_dirty(right);
1758
1759 btrfs_item_key(right, &disk_key, 0);
1760 btrfs_set_node_key(upper, &disk_key, slot + 1);
1761 btrfs_mark_buffer_dirty(upper);
1762
1763 /* then fixup the leaf pointer in the path */
1764 if (path->slots[0] >= left_nritems) {
1765 path->slots[0] -= left_nritems;
1766 free_extent_buffer(path->nodes[0]);
1767 path->nodes[0] = right;
1768 path->slots[1] += 1;
1769 } else {
1770 free_extent_buffer(right);
1771 }
1772 return 0;
1773 }
1774 /*
1775 * push some data in the path leaf to the left, trying to free up at
1776 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1777 */
1778 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1779 *root, struct btrfs_path *path, int data_size,
1780 int empty)
1781 {
1782 struct btrfs_disk_key disk_key;
1783 struct extent_buffer *right = path->nodes[0];
1784 struct extent_buffer *left;
1785 int slot;
1786 int i;
1787 int free_space;
1788 int push_space = 0;
1789 int push_items = 0;
1790 struct btrfs_item *item;
1791 u32 old_left_nritems;
1792 u32 right_nritems;
1793 u32 nr;
1794 int ret = 0;
1795 u32 this_item_size;
1796 u32 old_left_item_size;
1797
1798 slot = path->slots[1];
1799 if (slot == 0)
1800 return 1;
1801 if (!path->nodes[1])
1802 return 1;
1803
1804 right_nritems = btrfs_header_nritems(right);
1805 if (right_nritems == 0) {
1806 return 1;
1807 }
1808
1809 left = read_node_slot(root, path->nodes[1], slot - 1);
1810 free_space = btrfs_leaf_free_space(root, left);
1811 if (free_space < data_size) {
1812 free_extent_buffer(left);
1813 return 1;
1814 }
1815
1816 /* cow and double check */
1817 ret = btrfs_cow_block(trans, root, left,
1818 path->nodes[1], slot - 1, &left);
1819 if (ret) {
1820 /* we hit -ENOSPC, but it isn't fatal here */
1821 free_extent_buffer(left);
1822 return 1;
1823 }
1824
1825 free_space = btrfs_leaf_free_space(root, left);
1826 if (free_space < data_size) {
1827 free_extent_buffer(left);
1828 return 1;
1829 }
1830
1831 if (empty)
1832 nr = right_nritems;
1833 else
1834 nr = right_nritems - 1;
1835
1836 for (i = 0; i < nr; i++) {
1837 item = btrfs_item_nr(i);
1838
1839 if (path->slots[0] == i)
1840 push_space += data_size + sizeof(*item);
1841
1842 this_item_size = btrfs_item_size(right, item);
1843 if (this_item_size + sizeof(*item) + push_space > free_space)
1844 break;
1845
1846 push_items++;
1847 push_space += this_item_size + sizeof(*item);
1848 }
1849
1850 if (push_items == 0) {
1851 free_extent_buffer(left);
1852 return 1;
1853 }
1854 if (!empty && push_items == btrfs_header_nritems(right))
1855 WARN_ON(1);
1856
1857 /* push data from right to left */
1858 copy_extent_buffer(left, right,
1859 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1860 btrfs_item_nr_offset(0),
1861 push_items * sizeof(struct btrfs_item));
1862
1863 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1864 btrfs_item_offset_nr(right, push_items -1);
1865
1866 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1867 leaf_data_end(root, left) - push_space,
1868 btrfs_leaf_data(right) +
1869 btrfs_item_offset_nr(right, push_items - 1),
1870 push_space);
1871 old_left_nritems = btrfs_header_nritems(left);
1872 BUG_ON(old_left_nritems == 0);
1873
1874 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1875 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1876 u32 ioff;
1877
1878 item = btrfs_item_nr(i);
1879 ioff = btrfs_item_offset(left, item);
1880 btrfs_set_item_offset(left, item,
1881 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1882 }
1883 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1884
1885 /* fixup right node */
1886 if (push_items > right_nritems) {
1887 printk("push items %d nr %u\n", push_items, right_nritems);
1888 WARN_ON(1);
1889 }
1890
1891 if (push_items < right_nritems) {
1892 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1893 leaf_data_end(root, right);
1894 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1895 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1896 btrfs_leaf_data(right) +
1897 leaf_data_end(root, right), push_space);
1898
1899 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1900 btrfs_item_nr_offset(push_items),
1901 (btrfs_header_nritems(right) - push_items) *
1902 sizeof(struct btrfs_item));
1903 }
1904 right_nritems -= push_items;
1905 btrfs_set_header_nritems(right, right_nritems);
1906 push_space = BTRFS_LEAF_DATA_SIZE(root);
1907 for (i = 0; i < right_nritems; i++) {
1908 item = btrfs_item_nr(i);
1909 push_space = push_space - btrfs_item_size(right, item);
1910 btrfs_set_item_offset(right, item, push_space);
1911 }
1912
1913 btrfs_mark_buffer_dirty(left);
1914 if (right_nritems)
1915 btrfs_mark_buffer_dirty(right);
1916
1917 btrfs_item_key(right, &disk_key, 0);
1918 btrfs_fixup_low_keys(root, path, &disk_key, 1);
1919
1920 /* then fixup the leaf pointer in the path */
1921 if (path->slots[0] < push_items) {
1922 path->slots[0] += old_left_nritems;
1923 free_extent_buffer(path->nodes[0]);
1924 path->nodes[0] = left;
1925 path->slots[1] -= 1;
1926 } else {
1927 free_extent_buffer(left);
1928 path->slots[0] -= push_items;
1929 }
1930 BUG_ON(path->slots[0] < 0);
1931 return ret;
1932 }
1933
1934 /*
1935 * split the path's leaf in two, making sure there is at least data_size
1936 * available for the resulting leaf level of the path.
