?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Don't empty the middle buffer in push_nodes_for_insert
[btrfs-progs-unstable.git] / ctree.c
blobeb645da9b131b48962477cce7f30be0780ccf0a3
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
23 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
24 *root, struct btrfs_path *path, int level);
25 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
26 *root, struct btrfs_key *ins_key,
27 struct btrfs_path *path, int data_size, int extend);
28 static int push_node_left(struct btrfs_trans_handle *trans,
29 struct btrfs_root *root, struct extent_buffer *dst,
30 struct extent_buffer *src, int empty);
31 static int balance_node_right(struct btrfs_trans_handle *trans,
32 struct btrfs_root *root,
33 struct extent_buffer *dst_buf,
34 struct extent_buffer *src_buf);
35 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
36 struct btrfs_path *path, int level, int slot);
37
38 inline void btrfs_init_path(struct btrfs_path *p)
39 {
40 memset(p, 0, sizeof(*p));
41 }
42
43 struct btrfs_path *btrfs_alloc_path(void)
44 {
45 struct btrfs_path *path;
46 path = kmalloc(sizeof(struct btrfs_path), GFP_NOFS);
47 if (path) {
48 btrfs_init_path(path);
49 path->reada = 0;
50 }
51 return path;
52 }
53
54 void btrfs_free_path(struct btrfs_path *p)
55 {
56 btrfs_release_path(NULL, p);
57 kfree(p);
58 }
59
60 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
61 {
62 int i;
63 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
64 if (!p->nodes[i])
65 break;
66 free_extent_buffer(p->nodes[i]);
67 }
68 memset(p, 0, sizeof(*p));
69 }
70
71 static void add_root_to_dirty_list(struct btrfs_root *root)
72 {
73 if (root->track_dirty && list_empty(&root->dirty_list)) {
74 list_add(&root->dirty_list,
75 &root->fs_info->dirty_cowonly_roots);
76 }
77 }
78
79 int btrfs_copy_root(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 struct extent_buffer *buf,
82 struct extent_buffer **cow_ret, u64 new_root_objectid)
83 {
84 struct extent_buffer *cow;
85 u32 nritems;
86 int ret = 0;
87 int level;
88 struct btrfs_key first_key;
89 struct btrfs_root *new_root;
90
91 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
92 if (!new_root)
93 return -ENOMEM;
94
95 memcpy(new_root, root, sizeof(*new_root));
96 new_root->root_key.objectid = new_root_objectid;
97
98 WARN_ON(root->ref_cows && trans->transid !=
99 root->fs_info->running_transaction->transid);
100 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
101
102 level = btrfs_header_level(buf);
103 nritems = btrfs_header_nritems(buf);
104 if (nritems) {
105 if (level == 0)
106 btrfs_item_key_to_cpu(buf, &first_key, 0);
107 else
108 btrfs_node_key_to_cpu(buf, &first_key, 0);
109 } else {
110 first_key.objectid = 0;
111 }
112 cow = __btrfs_alloc_free_block(trans, new_root, buf->len,
113 new_root_objectid,
114 trans->transid, first_key.objectid,
115 level, buf->start, 0);
116 if (IS_ERR(cow)) {
117 kfree(new_root);
118 return PTR_ERR(cow);
119 }
120
121 copy_extent_buffer(cow, buf, 0, 0, cow->len);
122 btrfs_set_header_bytenr(cow, cow->start);
123 btrfs_set_header_generation(cow, trans->transid);
124 btrfs_set_header_owner(cow, new_root_objectid);
125 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
126
127 WARN_ON(btrfs_header_generation(buf) > trans->transid);
128 ret = btrfs_inc_ref(trans, new_root, buf);
129 kfree(new_root);
130
131 if (ret)
132 return ret;
133
134 btrfs_mark_buffer_dirty(cow);
135 *cow_ret = cow;
136 return 0;
137 }
138
139 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
140 struct btrfs_root *root,
141 struct extent_buffer *buf,
142 struct extent_buffer *parent, int parent_slot,
143 struct extent_buffer **cow_ret,
144 u64 search_start, u64 empty_size)
145 {
146 u64 root_gen;
147 struct extent_buffer *cow;
148 u32 nritems;
149 int ret = 0;
150 int different_trans = 0;
151 int level;
152 struct btrfs_key first_key;
153
154 if (root->ref_cows) {
155 root_gen = trans->transid;
156 } else {
157 root_gen = 0;
158 }
159
160 WARN_ON(root->ref_cows && trans->transid !=
161 root->fs_info->running_transaction->transid);
162 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
163
164 level = btrfs_header_level(buf);
165 nritems = btrfs_header_nritems(buf);
166 if (nritems) {
167 if (level == 0)
168 btrfs_item_key_to_cpu(buf, &first_key, 0);
169 else
170 btrfs_node_key_to_cpu(buf, &first_key, 0);
171 } else {
172 first_key.objectid = 0;
173 }
174 cow = __btrfs_alloc_free_block(trans, root, buf->len,
175 root->root_key.objectid,
176 root_gen, first_key.objectid, level,
177 search_start, empty_size);
178 if (IS_ERR(cow))
179 return PTR_ERR(cow);
180
181 copy_extent_buffer(cow, buf, 0, 0, cow->len);
182 btrfs_set_header_bytenr(cow, cow->start);
183 btrfs_set_header_generation(cow, trans->transid);
184 btrfs_set_header_owner(cow, root->root_key.objectid);
185 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
186
187 WARN_ON(btrfs_header_generation(buf) > trans->transid);
188 if (btrfs_header_generation(buf) != trans->transid) {
189 different_trans = 1;
190 ret = btrfs_inc_ref(trans, root, buf);
191 if (ret)
192 return ret;
193 } else {
194 clean_tree_block(trans, root, buf);
195 }
196
197 if (buf == root->node) {
198 root_gen = btrfs_header_generation(buf);
199 root->node = cow;
200 extent_buffer_get(cow);
201 if (buf != root->commit_root) {
202 btrfs_free_extent(trans, root, buf->start,
203 buf->len, root->root_key.objectid,
204 root_gen, 0, 0, 1);
205 }
206 free_extent_buffer(buf);
207 add_root_to_dirty_list(root);
208 } else {
209 root_gen = btrfs_header_generation(parent);
210 btrfs_set_node_blockptr(parent, parent_slot,
211 cow->start);
212 WARN_ON(trans->transid == 0);
213 btrfs_set_node_ptr_generation(parent, parent_slot,
214 trans->transid);
215 btrfs_mark_buffer_dirty(parent);
216 WARN_ON(btrfs_header_generation(parent) != trans->transid);
217 btrfs_free_extent(trans, root, buf->start, buf->len,
218 btrfs_header_owner(parent), root_gen,
219 0, 0, 1);
220 }
221 free_extent_buffer(buf);
222 btrfs_mark_buffer_dirty(cow);
223 *cow_ret = cow;
224 return 0;
225 }
226
227 int btrfs_cow_block(struct btrfs_trans_handle *trans,
228 struct btrfs_root *root, struct extent_buffer *buf,
229 struct extent_buffer *parent, int parent_slot,
230 struct extent_buffer **cow_ret)
231 {
232 u64 search_start;
233 int ret;
234 /*
235 if (trans->transaction != root->fs_info->running_transaction) {
236 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
237 root->fs_info->running_transaction->transid);
238 WARN_ON(1);
239 }
240 */
241 if (trans->transid != root->fs_info->generation) {
242 printk(KERN_CRIT "trans %llu running %llu\n",
243 (unsigned long long)trans->transid,
244 (unsigned long long)root->fs_info->generation);
245 WARN_ON(1);
246 }
247 if (btrfs_header_generation(buf) == trans->transid &&
248 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
249 *cow_ret = buf;
250 return 0;
251 }
252
253 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
254 ret = __btrfs_cow_block(trans, root, buf, parent,
255 parent_slot, cow_ret, search_start, 0);
256 return ret;
257 }
258
259 /*
260 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
261 {
262 if (blocknr < other && other - (blocknr + blocksize) < 32768)
263 return 1;
264 if (blocknr > other && blocknr - (other + blocksize) < 32768)
265 return 1;
266 return 0;
267 }
268 */
269
270 /*
271 * compare two keys in a memcmp fashion
272 */
273 int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
274 {
275 struct btrfs_key k1;
276
277 btrfs_disk_key_to_cpu(&k1, disk);
278
279 if (k1.objectid > k2->objectid)
280 return 1;
281 if (k1.objectid < k2->objectid)
282 return -1;
283 if (k1.type > k2->type)
284 return 1;
285 if (k1.type < k2->type)
286 return -1;
287 if (k1.offset > k2->offset)
288 return 1;
289 if (k1.offset < k2->offset)
290 return -1;
291 return 0;
292 }
293
294
295 #if 0
296 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
297 struct btrfs_root *root, struct extent_buffer *parent,
298 int start_slot, int cache_only, u64 *last_ret,
299 struct btrfs_key *progress)
300 {
301 struct extent_buffer *cur;
302 struct extent_buffer *tmp;
303 u64 blocknr;
304 u64 search_start = *last_ret;
305 u64 last_block = 0;
306 u64 other;
307 u32 parent_nritems;
308 int end_slot;
309 int i;
310 int err = 0;
311 int parent_level;
312 int uptodate;
313 u32 blocksize;
314 int progress_passed = 0;
315 struct btrfs_disk_key disk_key;
316
317 parent_level = btrfs_header_level(parent);
318 if (cache_only && parent_level != 1)
319 return 0;
320
321 if (trans->transaction != root->fs_info->running_transaction) {
322 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
323 root->fs_info->running_transaction->transid);
324 WARN_ON(1);
325 }
326 if (trans->transid != root->fs_info->generation) {
327 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
328 root->fs_info->generation);
329 WARN_ON(1);
330 }
331
332 parent_nritems = btrfs_header_nritems(parent);
333 blocksize = btrfs_level_size(root, parent_level - 1);
334 end_slot = parent_nritems;
335
336 if (parent_nritems == 1)
337 return 0;
338
339 for (i = start_slot; i < end_slot; i++) {
340 int close = 1;
341
342 if (!parent->map_token) {
343 map_extent_buffer(parent,
344 btrfs_node_key_ptr_offset(i),
345 sizeof(struct btrfs_key_ptr),
346 &parent->map_token, &parent->kaddr,
347 &parent->map_start, &parent->map_len,
348 KM_USER1);
349 }
350 btrfs_node_key(parent, &disk_key, i);
351 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
352 continue;
353
354 progress_passed = 1;
355 blocknr = btrfs_node_blockptr(parent, i);
356 if (last_block == 0)
357 last_block = blocknr;
358
359 if (i > 0) {
360 other = btrfs_node_blockptr(parent, i - 1);
361 close = close_blocks(blocknr, other, blocksize);
362 }
363 if (close && i < end_slot - 2) {
364 other = btrfs_node_blockptr(parent, i + 1);
365 close = close_blocks(blocknr, other, blocksize);
366 }
367 if (close) {
368 last_block = blocknr;
369 continue;
370 }
371 if (parent->map_token) {
372 unmap_extent_buffer(parent, parent->map_token,
373 KM_USER1);
374 parent->map_token = NULL;
375 }
376
377 cur = btrfs_find_tree_block(root, blocknr, blocksize);
378 if (cur)
379 uptodate = btrfs_buffer_uptodate(cur);
380 else
381 uptodate = 0;
382 if (!cur || !uptodate) {
383 if (cache_only) {
384 free_extent_buffer(cur);
385 continue;
386 }
387 if (!cur) {
388 cur = read_tree_block(root, blocknr,
389 blocksize);
390 } else if (!uptodate) {
391 btrfs_read_buffer(cur);
392 }
393 }
394 if (search_start == 0)
395 search_start = last_block;
396
397 err = __btrfs_cow_block(trans, root, cur, parent, i,
398 &tmp, search_start,
399 min(16 * blocksize,
400 (end_slot - i) * blocksize));
401 if (err) {
402 free_extent_buffer(cur);
403 break;
404 }
405 search_start = tmp->start;
406 last_block = tmp->start;
407 *last_ret = search_start;
408 if (parent_level == 1)
409 btrfs_clear_buffer_defrag(tmp);
410 free_extent_buffer(tmp);
411 }
412 if (parent->map_token) {
413 unmap_extent_buffer(parent, parent->map_token,
414 KM_USER1);
415 parent->map_token = NULL;
416 }
417 return err;
418 }
419 #endif
420
421 /*
422 * The leaf data grows from end-to-front in the node.
