?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ioctls to scan for btrfs filesystems
[btrfs-progs-unstable.git] / ctree.c
blob88ebd9ec0c13ba2735c4959af1d77654b7bb6d65
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);
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
126 WARN_ON(btrfs_header_generation(buf) > trans->transid);
127 ret = btrfs_inc_ref(trans, new_root, buf);
128 kfree(new_root);
129
130 if (ret)
131 return ret;
132
133 btrfs_mark_buffer_dirty(cow);
134 *cow_ret = cow;
135 return 0;
136 }
137
138 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
139 struct btrfs_root *root,
140 struct extent_buffer *buf,
141 struct extent_buffer *parent, int parent_slot,
142 struct extent_buffer **cow_ret,
143 u64 search_start, u64 empty_size)
144 {
145 u64 root_gen;
146 struct extent_buffer *cow;
147 u32 nritems;
148 int ret = 0;
149 int different_trans = 0;
150 int level;
151 struct btrfs_key first_key;
152
153 if (root->ref_cows) {
154 root_gen = trans->transid;
155 } else {
156 root_gen = 0;
157 }
158
159 WARN_ON(root->ref_cows && trans->transid !=
160 root->fs_info->running_transaction->transid);
161 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
162
163 level = btrfs_header_level(buf);
164 nritems = btrfs_header_nritems(buf);
165 if (nritems) {
166 if (level == 0)
167 btrfs_item_key_to_cpu(buf, &first_key, 0);
168 else
169 btrfs_node_key_to_cpu(buf, &first_key, 0);
170 } else {
171 first_key.objectid = 0;
172 }
173 cow = __btrfs_alloc_free_block(trans, root, buf->len,
174 root->root_key.objectid,
175 root_gen, first_key.objectid, level,
176 search_start, empty_size);
177 if (IS_ERR(cow))
178 return PTR_ERR(cow);
179
180 copy_extent_buffer(cow, buf, 0, 0, cow->len);
181 btrfs_set_header_bytenr(cow, cow->start);
182 btrfs_set_header_generation(cow, trans->transid);
183 btrfs_set_header_owner(cow, root->root_key.objectid);
184
185 WARN_ON(btrfs_header_generation(buf) > trans->transid);
186 if (btrfs_header_generation(buf) != trans->transid) {
187 different_trans = 1;
188 ret = btrfs_inc_ref(trans, root, buf);
189 if (ret)
190 return ret;
191 } else {
192 clean_tree_block(trans, root, buf);
193 }
194
195 if (buf == root->node) {
196 root_gen = btrfs_header_generation(buf);
197 root->node = cow;
198 extent_buffer_get(cow);
199 if (buf != root->commit_root) {
200 btrfs_free_extent(trans, root, buf->start,
201 buf->len, root->root_key.objectid,
202 root_gen, 0, 0, 1);
203 }
204 free_extent_buffer(buf);
205 add_root_to_dirty_list(root);
206 } else {
207 root_gen = btrfs_header_generation(parent);
208 btrfs_set_node_blockptr(parent, parent_slot,
209 cow->start);
210 WARN_ON(trans->transid == 0);
211 btrfs_set_node_ptr_generation(parent, parent_slot,
212 trans->transid);
213 btrfs_mark_buffer_dirty(parent);
214 WARN_ON(btrfs_header_generation(parent) != trans->transid);
215 btrfs_free_extent(trans, root, buf->start, buf->len,
216 btrfs_header_owner(parent), root_gen,
217 0, 0, 1);
218 }
219 free_extent_buffer(buf);
220 btrfs_mark_buffer_dirty(cow);
221 *cow_ret = cow;
222 return 0;
223 }
224
225 int btrfs_cow_block(struct btrfs_trans_handle *trans,
226 struct btrfs_root *root, struct extent_buffer *buf,
227 struct extent_buffer *parent, int parent_slot,
228 struct extent_buffer **cow_ret)
229 {
230 u64 search_start;
231 int ret;
232 /*
233 if (trans->transaction != root->fs_info->running_transaction) {
234 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
235 root->fs_info->running_transaction->transid);
236 WARN_ON(1);
237 }
238 */
239 if (trans->transid != root->fs_info->generation) {
240 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
241 root->fs_info->generation);
242 WARN_ON(1);
243 }
244 if (btrfs_header_generation(buf) == trans->transid) {
245 *cow_ret = buf;
246 return 0;
247 }
248
249 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
250 ret = __btrfs_cow_block(trans, root, buf, parent,
251 parent_slot, cow_ret, search_start, 0);
252 return ret;
253 }
254
255 /*
256 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
257 {
258 if (blocknr < other && other - (blocknr + blocksize) < 32768)
259 return 1;
260 if (blocknr > other && blocknr - (other + blocksize) < 32768)
261 return 1;
262 return 0;
263 }
264 */
265
266 /*
267 * compare two keys in a memcmp fashion
268 */
269 int btrfs_comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
270 {
271 struct btrfs_key k1;
272
273 btrfs_disk_key_to_cpu(&k1, disk);
274
275 if (k1.objectid > k2->objectid)
276 return 1;
277 if (k1.objectid < k2->objectid)
278 return -1;
279 if (k1.type > k2->type)
280 return 1;
281 if (k1.type < k2->type)
282 return -1;
283 if (k1.offset > k2->offset)
284 return 1;
285 if (k1.offset < k2->offset)
286 return -1;
287 return 0;
288 }
289
290
291 #if 0
292 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
293 struct btrfs_root *root, struct extent_buffer *parent,
294 int start_slot, int cache_only, u64 *last_ret,
295 struct btrfs_key *progress)
296 {
297 struct extent_buffer *cur;
298 struct extent_buffer *tmp;
299 u64 blocknr;
300 u64 search_start = *last_ret;
301 u64 last_block = 0;
302 u64 other;
303 u32 parent_nritems;
304 int end_slot;
305 int i;
306 int err = 0;
307 int parent_level;
308 int uptodate;
309 u32 blocksize;
310 int progress_passed = 0;
311 struct btrfs_disk_key disk_key;
312
313 parent_level = btrfs_header_level(parent);
314 if (cache_only && parent_level != 1)
315 return 0;
316
317 if (trans->transaction != root->fs_info->running_transaction) {
318 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
319 root->fs_info->running_transaction->transid);
320 WARN_ON(1);
321 }
322 if (trans->transid != root->fs_info->generation) {
323 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
324 root->fs_info->generation);
325 WARN_ON(1);
326 }
327
328 parent_nritems = btrfs_header_nritems(parent);
329 blocksize = btrfs_level_size(root, parent_level - 1);
330 end_slot = parent_nritems;
331
332 if (parent_nritems == 1)
333 return 0;
334
335 for (i = start_slot; i < end_slot; i++) {
336 int close = 1;
337
338 if (!parent->map_token) {
339 map_extent_buffer(parent,
340 btrfs_node_key_ptr_offset(i),
341 sizeof(struct btrfs_key_ptr),
342 &parent->map_token, &parent->kaddr,
343 &parent->map_start, &parent->map_len,
344 KM_USER1);
345 }
346 btrfs_node_key(parent, &disk_key, i);
347 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
348 continue;
349
350 progress_passed = 1;
351 blocknr = btrfs_node_blockptr(parent, i);
352 if (last_block == 0)
353 last_block = blocknr;
354
355 if (i > 0) {
356 other = btrfs_node_blockptr(parent, i - 1);
357 close = close_blocks(blocknr, other, blocksize);
358 }
359 if (close && i < end_slot - 2) {
360 other = btrfs_node_blockptr(parent, i + 1);
361 close = close_blocks(blocknr, other, blocksize);
362 }
363 if (close) {
364 last_block = blocknr;
365 continue;
366 }
367 if (parent->map_token) {
368 unmap_extent_buffer(parent, parent->map_token,
369 KM_USER1);
370 parent->map_token = NULL;
371 }
372
373 cur = btrfs_find_tree_block(root, blocknr, blocksize);
374 if (cur)
375 uptodate = btrfs_buffer_uptodate(cur);
376 else
377 uptodate = 0;
378 if (!cur || !uptodate) {
379 if (cache_only) {
380 free_extent_buffer(cur);
381 continue;
382 }
383 if (!cur) {
384 cur = read_tree_block(root, blocknr,
385 blocksize);
386 } else if (!uptodate) {
387 btrfs_read_buffer(cur);
388 }
389 }
390 if (search_start == 0)
391 search_start = last_block;
392
393 err = __btrfs_cow_block(trans, root, cur, parent, i,
394 &tmp, search_start,
395 min(16 * blocksize,
396 (end_slot - i) * blocksize));
397 if (err) {
398 free_extent_buffer(cur);
399 break;
400 }
401 search_start = tmp->start;
402 last_block = tmp->start;
403 *last_ret = search_start;
404 if (parent_level == 1)
405 btrfs_clear_buffer_defrag(tmp);
406 free_extent_buffer(tmp);
407 }
408 if (parent->map_token) {
409 unmap_extent_buffer(parent, parent->map_token,
410 KM_USER1);
411 parent->map_token = NULL;
412 }
413 return err;
414 }
415 #endif
416
417 /*
418 * The leaf data grows from end-to-front in the node.