1937 *
1938 * returns 0 if all went well and < 0 on failure.
1939 */
1940 static noinline int copy_for_split(struct btrfs_trans_handle *trans,
1941 struct btrfs_root *root,
1942 struct btrfs_path *path,
1943 struct extent_buffer *l,
1944 struct extent_buffer *right,
1945 int slot, int mid, int nritems)
1946 {
1947 int data_copy_size;
1948 int rt_data_off;
1949 int i;
1950 int ret = 0;
1951 int wret;
1952 struct btrfs_disk_key disk_key;
1953
1954 nritems = nritems - mid;
1955 btrfs_set_header_nritems(right, nritems);
1956 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
1957
1958 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
1959 btrfs_item_nr_offset(mid),
1960 nritems * sizeof(struct btrfs_item));
1961
1962 copy_extent_buffer(right, l,
1963 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
1964 data_copy_size, btrfs_leaf_data(l) +
1965 leaf_data_end(root, l), data_copy_size);
1966
1967 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
1968 btrfs_item_end_nr(l, mid);
1969
1970 for (i = 0; i < nritems; i++) {
1971 struct btrfs_item *item = btrfs_item_nr(i);
1972 u32 ioff = btrfs_item_offset(right, item);
1973 btrfs_set_item_offset(right, item, ioff + rt_data_off);
1974 }
1975
1976 btrfs_set_header_nritems(l, mid);
1977 ret = 0;
1978 btrfs_item_key(right, &disk_key, 0);
1979 wret = insert_ptr(trans, root, path, &disk_key, right->start,
1980 path->slots[1] + 1, 1);
1981 if (wret)
1982 ret = wret;
1983
1984 btrfs_mark_buffer_dirty(right);
1985 btrfs_mark_buffer_dirty(l);
1986 BUG_ON(path->slots[0] != slot);
1987
1988 if (mid <= slot) {
1989 free_extent_buffer(path->nodes[0]);
1990 path->nodes[0] = right;
1991 path->slots[0] -= mid;
1992 path->slots[1] += 1;
1993 } else {
1994 free_extent_buffer(right);
1995 }
1996
1997 BUG_ON(path->slots[0] < 0);
1998
1999 return ret;
2000 }
2001
2002 /*
2003 * split the path's leaf in two, making sure there is at least data_size
2004 * available for the resulting leaf level of the path.
2005 *
2006 * returns 0 if all went well and < 0 on failure.