423 * this returns the address of the start of the last item,
424 * which is the stop of the leaf data stack
425 */
426 static inline unsigned int leaf_data_end(struct btrfs_root *root,
427 struct extent_buffer *leaf)
428 {
429 u32 nr = btrfs_header_nritems(leaf);
430 if (nr == 0)
431 return BTRFS_LEAF_DATA_SIZE(root);
432 return btrfs_item_offset_nr(leaf, nr - 1);
433 }
434
435 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
436 int level)
437 {
438 struct extent_buffer *parent = NULL;
439 struct extent_buffer *node = path->nodes[level];
440 struct btrfs_disk_key parent_key;
441 struct btrfs_disk_key node_key;
442 int parent_slot;
443 int slot;
444 struct btrfs_key cpukey;
445 u32 nritems = btrfs_header_nritems(node);
446
447 if (path->nodes[level + 1])
448 parent = path->nodes[level + 1];
449
450 slot = path->slots[level];
451 BUG_ON(nritems == 0);
452 if (parent) {
453 parent_slot = path->slots[level + 1];
454 btrfs_node_key(parent, &parent_key, parent_slot);
455 btrfs_node_key(node, &node_key, 0);
456 BUG_ON(memcmp(&parent_key, &node_key,
457 sizeof(struct btrfs_disk_key)));
458 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
459 btrfs_header_bytenr(node));
460 }
461 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
462 if (slot != 0) {
463 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
464 btrfs_node_key(node, &node_key, slot);
465 BUG_ON(btrfs_comp_keys(&node_key, &cpukey) <= 0);
466 }
467 if (slot < nritems - 1) {
468 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
469 btrfs_node_key(node, &node_key, slot);
470 BUG_ON(btrfs_comp_keys(&node_key, &cpukey) >= 0);
471 }
472 return 0;
473 }
474
475 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
476 int level)
477 {
478 struct extent_buffer *leaf = path->nodes[level];
479 struct extent_buffer *parent = NULL;
480 int parent_slot;
481 struct btrfs_key cpukey;
482 struct btrfs_disk_key parent_key;
483 struct btrfs_disk_key leaf_key;
484 int slot = path->slots[0];
485
486 u32 nritems = btrfs_header_nritems(leaf);
487
488 if (path->nodes[level + 1])
489 parent = path->nodes[level + 1];
490
491 if (nritems == 0)
492 return 0;
493
494 if (parent) {
495 parent_slot = path->slots[level + 1];
496 btrfs_node_key(parent, &parent_key, parent_slot);
497 btrfs_item_key(leaf, &leaf_key, 0);
498
499 BUG_ON(memcmp(&parent_key, &leaf_key,
500 sizeof(struct btrfs_disk_key)));
501 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
502 btrfs_header_bytenr(leaf));
503 }
504 #if 0
505 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
506 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
507 btrfs_item_key(leaf, &leaf_key, i);
508 if (comp_keys(&leaf_key, &cpukey) >= 0) {
509 btrfs_print_leaf(root, leaf);
510 printk("slot %d offset bad key\n", i);
511 BUG_ON(1);
512 }
513 if (btrfs_item_offset_nr(leaf, i) !=
514 btrfs_item_end_nr(leaf, i + 1)) {
515 btrfs_print_leaf(root, leaf);
516 printk("slot %d offset bad\n", i);
517 BUG_ON(1);
518 }
519 if (i == 0) {
520 if (btrfs_item_offset_nr(leaf, i) +
521 btrfs_item_size_nr(leaf, i) !=
522 BTRFS_LEAF_DATA_SIZE(root)) {
523 btrfs_print_leaf(root, leaf);
524 printk("slot %d first offset bad\n", i);
525 BUG_ON(1);
526 }
527 }
528 }
529 if (nritems > 0) {
530 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
531 btrfs_print_leaf(root, leaf);
532 printk("slot %d bad size \n", nritems - 1);
533 BUG_ON(1);
534 }
535 }
536 #endif
537 if (slot != 0 && slot < nritems - 1) {
538 btrfs_item_key(leaf, &leaf_key, slot);
539 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
540 if (btrfs_comp_keys(&leaf_key, &cpukey) <= 0) {
541 btrfs_print_leaf(root, leaf);
542 printk("slot %d offset bad key\n", slot);
543 BUG_ON(1);
544 }
545 if (btrfs_item_offset_nr(leaf, slot - 1) !=
546 btrfs_item_end_nr(leaf, slot)) {
547 btrfs_print_leaf(root, leaf);
548 printk("slot %d offset bad\n", slot);
549 BUG_ON(1);
550 }
551 }
552 if (slot < nritems - 1) {
553 btrfs_item_key(leaf, &leaf_key, slot);
554 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
555 BUG_ON(btrfs_comp_keys(&leaf_key, &cpukey) >= 0);
556 if (btrfs_item_offset_nr(leaf, slot) !=
557 btrfs_item_end_nr(leaf, slot + 1)) {
558 btrfs_print_leaf(root, leaf);
559 printk("slot %d offset bad\n", slot);
560 BUG_ON(1);
561 }
562 }
563 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
564 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
565 return 0;
566 }
567
568 static int noinline check_block(struct btrfs_root *root,
569 struct btrfs_path *path, int level)
570 {
571 return 0;
572 #if 0
573 struct extent_buffer *buf = path->nodes[level];
574
575 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
576 (unsigned long)btrfs_header_fsid(buf),
577 BTRFS_FSID_SIZE)) {
578 printk("warning bad block %Lu\n", buf->start);
579 return 1;
580 }
581 #endif
582 if (level == 0)
583 return check_leaf(root, path, level);
584 return check_node(root, path, level);
585 }
586
587 /*
588 * search for key in the extent_buffer. The items start at offset p,
589 * and they are item_size apart. There are 'max' items in p.
590 *
591 * the slot in the array is returned via slot, and it points to
592 * the place where you would insert key if it is not found in
593 * the array.