419 * this returns the address of the start of the last item,
420 * which is the stop of the leaf data stack
421 */
422 static inline unsigned int leaf_data_end(struct btrfs_root *root,
423 struct extent_buffer *leaf)
424 {
425 u32 nr = btrfs_header_nritems(leaf);
426 if (nr == 0)
427 return BTRFS_LEAF_DATA_SIZE(root);
428 return btrfs_item_offset_nr(leaf, nr - 1);
429 }
430
431 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
432 int level)
433 {
434 struct extent_buffer *parent = NULL;
435 struct extent_buffer *node = path->nodes[level];
436 struct btrfs_disk_key parent_key;
437 struct btrfs_disk_key node_key;
438 int parent_slot;
439 int slot;
440 struct btrfs_key cpukey;
441 u32 nritems = btrfs_header_nritems(node);
442
443 if (path->nodes[level + 1])
444 parent = path->nodes[level + 1];
445
446 slot = path->slots[level];
447 BUG_ON(nritems == 0);
448 if (parent) {
449 parent_slot = path->slots[level + 1];
450 btrfs_node_key(parent, &parent_key, parent_slot);
451 btrfs_node_key(node, &node_key, 0);
452 BUG_ON(memcmp(&parent_key, &node_key,
453 sizeof(struct btrfs_disk_key)));
454 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
455 btrfs_header_bytenr(node));
456 }
457 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
458 if (slot != 0) {
459 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
460 btrfs_node_key(node, &node_key, slot);
461 BUG_ON(btrfs_comp_keys(&node_key, &cpukey) <= 0);
462 }
463 if (slot < nritems - 1) {
464 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
465 btrfs_node_key(node, &node_key, slot);
466 BUG_ON(btrfs_comp_keys(&node_key, &cpukey) >= 0);
467 }
468 return 0;
469 }
470
471 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
472 int level)
473 {
474 struct extent_buffer *leaf = path->nodes[level];
475 struct extent_buffer *parent = NULL;
476 int parent_slot;
477 struct btrfs_key cpukey;
478 struct btrfs_disk_key parent_key;
479 struct btrfs_disk_key leaf_key;
480 int slot = path->slots[0];
481
482 u32 nritems = btrfs_header_nritems(leaf);
483
484 if (path->nodes[level + 1])
485 parent = path->nodes[level + 1];
486
487 if (nritems == 0)
488 return 0;
489
490 if (parent) {
491 parent_slot = path->slots[level + 1];
492 btrfs_node_key(parent, &parent_key, parent_slot);
493 btrfs_item_key(leaf, &leaf_key, 0);
494
495 BUG_ON(memcmp(&parent_key, &leaf_key,
496 sizeof(struct btrfs_disk_key)));
497 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
498 btrfs_header_bytenr(leaf));
499 }
500 #if 0
501 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
502 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
503 btrfs_item_key(leaf, &leaf_key, i);
504 if (comp_keys(&leaf_key, &cpukey) >= 0) {
505 btrfs_print_leaf(root, leaf);
506 printk("slot %d offset bad key\n", i);
507 BUG_ON(1);
508 }
509 if (btrfs_item_offset_nr(leaf, i) !=
510 btrfs_item_end_nr(leaf, i + 1)) {
511 btrfs_print_leaf(root, leaf);
512 printk("slot %d offset bad\n", i);
513 BUG_ON(1);
514 }
515 if (i == 0) {
516 if (btrfs_item_offset_nr(leaf, i) +
517 btrfs_item_size_nr(leaf, i) !=
518 BTRFS_LEAF_DATA_SIZE(root)) {
519 btrfs_print_leaf(root, leaf);
520 printk("slot %d first offset bad\n", i);
521 BUG_ON(1);
522 }
523 }
524 }
525 if (nritems > 0) {
526 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
527 btrfs_print_leaf(root, leaf);
528 printk("slot %d bad size \n", nritems - 1);
529 BUG_ON(1);
530 }
531 }
532 #endif
533 if (slot != 0 && slot < nritems - 1) {
534 btrfs_item_key(leaf, &leaf_key, slot);
535 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
536 if (btrfs_comp_keys(&leaf_key, &cpukey) <= 0) {
537 btrfs_print_leaf(root, leaf);
538 printk("slot %d offset bad key\n", slot);
539 BUG_ON(1);
540 }
541 if (btrfs_item_offset_nr(leaf, slot - 1) !=
542 btrfs_item_end_nr(leaf, slot)) {
543 btrfs_print_leaf(root, leaf);
544 printk("slot %d offset bad\n", slot);
545 BUG_ON(1);
546 }
547 }
548 if (slot < nritems - 1) {
549 btrfs_item_key(leaf, &leaf_key, slot);
550 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
551 BUG_ON(btrfs_comp_keys(&leaf_key, &cpukey) >= 0);
552 if (btrfs_item_offset_nr(leaf, slot) !=
553 btrfs_item_end_nr(leaf, slot + 1)) {
554 btrfs_print_leaf(root, leaf);
555 printk("slot %d offset bad\n", slot);
556 BUG_ON(1);
557 }
558 }
559 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
560 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
561 return 0;
562 }
563
564 static int noinline check_block(struct btrfs_root *root,
565 struct btrfs_path *path, int level)
566 {
567 return 0;
568 #if 0
569 struct extent_buffer *buf = path->nodes[level];
570
571 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
572 (unsigned long)btrfs_header_fsid(buf),
573 BTRFS_FSID_SIZE)) {
574 printk("warning bad block %Lu\n", buf->start);
575 return 1;
576 }
577 #endif
578 if (level == 0)
579 return check_leaf(root, path, level);
580 return check_node(root, path, level);
581 }
582
583 /*
584 * search for key in the extent_buffer. The items start at offset p,
585 * and they are item_size apart. There are 'max' items in p.
586 *
587 * the slot in the array is returned via slot, and it points to
588 * the place where you would insert key if it is not found in
589 * the array.