2007 */
2008 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2009 struct btrfs_root *root,
2010 struct btrfs_key *ins_key,
2011 struct btrfs_path *path, int data_size,
2012 int extend)
2013 {
2014 struct btrfs_disk_key disk_key;
2015 struct extent_buffer *l;
2016 u32 nritems;
2017 int mid;
2018 int slot;
2019 struct extent_buffer *right;
2020 int ret = 0;
2021 int wret;
2022 int split;
2023 int num_doubles = 0;
2024
2025 l = path->nodes[0];
2026 slot = path->slots[0];
2027 if (extend && data_size + btrfs_item_size_nr(l, slot) +
2028 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
2029 return -EOVERFLOW;
2030
2031 /* first try to make some room by pushing left and right */
2032 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2033 wret = push_leaf_right(trans, root, path, data_size, 0);
2034 if (wret < 0)
2035 return wret;
2036 if (wret) {
2037 wret = push_leaf_left(trans, root, path, data_size, 0);
2038 if (wret < 0)
2039 return wret;
2040 }
2041 l = path->nodes[0];
2042
2043 /* did the pushes work? */
2044 if (btrfs_leaf_free_space(root, l) >= data_size)
2045 return 0;
2046 }
2047
2048 if (!path->nodes[1]) {
2049 ret = insert_new_root(trans, root, path, 1);
2050 if (ret)
2051 return ret;
2052 }
2053 again:
2054 split = 1;
2055 l = path->nodes[0];
2056 slot = path->slots[0];
2057 nritems = btrfs_header_nritems(l);
2058 mid = (nritems + 1) / 2;
2059
2060 if (mid <= slot) {
2061 if (nritems == 1 ||
2062 leaf_space_used(l, mid, nritems - mid) + data_size >
2063 BTRFS_LEAF_DATA_SIZE(root)) {
2064 if (slot >= nritems) {
2065 split = 0;
2066 } else {
2067 mid = slot;
2068 if (mid != nritems &&
2069 leaf_space_used(l, mid, nritems - mid) +
2070 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2071 split = 2;
2072 }
2073 }
2074 }
2075 } else {
2076 if (leaf_space_used(l, 0, mid) + data_size >
2077 BTRFS_LEAF_DATA_SIZE(root)) {
2078 if (!extend && data_size && slot == 0) {
2079 split = 0;
2080 } else if ((extend || !data_size) && slot == 0) {
2081 mid = 1;
2082 } else {
2083 mid = slot;
2084 if (mid != nritems &&
2085 leaf_space_used(l, mid, nritems - mid) +
2086 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2087 split = 2 ;
2088 }
2089 }
2090 }
2091 }
2092
2093 if (split == 0)
2094 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2095 else
2096 btrfs_item_key(l, &disk_key, mid);
2097
2098 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2099 root->root_key.objectid,
2100 &disk_key, 0, l->start, 0);
2101 if (IS_ERR(right)) {
2102 BUG_ON(1);
2103 return PTR_ERR(right);
2104 }
2105
2106 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2107 btrfs_set_header_bytenr(right, right->start);
2108 btrfs_set_header_generation(right, trans->transid);
2109 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
2110 btrfs_set_header_owner(right, root->root_key.objectid);
2111 btrfs_set_header_level(right, 0);
2112 write_extent_buffer(right, root->fs_info->fsid,
2113 btrfs_header_fsid(), BTRFS_FSID_SIZE);
2114
2115 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2116 btrfs_header_chunk_tree_uuid(right),
2117 BTRFS_UUID_SIZE);
2118
2119 if (split == 0) {
2120 if (mid <= slot) {
2121 btrfs_set_header_nritems(right, 0);
2122 wret = insert_ptr(trans, root, path,
2123 &disk_key, right->start,
2124 path->slots[1] + 1, 1);
2125 if (wret)
2126 ret = wret;
2127
2128 free_extent_buffer(path->nodes[0]);
2129 path->nodes[0] = right;
2130 path->slots[0] = 0;
2131 path->slots[1] += 1;
2132 } else {
2133 btrfs_set_header_nritems(right, 0);
2134 wret = insert_ptr(trans, root, path,
2135 &disk_key,
2136 right->start,
2137 path->slots[1], 1);
2138 if (wret)
2139 ret = wret;
2140 free_extent_buffer(path->nodes[0]);
2141 path->nodes[0] = right;
2142 path->slots[0] = 0;
2143 if (path->slots[1] == 0) {
2144 btrfs_fixup_low_keys(root, path,
2145 &disk_key, 1);
2146 }
2147 }
2148 btrfs_mark_buffer_dirty(right);
2149 return ret;
2150 }
2151
2152 ret = copy_for_split(trans, root, path, l, right, slot, mid, nritems);
2153 BUG_ON(ret);
2154
2155 if (split == 2) {
2156 BUG_ON(num_doubles != 0);
2157 num_doubles++;
2158 goto again;
2159 }
2160
2161 return ret;
2162 }
2163
2164 /*
2165 * This function splits a single item into two items,
2166 * giving 'new_key' to the new item and splitting the
2167 * old one at split_offset (from the start of the item).
2168 *
2169 * The path may be released by this operation. After
2170 * the split, the path is pointing to the old item. The
2171 * new item is going to be in the same node as the old one.
2172 *
2173 * Note, the item being split must be smaller enough to live alone on
2174 * a tree block with room for one extra struct btrfs_item
2175 *
2176 * This allows us to split the item in place, keeping a lock on the
2177 * leaf the entire time.