594 *
595 * slot may point to max if the key is bigger than all of the keys
596 */
597 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
598 int item_size, struct btrfs_key *key,
599 int max, int *slot)
600 {
601 int low = 0;
602 int high = max;
603 int mid;
604 int ret;
605 unsigned long offset;
606 struct btrfs_disk_key *tmp;
607
608 while(low < high) {
609 mid = (low + high) / 2;
610 offset = p + mid * item_size;
611
612 tmp = (struct btrfs_disk_key *)(eb->data + offset);
613 ret = btrfs_comp_keys(tmp, key);
614
615 if (ret < 0)
616 low = mid + 1;
617 else if (ret > 0)
618 high = mid;
619 else {
620 *slot = mid;
621 return 0;
622 }
623 }
624 *slot = low;
625 return 1;
626 }
627
628 /*
629 * simple bin_search frontend that does the right thing for
630 * leaves vs nodes
631 */
632 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
633 int level, int *slot)
634 {
635 if (level == 0) {
636 return generic_bin_search(eb,
637 offsetof(struct btrfs_leaf, items),
638 sizeof(struct btrfs_item),
639 key, btrfs_header_nritems(eb),
640 slot);
641 } else {
642 return generic_bin_search(eb,
643 offsetof(struct btrfs_node, ptrs),
644 sizeof(struct btrfs_key_ptr),
645 key, btrfs_header_nritems(eb),
646 slot);
647 }
648 return -1;
649 }
650
651 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
652 struct extent_buffer *parent, int slot)
653 {
654 if (slot < 0)
655 return NULL;
656 if (slot >= btrfs_header_nritems(parent))
657 return NULL;
658 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
659 btrfs_level_size(root, btrfs_header_level(parent) - 1));
660 }
661
662 static int balance_level(struct btrfs_trans_handle *trans,
663 struct btrfs_root *root,
664 struct btrfs_path *path, int level)
665 {
666 struct extent_buffer *right = NULL;
667 struct extent_buffer *mid;
668 struct extent_buffer *left = NULL;
669 struct extent_buffer *parent = NULL;
670 int ret = 0;
671 int wret;
672 int pslot;
673 int orig_slot = path->slots[level];
674 int err_on_enospc = 0;
675 u64 orig_ptr;
676
677 if (level == 0)
678 return 0;
679
680 mid = path->nodes[level];
681 WARN_ON(btrfs_header_generation(mid) != trans->transid);
682
683 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
684
685 if (level < BTRFS_MAX_LEVEL - 1)
686 parent = path->nodes[level + 1];
687 pslot = path->slots[level + 1];
688
689 /*
690 * deal with the case where there is only one pointer in the root
691 * by promoting the node below to a root
692 */
693 if (!parent) {
694 struct extent_buffer *child;
695
696 if (btrfs_header_nritems(mid) != 1)
697 return 0;
698
699 /* promote the child to a root */
700 child = read_node_slot(root, mid, 0);
701 BUG_ON(!child);
702 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
703 BUG_ON(ret);
704
705 root->node = child;
706 add_root_to_dirty_list(root);
707 path->nodes[level] = NULL;
708 clean_tree_block(trans, root, mid);
709 wait_on_tree_block_writeback(root, mid);
710 /* once for the path */
711 free_extent_buffer(mid);
712 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
713 root->root_key.objectid,
714 btrfs_header_generation(mid), 0, 0, 1);
715 /* once for the root ptr */
716 free_extent_buffer(mid);
717 return ret;
718 }
719 if (btrfs_header_nritems(mid) >
720 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
721 return 0;
722
723 if (btrfs_header_nritems(mid) < 2)
724 err_on_enospc = 1;
725
726 left = read_node_slot(root, parent, pslot - 1);
727 if (left) {
728 wret = btrfs_cow_block(trans, root, left,
729 parent, pslot - 1, &left);
730 if (wret) {
731 ret = wret;
732 goto enospc;
733 }
734 }
735 right = read_node_slot(root, parent, pslot + 1);
736 if (right) {
737 wret = btrfs_cow_block(trans, root, right,
738 parent, pslot + 1, &right);
739 if (wret) {
740 ret = wret;
741 goto enospc;
742 }
743 }
744
745 /* first, try to make some room in the middle buffer */
746 if (left) {
747 orig_slot += btrfs_header_nritems(left);
748 wret = push_node_left(trans, root, left, mid, 0);
749 if (wret < 0)
750 ret = wret;
751 if (btrfs_header_nritems(mid) < 2)
752 err_on_enospc = 1;
753 }
754
755 /*
756 * then try to empty the right most buffer into the middle
757 */
758 if (right) {
759 wret = push_node_left(trans, root, mid, right, 1);
760 if (wret < 0 && wret != -ENOSPC)
761 ret = wret;
762 if (btrfs_header_nritems(right) == 0) {
763 u64 bytenr = right->start;
764 u64 generation = btrfs_header_generation(parent);
765 u32 blocksize = right->len;
766
767 clean_tree_block(trans, root, right);
768 wait_on_tree_block_writeback(root, right);
769 free_extent_buffer(right);
770 right = NULL;
771 wret = del_ptr(trans, root, path, level + 1, pslot +
772 1);
773 if (wret)
774 ret = wret;
775 wret = btrfs_free_extent(trans, root, bytenr,
776 blocksize,
777 btrfs_header_owner(parent),
778 generation, 0, 0, 1);
779 if (wret)
780 ret = wret;
781 } else {
782 struct btrfs_disk_key right_key;
783 btrfs_node_key(right, &right_key, 0);
784 btrfs_set_node_key(parent, &right_key, pslot + 1);
785 btrfs_mark_buffer_dirty(parent);
786 }
787 }
788 if (btrfs_header_nritems(mid) == 1) {
789 /*
790 * we're not allowed to leave a node with one item in the
791 * tree during a delete. A deletion from lower in the tree
792 * could try to delete the only pointer in this node.
793 * So, pull some keys from the left.
794 * There has to be a left pointer at this point because
795 * otherwise we would have pulled some pointers from the
796 * right
797 */
798 BUG_ON(!left);
799 wret = balance_node_right(trans, root, mid, left);
800 if (wret < 0) {
801 ret = wret;
802 goto enospc;
803 }
804 BUG_ON(wret == 1);
805 }
806 if (btrfs_header_nritems(mid) == 0) {
807 /* we've managed to empty the middle node, drop it */
808 u64 root_gen = btrfs_header_generation(parent);
809 u64 bytenr = mid->start;
810 u32 blocksize = mid->len;
811 clean_tree_block(trans, root, mid);
812 wait_on_tree_block_writeback(root, mid);
813 free_extent_buffer(mid);
814 mid = NULL;
815 wret = del_ptr(trans, root, path, level + 1, pslot);
816 if (wret)
817 ret = wret;
818 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
819 btrfs_header_owner(parent),
820 root_gen, 0, 0, 1);
821 if (wret)
822 ret = wret;
823 } else {
824 /* update the parent key to reflect our changes */
825 struct btrfs_disk_key mid_key;
826 btrfs_node_key(mid, &mid_key, 0);
827 btrfs_set_node_key(parent, &mid_key, pslot);
828 btrfs_mark_buffer_dirty(parent);
829 }
830
831 /* update the path */
832 if (left) {
833 if (btrfs_header_nritems(left) > orig_slot) {
834 extent_buffer_get(left);
835 path->nodes[level] = left;
836 path->slots[level + 1] -= 1;
837 path->slots[level] = orig_slot;
838 if (mid)
839 free_extent_buffer(mid);
840 } else {
841 orig_slot -= btrfs_header_nritems(left);
842 path->slots[level] = orig_slot;
843 }
844 }
845 /* double check we haven't messed things up */
846 check_block(root, path, level);
847 if (orig_ptr !=
848 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
849 BUG();
850 enospc:
851 if (right)
852 free_extent_buffer(right);
853 if (left)
854 free_extent_buffer(left);
855 return ret;
856 }
857
858 /* returns zero if the push worked, non-zero otherwise */
859 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
860 struct btrfs_root *root,
861 struct btrfs_path *path, int level)
862 {
863 struct extent_buffer *right = NULL;
864 struct extent_buffer *mid;
865 struct extent_buffer *left = NULL;
866 struct extent_buffer *parent = NULL;
867 int ret = 0;
868 int wret;
869 int pslot;
870 int orig_slot = path->slots[level];
871 u64 orig_ptr;
872
873 if (level == 0)
874 return 1;
875
876 mid = path->nodes[level];
877 WARN_ON(btrfs_header_generation(mid) != trans->transid);
878 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
879
880 if (level < BTRFS_MAX_LEVEL - 1)
881 parent = path->nodes[level + 1];
882 pslot = path->slots[level + 1];
883
884 if (!parent)
885 return 1;
886
887 left = read_node_slot(root, parent, pslot - 1);
888
889 /* first, try to make some room in the middle buffer */
890 if (left) {
891 u32 left_nr;
892 left_nr = btrfs_header_nritems(left);
893 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
894 wret = 1;
895 } else {
896 ret = btrfs_cow_block(trans, root, left, parent,
897 pslot - 1, &left);
898 if (ret)
899 wret = 1;
900 else {
901 wret = push_node_left(trans, root,
902 left, mid, 0);
903 }
904 }
905 if (wret < 0)
906 ret = wret;
907 if (wret == 0) {
908 struct btrfs_disk_key disk_key;
909 orig_slot += left_nr;
910 btrfs_node_key(mid, &disk_key, 0);
911 btrfs_set_node_key(parent, &disk_key, pslot);
912 btrfs_mark_buffer_dirty(parent);
913 if (btrfs_header_nritems(left) > orig_slot) {
914 path->nodes[level] = left;
915 path->slots[level + 1] -= 1;
916 path->slots[level] = orig_slot;
917 free_extent_buffer(mid);
918 } else {
919 orig_slot -=
920 btrfs_header_nritems(left);
921 path->slots[level] = orig_slot;
922 free_extent_buffer(left);
923 }
924 return 0;
925 }
926 free_extent_buffer(left);
927 }
928 right= read_node_slot(root, parent, pslot + 1);
929
930 /*
931 * then try to empty the right most buffer into the middle
932 */
933 if (right) {
934 u32 right_nr;
935 right_nr = btrfs_header_nritems(right);
936 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
937 wret = 1;
938 } else {
939 ret = btrfs_cow_block(trans, root, right,
940 parent, pslot + 1,
941 &right);
942 if (ret)
943 wret = 1;
944 else {
945 wret = balance_node_right(trans, root,
946 right, mid);
947 }
948 }
949 if (wret < 0)
950 ret = wret;
951 if (wret == 0) {
952 struct btrfs_disk_key disk_key;
953
954 btrfs_node_key(right, &disk_key, 0);
955 btrfs_set_node_key(parent, &disk_key, pslot + 1);
956 btrfs_mark_buffer_dirty(parent);
957
958 if (btrfs_header_nritems(mid) <= orig_slot) {
959 path->nodes[level] = right;
960 path->slots[level + 1] += 1;
961 path->slots[level] = orig_slot -
962 btrfs_header_nritems(mid);
963 free_extent_buffer(mid);
964 } else {
965 free_extent_buffer(right);
966 }
967 return 0;
968 }
969 free_extent_buffer(right);
970 }
971 return 1;
972 }
973
974 /*
975 * readahead one full node of leaves
976 */
977 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
978 int level, int slot, u64 objectid)
979 {
980 struct extent_buffer *node;
981 struct btrfs_disk_key disk_key;
982 u32 nritems;
983 u64 search;
984 u64 lowest_read;
985 u64 highest_read;
986 u64 nread = 0;
987 int direction = path->reada;
988 struct extent_buffer *eb;
989 u32 nr;
990 u32 blocksize;
991 u32 nscan = 0;
992
993 if (level != 1)
994 return;
995
996 if (!path->nodes[level])
997 return;
998
999 node = path->nodes[level];
1000 search = btrfs_node_blockptr(node, slot);
1001 blocksize = btrfs_level_size(root, level - 1);
1002 eb = btrfs_find_tree_block(root, search, blocksize);
1003 if (eb) {
1004 free_extent_buffer(eb);
1005 return;
1006 }
1007
1008 highest_read = search;
1009 lowest_read = search;
1010
1011 nritems = btrfs_header_nritems(node);
1012 nr = slot;
1013 while(1) {
1014 if (direction < 0) {
1015 if (nr == 0)
1016 break;
1017 nr--;
1018 } else if (direction > 0) {
1019 nr++;
1020 if (nr >= nritems)
1021 break;
1022 }
1023 if (path->reada < 0 && objectid) {
1024 btrfs_node_key(node, &disk_key, nr);
1025 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1026 break;
1027 }
1028 search = btrfs_node_blockptr(node, nr);
1029 if ((search >= lowest_read && search <= highest_read) ||
1030 (search < lowest_read && lowest_read - search <= 32768) ||
1031 (search > highest_read && search - highest_read <= 32768)) {
1032 readahead_tree_block(root, search, blocksize);
1033 nread += blocksize;
1034 }
1035 nscan++;
1036 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1037 break;
1038 if(nread > (1024 * 1024) || nscan > 128)
1039 break;
1040
1041 if (search < lowest_read)
1042 lowest_read = search;
1043 if (search > highest_read)
1044 highest_read = search;
1045 }
1046 }
1047
1048 /*
1049 * look for key in the tree. path is filled in with nodes along the way
1050 * if key is found, we return zero and you can find the item in the leaf
1051 * level of the path (level 0)
1052 *
1053 * If the key isn't found, the path points to the slot where it should
1054 * be inserted, and 1 is returned. If there are other errors during the
1055 * search a negative error number is returned.