590 *
591 * slot may point to max if the key is bigger than all of the keys
592 */
593 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
594 int item_size, struct btrfs_key *key,
595 int max, int *slot)
596 {
597 int low = 0;
598 int high = max;
599 int mid;
600 int ret;
601 unsigned long offset;
602 struct btrfs_disk_key *tmp;
603
604 while(low < high) {
605 mid = (low + high) / 2;
606 offset = p + mid * item_size;
607
608 tmp = (struct btrfs_disk_key *)(eb->data + offset);
609 ret = btrfs_comp_keys(tmp, key);
610
611 if (ret < 0)
612 low = mid + 1;
613 else if (ret > 0)
614 high = mid;
615 else {
616 *slot = mid;
617 return 0;
618 }
619 }
620 *slot = low;
621 return 1;
622 }
623
624 /*
625 * simple bin_search frontend that does the right thing for
626 * leaves vs nodes
627 */
628 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
629 int level, int *slot)
630 {
631 if (level == 0) {
632 return generic_bin_search(eb,
633 offsetof(struct btrfs_leaf, items),
634 sizeof(struct btrfs_item),
635 key, btrfs_header_nritems(eb),
636 slot);
637 } else {
638 return generic_bin_search(eb,
639 offsetof(struct btrfs_node, ptrs),
640 sizeof(struct btrfs_key_ptr),
641 key, btrfs_header_nritems(eb),
642 slot);
643 }
644 return -1;
645 }
646
647 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
648 struct extent_buffer *parent, int slot)
649 {
650 if (slot < 0)
651 return NULL;
652 if (slot >= btrfs_header_nritems(parent))
653 return NULL;
654 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
655 btrfs_level_size(root, btrfs_header_level(parent) - 1));
656 }
657
658 static int balance_level(struct btrfs_trans_handle *trans,
659 struct btrfs_root *root,
660 struct btrfs_path *path, int level)
661 {
662 struct extent_buffer *right = NULL;
663 struct extent_buffer *mid;
664 struct extent_buffer *left = NULL;
665 struct extent_buffer *parent = NULL;
666 int ret = 0;
667 int wret;
668 int pslot;
669 int orig_slot = path->slots[level];
670 int err_on_enospc = 0;
671 u64 orig_ptr;
672
673 if (level == 0)
674 return 0;
675
676 mid = path->nodes[level];
677 WARN_ON(btrfs_header_generation(mid) != trans->transid);
678
679 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
680
681 if (level < BTRFS_MAX_LEVEL - 1)
682 parent = path->nodes[level + 1];
683 pslot = path->slots[level + 1];
684
685 /*
686 * deal with the case where there is only one pointer in the root
687 * by promoting the node below to a root
688 */
689 if (!parent) {
690 struct extent_buffer *child;
691
692 if (btrfs_header_nritems(mid) != 1)
693 return 0;
694
695 /* promote the child to a root */
696 child = read_node_slot(root, mid, 0);
697 BUG_ON(!child);
698 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
699 BUG_ON(ret);
700
701 root->node = child;
702 add_root_to_dirty_list(root);
703 path->nodes[level] = NULL;
704 clean_tree_block(trans, root, mid);
705 wait_on_tree_block_writeback(root, mid);
706 /* once for the path */
707 free_extent_buffer(mid);
708 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
709 root->root_key.objectid,
710 btrfs_header_generation(mid), 0, 0, 1);
711 /* once for the root ptr */
712 free_extent_buffer(mid);
713 return ret;
714 }
715 if (btrfs_header_nritems(mid) >
716 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
717 return 0;
718
719 if (btrfs_header_nritems(mid) < 2)
720 err_on_enospc = 1;
721
722 left = read_node_slot(root, parent, pslot - 1);
723 if (left) {
724 wret = btrfs_cow_block(trans, root, left,
725 parent, pslot - 1, &left);
726 if (wret) {
727 ret = wret;
728 goto enospc;
729 }
730 }
731 right = read_node_slot(root, parent, pslot + 1);
732 if (right) {
733 wret = btrfs_cow_block(trans, root, right,
734 parent, pslot + 1, &right);
735 if (wret) {
736 ret = wret;
737 goto enospc;
738 }
739 }
740
741 /* first, try to make some room in the middle buffer */
742 if (left) {
743 orig_slot += btrfs_header_nritems(left);
744 wret = push_node_left(trans, root, left, mid);
745 if (wret < 0)
746 ret = wret;
747 if (btrfs_header_nritems(mid) < 2)
748 err_on_enospc = 1;
749 }
750
751 /*
752 * then try to empty the right most buffer into the middle
753 */
754 if (right) {
755 wret = push_node_left(trans, root, mid, right);
756 if (wret < 0 && wret != -ENOSPC)
757 ret = wret;
758 if (btrfs_header_nritems(right) == 0) {
759 u64 bytenr = right->start;
760 u64 generation = btrfs_header_generation(parent);
761 u32 blocksize = right->len;
762
763 clean_tree_block(trans, root, right);
764 wait_on_tree_block_writeback(root, right);
765 free_extent_buffer(right);
766 right = NULL;
767 wret = del_ptr(trans, root, path, level + 1, pslot +
768 1);
769 if (wret)
770 ret = wret;
771 wret = btrfs_free_extent(trans, root, bytenr,
772 blocksize,
773 btrfs_header_owner(parent),
774 generation, 0, 0, 1);
775 if (wret)
776 ret = wret;
777 } else {
778 struct btrfs_disk_key right_key;
779 btrfs_node_key(right, &right_key, 0);
780 btrfs_set_node_key(parent, &right_key, pslot + 1);
781 btrfs_mark_buffer_dirty(parent);
782 }
783 }
784 if (btrfs_header_nritems(mid) == 1) {
785 /*
786 * we're not allowed to leave a node with one item in the
787 * tree during a delete. A deletion from lower in the tree
788 * could try to delete the only pointer in this node.
789 * So, pull some keys from the left.
790 * There has to be a left pointer at this point because
791 * otherwise we would have pulled some pointers from the
792 * right
793 */
794 BUG_ON(!left);
795 wret = balance_node_right(trans, root, mid, left);
796 if (wret < 0) {
797 ret = wret;
798 goto enospc;
799 }
800 BUG_ON(wret == 1);
801 }
802 if (btrfs_header_nritems(mid) == 0) {
803 /* we've managed to empty the middle node, drop it */
804 u64 root_gen = btrfs_header_generation(parent);
805 u64 bytenr = mid->start;
806 u32 blocksize = mid->len;
807 clean_tree_block(trans, root, mid);
808 wait_on_tree_block_writeback(root, mid);
809 free_extent_buffer(mid);
810 mid = NULL;
811 wret = del_ptr(trans, root, path, level + 1, pslot);
812 if (wret)
813 ret = wret;
814 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
815 btrfs_header_owner(parent),
816 root_gen, 0, 0, 1);
817 if (wret)
818 ret = wret;
819 } else {
820 /* update the parent key to reflect our changes */
821 struct btrfs_disk_key mid_key;
822 btrfs_node_key(mid, &mid_key, 0);
823 btrfs_set_node_key(parent, &mid_key, pslot);
824 btrfs_mark_buffer_dirty(parent);
825 }
826
827 /* update the path */
828 if (left) {
829 if (btrfs_header_nritems(left) > orig_slot) {
830 extent_buffer_get(left);
831 path->nodes[level] = left;
832 path->slots[level + 1] -= 1;
833 path->slots[level] = orig_slot;
834 if (mid)
835 free_extent_buffer(mid);
836 } else {
837 orig_slot -= btrfs_header_nritems(left);
838 path->slots[level] = orig_slot;
839 }
840 }
841 /* double check we haven't messed things up */
842 check_block(root, path, level);
843 if (orig_ptr !=
844 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
845 BUG();
846 enospc:
847 if (right)
848 free_extent_buffer(right);
849 if (left)
850 free_extent_buffer(left);
851 return ret;
852 }
853
854 /* returns zero if the push worked, non-zero otherwise */
855 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
856 struct btrfs_root *root,
857 struct btrfs_path *path, int level)
858 {
859 struct extent_buffer *right = NULL;
860 struct extent_buffer *mid;
861 struct extent_buffer *left = NULL;
862 struct extent_buffer *parent = NULL;
863 int ret = 0;
864 int wret;
865 int pslot;
866 int orig_slot = path->slots[level];
867 u64 orig_ptr;
868
869 if (level == 0)
870 return 1;
871
872 mid = path->nodes[level];
873 WARN_ON(btrfs_header_generation(mid) != trans->transid);
874 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
875
876 if (level < BTRFS_MAX_LEVEL - 1)
877 parent = path->nodes[level + 1];