2178 */
2179 int btrfs_split_item(struct btrfs_trans_handle *trans,
2180 struct btrfs_root *root,
2181 struct btrfs_path *path,
2182 struct btrfs_key *new_key,
2183 unsigned long split_offset)
2184 {
2185 u32 item_size;
2186 struct extent_buffer *leaf;
2187 struct btrfs_key orig_key;
2188 struct btrfs_item *item;
2189 struct btrfs_item *new_item;
2190 int ret = 0;
2191 int slot;
2192 u32 nritems;
2193 u32 orig_offset;
2194 struct btrfs_disk_key disk_key;
2195 char *buf;
2196
2197 leaf = path->nodes[0];
2198 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2199 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2200 goto split;
2201
2202 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2203 btrfs_release_path(path);
2204
2205 path->search_for_split = 1;
2206
2207 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2208 path->search_for_split = 0;
2209
2210 /* if our item isn't there or got smaller, return now */
2211 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2212 path->slots[0])) {
2213 return -EAGAIN;
2214 }
2215
2216 ret = split_leaf(trans, root, &orig_key, path, 0, 0);
2217 BUG_ON(ret);
2218
2219 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2220 leaf = path->nodes[0];
2221
2222 split:
2223 item = btrfs_item_nr(path->slots[0]);
2224 orig_offset = btrfs_item_offset(leaf, item);
2225 item_size = btrfs_item_size(leaf, item);
2226
2227
2228 buf = kmalloc(item_size, GFP_NOFS);
2229 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2230 path->slots[0]), item_size);
2231 slot = path->slots[0] + 1;
2232 leaf = path->nodes[0];
2233
2234 nritems = btrfs_header_nritems(leaf);
2235
2236 if (slot != nritems) {
2237 /* shift the items */
2238 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2239 btrfs_item_nr_offset(slot),
2240 (nritems - slot) * sizeof(struct btrfs_item));
2241
2242 }
2243
2244 btrfs_cpu_key_to_disk(&disk_key, new_key);
2245 btrfs_set_item_key(leaf, &disk_key, slot);
2246
2247 new_item = btrfs_item_nr(slot);
2248
2249 btrfs_set_item_offset(leaf, new_item, orig_offset);
2250 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
2251
2252 btrfs_set_item_offset(leaf, item,
2253 orig_offset + item_size - split_offset);
2254 btrfs_set_item_size(leaf, item, split_offset);
2255
2256 btrfs_set_header_nritems(leaf, nritems + 1);
2257
2258 /* write the data for the start of the original item */
2259 write_extent_buffer(leaf, buf,
2260 btrfs_item_ptr_offset(leaf, path->slots[0]),
2261 split_offset);
2262
2263 /* write the data for the new item */
2264 write_extent_buffer(leaf, buf + split_offset,
2265 btrfs_item_ptr_offset(leaf, slot),
2266 item_size - split_offset);
2267 btrfs_mark_buffer_dirty(leaf);
2268
2269 ret = 0;
2270 if (btrfs_leaf_free_space(root, leaf) < 0) {
2271 btrfs_print_leaf(root, leaf);
2272 BUG();
2273 }
2274 kfree(buf);
2275 return ret;
2276 }
2277
2278 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2279 struct btrfs_root *root,
2280 struct btrfs_path *path,
2281 u32 new_size, int from_end)
2282 {
2283 int ret = 0;
2284 int slot;
2285 struct extent_buffer *leaf;
2286 struct btrfs_item *item;
2287 u32 nritems;
2288 unsigned int data_end;
2289 unsigned int old_data_start;
2290 unsigned int old_size;
2291 unsigned int size_diff;
2292 int i;
2293
2294 leaf = path->nodes[0];
2295 slot = path->slots[0];
2296
2297 old_size = btrfs_item_size_nr(leaf, slot);
2298 if (old_size == new_size)
2299 return 0;
2300
2301 nritems = btrfs_header_nritems(leaf);
2302 data_end = leaf_data_end(root, leaf);
2303
2304 old_data_start = btrfs_item_offset_nr(leaf, slot);
2305
2306 size_diff = old_size - new_size;
2307
2308 BUG_ON(slot < 0);
2309 BUG_ON(slot >= nritems);
2310
2311 /*