1056 *
1057 * if ins_len > 0, nodes and leaves will be split as we walk down the
1058 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1059 * possible)
1060 */
1061 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1062 *root, struct btrfs_key *key, struct btrfs_path *p, int
1063 ins_len, int cow)
1064 {
1065 struct extent_buffer *b;
1066 u64 bytenr;
1067 u64 ptr_gen;
1068 int slot;
1069 int ret;
1070 int level;
1071 int should_reada = p->reada;
1072 u8 lowest_level = 0;
1073
1074 lowest_level = p->lowest_level;
1075 WARN_ON(lowest_level && ins_len);
1076 WARN_ON(p->nodes[0] != NULL);
1077 /*
1078 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1079 */
1080 again:
1081 b = root->node;
1082 extent_buffer_get(b);
1083 while (b) {
1084 level = btrfs_header_level(b);
1085 if (cow) {
1086 int wret;
1087 wret = btrfs_cow_block(trans, root, b,
1088 p->nodes[level + 1],
1089 p->slots[level + 1],
1090 &b);
1091 if (wret) {
1092 free_extent_buffer(b);
1093 return wret;
1094 }
1095 }
1096 BUG_ON(!cow && ins_len);
1097 if (level != btrfs_header_level(b))
1098 WARN_ON(1);
1099 level = btrfs_header_level(b);
1100 p->nodes[level] = b;
1101 ret = check_block(root, p, level);
1102 if (ret)
1103 return -1;
1104 ret = bin_search(b, key, level, &slot);
1105 if (level != 0) {
1106 if (ret && slot > 0)
1107 slot -= 1;
1108 p->slots[level] = slot;
1109 if (ins_len > 0 && btrfs_header_nritems(b) >=
1110 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1111 int sret = split_node(trans, root, p, level);
1112 BUG_ON(sret > 0);
1113 if (sret)
1114 return sret;
1115 b = p->nodes[level];
1116 slot = p->slots[level];
1117 } else if (ins_len < 0) {
1118 int sret = balance_level(trans, root, p,
1119 level);
1120 if (sret)
1121 return sret;
1122 b = p->nodes[level];
1123 if (!b) {
1124 btrfs_release_path(NULL, p);
1125 goto again;
1126 }
1127 slot = p->slots[level];
1128 BUG_ON(btrfs_header_nritems(b) == 1);
1129 }
1130 /* this is only true while dropping a snapshot */
1131 if (level == lowest_level)
1132 break;
1133 bytenr = btrfs_node_blockptr(b, slot);
1134 ptr_gen = btrfs_node_ptr_generation(b, slot);
1135 if (should_reada)
1136 reada_for_search(root, p, level, slot,
1137 key->objectid);
1138 b = read_tree_block(root, bytenr,
1139 btrfs_level_size(root, level - 1));
1140 if (ptr_gen != btrfs_header_generation(b)) {
1141 printk("block %llu bad gen wanted %llu "
1142 "found %llu\n",
1143 (unsigned long long)b->start,
1144 (unsigned long long)ptr_gen,
1145 (unsigned long long)btrfs_header_generation(b));
1146 }
1147 } else {
1148 p->slots[level] = slot;
1149 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1150 sizeof(struct btrfs_item) + ins_len) {
1151 int sret = split_leaf(trans, root, key,
1152 p, ins_len, ret == 0);
1153 BUG_ON(sret > 0);
1154 if (sret)
1155 return sret;
1156 }
1157 return ret;
1158 }
1159 }
1160 return 1;
1161 }
1162
1163 /*
1164 * adjust the pointers going up the tree, starting at level
1165 * making sure the right key of each node is points to 'key'.
1166 * This is used after shifting pointers to the left, so it stops
1167 * fixing up pointers when a given leaf/node is not in slot 0 of the
1168 * higher levels
1169 *
1170 * If this fails to write a tree block, it returns -1, but continues
1171 * fixing up the blocks in ram so the tree is consistent.
1172 */
1173 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1174 struct btrfs_root *root, struct btrfs_path *path,
1175 struct btrfs_disk_key *key, int level)
1176 {
1177 int i;
1178 int ret = 0;
1179 struct extent_buffer *t;
1180
1181 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1182 int tslot = path->slots[i];
1183 if (!path->nodes[i])
1184 break;
1185 t = path->nodes[i];
1186 btrfs_set_node_key(t, key, tslot);
1187 btrfs_mark_buffer_dirty(path->nodes[i]);
1188 if (tslot != 0)
1189 break;
1190 }
1191 return ret;
1192 }
1193
1194 /*
1195 * try to push data from one node into the next node left in the
1196 * tree.
1197 *
1198 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1199 * error, and > 0 if there was no room in the left hand block.
1200 */
1201 static int push_node_left(struct btrfs_trans_handle *trans,
1202 struct btrfs_root *root, struct extent_buffer *dst,
1203 struct extent_buffer *src, int empty)
1204 {
1205 int push_items = 0;
1206 int src_nritems;
1207 int dst_nritems;
1208 int ret = 0;
1209
1210 src_nritems = btrfs_header_nritems(src);
1211 dst_nritems = btrfs_header_nritems(dst);
1212 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1213 WARN_ON(btrfs_header_generation(src) != trans->transid);
1214 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1215
1216 if (!empty && src_nritems <= 2)
1217 return 1;
1218
1219 if (push_items <= 0) {
1220 return 1;
1221 }
1222
1223 if (empty)
1224 push_items = min(src_nritems, push_items);
1225 else
1226 push_items = min(src_nritems - 2, push_items);
1227
1228 copy_extent_buffer(dst, src,
1229 btrfs_node_key_ptr_offset(dst_nritems),
1230 btrfs_node_key_ptr_offset(0),
1231 push_items * sizeof(struct btrfs_key_ptr));
1232
1233 if (push_items < src_nritems) {
1234 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1235 btrfs_node_key_ptr_offset(push_items),
1236 (src_nritems - push_items) *
1237 sizeof(struct btrfs_key_ptr));
1238 }
1239 btrfs_set_header_nritems(src, src_nritems - push_items);
1240 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1241 btrfs_mark_buffer_dirty(src);
1242 btrfs_mark_buffer_dirty(dst);
1243 return ret;
1244 }
1245
1246 /*
1247 * try to push data from one node into the next node right in the
1248 * tree.
1249 *
1250 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1251 * error, and > 0 if there was no room in the right hand block.
1252 *
1253 * this will only push up to 1/2 the contents of the left node over
1254 */
1255 static int balance_node_right(struct btrfs_trans_handle *trans,
1256 struct btrfs_root *root,
1257 struct extent_buffer *dst,
1258 struct extent_buffer *src)
1259 {
1260 int push_items = 0;
1261 int max_push;
1262 int src_nritems;
1263 int dst_nritems;
1264 int ret = 0;
1265
1266 WARN_ON(btrfs_header_generation(src) != trans->transid);
1267 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1268
1269 src_nritems = btrfs_header_nritems(src);
1270 dst_nritems = btrfs_header_nritems(dst);
1271 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1272 if (push_items <= 0)
1273 return 1;
1274
1275 max_push = src_nritems / 2 + 1;
1276 /* don't try to empty the node */
1277 if (max_push >= src_nritems)
1278 return 1;
1279
1280 if (max_push < push_items)
1281 push_items = max_push;
1282
1283 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1284 btrfs_node_key_ptr_offset(0),
1285 (dst_nritems) *
1286 sizeof(struct btrfs_key_ptr));
1287
1288 copy_extent_buffer(dst, src,
1289 btrfs_node_key_ptr_offset(0),
1290 btrfs_node_key_ptr_offset(src_nritems - push_items),
1291 push_items * sizeof(struct btrfs_key_ptr));
1292
1293 btrfs_set_header_nritems(src, src_nritems - push_items);
1294 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1295
1296 btrfs_mark_buffer_dirty(src);
1297 btrfs_mark_buffer_dirty(dst);
1298 return ret;
1299 }
1300
1301 /*
1302 * helper function to insert a new root level in the tree.
1303 * A new node is allocated, and a single item is inserted to
1304 * point to the existing root
1305 *
1306 * returns zero on success or < 0 on failure.
1307 */
1308 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1309 struct btrfs_root *root,
1310 struct btrfs_path *path, int level)
1311 {
1312 u64 root_gen;
1313 u64 lower_gen;
1314 struct extent_buffer *lower;
1315 struct extent_buffer *c;
1316 struct btrfs_disk_key lower_key;
1317
1318 BUG_ON(path->nodes[level]);
1319 BUG_ON(path->nodes[level-1] != root->node);
1320
1321 if (root->ref_cows)
1322 root_gen = trans->transid;
1323 else
1324 root_gen = 0;
1325
1326 lower = path->nodes[level-1];
1327 if (level == 1)
1328 btrfs_item_key(lower, &lower_key, 0);
1329 else
1330 btrfs_node_key(lower, &lower_key, 0);
1331
1332 c = __btrfs_alloc_free_block(trans, root, root->nodesize,
1333 root->root_key.objectid,
1334 root_gen, lower_key.objectid, level,
1335 root->node->start, 0);
1336 if (IS_ERR(c))
1337 return PTR_ERR(c);
1338 memset_extent_buffer(c, 0, 0, root->nodesize);
1339 btrfs_set_header_nritems(c, 1);
1340 btrfs_set_header_level(c, level);
1341 btrfs_set_header_bytenr(c, c->start);
1342 btrfs_set_header_generation(c, trans->transid);
1343 btrfs_set_header_owner(c, root->root_key.objectid);
1344
1345 write_extent_buffer(c, root->fs_info->fsid,
1346 (unsigned long)btrfs_header_fsid(c),
1347 BTRFS_FSID_SIZE);
1348
1349 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1350 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1351 BTRFS_UUID_SIZE);
1352
1353 btrfs_set_node_key(c, &lower_key, 0);
1354 btrfs_set_node_blockptr(c, 0, lower->start);
1355 lower_gen = btrfs_header_generation(lower);
1356 WARN_ON(lower_gen == 0);
1357
1358 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1359
1360 btrfs_mark_buffer_dirty(c);
1361
1362 /* the super has an extra ref to root->node */
1363 free_extent_buffer(root->node);
1364 root->node = c;
1365 add_root_to_dirty_list(root);
1366 extent_buffer_get(c);
1367 path->nodes[level] = c;
1368 path->slots[level] = 0;
1369
1370 if (root->ref_cows && lower_gen != trans->transid) {
1371 struct btrfs_path *back_path = btrfs_alloc_path();
1372 int ret;
1373 ret = btrfs_insert_extent_backref(trans,
1374 root->fs_info->extent_root,
1375 path, lower->start,
1376 root->root_key.objectid,
1377 trans->transid, 0, 0);
1378 BUG_ON(ret);
1379 btrfs_free_path(back_path);
1380 }
1381 return 0;
1382 }
1383
1384 /*
1385 * worker function to insert a single pointer in a node.