878 pslot = path->slots[level + 1];
879
880 if (!parent)
881 return 1;
882
883 left = read_node_slot(root, parent, pslot - 1);
884
885 /* first, try to make some room in the middle buffer */
886 if (left) {
887 u32 left_nr;
888 left_nr = btrfs_header_nritems(left);
889 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
890 wret = 1;
891 } else {
892 ret = btrfs_cow_block(trans, root, left, parent,
893 pslot - 1, &left);
894 if (ret)
895 wret = 1;
896 else {
897 wret = push_node_left(trans, root,
898 left, mid);
899 }
900 }
901 if (wret < 0)
902 ret = wret;
903 if (wret == 0) {
904 struct btrfs_disk_key disk_key;
905 orig_slot += left_nr;
906 btrfs_node_key(mid, &disk_key, 0);
907 btrfs_set_node_key(parent, &disk_key, pslot);
908 btrfs_mark_buffer_dirty(parent);
909 if (btrfs_header_nritems(left) > orig_slot) {
910 path->nodes[level] = left;
911 path->slots[level + 1] -= 1;
912 path->slots[level] = orig_slot;
913 free_extent_buffer(mid);
914 } else {
915 orig_slot -=
916 btrfs_header_nritems(left);
917 path->slots[level] = orig_slot;
918 free_extent_buffer(left);
919 }
920 return 0;
921 }
922 free_extent_buffer(left);
923 }
924 right= read_node_slot(root, parent, pslot + 1);
925
926 /*
927 * then try to empty the right most buffer into the middle
928 */
929 if (right) {
930 u32 right_nr;
931 right_nr = btrfs_header_nritems(right);
932 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
933 wret = 1;
934 } else {
935 ret = btrfs_cow_block(trans, root, right,
936 parent, pslot + 1,
937 &right);
938 if (ret)
939 wret = 1;
940 else {
941 wret = balance_node_right(trans, root,
942 right, mid);
943 }
944 }
945 if (wret < 0)
946 ret = wret;
947 if (wret == 0) {
948 struct btrfs_disk_key disk_key;
949
950 btrfs_node_key(right, &disk_key, 0);
951 btrfs_set_node_key(parent, &disk_key, pslot + 1);
952 btrfs_mark_buffer_dirty(parent);
953
954 if (btrfs_header_nritems(mid) <= orig_slot) {
955 path->nodes[level] = right;
956 path->slots[level + 1] += 1;
957 path->slots[level] = orig_slot -
958 btrfs_header_nritems(mid);
959 free_extent_buffer(mid);
960 } else {
961 free_extent_buffer(right);
962 }
963 return 0;
964 }
965 free_extent_buffer(right);
966 }
967 return 1;
968 }
969
970 /*
971 * readahead one full node of leaves
972 */
973 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
974 int level, int slot, u64 objectid)
975 {
976 struct extent_buffer *node;
977 struct btrfs_disk_key disk_key;
978 u32 nritems;
979 u64 search;
980 u64 lowest_read;
981 u64 highest_read;
982 u64 nread = 0;
983 int direction = path->reada;
984 struct extent_buffer *eb;
985 u32 nr;
986 u32 blocksize;
987 u32 nscan = 0;
988
989 if (level != 1)
990 return;
991
992 if (!path->nodes[level])
993 return;
994
995 node = path->nodes[level];
996 search = btrfs_node_blockptr(node, slot);
997 blocksize = btrfs_level_size(root, level - 1);
998 eb = btrfs_find_tree_block(root, search, blocksize);
999 if (eb) {
1000 free_extent_buffer(eb);
1001 return;
1002 }
1003
1004 highest_read = search;
1005 lowest_read = search;
1006
1007 nritems = btrfs_header_nritems(node);
1008 nr = slot;
1009 while(1) {
1010 if (direction < 0) {
1011 if (nr == 0)
1012 break;
1013 nr--;
1014 } else if (direction > 0) {
1015 nr++;
1016 if (nr >= nritems)
1017 break;
1018 }
1019 if (path->reada < 0 && objectid) {
1020 btrfs_node_key(node, &disk_key, nr);
1021 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1022 break;
1023 }
1024 search = btrfs_node_blockptr(node, nr);
1025 if ((search >= lowest_read && search <= highest_read) ||
1026 (search < lowest_read && lowest_read - search <= 32768) ||
1027 (search > highest_read && search - highest_read <= 32768)) {
1028 readahead_tree_block(root, search, blocksize);
1029 nread += blocksize;
1030 }
1031 nscan++;
1032 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1033 break;
1034 if(nread > (1024 * 1024) || nscan > 128)
1035 break;
1036
1037 if (search < lowest_read)
1038 lowest_read = search;
1039 if (search > highest_read)
1040 highest_read = search;
1041 }
1042 }
1043
1044 /*
1045 * look for key in the tree. path is filled in with nodes along the way
1046 * if key is found, we return zero and you can find the item in the leaf
1047 * level of the path (level 0)
1048 *
1049 * If the key isn't found, the path points to the slot where it should
1050 * be inserted, and 1 is returned. If there are other errors during the
1051 * search a negative error number is returned.
1052 *
1053 * if ins_len > 0, nodes and leaves will be split as we walk down the
1054 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1055 * possible)
1056 */
1057 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1058 *root, struct btrfs_key *key, struct btrfs_path *p, int
1059 ins_len, int cow)
1060 {
1061 struct extent_buffer *b;
1062 u64 bytenr;
1063 u64 ptr_gen;
1064 int slot;
1065 int ret;
1066 int level;
1067 int should_reada = p->reada;
1068 u8 lowest_level = 0;
1069
1070 lowest_level = p->lowest_level;
1071 WARN_ON(lowest_level && ins_len);
1072 WARN_ON(p->nodes[0] != NULL);
1073 /*
1074 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1075 */
1076 again:
1077 b = root->node;
1078 extent_buffer_get(b);
1079 while (b) {
1080 level = btrfs_header_level(b);
1081 if (cow) {
1082 int wret;
1083 wret = btrfs_cow_block(trans, root, b,
1084 p->nodes[level + 1],
1085 p->slots[level + 1],
1086 &b);
1087 if (wret) {
1088 free_extent_buffer(b);
1089 return wret;
1090 }
1091 }
1092 BUG_ON(!cow && ins_len);
1093 if (level != btrfs_header_level(b))
1094 WARN_ON(1);
1095 level = btrfs_header_level(b);
1096 p->nodes[level] = b;
1097 ret = check_block(root, p, level);
1098 if (ret)
1099 return -1;
1100 ret = bin_search(b, key, level, &slot);
1101 if (level != 0) {
1102 if (ret && slot > 0)
1103 slot -= 1;
1104 p->slots[level] = slot;
1105 if (ins_len > 0 && btrfs_header_nritems(b) >=
1106 BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1107 int sret = split_node(trans, root, p, level);
1108 BUG_ON(sret > 0);
1109 if (sret)
1110 return sret;
1111 b = p->nodes[level];
1112 slot = p->slots[level];
1113 } else if (ins_len < 0) {
1114 int sret = balance_level(trans, root, p,
1115 level);
1116 if (sret)
1117 return sret;
1118 b = p->nodes[level];
1119 if (!b) {
1120 btrfs_release_path(NULL, p);
1121 goto again;
1122 }
1123 slot = p->slots[level];
1124 BUG_ON(btrfs_header_nritems(b) == 1);
1125 }
1126 /* this is only true while dropping a snapshot */
1127 if (level == lowest_level)
1128 break;
1129 bytenr = btrfs_node_blockptr(b, slot);
1130 ptr_gen = btrfs_node_ptr_generation(b, slot);
1131 if (should_reada)
1132 reada_for_search(root, p, level, slot,
1133 key->objectid);
1134 b = read_tree_block(root, bytenr,
1135 btrfs_level_size(root, level - 1));
1136 if (ptr_gen != btrfs_header_generation(b)) {
1137 printk("block %llu bad gen wanted %llu "
1138 "found %llu\n",
1139 (unsigned long long)b->start,
1140 (unsigned long long)ptr_gen,
1141 (unsigned long long)btrfs_header_generation(b));
1142 }
1143 } else {
1144 p->slots[level] = slot;
1145 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1146 sizeof(struct btrfs_item) + ins_len) {
1147 int sret = split_leaf(trans, root, key,
1148 p, ins_len, ret == 0);
1149 BUG_ON(sret > 0);
1150 if (sret)
1151 return sret;
1152 }
1153 return ret;
1154 }
1155 }
1156 return 1;
1157 }
1158
1159 /*
1160 * adjust the pointers going up the tree, starting at level
1161 * making sure the right key of each node is points to 'key'.
1162 * This is used after shifting pointers to the left, so it stops
1163 * fixing up pointers when a given leaf/node is not in slot 0 of the
1164 * higher levels
1165 *
1166 * If this fails to write a tree block, it returns -1, but continues
1167 * fixing up the blocks in ram so the tree is consistent.