2312 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2313 */
2314 /* first correct the data pointers */
2315 for (i = slot; i < nritems; i++) {
2316 u32 ioff;
2317 item = btrfs_item_nr(i);
2318 ioff = btrfs_item_offset(leaf, item);
2319 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2320 }
2321
2322 /* shift the data */
2323 if (from_end) {
2324 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2325 data_end + size_diff, btrfs_leaf_data(leaf) +
2326 data_end, old_data_start + new_size - data_end);
2327 } else {
2328 struct btrfs_disk_key disk_key;
2329 u64 offset;
2330
2331 btrfs_item_key(leaf, &disk_key, slot);
2332
2333 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2334 unsigned long ptr;
2335 struct btrfs_file_extent_item *fi;
2336
2337 fi = btrfs_item_ptr(leaf, slot,
2338 struct btrfs_file_extent_item);
2339 fi = (struct btrfs_file_extent_item *)(
2340 (unsigned long)fi - size_diff);
2341
2342 if (btrfs_file_extent_type(leaf, fi) ==
2343 BTRFS_FILE_EXTENT_INLINE) {
2344 ptr = btrfs_item_ptr_offset(leaf, slot);
2345 memmove_extent_buffer(leaf, ptr,
2346 (unsigned long)fi,
2347 offsetof(struct btrfs_file_extent_item,
2348 disk_bytenr));
2349 }
2350 }
2351
2352 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2353 data_end + size_diff, btrfs_leaf_data(leaf) +
2354 data_end, old_data_start - data_end);
2355
2356 offset = btrfs_disk_key_offset(&disk_key);
2357 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2358 btrfs_set_item_key(leaf, &disk_key, slot);
2359 if (slot == 0)
2360 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2361 }
2362
2363 item = btrfs_item_nr(slot);
2364 btrfs_set_item_size(leaf, item, new_size);
2365 btrfs_mark_buffer_dirty(leaf);
2366
2367 ret = 0;
2368 if (btrfs_leaf_free_space(root, leaf) < 0) {
2369 btrfs_print_leaf(root, leaf);
2370 BUG();
2371 }
2372 return ret;
2373 }
2374
2375 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *root, struct btrfs_path *path,
2377 u32 data_size)
2378 {
2379 int ret = 0;
2380 int slot;
2381 struct extent_buffer *leaf;
2382 struct btrfs_item *item;
2383 u32 nritems;
2384 unsigned int data_end;
2385 unsigned int old_data;
2386 unsigned int old_size;
2387 int i;
2388
2389 leaf = path->nodes[0];
2390
2391 nritems = btrfs_header_nritems(leaf);
2392 data_end = leaf_data_end(root, leaf);
2393
2394 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2395 btrfs_print_leaf(root, leaf);
2396 BUG();
2397 }
2398 slot = path->slots[0];
2399 old_data = btrfs_item_end_nr(leaf, slot);
2400
2401 BUG_ON(slot < 0);
2402 if (slot >= nritems) {
2403 btrfs_print_leaf(root, leaf);
2404 printk("slot %d too large, nritems %d\n", slot, nritems);
2405 BUG_ON(1);
2406 }
2407
2408 /*
2409 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2410 */
2411 /* first correct the data pointers */
2412 for (i = slot; i < nritems; i++) {
2413 u32 ioff;
2414 item = btrfs_item_nr(i);
2415 ioff = btrfs_item_offset(leaf, item);
2416 btrfs_set_item_offset(leaf, item, ioff - data_size);
2417 }
2418
2419 /* shift the data */
2420 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2421 data_end - data_size, btrfs_leaf_data(leaf) +
2422 data_end, old_data - data_end);
2423
2424 data_end = old_data;
2425 old_size = btrfs_item_size_nr(leaf, slot);
2426 item = btrfs_item_nr(slot);
2427 btrfs_set_item_size(leaf, item, old_size + data_size);
2428 btrfs_mark_buffer_dirty(leaf);
2429
2430 ret = 0;
2431 if (btrfs_leaf_free_space(root, leaf) < 0) {
2432 btrfs_print_leaf(root, leaf);
2433 BUG();
2434 }
2435 return ret;
2436 }
2437
2438 /*
2439 * Given a key and some data, insert an item into the tree.
2440 * This does all the path init required, making room in the tree if needed.