1386 * the node should have enough room for the pointer already
1387 *
1388 * slot and level indicate where you want the key to go, and
1389 * blocknr is the block the key points to.
1390 *
1391 * returns zero on success and < 0 on any error
1392 */
1393 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1394 *root, struct btrfs_path *path, struct btrfs_disk_key
1395 *key, u64 bytenr, int slot, int level)
1396 {
1397 struct extent_buffer *lower;
1398 int nritems;
1399
1400 BUG_ON(!path->nodes[level]);
1401 lower = path->nodes[level];
1402 nritems = btrfs_header_nritems(lower);
1403 if (slot > nritems)
1404 BUG();
1405 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1406 BUG();
1407 if (slot != nritems) {
1408 memmove_extent_buffer(lower,
1409 btrfs_node_key_ptr_offset(slot + 1),
1410 btrfs_node_key_ptr_offset(slot),
1411 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1412 }
1413 btrfs_set_node_key(lower, key, slot);
1414 btrfs_set_node_blockptr(lower, slot, bytenr);
1415 WARN_ON(trans->transid == 0);
1416 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1417 btrfs_set_header_nritems(lower, nritems + 1);
1418 btrfs_mark_buffer_dirty(lower);
1419 return 0;
1420 }
1421
1422 /*
1423 * split the node at the specified level in path in two.
1424 * The path is corrected to point to the appropriate node after the split
1425 *
1426 * Before splitting this tries to make some room in the node by pushing
1427 * left and right, if either one works, it returns right away.
1428 *
1429 * returns 0 on success and < 0 on failure
1430 */
1431 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1432 *root, struct btrfs_path *path, int level)
1433 {
1434 u64 root_gen;
1435 struct extent_buffer *c;
1436 struct extent_buffer *split;
1437 struct btrfs_disk_key disk_key;
1438 int mid;
1439 int ret;
1440 int wret;
1441 u32 c_nritems;
1442
1443 c = path->nodes[level];
1444 WARN_ON(btrfs_header_generation(c) != trans->transid);
1445 if (c == root->node) {
1446 /* trying to split the root, lets make a new one */
1447 ret = insert_new_root(trans, root, path, level + 1);
1448 if (ret)
1449 return ret;
1450 } else {
1451 ret = push_nodes_for_insert(trans, root, path, level);
1452 c = path->nodes[level];
1453 if (!ret && btrfs_header_nritems(c) <
1454 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1455 return 0;
1456 if (ret < 0)
1457 return ret;
1458 }
1459
1460 c_nritems = btrfs_header_nritems(c);
1461 if (root->ref_cows)
1462 root_gen = trans->transid;
1463 else
1464 root_gen = 0;
1465
1466 btrfs_node_key(c, &disk_key, 0);
1467 split = __btrfs_alloc_free_block(trans, root, root->nodesize,
1468 root->root_key.objectid,
1469 root_gen,
1470 btrfs_disk_key_objectid(&disk_key),
1471 level, c->start, 0);
1472 if (IS_ERR(split))
1473 return PTR_ERR(split);
1474
1475 btrfs_set_header_flags(split, btrfs_header_flags(c));
1476 btrfs_set_header_level(split, btrfs_header_level(c));
1477 btrfs_set_header_bytenr(split, split->start);
1478 btrfs_set_header_generation(split, trans->transid);
1479 btrfs_set_header_owner(split, root->root_key.objectid);
1480 btrfs_set_header_flags(split, 0);
1481 write_extent_buffer(split, root->fs_info->fsid,
1482 (unsigned long)btrfs_header_fsid(split),
1483 BTRFS_FSID_SIZE);
1484 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1485 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1486 BTRFS_UUID_SIZE);
1487
1488 mid = (c_nritems + 1) / 2;
1489
1490 copy_extent_buffer(split, c,
1491 btrfs_node_key_ptr_offset(0),
1492 btrfs_node_key_ptr_offset(mid),
1493 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1494 btrfs_set_header_nritems(split, c_nritems - mid);
1495 btrfs_set_header_nritems(c, mid);
1496 ret = 0;
1497
1498 btrfs_mark_buffer_dirty(c);
1499 btrfs_mark_buffer_dirty(split);
1500
1501 btrfs_node_key(split, &disk_key, 0);
1502 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1503 path->slots[level + 1] + 1,
1504 level + 1);
1505 if (wret)
1506 ret = wret;
1507
1508 if (path->slots[level] >= mid) {
1509 path->slots[level] -= mid;
1510 free_extent_buffer(c);
1511 path->nodes[level] = split;
1512 path->slots[level + 1] += 1;
1513 } else {
1514 free_extent_buffer(split);
1515 }
1516 return ret;
1517 }
1518
1519 /*
1520 * how many bytes are required to store the items in a leaf. start
1521 * and nr indicate which items in the leaf to check. This totals up the
1522 * space used both by the item structs and the item data
1523 */
1524 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1525 {
1526 int data_len;
1527 int nritems = btrfs_header_nritems(l);
1528 int end = min(nritems, start + nr) - 1;
1529
1530 if (!nr)
1531 return 0;
1532 data_len = btrfs_item_end_nr(l, start);
1533 data_len = data_len - btrfs_item_offset_nr(l, end);
1534 data_len += sizeof(struct btrfs_item) * nr;
1535 WARN_ON(data_len < 0);
1536 return data_len;
1537 }
1538
1539 /*
1540 * The space between the end of the leaf items and
1541 * the start of the leaf data. IOW, how much room
1542 * the leaf has left for both items and data
1543 */
1544 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1545 {
1546 int nritems = btrfs_header_nritems(leaf);
1547 int ret;
1548 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1549 if (ret < 0) {
1550 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1551 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1552 leaf_space_used(leaf, 0, nritems), nritems);
1553 }
1554 return ret;
1555 }
1556
1557 /*
1558 * push some data in the path leaf to the right, trying to free up at
1559 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1560 *
1561 * returns 1 if the push failed because the other node didn't have enough
1562 * room, 0 if everything worked out and < 0 if there were major errors.
1563 */
1564 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1565 *root, struct btrfs_path *path, int data_size,
1566 int empty)
1567 {
1568 struct extent_buffer *left = path->nodes[0];
1569 struct extent_buffer *right;
1570 struct extent_buffer *upper;
1571 struct btrfs_disk_key disk_key;
1572 int slot;
1573 u32 i;
1574 int free_space;
1575 int push_space = 0;
1576 int push_items = 0;
1577 struct btrfs_item *item;
1578 u32 left_nritems;
1579 u32 nr;
1580 u32 right_nritems;
1581 u32 data_end;
1582 u32 this_item_size;
1583 int ret;
1584
1585 slot = path->slots[1];
1586 if (!path->nodes[1]) {
1587 return 1;
1588 }
1589 upper = path->nodes[1];
1590 if (slot >= btrfs_header_nritems(upper) - 1)
1591 return 1;
1592
1593 right = read_tree_block(root, btrfs_node_blockptr(upper, slot + 1),
1594 root->leafsize);
1595 free_space = btrfs_leaf_free_space(root, right);
1596 if (free_space < data_size + sizeof(struct btrfs_item)) {
1597 free_extent_buffer(right);
1598 return 1;
1599 }
1600
1601 /* cow and double check */
1602 ret = btrfs_cow_block(trans, root, right, upper,
1603 slot + 1, &right);
1604 if (ret) {
1605 free_extent_buffer(right);
1606 return 1;
1607 }
1608 free_space = btrfs_leaf_free_space(root, right);
1609 if (free_space < data_size + sizeof(struct btrfs_item)) {
1610 free_extent_buffer(right);
1611 return 1;
1612 }
1613
1614 left_nritems = btrfs_header_nritems(left);
1615 if (left_nritems == 0) {
1616 free_extent_buffer(right);
1617 return 1;
1618 }
1619
1620 if (empty)
1621 nr = 0;
1622 else
1623 nr = 1;
1624
1625 i = left_nritems - 1;
1626 while (i >= nr) {
1627 item = btrfs_item_nr(left, i);
1628
1629 if (path->slots[0] == i)
1630 push_space += data_size + sizeof(*item);
1631
1632 this_item_size = btrfs_item_size(left, item);
1633 if (this_item_size + sizeof(*item) + push_space > free_space)
1634 break;
1635 push_items++;
1636 push_space += this_item_size + sizeof(*item);
1637 if (i == 0)
1638 break;
1639 i--;
1640 }
1641
1642 if (push_items == 0) {
1643 free_extent_buffer(right);
1644 return 1;
1645 }
1646
1647 if (!empty && push_items == left_nritems)
1648 WARN_ON(1);
1649
1650 /* push left to right */
1651 right_nritems = btrfs_header_nritems(right);
1652
1653 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1654 push_space -= leaf_data_end(root, left);
1655
1656 /* make room in the right data area */
1657 data_end = leaf_data_end(root, right);
1658 memmove_extent_buffer(right,
1659 btrfs_leaf_data(right) + data_end - push_space,
1660 btrfs_leaf_data(right) + data_end,
1661 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1662
1663 /* copy from the left data area */
1664 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1665 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1666 btrfs_leaf_data(left) + leaf_data_end(root, left),
1667 push_space);
1668
1669 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1670 btrfs_item_nr_offset(0),
1671 right_nritems * sizeof(struct btrfs_item));
1672
1673 /* copy the items from left to right */
1674 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1675 btrfs_item_nr_offset(left_nritems - push_items),
1676 push_items * sizeof(struct btrfs_item));
1677
1678 /* update the item pointers */
1679 right_nritems += push_items;
1680 btrfs_set_header_nritems(right, right_nritems);
1681 push_space = BTRFS_LEAF_DATA_SIZE(root);
1682 for (i = 0; i < right_nritems; i++) {
1683 item = btrfs_item_nr(right, i);
1684 push_space -= btrfs_item_size(right, item);
1685 btrfs_set_item_offset(right, item, push_space);
1686 }
1687
1688 left_nritems -= push_items;
1689 btrfs_set_header_nritems(left, left_nritems);
1690