1168 */
1169 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1170 struct btrfs_root *root, struct btrfs_path *path,
1171 struct btrfs_disk_key *key, int level)
1172 {
1173 int i;
1174 int ret = 0;
1175 struct extent_buffer *t;
1176
1177 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1178 int tslot = path->slots[i];
1179 if (!path->nodes[i])
1180 break;
1181 t = path->nodes[i];
1182 btrfs_set_node_key(t, key, tslot);
1183 btrfs_mark_buffer_dirty(path->nodes[i]);
1184 if (tslot != 0)
1185 break;
1186 }
1187 return ret;
1188 }
1189
1190 /*
1191 * try to push data from one node into the next node left in the
1192 * tree.
1193 *
1194 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1195 * error, and > 0 if there was no room in the left hand block.
1196 */
1197 static int push_node_left(struct btrfs_trans_handle *trans,
1198 struct btrfs_root *root, struct extent_buffer *dst,
1199 struct extent_buffer *src)
1200 {
1201 int push_items = 0;
1202 int src_nritems;
1203 int dst_nritems;
1204 int ret = 0;
1205
1206 src_nritems = btrfs_header_nritems(src);
1207 dst_nritems = btrfs_header_nritems(dst);
1208 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1209 WARN_ON(btrfs_header_generation(src) != trans->transid);
1210 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1211
1212 if (push_items <= 0) {
1213 return 1;
1214 }
1215
1216 if (src_nritems < push_items)
1217 push_items = src_nritems;
1218
1219 copy_extent_buffer(dst, src,
1220 btrfs_node_key_ptr_offset(dst_nritems),
1221 btrfs_node_key_ptr_offset(0),
1222 push_items * sizeof(struct btrfs_key_ptr));
1223
1224 if (push_items < src_nritems) {
1225 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1226 btrfs_node_key_ptr_offset(push_items),
1227 (src_nritems - push_items) *
1228 sizeof(struct btrfs_key_ptr));
1229 }
1230 btrfs_set_header_nritems(src, src_nritems - push_items);
1231 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1232 btrfs_mark_buffer_dirty(src);
1233 btrfs_mark_buffer_dirty(dst);
1234 return ret;
1235 }
1236
1237 /*
1238 * try to push data from one node into the next node right in the
1239 * tree.
1240 *
1241 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1242 * error, and > 0 if there was no room in the right hand block.
1243 *
1244 * this will only push up to 1/2 the contents of the left node over
1245 */
1246 static int balance_node_right(struct btrfs_trans_handle *trans,
1247 struct btrfs_root *root,
1248 struct extent_buffer *dst,
1249 struct extent_buffer *src)
1250 {
1251 int push_items = 0;
1252 int max_push;
1253 int src_nritems;
1254 int dst_nritems;
1255 int ret = 0;
1256
1257 WARN_ON(btrfs_header_generation(src) != trans->transid);
1258 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1259
1260 src_nritems = btrfs_header_nritems(src);
1261 dst_nritems = btrfs_header_nritems(dst);
1262 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1263 if (push_items <= 0)
1264 return 1;
1265
1266 max_push = src_nritems / 2 + 1;
1267 /* don't try to empty the node */
1268 if (max_push >= src_nritems)
1269 return 1;
1270
1271 if (max_push < push_items)
1272 push_items = max_push;
1273
1274 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1275 btrfs_node_key_ptr_offset(0),
1276 (dst_nritems) *
1277 sizeof(struct btrfs_key_ptr));
1278
1279 copy_extent_buffer(dst, src,
1280 btrfs_node_key_ptr_offset(0),
1281 btrfs_node_key_ptr_offset(src_nritems - push_items),
1282 push_items * sizeof(struct btrfs_key_ptr));
1283
1284 btrfs_set_header_nritems(src, src_nritems - push_items);
1285 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1286
1287 btrfs_mark_buffer_dirty(src);
1288 btrfs_mark_buffer_dirty(dst);
1289 return ret;
1290 }
1291
1292 /*
1293 * helper function to insert a new root level in the tree.
1294 * A new node is allocated, and a single item is inserted to
1295 * point to the existing root
1296 *
1297 * returns zero on success or < 0 on failure.
1298 */
1299 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1300 struct btrfs_root *root,
1301 struct btrfs_path *path, int level)
1302 {
1303 u64 root_gen;
1304 u64 lower_gen;
1305 struct extent_buffer *lower;
1306 struct extent_buffer *c;
1307 struct btrfs_disk_key lower_key;
1308
1309 BUG_ON(path->nodes[level]);
1310 BUG_ON(path->nodes[level-1] != root->node);
1311
1312 if (root->ref_cows)
1313 root_gen = trans->transid;
1314 else
1315 root_gen = 0;
1316
1317 lower = path->nodes[level-1];
1318 if (level == 1)
1319 btrfs_item_key(lower, &lower_key, 0);
1320 else
1321 btrfs_node_key(lower, &lower_key, 0);
1322
1323 c = __btrfs_alloc_free_block(trans, root, root->nodesize,
1324 root->root_key.objectid,
1325 root_gen, lower_key.objectid, level,
1326 root->node->start, 0);
1327 if (IS_ERR(c))
1328 return PTR_ERR(c);
1329 memset_extent_buffer(c, 0, 0, root->nodesize);
1330 btrfs_set_header_nritems(c, 1);
1331 btrfs_set_header_level(c, level);
1332 btrfs_set_header_bytenr(c, c->start);
1333 btrfs_set_header_generation(c, trans->transid);
1334 btrfs_set_header_owner(c, root->root_key.objectid);
1335
1336 write_extent_buffer(c, root->fs_info->fsid,
1337 (unsigned long)btrfs_header_fsid(c),
1338 BTRFS_FSID_SIZE);
1339 btrfs_set_node_key(c, &lower_key, 0);
1340 btrfs_set_node_blockptr(c, 0, lower->start);
1341 lower_gen = btrfs_header_generation(lower);
1342 WARN_ON(lower_gen == 0);
1343
1344 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1345
1346 btrfs_mark_buffer_dirty(c);
1347
1348 /* the super has an extra ref to root->node */
1349 free_extent_buffer(root->node);
1350 root->node = c;
1351 add_root_to_dirty_list(root);
1352 extent_buffer_get(c);
1353 path->nodes[level] = c;
1354 path->slots[level] = 0;
1355
1356 if (root->ref_cows && lower_gen != trans->transid) {
1357 struct btrfs_path *back_path = btrfs_alloc_path();
1358 int ret;
1359 ret = btrfs_insert_extent_backref(trans,
1360 root->fs_info->extent_root,
1361 path, lower->start,
1362 root->root_key.objectid,
1363 trans->transid, 0, 0);
1364 BUG_ON(ret);
1365 btrfs_free_path(back_path);
1366 }
1367 return 0;
1368 }
1369
1370 /*
1371 * worker function to insert a single pointer in a node.
1372 * the node should have enough room for the pointer already
1373 *
1374 * slot and level indicate where you want the key to go, and
1375 * blocknr is the block the key points to.
1376 *
1377 * returns zero on success and < 0 on any error
1378 */
1379 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1380 *root, struct btrfs_path *path, struct btrfs_disk_key
1381 *key, u64 bytenr, int slot, int level)
1382 {
1383 struct extent_buffer *lower;
1384 int nritems;
1385
1386 BUG_ON(!path->nodes[level]);
1387 lower = path->nodes[level];
1388 nritems = btrfs_header_nritems(lower);
1389 if (slot > nritems)
1390 BUG();
1391 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1392 BUG();
1393 if (slot != nritems) {
1394 memmove_extent_buffer(lower,
1395 btrfs_node_key_ptr_offset(slot + 1),
1396 btrfs_node_key_ptr_offset(slot),
1397 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1398 }
1399 btrfs_set_node_key(lower, key, slot);
1400 btrfs_set_node_blockptr(lower, slot, bytenr);
1401 WARN_ON(trans->transid == 0);
1402 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1403 btrfs_set_header_nritems(lower, nritems + 1);
1404 btrfs_mark_buffer_dirty(lower);
1405 return 0;
1406 }
1407
1408 /*
1409 * split the node at the specified level in path in two.
1410 * The path is corrected to point to the appropriate node after the split
1411 *
1412 * Before splitting this tries to make some room in the node by pushing
1413 * left and right, if either one works, it returns right away.