2441 */
2442 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2443 struct btrfs_root *root,
2444 struct btrfs_path *path,
2445 struct btrfs_key *cpu_key, u32 *data_size,
2446 int nr)
2447 {
2448 struct extent_buffer *leaf;
2449 struct btrfs_item *item;
2450 int ret = 0;
2451 int slot;
2452 int i;
2453 u32 nritems;
2454 u32 total_size = 0;
2455 u32 total_data = 0;
2456 unsigned int data_end;
2457 struct btrfs_disk_key disk_key;
2458
2459 for (i = 0; i < nr; i++) {
2460 total_data += data_size[i];
2461 }
2462
2463 /* create a root if there isn't one */
2464 if (!root->node)
2465 BUG();
2466
2467 total_size = total_data + nr * sizeof(struct btrfs_item);
2468 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2469 if (ret == 0) {
2470 return -EEXIST;
2471 }
2472 if (ret < 0)
2473 goto out;
2474
2475 leaf = path->nodes[0];
2476
2477 nritems = btrfs_header_nritems(leaf);
2478 data_end = leaf_data_end(root, leaf);
2479
2480 if (btrfs_leaf_free_space(root, leaf) < total_size) {
2481 btrfs_print_leaf(root, leaf);
2482 printk("not enough freespace need %u have %d\n",
2483 total_size, btrfs_leaf_free_space(root, leaf));
2484 BUG();
2485 }
2486
2487 slot = path->slots[0];
2488 BUG_ON(slot < 0);
2489
2490 if (slot != nritems) {
2491 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2492
2493 if (old_data < data_end) {
2494 btrfs_print_leaf(root, leaf);
2495 printk("slot %d old_data %d data_end %d\n",
2496 slot, old_data, data_end);
2497 BUG_ON(1);
2498 }
2499 /*
2500 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2501 */
2502 /* first correct the data pointers */
2503 for (i = slot; i < nritems; i++) {
2504 u32 ioff;
2505
2506 item = btrfs_item_nr(i);
2507 ioff = btrfs_item_offset(leaf, item);
2508 btrfs_set_item_offset(leaf, item, ioff - total_data);
2509 }
2510
2511 /* shift the items */
2512 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2513 btrfs_item_nr_offset(slot),
2514 (nritems - slot) * sizeof(struct btrfs_item));
2515
2516 /* shift the data */
2517 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2518 data_end - total_data, btrfs_leaf_data(leaf) +
2519 data_end, old_data - data_end);
2520 data_end = old_data;
2521 }
2522
2523 /* setup the item for the new data */
2524 for (i = 0; i < nr; i++) {
2525 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2526 btrfs_set_item_key(leaf, &disk_key, slot + i);
2527 item = btrfs_item_nr(slot + i);
2528 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2529 data_end -= data_size[i];
2530 btrfs_set_item_size(leaf, item, data_size[i]);
2531 }
2532 btrfs_set_header_nritems(leaf, nritems + nr);
2533 btrfs_mark_buffer_dirty(leaf);
2534
2535 ret = 0;
2536 if (slot == 0) {
2537 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2538 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2539 }
2540
2541 if (btrfs_leaf_free_space(root, leaf) < 0) {
2542 btrfs_print_leaf(root, leaf);
2543 BUG();
2544 }
2545
2546 out:
2547 return ret;
2548 }
2549
2550 /*
2551 * Given a key and some data, insert an item into the tree.
2552 * This does all the path init required, making room in the tree if needed.
2553 */
2554 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2555 *root, struct btrfs_key *cpu_key, void *data, u32
2556 data_size)
2557 {
2558 int ret = 0;
2559 struct btrfs_path *path;
2560 struct extent_buffer *leaf;
2561 unsigned long ptr;
2562
2563 path = btrfs_alloc_path();
2564 BUG_ON(!path);
2565 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2566 if (!ret) {
2567 leaf = path->nodes[0];
2568 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2569 write_extent_buffer(leaf, data, ptr, data_size);
2570 btrfs_mark_buffer_dirty(leaf);
2571 }
2572 btrfs_free_path(path);
2573 return ret;
2574 }
2575
2576 /*
2577 * delete the pointer from a given node.
2578 *
2579 * If the delete empties a node, the node is removed from the tree,
2580 * continuing all the way the root if required. The root is converted into
2581 * a leaf if all the nodes are emptied.
2582 */
2583 int btrfs_del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2584 struct btrfs_path *path, int level, int slot)
2585 {
2586 struct extent_buffer *parent = path->nodes[level];
2587 u32 nritems;
2588 int ret = 0;
2589
2590 nritems = btrfs_header_nritems(parent);
2591 if (slot != nritems -1) {
2592 memmove_extent_buffer(parent,
2593 btrfs_node_key_ptr_offset(slot),
2594 btrfs_node_key_ptr_offset(slot + 1),
2595 sizeof(struct btrfs_key_ptr) *
2596 (nritems - slot - 1));
2597 }
2598 nritems--;
2599 btrfs_set_header_nritems(parent, nritems);
2600 if (nritems == 0 && parent == root->node) {
2601 BUG_ON(btrfs_header_level(root->node) != 1);
2602 /* just turn the root into a leaf and break */
2603 btrfs_set_header_level(root->node, 0);
2604 } else if (slot == 0) {
2605 struct btrfs_disk_key disk_key;
2606
2607 btrfs_node_key(parent, &disk_key, 0);
2608 btrfs_fixup_low_keys(root, path, &disk_key, level + 1);
2609 }
2610 btrfs_mark_buffer_dirty(parent);
2611 return ret;
2612 }
2613
2614 /*
2615 * a helper function to delete the leaf pointed to by path->slots[1] and
2616 * path->nodes[1].
2617 *
2618 * This deletes the pointer in path->nodes[1] and frees the leaf
2619 * block extent. zero is returned if it all worked out, < 0 otherwise.
2620 *
2621 * The path must have already been setup for deleting the leaf, including
2622 * all the proper balancing. path->nodes[1] must be locked.