1691 if (left_nritems)
1692 btrfs_mark_buffer_dirty(left);
1693 btrfs_mark_buffer_dirty(right);
1694
1695 btrfs_item_key(right, &disk_key, 0);
1696 btrfs_set_node_key(upper, &disk_key, slot + 1);
1697 btrfs_mark_buffer_dirty(upper);
1698
1699 /* then fixup the leaf pointer in the path */
1700 if (path->slots[0] >= left_nritems) {
1701 path->slots[0] -= left_nritems;
1702 free_extent_buffer(path->nodes[0]);
1703 path->nodes[0] = right;
1704 path->slots[1] += 1;
1705 } else {
1706 free_extent_buffer(right);
1707 }
1708 return 0;
1709 }
1710 /*
1711 * push some data in the path leaf to the left, trying to free up at
1712 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1713 */
1714 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1715 *root, struct btrfs_path *path, int data_size,
1716 int empty)
1717 {
1718 struct btrfs_disk_key disk_key;
1719 struct extent_buffer *right = path->nodes[0];
1720 struct extent_buffer *left;
1721 int slot;
1722 int i;
1723 int free_space;
1724 int push_space = 0;
1725 int push_items = 0;
1726 struct btrfs_item *item;
1727 u32 old_left_nritems;
1728 u32 right_nritems;
1729 u32 nr;
1730 int ret = 0;
1731 int wret;
1732 u32 this_item_size;
1733 u32 old_left_item_size;
1734
1735 slot = path->slots[1];
1736 if (slot == 0)
1737 return 1;
1738 if (!path->nodes[1])
1739 return 1;
1740
1741 right_nritems = btrfs_header_nritems(right);
1742 if (right_nritems == 0) {
1743 return 1;
1744 }
1745
1746 left = read_tree_block(root, btrfs_node_blockptr(path->nodes[1],
1747 slot - 1), root->leafsize);
1748 free_space = btrfs_leaf_free_space(root, left);
1749 if (free_space < data_size + sizeof(struct btrfs_item)) {
1750 free_extent_buffer(left);
1751 return 1;
1752 }
1753
1754 /* cow and double check */
1755 ret = btrfs_cow_block(trans, root, left,
1756 path->nodes[1], slot - 1, &left);
1757 if (ret) {
1758 /* we hit -ENOSPC, but it isn't fatal here */
1759 free_extent_buffer(left);
1760 return 1;
1761 }
1762
1763 free_space = btrfs_leaf_free_space(root, left);
1764 if (free_space < data_size + sizeof(struct btrfs_item)) {
1765 free_extent_buffer(left);
1766 return 1;
1767 }
1768
1769 if (empty)
1770 nr = right_nritems;
1771 else
1772 nr = right_nritems - 1;
1773
1774 for (i = 0; i < nr; i++) {
1775 item = btrfs_item_nr(right, i);
1776
1777 if (path->slots[0] == i)
1778 push_space += data_size + sizeof(*item);
1779
1780 this_item_size = btrfs_item_size(right, item);
1781 if (this_item_size + sizeof(*item) + push_space > free_space)
1782 break;
1783
1784 push_items++;
1785 push_space += this_item_size + sizeof(*item);
1786 }
1787
1788 if (push_items == 0) {
1789 free_extent_buffer(left);
1790 return 1;
1791 }
1792 if (!empty && push_items == btrfs_header_nritems(right))
1793 WARN_ON(1);
1794
1795 /* push data from right to left */
1796 copy_extent_buffer(left, right,
1797 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1798 btrfs_item_nr_offset(0),
1799 push_items * sizeof(struct btrfs_item));
1800
1801 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1802 btrfs_item_offset_nr(right, push_items -1);
1803
1804 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1805 leaf_data_end(root, left) - push_space,
1806 btrfs_leaf_data(right) +
1807 btrfs_item_offset_nr(right, push_items - 1),
1808 push_space);
1809 old_left_nritems = btrfs_header_nritems(left);
1810 BUG_ON(old_left_nritems < 0);
1811
1812 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1813 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1814 u32 ioff;
1815
1816 item = btrfs_item_nr(left, i);
1817 ioff = btrfs_item_offset(left, item);
1818 btrfs_set_item_offset(left, item,
1819 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1820 }
1821 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1822
1823 /* fixup right node */
1824 if (push_items > right_nritems) {
1825 printk("push items %d nr %u\n", push_items, right_nritems);
1826 WARN_ON(1);
1827 }
1828
1829 if (push_items < right_nritems) {
1830 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1831 leaf_data_end(root, right);
1832 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1833 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1834 btrfs_leaf_data(right) +
1835 leaf_data_end(root, right), push_space);
1836
1837 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1838 btrfs_item_nr_offset(push_items),
1839 (btrfs_header_nritems(right) - push_items) *
1840 sizeof(struct btrfs_item));
1841 }
1842 right_nritems -= push_items;
1843 btrfs_set_header_nritems(right, right_nritems);
1844 push_space = BTRFS_LEAF_DATA_SIZE(root);
1845 for (i = 0; i < right_nritems; i++) {
1846 item = btrfs_item_nr(right, i);
1847 push_space = push_space - btrfs_item_size(right, item);
1848 btrfs_set_item_offset(right, item, push_space);
1849 }
1850
1851 btrfs_mark_buffer_dirty(left);
1852 if (right_nritems)
1853 btrfs_mark_buffer_dirty(right);
1854
1855 btrfs_item_key(right, &disk_key, 0);
1856 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
1857 if (wret)
1858 ret = wret;
1859
1860 /* then fixup the leaf pointer in the path */
1861 if (path->slots[0] < push_items) {
1862 path->slots[0] += old_left_nritems;
1863 free_extent_buffer(path->nodes[0]);
1864 path->nodes[0] = left;
1865 path->slots[1] -= 1;
1866 } else {
1867 free_extent_buffer(left);
1868 path->slots[0] -= push_items;
1869 }
1870 BUG_ON(path->slots[0] < 0);
1871 return ret;
1872 }
1873
1874 /*
1875 * split the path's leaf in two, making sure there is at least data_size
1876 * available for the resulting leaf level of the path.
1877 *
1878 * returns 0 if all went well and < 0 on failure.
1879 */
1880 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
1881 *root, struct btrfs_key *ins_key,
1882 struct btrfs_path *path, int data_size, int extend)
1883 {
1884 u64 root_gen;
1885 struct extent_buffer *l;
1886 u32 nritems;
1887 int mid;
1888 int slot;
1889 struct extent_buffer *right;
1890 int space_needed = data_size + sizeof(struct btrfs_item);
1891 int data_copy_size;
1892 int rt_data_off;
1893 int i;
1894 int ret = 0;
1895 int wret;
1896 int double_split;
1897 int num_doubles = 0;
1898 struct btrfs_disk_key disk_key;
1899
1900 if (extend)
1901 space_needed = data_size;
1902
1903 if (root->ref_cows)
1904 root_gen = trans->transid;
1905 else
1906 root_gen = 0;
1907
1908 /* first try to make some room by pushing left and right */
1909 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
1910 wret = push_leaf_right(trans, root, path, data_size, 0);
1911 if (wret < 0) {
1912 return wret;
1913 }
1914 if (wret) {
1915 wret = push_leaf_left(trans, root, path, data_size, 0);
1916 if (wret < 0)
1917 return wret;
1918 }
1919 l = path->nodes[0];
1920
1921 /* did the pushes work? */
1922 if (btrfs_leaf_free_space(root, l) >= space_needed)
1923 return 0;
1924 }
1925
1926 if (!path->nodes[1]) {
1927 ret = insert_new_root(trans, root, path, 1);
1928 if (ret)
1929 return ret;
1930 }
1931 again:
1932 double_split = 0;
1933 l = path->nodes[0];
1934 slot = path->slots[0];
1935 nritems = btrfs_header_nritems(l);
1936 mid = (nritems + 1)/ 2;
1937
1938 btrfs_item_key(l, &disk_key, 0);
1939
1940 right = __btrfs_alloc_free_block(trans, root, root->leafsize,
1941 root->root_key.objectid,
1942 root_gen, disk_key.objectid, 0,
1943 l->start, 0);
1944 if (IS_ERR(right)) {
1945 BUG_ON(1);
1946 return PTR_ERR(right);
1947 }
1948
1949 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
1950 btrfs_set_header_bytenr(right, right->start);
1951 btrfs_set_header_generation(right, trans->transid);
1952 btrfs_set_header_owner(right, root->root_key.objectid);
1953 btrfs_set_header_level(right, 0);
1954 write_extent_buffer(right, root->fs_info->fsid,
1955 (unsigned long)btrfs_header_fsid(right),
1956 BTRFS_FSID_SIZE);
1957
1958 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
1959 (unsigned long)btrfs_header_chunk_tree_uuid(right),
1960 BTRFS_UUID_SIZE);
1961 if (mid <= slot) {
1962 if (nritems == 1 ||
1963 leaf_space_used(l, mid, nritems - mid) + space_needed >
1964 BTRFS_LEAF_DATA_SIZE(root)) {
1965 if (slot >= nritems) {
1966 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1967 btrfs_set_header_nritems(right, 0);
1968 wret = insert_ptr(trans, root, path,
1969 &disk_key, right->start,
1970 path->slots[1] + 1, 1);
1971 if (wret)
1972 ret = wret;
1973 free_extent_buffer(path->nodes[0]);
1974 path->nodes[0] = right;
1975 path->slots[0] = 0;
1976 path->slots[1] += 1;
1977 return ret;
1978 }
1979 mid = slot;
1980 if (mid != nritems &&
1981 leaf_space_used(l, mid, nritems - mid) +
1982 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
1983 double_split = 1;
1984 }
1985 }
1986 } else {
1987 if (leaf_space_used(l, 0, mid + 1) + space_needed >
1988 BTRFS_LEAF_DATA_SIZE(root)) {
1989 if (!extend && slot == 0) {
1990 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1991 btrfs_set_header_nritems(right, 0);
1992 wret = insert_ptr(trans, root, path,
1993 &disk_key,
1994 right->start,
1995 path->slots[1], 1);
1996 if (wret)
1997 ret = wret;
1998 free_extent_buffer(path->nodes[0]);
1999 path->nodes[0] = right;
2000 path->slots[0] = 0;
2001 if (path->slots[1] == 0) {
2002 wret = fixup_low_keys(trans, root,
2003 path, &disk_key, 1);
2004 if (wret)
2005 ret = wret;
2006 }
2007 return ret;
2008 } else if (extend && slot == 0) {
2009 mid = 1;
2010 } else {
2011 mid = slot;
2012 if (mid != nritems &&
2013 leaf_space_used(l, mid, nritems - mid) +