1414 *
1415 * returns 0 on success and < 0 on failure
1416 */
1417 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1418 *root, struct btrfs_path *path, int level)
1419 {
1420 u64 root_gen;
1421 struct extent_buffer *c;
1422 struct extent_buffer *split;
1423 struct btrfs_disk_key disk_key;
1424 int mid;
1425 int ret;
1426 int wret;
1427 u32 c_nritems;
1428
1429 c = path->nodes[level];
1430 WARN_ON(btrfs_header_generation(c) != trans->transid);
1431 if (c == root->node) {
1432 /* trying to split the root, lets make a new one */
1433 ret = insert_new_root(trans, root, path, level + 1);
1434 if (ret)
1435 return ret;
1436 } else {
1437 ret = push_nodes_for_insert(trans, root, path, level);
1438 c = path->nodes[level];
1439 if (!ret && btrfs_header_nritems(c) <
1440 BTRFS_NODEPTRS_PER_BLOCK(root) - 1)
1441 return 0;
1442 if (ret < 0)
1443 return ret;
1444 }
1445
1446 c_nritems = btrfs_header_nritems(c);
1447 if (root->ref_cows)
1448 root_gen = trans->transid;
1449 else
1450 root_gen = 0;
1451
1452 btrfs_node_key(c, &disk_key, 0);
1453 split = __btrfs_alloc_free_block(trans, root, root->nodesize,
1454 root->root_key.objectid,
1455 root_gen,
1456 btrfs_disk_key_objectid(&disk_key),
1457 level, c->start, 0);
1458 if (IS_ERR(split))
1459 return PTR_ERR(split);
1460
1461 btrfs_set_header_flags(split, btrfs_header_flags(c));
1462 btrfs_set_header_level(split, btrfs_header_level(c));
1463 btrfs_set_header_bytenr(split, split->start);
1464 btrfs_set_header_generation(split, trans->transid);
1465 btrfs_set_header_owner(split, root->root_key.objectid);
1466 write_extent_buffer(split, root->fs_info->fsid,
1467 (unsigned long)btrfs_header_fsid(split),
1468 BTRFS_FSID_SIZE);
1469
1470 mid = (c_nritems + 1) / 2;
1471
1472 copy_extent_buffer(split, c,
1473 btrfs_node_key_ptr_offset(0),
1474 btrfs_node_key_ptr_offset(mid),
1475 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1476 btrfs_set_header_nritems(split, c_nritems - mid);
1477 btrfs_set_header_nritems(c, mid);
1478 ret = 0;
1479
1480 btrfs_mark_buffer_dirty(c);
1481 btrfs_mark_buffer_dirty(split);
1482
1483 btrfs_node_key(split, &disk_key, 0);
1484 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1485 path->slots[level + 1] + 1,
1486 level + 1);
1487 if (wret)
1488 ret = wret;
1489
1490 if (path->slots[level] >= mid) {
1491 path->slots[level] -= mid;
1492 free_extent_buffer(c);
1493 path->nodes[level] = split;
1494 path->slots[level + 1] += 1;
1495 } else {
1496 free_extent_buffer(split);
1497 }
1498 return ret;
1499 }
1500
1501 /*
1502 * how many bytes are required to store the items in a leaf. start
1503 * and nr indicate which items in the leaf to check. This totals up the
1504 * space used both by the item structs and the item data
1505 */
1506 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1507 {
1508 int data_len;
1509 int nritems = btrfs_header_nritems(l);
1510 int end = min(nritems, start + nr) - 1;
1511
1512 if (!nr)
1513 return 0;
1514 data_len = btrfs_item_end_nr(l, start);
1515 data_len = data_len - btrfs_item_offset_nr(l, end);
1516 data_len += sizeof(struct btrfs_item) * nr;
1517 WARN_ON(data_len < 0);
1518 return data_len;
1519 }
1520
1521 /*
1522 * The space between the end of the leaf items and
1523 * the start of the leaf data. IOW, how much room
1524 * the leaf has left for both items and data
1525 */
1526 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1527 {
1528 int nritems = btrfs_header_nritems(leaf);
1529 int ret;
1530 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1531 if (ret < 0) {
1532 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1533 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1534 leaf_space_used(leaf, 0, nritems), nritems);
1535 }
1536 return ret;
1537 }
1538
1539 /*
1540 * push some data in the path leaf to the right, trying to free up at
1541 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1542 *
1543 * returns 1 if the push failed because the other node didn't have enough
1544 * room, 0 if everything worked out and < 0 if there were major errors.
1545 */
1546 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1547 *root, struct btrfs_path *path, int data_size,
1548 int empty)
1549 {
1550 struct extent_buffer *left = path->nodes[0];
1551 struct extent_buffer *right;
1552 struct extent_buffer *upper;
1553 struct btrfs_disk_key disk_key;
1554 int slot;
1555 u32 i;
1556 int free_space;
1557 int push_space = 0;
1558 int push_items = 0;
1559 struct btrfs_item *item;
1560 u32 left_nritems;
1561 u32 nr;
1562 u32 right_nritems;
1563 u32 data_end;
1564 u32 this_item_size;
1565 int ret;
1566
1567 slot = path->slots[1];
1568 if (!path->nodes[1]) {
1569 return 1;
1570 }
1571 upper = path->nodes[1];
1572 if (slot >= btrfs_header_nritems(upper) - 1)
1573 return 1;
1574
1575 right = read_tree_block(root, btrfs_node_blockptr(upper, slot + 1),
1576 root->leafsize);
1577 free_space = btrfs_leaf_free_space(root, right);
1578 if (free_space < data_size + sizeof(struct btrfs_item)) {
1579 free_extent_buffer(right);
1580 return 1;
1581 }
1582
1583 /* cow and double check */
1584 ret = btrfs_cow_block(trans, root, right, upper,
1585 slot + 1, &right);
1586 if (ret) {
1587 free_extent_buffer(right);
1588 return 1;
1589 }
1590 free_space = btrfs_leaf_free_space(root, right);
1591 if (free_space < data_size + sizeof(struct btrfs_item)) {
1592 free_extent_buffer(right);
1593 return 1;
1594 }
1595
1596 left_nritems = btrfs_header_nritems(left);
1597 if (left_nritems == 0) {
1598 free_extent_buffer(right);
1599 return 1;
1600 }
1601
1602 if (empty)
1603 nr = 0;
1604 else
1605 nr = 1;
1606
1607 i = left_nritems - 1;
1608 while (i >= nr) {
1609 item = btrfs_item_nr(left, i);
1610
1611 if (path->slots[0] == i)
1612 push_space += data_size + sizeof(*item);
1613
1614 this_item_size = btrfs_item_size(left, item);
1615 if (this_item_size + sizeof(*item) + push_space > free_space)
1616 break;
1617 push_items++;
1618 push_space += this_item_size + sizeof(*item);
1619 if (i == 0)
1620 break;
1621 i--;
1622 }
1623
1624 if (push_items == 0) {
1625 free_extent_buffer(right);
1626 return 1;
1627 }
1628
1629 if (!empty && push_items == left_nritems)
1630 WARN_ON(1);
1631
1632 /* push left to right */
1633 right_nritems = btrfs_header_nritems(right);
1634
1635 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1636 push_space -= leaf_data_end(root, left);
1637
1638 /* make room in the right data area */
1639 data_end = leaf_data_end(root, right);
1640 memmove_extent_buffer(right,
1641 btrfs_leaf_data(right) + data_end - push_space,
1642 btrfs_leaf_data(right) + data_end,
1643 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1644