2623 */
2624 static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
2625 struct btrfs_root *root,
2626 struct btrfs_path *path,
2627 struct extent_buffer *leaf)
2628 {
2629 int ret;
2630
2631 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
2632 ret = btrfs_del_ptr(trans, root, path, 1, path->slots[1]);
2633 if (ret)
2634 return ret;
2635
2636 ret = btrfs_free_extent(trans, root, leaf->start, leaf->len,
2637 0, root->root_key.objectid, 0, 0);
2638 return ret;
2639 }
2640
2641 /*
2642 * delete the item at the leaf level in path. If that empties
2643 * the leaf, remove it from the tree
2644 */
2645 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2646 struct btrfs_path *path, int slot, int nr)
2647 {
2648 struct extent_buffer *leaf;
2649 struct btrfs_item *item;
2650 int last_off;
2651 int dsize = 0;
2652 int ret = 0;
2653 int wret;
2654 int i;
2655 u32 nritems;
2656
2657 leaf = path->nodes[0];
2658 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2659
2660 for (i = 0; i < nr; i++)
2661 dsize += btrfs_item_size_nr(leaf, slot + i);
2662
2663 nritems = btrfs_header_nritems(leaf);
2664
2665 if (slot + nr != nritems) {
2666 int data_end = leaf_data_end(root, leaf);
2667
2668 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2669 data_end + dsize,
2670 btrfs_leaf_data(leaf) + data_end,
2671 last_off - data_end);
2672
2673 for (i = slot + nr; i < nritems; i++) {
2674 u32 ioff;
2675
2676 item = btrfs_item_nr(i);
2677 ioff = btrfs_item_offset(leaf, item);
2678 btrfs_set_item_offset(leaf, item, ioff + dsize);
2679 }
2680
2681 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2682 btrfs_item_nr_offset(slot + nr),
2683 sizeof(struct btrfs_item) *
2684 (nritems - slot - nr));
2685 }
2686 btrfs_set_header_nritems(leaf, nritems - nr);
2687 nritems -= nr;
2688
2689 /* delete the leaf if we've emptied it */
2690 if (nritems == 0) {
2691 if (leaf == root->node) {
2692 btrfs_set_header_level(leaf, 0);
2693 } else {
2694 clean_tree_block(trans, root, leaf);
2695 wait_on_tree_block_writeback(root, leaf);
2696
2697 wret = btrfs_del_leaf(trans, root, path, leaf);
2698 BUG_ON(ret);
2699 if (wret)
2700 ret = wret;
2701 }
2702 } else {
2703 int used = leaf_space_used(leaf, 0, nritems);
2704 if (slot == 0) {
2705 struct btrfs_disk_key disk_key;
2706
2707 btrfs_item_key(leaf, &disk_key, 0);
2708 btrfs_fixup_low_keys(root, path, &disk_key, 1);
2709 }
2710
2711 /* delete the leaf if it is mostly empty */
2712 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2713 /* push_leaf_left fixes the path.
2714 * make sure the path still points to our leaf
2715 * for possible call to del_ptr below
2716 */
2717 slot = path->slots[1];
2718 extent_buffer_get(leaf);
2719
2720 wret = push_leaf_left(trans, root, path, 1, 1);
2721 if (wret < 0 && wret != -ENOSPC)
2722 ret = wret;
2723
2724 if (path->nodes[0] == leaf &&
2725 btrfs_header_nritems(leaf)) {
2726 wret = push_leaf_right(trans, root, path, 1, 1);
2727 if (wret < 0 && wret != -ENOSPC)
2728 ret = wret;
2729 }
2730
2731 if (btrfs_header_nritems(leaf) == 0) {
2732 clean_tree_block(trans, root, leaf);
2733 wait_on_tree_block_writeback(root, leaf);
2734
2735 path->slots[1] = slot;
2736 ret = btrfs_del_leaf(trans, root, path, leaf);
2737 BUG_ON(ret);
2738 free_extent_buffer(leaf);
2739
2740 } else {
2741 btrfs_mark_buffer_dirty(leaf);
2742 free_extent_buffer(leaf);
2743 }
2744 } else {
2745 btrfs_mark_buffer_dirty(leaf);
2746 }
2747 }
2748 return ret;
2749 }
2750
2751 /*
2752 * walk up the tree as far as required to find the previous leaf.
2753 * returns 0 if it found something or 1 if there are no lesser leaves.
2754 * returns < 0 on io errors.