2014 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2015 double_split = 1;
2016 }
2017 }
2018 }
2019 }
2020 nritems = nritems - mid;
2021 btrfs_set_header_nritems(right, nritems);
2022 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2023
2024 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2025 btrfs_item_nr_offset(mid),
2026 nritems * sizeof(struct btrfs_item));
2027
2028 copy_extent_buffer(right, l,
2029 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2030 data_copy_size, btrfs_leaf_data(l) +
2031 leaf_data_end(root, l), data_copy_size);
2032
2033 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2034 btrfs_item_end_nr(l, mid);
2035
2036 for (i = 0; i < nritems; i++) {
2037 struct btrfs_item *item = btrfs_item_nr(right, i);
2038 u32 ioff = btrfs_item_offset(right, item);
2039 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2040 }
2041
2042 btrfs_set_header_nritems(l, mid);
2043 ret = 0;
2044 btrfs_item_key(right, &disk_key, 0);
2045 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2046 path->slots[1] + 1, 1);
2047 if (wret)
2048 ret = wret;
2049
2050 btrfs_mark_buffer_dirty(right);
2051 btrfs_mark_buffer_dirty(l);
2052 BUG_ON(path->slots[0] != slot);
2053
2054 if (mid <= slot) {
2055 free_extent_buffer(path->nodes[0]);
2056 path->nodes[0] = right;
2057 path->slots[0] -= mid;
2058 path->slots[1] += 1;
2059 } else
2060 free_extent_buffer(right);
2061
2062 BUG_ON(path->slots[0] < 0);
2063
2064 if (double_split) {
2065 BUG_ON(num_doubles != 0);
2066 num_doubles++;
2067 goto again;
2068 }
2069 return ret;
2070 }
2071
2072 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2073 struct btrfs_root *root,
2074 struct btrfs_path *path,
2075 u32 new_size, int from_end)
2076 {
2077 int ret = 0;
2078 int slot;
2079 int slot_orig;
2080 struct extent_buffer *leaf;
2081 struct btrfs_item *item;
2082 u32 nritems;
2083 unsigned int data_end;
2084 unsigned int old_data_start;
2085 unsigned int old_size;
2086 unsigned int size_diff;
2087 int i;
2088
2089 slot_orig = path->slots[0];
2090 leaf = path->nodes[0];
2091 slot = path->slots[0];
2092
2093 old_size = btrfs_item_size_nr(leaf, slot);
2094 if (old_size == new_size)
2095 return 0;
2096
2097 nritems = btrfs_header_nritems(leaf);
2098 data_end = leaf_data_end(root, leaf);
2099
2100 old_data_start = btrfs_item_offset_nr(leaf, slot);
2101
2102 size_diff = old_size - new_size;
2103
2104 BUG_ON(slot < 0);
2105 BUG_ON(slot >= nritems);
2106
2107 /*
2108 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2109 */
2110 /* first correct the data pointers */
2111 for (i = slot; i < nritems; i++) {
2112 u32 ioff;
2113 item = btrfs_item_nr(leaf, i);
2114 ioff = btrfs_item_offset(leaf, item);
2115 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2116 }
2117
2118 /* shift the data */
2119 if (from_end) {
2120 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2121 data_end + size_diff, btrfs_leaf_data(leaf) +
2122 data_end, old_data_start + new_size - data_end);
2123 } else {
2124 struct btrfs_disk_key disk_key;
2125 u64 offset;
2126
2127 btrfs_item_key(leaf, &disk_key, slot);
2128
2129 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2130 unsigned long ptr;
2131 struct btrfs_file_extent_item *fi;
2132
2133 fi = btrfs_item_ptr(leaf, slot,
2134 struct btrfs_file_extent_item);
2135 fi = (struct btrfs_file_extent_item *)(
2136 (unsigned long)fi - size_diff);
2137
2138 if (btrfs_file_extent_type(leaf, fi) ==
2139 BTRFS_FILE_EXTENT_INLINE) {
2140 ptr = btrfs_item_ptr_offset(leaf, slot);
2141 memmove_extent_buffer(leaf, ptr,
2142 (unsigned long)fi,
2143 offsetof(struct btrfs_file_extent_item,
2144 disk_bytenr));
2145 }
2146 }
2147
2148 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2149 data_end + size_diff, btrfs_leaf_data(leaf) +
2150 data_end, old_data_start - data_end);
2151
2152 offset = btrfs_disk_key_offset(&disk_key);
2153 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2154 btrfs_set_item_key(leaf, &disk_key, slot);
2155 if (slot == 0)
2156 fixup_low_keys(trans, root, path, &disk_key, 1);
2157 }
2158
2159 item = btrfs_item_nr(leaf, slot);
2160 btrfs_set_item_size(leaf, item, new_size);
2161 btrfs_mark_buffer_dirty(leaf);
2162
2163 ret = 0;
2164 if (btrfs_leaf_free_space(root, leaf) < 0) {
2165 btrfs_print_leaf(root, leaf);
2166 BUG();
2167 }
2168 return ret;
2169 }
2170
2171 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2172 struct btrfs_root *root, struct btrfs_path *path,
2173 u32 data_size)
2174 {
2175 int ret = 0;
2176 int slot;
2177 int slot_orig;
2178 struct extent_buffer *leaf;
2179 struct btrfs_item *item;
2180 u32 nritems;
2181 unsigned int data_end;
2182 unsigned int old_data;
2183 unsigned int old_size;
2184 int i;
2185
2186 slot_orig = path->slots[0];
2187 leaf = path->nodes[0];
2188
2189 nritems = btrfs_header_nritems(leaf);
2190 data_end = leaf_data_end(root, leaf);
2191
2192 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2193 btrfs_print_leaf(root, leaf);
2194 BUG();
2195 }
2196 slot = path->slots[0];
2197 old_data = btrfs_item_end_nr(leaf, slot);
2198
2199 BUG_ON(slot < 0);
2200 if (slot >= nritems) {
2201 btrfs_print_leaf(root, leaf);
2202 printk("slot %d too large, nritems %d\n", slot, nritems);
2203 BUG_ON(1);
2204 }
2205
2206 /*
2207 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2208 */
2209 /* first correct the data pointers */
2210 for (i = slot; i < nritems; i++) {
2211 u32 ioff;
2212 item = btrfs_item_nr(leaf, i);
2213 ioff = btrfs_item_offset(leaf, item);
2214 btrfs_set_item_offset(leaf, item, ioff - data_size);
2215 }
2216
2217 /* shift the data */
2218 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2219 data_end - data_size, btrfs_leaf_data(leaf) +
2220 data_end, old_data - data_end);
2221
2222 data_end = old_data;
2223 old_size = btrfs_item_size_nr(leaf, slot);
2224 item = btrfs_item_nr(leaf, slot);
2225 btrfs_set_item_size(leaf, item, old_size + data_size);
2226 btrfs_mark_buffer_dirty(leaf);
2227
2228 ret = 0;
2229 if (btrfs_leaf_free_space(root, leaf) < 0) {
2230 btrfs_print_leaf(root, leaf);
2231 BUG();
2232 }
2233 return ret;
2234 }
2235
2236 /*
2237 * Given a key and some data, insert an item into the tree.
2238 * This does all the path init required, making room in the tree if needed.
2239 */
2240 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2241 struct btrfs_root *root,
2242 struct btrfs_path *path,
2243 struct btrfs_key *cpu_key, u32 *data_size,
2244 int nr)
2245 {
2246 struct extent_buffer *leaf;
2247 struct btrfs_item *item;
2248 int ret = 0;
2249 int slot;
2250 int slot_orig;
2251 int i;
2252 u32 nritems;
2253 u32 total_size = 0;
2254 u32 total_data = 0;
2255 unsigned int data_end;
2256 struct btrfs_disk_key disk_key;
2257
2258 for (i = 0; i < nr; i++) {
2259 total_data += data_size[i];
2260 }
2261
2262 /* create a root if there isn't one */
2263 if (!root->node)
2264 BUG();
2265
2266 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2267 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2268 if (ret == 0) {
2269 return -EEXIST;
2270 }
2271 if (ret < 0)
2272 goto out;
2273
2274 slot_orig = path->slots[0];
2275 leaf = path->nodes[0];
2276
2277 nritems = btrfs_header_nritems(leaf);
2278 data_end = leaf_data_end(root, leaf);
2279
2280 if (btrfs_leaf_free_space(root, leaf) <
2281 sizeof(struct btrfs_item) + total_size) {
2282 btrfs_print_leaf(root, leaf);
2283 printk("not enough freespace need %u have %d\n",
2284 total_size, btrfs_leaf_free_space(root, leaf));
2285 BUG();
2286 }
2287
2288 slot = path->slots[0];
2289 BUG_ON(slot < 0);
2290
2291 if (slot != nritems) {
2292 int i;
2293 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2294
2295 if (old_data < data_end) {
2296 btrfs_print_leaf(root, leaf);
2297 printk("slot %d old_data %d data_end %d\n",
2298 slot, old_data, data_end);
2299 BUG_ON(1);
2300 }
2301 /*
2302 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2303 */
2304 /* first correct the data pointers */
2305 for (i = slot; i < nritems; i++) {
2306 u32 ioff;
2307
2308 item = btrfs_item_nr(leaf, i);
2309 ioff = btrfs_item_offset(leaf, item);
2310 btrfs_set_item_offset(leaf, item, ioff - total_data);
2311 }
2312
2313 /* shift the items */
2314 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2315 btrfs_item_nr_offset(slot),
2316 (nritems - slot) * sizeof(struct btrfs_item));
2317
2318 /* shift the data */
2319 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2320 data_end - total_data, btrfs_leaf_data(leaf) +
2321 data_end, old_data - data_end);
2322 data_end = old_data;
2323 }
2324
2325 /* setup the item for the new data */
2326 for (i = 0; i < nr; i++) {
2327 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2328 btrfs_set_item_key(leaf, &disk_key, slot + i);
2329 item = btrfs_item_nr(leaf, slot + i);
2330 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2331 data_end -= data_size[i];
2332 btrfs_set_item_size(leaf, item, data_size[i]);
2333 }
2334 btrfs_set_header_nritems(leaf, nritems + nr);
2335 btrfs_mark_buffer_dirty(leaf);
2336
2337 ret = 0;
2338 if (slot == 0) {
2339 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2340 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2341 }
2342
2343 if (btrfs_leaf_free_space(root, leaf) < 0) {
2344 btrfs_print_leaf(root, leaf);
2345 BUG();
2346 }
2347
2348 out:
2349 return ret;
2350 }
2351
2352 /*
2353 * Given a key and some data, insert an item into the tree.