1645 /* copy from the left data area */
1646 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1647 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1648 btrfs_leaf_data(left) + leaf_data_end(root, left),
1649 push_space);
1650
1651 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1652 btrfs_item_nr_offset(0),
1653 right_nritems * sizeof(struct btrfs_item));
1654
1655 /* copy the items from left to right */
1656 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1657 btrfs_item_nr_offset(left_nritems - push_items),
1658 push_items * sizeof(struct btrfs_item));
1659
1660 /* update the item pointers */
1661 right_nritems += push_items;
1662 btrfs_set_header_nritems(right, right_nritems);
1663 push_space = BTRFS_LEAF_DATA_SIZE(root);
1664 for (i = 0; i < right_nritems; i++) {
1665 item = btrfs_item_nr(right, i);
1666 push_space -= btrfs_item_size(right, item);
1667 btrfs_set_item_offset(right, item, push_space);
1668 }
1669
1670 left_nritems -= push_items;
1671 btrfs_set_header_nritems(left, left_nritems);
1672
1673 if (left_nritems)
1674 btrfs_mark_buffer_dirty(left);
1675 btrfs_mark_buffer_dirty(right);
1676
1677 btrfs_item_key(right, &disk_key, 0);
1678 btrfs_set_node_key(upper, &disk_key, slot + 1);
1679 btrfs_mark_buffer_dirty(upper);
1680
1681 /* then fixup the leaf pointer in the path */
1682 if (path->slots[0] >= left_nritems) {
1683 path->slots[0] -= left_nritems;
1684 free_extent_buffer(path->nodes[0]);
1685 path->nodes[0] = right;
1686 path->slots[1] += 1;
1687 } else {
1688 free_extent_buffer(right);
1689 }
1690 return 0;
1691 }
1692 /*
1693 * push some data in the path leaf to the left, trying to free up at
1694 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1695 */
1696 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1697 *root, struct btrfs_path *path, int data_size,
1698 int empty)
1699 {
1700 struct btrfs_disk_key disk_key;
1701 struct extent_buffer *right = path->nodes[0];
1702 struct extent_buffer *left;
1703 int slot;
1704 int i;
1705 int free_space;
1706 int push_space = 0;
1707 int push_items = 0;
1708 struct btrfs_item *item;
1709 u32 old_left_nritems;
1710 u32 right_nritems;
1711 u32 nr;
1712 int ret = 0;
1713 int wret;
1714 u32 this_item_size;
1715 u32 old_left_item_size;
1716
1717 slot = path->slots[1];
1718 if (slot == 0)
1719 return 1;
1720 if (!path->nodes[1])
1721 return 1;
1722
1723 right_nritems = btrfs_header_nritems(right);
1724 if (right_nritems == 0) {
1725 return 1;
1726 }
1727
1728 left = read_tree_block(root, btrfs_node_blockptr(path->nodes[1],
1729 slot - 1), root->leafsize);
1730 free_space = btrfs_leaf_free_space(root, left);
1731 if (free_space < data_size + sizeof(struct btrfs_item)) {
1732 free_extent_buffer(left);
1733 return 1;
1734 }
1735
1736 /* cow and double check */
1737 ret = btrfs_cow_block(trans, root, left,
1738 path->nodes[1], slot - 1, &left);
1739 if (ret) {
1740 /* we hit -ENOSPC, but it isn't fatal here */
1741 free_extent_buffer(left);
1742 return 1;
1743 }
1744
1745 free_space = btrfs_leaf_free_space(root, left);
1746 if (free_space < data_size + sizeof(struct btrfs_item)) {
1747 free_extent_buffer(left);
1748 return 1;
1749 }
1750
1751 if (empty)
1752 nr = right_nritems;
1753 else
1754 nr = right_nritems - 1;
1755
1756 for (i = 0; i < nr; i++) {
1757 item = btrfs_item_nr(right, i);
1758
1759 if (path->slots[0] == i)
1760 push_space += data_size + sizeof(*item);
1761
1762 this_item_size = btrfs_item_size(right, item);
1763 if (this_item_size + sizeof(*item) + push_space > free_space)
1764 break;
1765
1766 push_items++;
1767 push_space += this_item_size + sizeof(*item);
1768 }
1769
1770 if (push_items == 0) {
1771 free_extent_buffer(left);
1772 return 1;
1773 }
1774 if (!empty && push_items == btrfs_header_nritems(right))
1775 WARN_ON(1);
1776
1777 /* push data from right to left */
1778 copy_extent_buffer(left, right,
1779 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1780 btrfs_item_nr_offset(0),
1781 push_items * sizeof(struct btrfs_item));
1782
1783 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1784 btrfs_item_offset_nr(right, push_items -1);
1785
1786 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1787 leaf_data_end(root, left) - push_space,
1788 btrfs_leaf_data(right) +
1789 btrfs_item_offset_nr(right, push_items - 1),
1790 push_space);
1791 old_left_nritems = btrfs_header_nritems(left);
1792 BUG_ON(old_left_nritems < 0);
1793
1794 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1795 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1796 u32 ioff;
1797
1798 item = btrfs_item_nr(left, i);
1799 ioff = btrfs_item_offset(left, item);
1800 btrfs_set_item_offset(left, item,
1801 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1802 }
1803 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1804
1805 /* fixup right node */
1806 if (push_items > right_nritems) {
1807 printk("push items %d nr %u\n", push_items, right_nritems);
1808 WARN_ON(1);
1809 }
1810
1811 if (push_items < right_nritems) {
1812 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1813 leaf_data_end(root, right);
1814 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1815 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1816 btrfs_leaf_data(right) +
1817 leaf_data_end(root, right), push_space);
1818
1819 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1820 btrfs_item_nr_offset(push_items),
1821 (btrfs_header_nritems(right) - push_items) *
1822 sizeof(struct btrfs_item));
1823 }
1824 right_nritems -= push_items;
1825 btrfs_set_header_nritems(right, right_nritems);
1826 push_space = BTRFS_LEAF_DATA_SIZE(root);
1827 for (i = 0; i < right_nritems; i++) {
1828 item = btrfs_item_nr(right, i);
1829 push_space = push_space - btrfs_item_size(right, item);
1830 btrfs_set_item_offset(right, item, push_space);
1831 }
1832
1833 btrfs_mark_buffer_dirty(left);
1834 if (right_nritems)
1835 btrfs_mark_buffer_dirty(right);
1836
1837 btrfs_item_key(right, &disk_key, 0);
1838 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
1839 if (wret)
1840 ret = wret;
1841
1842 /* then fixup the leaf pointer in the path */
1843 if (path->slots[0] < push_items) {
1844 path->slots[0] += old_left_nritems;
1845 free_extent_buffer(path->nodes[0]);
1846 path->nodes[0] = left;
1847 path->slots[1] -= 1;
1848 } else {
1849 free_extent_buffer(left);
1850 path->slots[0] -= push_items;
1851 }
1852 BUG_ON(path->slots[0] < 0);
1853 return ret;
1854 }
1855
1856 /*
1857 * split the path's leaf in two, making sure there is at least data_size
1858 * available for the resulting leaf level of the path.
1859 *
1860 * returns 0 if all went well and < 0 on failure.