2755 */
2756 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2757 {
2758 int slot;
2759 int level = 1;
2760 struct extent_buffer *c;
2761 struct extent_buffer *next = NULL;
2762
2763 while(level < BTRFS_MAX_LEVEL) {
2764 if (!path->nodes[level])
2765 return 1;
2766
2767 slot = path->slots[level];
2768 c = path->nodes[level];
2769 if (slot == 0) {
2770 level++;
2771 if (level == BTRFS_MAX_LEVEL)
2772 return 1;
2773 continue;
2774 }
2775 slot--;
2776
2777 next = read_node_slot(root, c, slot);
2778 if (!extent_buffer_uptodate(next)) {
2779 if (IS_ERR(next))
2780 return PTR_ERR(next);
2781 return -EIO;
2782 }
2783 break;
2784 }
2785 path->slots[level] = slot;
2786 while(1) {
2787 level--;
2788 c = path->nodes[level];
2789 free_extent_buffer(c);
2790 slot = btrfs_header_nritems(next);
2791 if (slot != 0)
2792 slot--;
2793 path->nodes[level] = next;
2794 path->slots[level] = slot;
2795 if (!level)
2796 break;
2797 next = read_node_slot(root, next, slot);
2798 if (!extent_buffer_uptodate(next)) {
2799 if (IS_ERR(next))
2800 return PTR_ERR(next);
2801 return -EIO;
2802 }
2803 }
2804 return 0;
2805 }
2806
2807 /*
2808 * walk up the tree as far as required to find the next leaf.
2809 * returns 0 if it found something or 1 if there are no greater leaves.
2810 * returns < 0 on io errors.
2811 */
2812 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2813 {
2814 int slot;
2815 int level = 1;
2816 struct extent_buffer *c;
2817 struct extent_buffer *next = NULL;
2818
2819 while(level < BTRFS_MAX_LEVEL) {
2820 if (!path->nodes[level])
2821 return 1;
2822
2823 slot = path->slots[level] + 1;
2824 c = path->nodes[level];
2825 if (slot >= btrfs_header_nritems(c)) {
2826 level++;
2827 if (level == BTRFS_MAX_LEVEL)
2828 return 1;
2829 continue;
2830 }
2831
2832 if (path->reada)
2833 reada_for_search(root, path, level, slot, 0);
2834
2835 next = read_node_slot(root, c, slot);
2836 if (!extent_buffer_uptodate(next))
2837 return -EIO;
2838 break;
2839 }
2840 path->slots[level] = slot;
2841 while(1) {
2842 level--;
2843 c = path->nodes[level];
2844 free_extent_buffer(c);
2845 path->nodes[level] = next;
2846 path->slots[level] = 0;
2847 if (!level)
2848 break;
2849 if (path->reada)
2850 reada_for_search(root, path, level, 0, 0);
2851 next = read_node_slot(root, next, 0);
2852 if (!extent_buffer_uptodate(next))
2853 return -EIO;
2854 }
2855 return 0;
2856 }
2857
2858 int btrfs_previous_item(struct btrfs_root *root,
2859 struct btrfs_path *path, u64 min_objectid,
2860 int type)
2861 {
2862 struct btrfs_key found_key;
2863 struct extent_buffer *leaf;
2864 int ret;
2865
2866 while(1) {
2867 if (path->slots[0] == 0) {
2868 ret = btrfs_prev_leaf(root, path);
2869 if (ret != 0)
2870 return ret;
2871 } else {
2872 path->slots[0]--;
2873 }
2874 leaf = path->nodes[0];
2875 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2876 if (found_key.type == type)
2877 return 0;
2878 }
2879 return 1;
2880 }
2881
2882 /*
2883 * search in extent tree to find a previous Metadata/Data extent item with
2884 * min objecitd.
2885 *
2886 * returns 0 if something is found, 1 if nothing was found and < 0 on error
2887 */
2888 int btrfs_previous_extent_item(struct btrfs_root *root,
2889 struct btrfs_path *path, u64 min_objectid)
2890 {
2891 struct btrfs_key found_key;
2892 struct extent_buffer *leaf;
2893 u32 nritems;
2894 int ret;
2895
2896 while (1) {
2897 if (path->slots[0] == 0) {
2898 ret = btrfs_prev_leaf(root, path);
2899 if (ret != 0)
2900 return ret;
2901 } else {
2902 path->slots[0]--;
2903 }
2904 leaf = path->nodes[0];
2905 nritems = btrfs_header_nritems(leaf);
2906 if (nritems == 0)
2907 return 1;
2908 if (path->slots[0] == nritems)
2909 path->slots[0]--;
2910
2911 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2912 if (found_key.objectid < min_objectid)
2913 break;
2914 if (found_key.type == BTRFS_EXTENT_ITEM_KEY ||
2915 found_key.type == BTRFS_METADATA_ITEM_KEY)
2916 return 0;
2917 if (found_key.objectid == min_objectid &&
2918 found_key.type < BTRFS_EXTENT_ITEM_KEY)
2919 break;
2920 }
2921 return 1;
2922 }
(deprecated) Btrfs progs developments and patch integration