2354 * This does all the path init required, making room in the tree if needed.
2355 */
2356 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2357 *root, struct btrfs_key *cpu_key, void *data, u32
2358 data_size)
2359 {
2360 int ret = 0;
2361 struct btrfs_path *path;
2362 struct extent_buffer *leaf;
2363 unsigned long ptr;
2364
2365 path = btrfs_alloc_path();
2366 BUG_ON(!path);
2367 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2368 if (!ret) {
2369 leaf = path->nodes[0];
2370 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2371 write_extent_buffer(leaf, data, ptr, data_size);
2372 btrfs_mark_buffer_dirty(leaf);
2373 }
2374 btrfs_free_path(path);
2375 return ret;
2376 }
2377
2378 /*
2379 * delete the pointer from a given node.
2380 *
2381 * If the delete empties a node, the node is removed from the tree,
2382 * continuing all the way the root if required. The root is converted into
2383 * a leaf if all the nodes are emptied.
2384 */
2385 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2386 struct btrfs_path *path, int level, int slot)
2387 {
2388 struct extent_buffer *parent = path->nodes[level];
2389 u32 nritems;
2390 int ret = 0;
2391 int wret;
2392
2393 nritems = btrfs_header_nritems(parent);
2394 if (slot != nritems -1) {
2395 memmove_extent_buffer(parent,
2396 btrfs_node_key_ptr_offset(slot),
2397 btrfs_node_key_ptr_offset(slot + 1),
2398 sizeof(struct btrfs_key_ptr) *
2399 (nritems - slot - 1));
2400 }
2401 nritems--;
2402 btrfs_set_header_nritems(parent, nritems);
2403 if (nritems == 0 && parent == root->node) {
2404 BUG_ON(btrfs_header_level(root->node) != 1);
2405 /* just turn the root into a leaf and break */
2406 btrfs_set_header_level(root->node, 0);
2407 } else if (slot == 0) {
2408 struct btrfs_disk_key disk_key;
2409
2410 btrfs_node_key(parent, &disk_key, 0);
2411 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2412 if (wret)
2413 ret = wret;
2414 }
2415 btrfs_mark_buffer_dirty(parent);
2416 return ret;
2417 }
2418
2419 /*
2420 * delete the item at the leaf level in path. If that empties
2421 * the leaf, remove it from the tree
2422 */
2423 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2424 struct btrfs_path *path, int slot, int nr)
2425 {
2426 struct extent_buffer *leaf;
2427 struct btrfs_item *item;
2428 int last_off;
2429 int dsize = 0;
2430 int ret = 0;
2431 int wret;
2432 int i;
2433 u32 nritems;
2434
2435 leaf = path->nodes[0];
2436 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2437
2438 for (i = 0; i < nr; i++)
2439 dsize += btrfs_item_size_nr(leaf, slot + i);
2440
2441 nritems = btrfs_header_nritems(leaf);
2442
2443 if (slot + nr != nritems) {
2444 int i;
2445 int data_end = leaf_data_end(root, leaf);
2446
2447 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2448 data_end + dsize,
2449 btrfs_leaf_data(leaf) + data_end,
2450 last_off - data_end);
2451
2452 for (i = slot + nr; i < nritems; i++) {
2453 u32 ioff;
2454
2455 item = btrfs_item_nr(leaf, i);
2456 ioff = btrfs_item_offset(leaf, item);
2457 btrfs_set_item_offset(leaf, item, ioff + dsize);
2458 }
2459
2460 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2461 btrfs_item_nr_offset(slot + nr),
2462 sizeof(struct btrfs_item) *
2463 (nritems - slot - nr));
2464 }
2465 btrfs_set_header_nritems(leaf, nritems - nr);
2466 nritems -= nr;
2467
2468 /* delete the leaf if we've emptied it */
2469 if (nritems == 0) {
2470 if (leaf == root->node) {
2471 btrfs_set_header_level(leaf, 0);
2472 } else {
2473 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2474 clean_tree_block(trans, root, leaf);
2475 wait_on_tree_block_writeback(root, leaf);
2476 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2477 if (wret)
2478 ret = wret;
2479 wret = btrfs_free_extent(trans, root,
2480 leaf->start, leaf->len,
2481 btrfs_header_owner(path->nodes[1]),
2482 root_gen, 0, 0, 1);
2483 if (wret)
2484 ret = wret;
2485 }
2486 } else {
2487 int used = leaf_space_used(leaf, 0, nritems);
2488 if (slot == 0) {
2489 struct btrfs_disk_key disk_key;
2490
2491 btrfs_item_key(leaf, &disk_key, 0);
2492 wret = fixup_low_keys(trans, root, path,
2493 &disk_key, 1);
2494 if (wret)
2495 ret = wret;
2496 }
2497
2498 /* delete the leaf if it is mostly empty */
2499 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2500 /* push_leaf_left fixes the path.
2501 * make sure the path still points to our leaf
2502 * for possible call to del_ptr below
2503 */
2504 slot = path->slots[1];
2505 extent_buffer_get(leaf);
2506
2507 wret = push_leaf_left(trans, root, path, 1, 1);
2508 if (wret < 0 && wret != -ENOSPC)
2509 ret = wret;
2510
2511 if (path->nodes[0] == leaf &&
2512 btrfs_header_nritems(leaf)) {
2513 wret = push_leaf_right(trans, root, path, 1, 1);
2514 if (wret < 0 && wret != -ENOSPC)
2515 ret = wret;
2516 }
2517
2518 if (btrfs_header_nritems(leaf) == 0) {
2519 u64 root_gen;
2520 u64 bytenr = leaf->start;
2521 u32 blocksize = leaf->len;
2522
2523 root_gen = btrfs_header_generation(
2524 path->nodes[1]);
2525
2526 clean_tree_block(trans, root, leaf);
2527 wait_on_tree_block_writeback(root, leaf);
2528
2529 wret = del_ptr(trans, root, path, 1, slot);
2530 if (wret)
2531 ret = wret;
2532
2533 free_extent_buffer(leaf);
2534 wret = btrfs_free_extent(trans, root, bytenr,
2535 blocksize,
2536 btrfs_header_owner(path->nodes[1]),
2537 root_gen, 0, 0, 1);
2538 if (wret)
2539 ret = wret;
2540 } else {
2541 btrfs_mark_buffer_dirty(leaf);
2542 free_extent_buffer(leaf);
2543 }
2544 } else {
2545 btrfs_mark_buffer_dirty(leaf);
2546 }
2547 }
2548 return ret;
2549 }
2550
2551 /*
2552 * walk up the tree as far as required to find the previous leaf.
2553 * returns 0 if it found something or 1 if there are no lesser leaves.
2554 * returns < 0 on io errors.
2555 */
2556 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2557 {
2558 u64 bytenr;
2559 int slot;
2560 int level = 1;
2561 struct extent_buffer *c;
2562 struct extent_buffer *next = NULL;
2563
2564 while(level < BTRFS_MAX_LEVEL) {
2565 if (!path->nodes[level])
2566 return 1;
2567
2568 slot = path->slots[level];
2569 c = path->nodes[level];
2570 if (slot == 0) {
2571 level++;
2572 if (level == BTRFS_MAX_LEVEL)
2573 return 1;
2574 continue;
2575 }
2576 slot--;
2577
2578 bytenr = btrfs_node_blockptr(c, slot);
2579 if (next)
2580 free_extent_buffer(next);
2581
2582 next = read_tree_block(root, bytenr,
2583 btrfs_level_size(root, level - 1));
2584 break;
2585 }
2586 path->slots[level] = slot;
2587 while(1) {
2588 level--;
2589 c = path->nodes[level];
2590 free_extent_buffer(c);
2591 slot = btrfs_header_nritems(next);
2592 if (slot != 0)
2593 slot--;
2594 path->nodes[level] = next;
2595 path->slots[level] = slot;
2596 if (!level)
2597 break;
2598 next = read_tree_block(root, btrfs_node_blockptr(next, slot),
2599 btrfs_level_size(root, level - 1));
2600 }
2601 return 0;
2602 }
2603
2604 /*
2605 * walk up the tree as far as required to find the next leaf.
2606 * returns 0 if it found something or 1 if there are no greater leaves.
2607 * returns < 0 on io errors.
2608 */
2609 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2610 {
2611 int slot;
2612 int level = 1;
2613 u64 bytenr;
2614 struct extent_buffer *c;
2615 struct extent_buffer *next = NULL;
2616
2617 while(level < BTRFS_MAX_LEVEL) {
2618 if (!path->nodes[level])
2619 return 1;
2620
2621 slot = path->slots[level] + 1;
2622 c = path->nodes[level];
2623 if (slot >= btrfs_header_nritems(c)) {
2624 level++;
2625 if (level == BTRFS_MAX_LEVEL)
2626 return 1;
2627 continue;
2628 }
2629
2630 bytenr = btrfs_node_blockptr(c, slot);
2631 if (next)
2632 free_extent_buffer(next);
2633
2634 if (path->reada)
2635 reada_for_search(root, path, level, slot, 0);
2636
2637 next = read_tree_block(root, bytenr,
2638 btrfs_level_size(root, level -1));
2639 break;
2640 }
2641 path->slots[level] = slot;
2642 while(1) {
2643 level--;
2644 c = path->nodes[level];
2645 free_extent_buffer(c);
2646 path->nodes[level] = next;
2647 path->slots[level] = 0;
2648 if (!level)
2649 break;
2650 if (path->reada)
2651 reada_for_search(root, path, level, 0, 0);
2652 next = read_tree_block(root, btrfs_node_blockptr(next, 0),
2653 btrfs_level_size(root, level - 1));
2654 }
2655 return 0;
2656 }
2657
2658 int btrfs_previous_item(struct btrfs_root *root,
2659 struct btrfs_path *path, u64 min_objectid,
2660 int type)
2661 {
2662 struct btrfs_key found_key;
2663 struct extent_buffer *leaf;
2664 int ret;
2665
2666 while(1) {
2667 if (path->slots[0] == 0) {
2668 ret = btrfs_prev_leaf(root, path);
2669 if (ret != 0)
2670 return ret;
2671 } else {
2672 path->slots[0]--;
2673 }
2674 leaf = path->nodes[0];
2675 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2676 if (found_key.type == type)
2677 return 0;
2678 }
2679 return 1;
2680 }
2681
Btrfs-progs-unstable mirror