1861 */
1862 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
1863 *root, struct btrfs_key *ins_key,
1864 struct btrfs_path *path, int data_size, int extend)
1865 {
1866 u64 root_gen;
1867 struct extent_buffer *l;
1868 u32 nritems;
1869 int mid;
1870 int slot;
1871 struct extent_buffer *right;
1872 int space_needed = data_size + sizeof(struct btrfs_item);
1873 int data_copy_size;
1874 int rt_data_off;
1875 int i;
1876 int ret = 0;
1877 int wret;
1878 int double_split;
1879 int num_doubles = 0;
1880 struct btrfs_disk_key disk_key;
1881
1882 if (extend)
1883 space_needed = data_size;
1884
1885 if (root->ref_cows)
1886 root_gen = trans->transid;
1887 else
1888 root_gen = 0;
1889
1890 /* first try to make some room by pushing left and right */
1891 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
1892 wret = push_leaf_right(trans, root, path, data_size, 0);
1893 if (wret < 0) {
1894 return wret;
1895 }
1896 if (wret) {
1897 wret = push_leaf_left(trans, root, path, data_size, 0);
1898 if (wret < 0)
1899 return wret;
1900 }
1901 l = path->nodes[0];
1902
1903 /* did the pushes work? */
1904 if (btrfs_leaf_free_space(root, l) >= space_needed)
1905 return 0;
1906 }
1907
1908 if (!path->nodes[1]) {
1909 ret = insert_new_root(trans, root, path, 1);
1910 if (ret)
1911 return ret;
1912 }
1913 again:
1914 double_split = 0;
1915 l = path->nodes[0];
1916 slot = path->slots[0];
1917 nritems = btrfs_header_nritems(l);
1918 mid = (nritems + 1)/ 2;
1919
1920 btrfs_item_key(l, &disk_key, 0);
1921
1922 right = __btrfs_alloc_free_block(trans, root, root->leafsize,
1923 root->root_key.objectid,
1924 root_gen, disk_key.objectid, 0,
1925 l->start, 0);
1926 if (IS_ERR(right))
1927 return PTR_ERR(right);
1928
1929 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
1930 btrfs_set_header_bytenr(right, right->start);
1931 btrfs_set_header_generation(right, trans->transid);
1932 btrfs_set_header_owner(right, root->root_key.objectid);
1933 btrfs_set_header_level(right, 0);
1934 write_extent_buffer(right, root->fs_info->fsid,
1935 (unsigned long)btrfs_header_fsid(right),
1936 BTRFS_FSID_SIZE);
1937 if (mid <= slot) {
1938 if (nritems == 1 ||
1939 leaf_space_used(l, mid, nritems - mid) + space_needed >
1940 BTRFS_LEAF_DATA_SIZE(root)) {
1941 if (slot >= nritems) {
1942 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1943 btrfs_set_header_nritems(right, 0);
1944 wret = insert_ptr(trans, root, path,
1945 &disk_key, right->start,
1946 path->slots[1] + 1, 1);
1947 if (wret)
1948 ret = wret;
1949 free_extent_buffer(path->nodes[0]);
1950 path->nodes[0] = right;
1951 path->slots[0] = 0;
1952 path->slots[1] += 1;
1953 return ret;
1954 }
1955 mid = slot;
1956 if (mid != nritems &&
1957 leaf_space_used(l, mid, nritems - mid) +
1958 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
1959 double_split = 1;
1960 }
1961 }
1962 } else {
1963 if (leaf_space_used(l, 0, mid + 1) + space_needed >
1964 BTRFS_LEAF_DATA_SIZE(root)) {
1965 if (!extend && slot == 0) {
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,
1970 right->start,
1971 path->slots[1], 1);
1972 if (wret)
1973 ret = wret;
1974 free_extent_buffer(path->nodes[0]);
1975 path->nodes[0] = right;
1976 path->slots[0] = 0;
1977 if (path->slots[1] == 0) {
1978 wret = fixup_low_keys(trans, root,
1979 path, &disk_key, 1);
1980 if (wret)
1981 ret = wret;
1982 }
1983 return ret;
1984 } else if (extend && slot == 0) {
1985 mid = 1;
1986 } else {
1987 mid = slot;
1988 if (mid != nritems &&
1989 leaf_space_used(l, mid, nritems - mid) +
1990 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
1991 double_split = 1;
1992 }
1993 }
1994 }
1995 }
1996 nritems = nritems - mid;
1997 btrfs_set_header_nritems(right, nritems);
1998 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
1999
2000 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2001 btrfs_item_nr_offset(mid),
2002 nritems * sizeof(struct btrfs_item));
2003
2004 copy_extent_buffer(right, l,
2005 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2006 data_copy_size, btrfs_leaf_data(l) +
2007 leaf_data_end(root, l), data_copy_size);
2008
2009 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2010 btrfs_item_end_nr(l, mid);
2011
2012 for (i = 0; i < nritems; i++) {
2013 struct btrfs_item *item = btrfs_item_nr(right, i);
2014 u32 ioff = btrfs_item_offset(right, item);
2015 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2016 }
2017
2018 btrfs_set_header_nritems(l, mid);
2019 ret = 0;
2020 btrfs_item_key(right, &disk_key, 0);
2021 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2022 path->slots[1] + 1, 1);
2023 if (wret)
2024 ret = wret;
2025
2026 btrfs_mark_buffer_dirty(right);
2027 btrfs_mark_buffer_dirty(l);
2028 BUG_ON(path->slots[0] != slot);
2029
2030 if (mid <= slot) {
2031 free_extent_buffer(path->nodes[0]);
2032 path->nodes[0] = right;
2033 path->slots[0] -= mid;
2034 path->slots[1] += 1;
2035 } else
2036 free_extent_buffer(right);
2037
2038 BUG_ON(path->slots[0] < 0);
2039
2040 if (double_split) {
2041 BUG_ON(num_doubles != 0);
2042 num_doubles++;
2043 goto again;
2044 }
2045 return ret;
2046 }
2047
2048 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2049 struct btrfs_root *root,
2050 struct btrfs_path *path,
2051 u32 new_size, int from_end)
2052 {
2053 int ret = 0;
2054 int slot;
2055 int slot_orig;
2056 struct extent_buffer *leaf;
2057 struct btrfs_item *item;
2058 u32 nritems;
2059 unsigned int data_end;
2060 unsigned int old_data_start;
2061 unsigned int old_size;
2062 unsigned int size_diff;
2063 int i;
2064
2065 slot_orig = path->slots[0];
2066 leaf = path->nodes[0];
2067 slot = path->slots[0];
2068
2069 old_size = btrfs_item_size_nr(leaf, slot);
2070 if (old_size == new_size)
2071 return 0;
2072
2073 nritems = btrfs_header_nritems(leaf);
2074 data_end = leaf_data_end(root, leaf);
2075
2076 old_data_start = btrfs_item_offset_nr(leaf, slot);
2077
2078 size_diff = old_size - new_size;
2079
2080 BUG_ON(slot < 0);
2081 BUG_ON(slot >= nritems);
2082
2083 /*
2084 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2085 */
2086 /* first correct the data pointers */
2087 for (i = slot; i < nritems; i++) {
2088 u32 ioff;
2089 item = btrfs_item_nr(leaf, i);
2090 ioff = btrfs_item_offset(leaf, item);
2091 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2092 }
2093
2094 /* shift the data */
2095 if (from_end) {
2096 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2097 data_end + size_diff, btrfs_leaf_data(leaf) +
2098 data_end, old_data_start + new_size - data_end);
2099 } else {
2100 struct btrfs_disk_key disk_key;
2101 u64 offset;
2102
2103 btrfs_item_key(leaf, &disk_key, slot);
2104
2105 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2106 unsigned long ptr;
2107 struct btrfs_file_extent_item *fi;
2108
2109 fi = btrfs_item_ptr(leaf, slot,
2110 struct btrfs_file_extent_item);
2111 fi = (struct btrfs_file_extent_item *)(
2112 (unsigned long)fi - size_diff);
2113
2114 if (btrfs_file_extent_type(leaf, fi) ==
2115 BTRFS_FILE_EXTENT_INLINE) {
2116 ptr = btrfs_item_ptr_offset(leaf, slot);
2117 memmove_extent_buffer(leaf, ptr,
2118 (unsigned long)fi,
2119 offsetof(struct btrfs_file_extent_item,
2120 disk_bytenr));
2121 }
2122 }
2123
2124 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2125 data_end + size_diff, btrfs_leaf_data(leaf) +
2126 data_end, old_data_start - data_end);
2127
2128 offset = btrfs_disk_key_offset(&disk_key);
2129 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2130 btrfs_set_item_key(leaf, &disk_key, slot);
2131 if (slot == 0)
2132 fixup_low_keys(trans, root, path, &disk_key, 1);
2133 }
2134
2135 item = btrfs_item_nr(leaf, slot);
2136 btrfs_set_item_size(leaf, item, new_size);
2137 btrfs_mark_buffer_dirty(leaf);
2138
2139 ret = 0;
2140 if (btrfs_leaf_free_space(root, leaf) < 0) {
2141 btrfs_print_leaf(root, leaf);
2142 BUG();
2143 }
2144 return ret;
2145 }
2146
2147 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2148 struct btrfs_root *root, struct btrfs_path *path,
2149 u32 data_size)
2150 {
2151 int ret = 0;
2152 int slot;
2153 int slot_orig;
2154 struct extent_buffer *leaf;
2155 struct btrfs_item *item;
2156 u32 nritems;
2157 unsigned int data_end;
2158 unsigned int old_data;
2159 unsigned int old_size;
2160 int i;
2161
2162 slot_orig = path->slots[0];
2163 leaf = path->nodes[0];
2164
2165 nritems = btrfs_header_nritems(leaf);
2166 data_end = leaf_data_end(root, leaf);
2167
2168 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2169 btrfs_print_leaf(root, leaf);
2170 BUG();
2171 }
2172 slot = path->slots[0];
2173 old_data = btrfs_item_end_nr(leaf, slot);
2174
2175 BUG_ON(slot < 0);
2176 if (slot >= nritems) {
2177 btrfs_print_leaf(root, leaf);
2178 printk("slot %d too large, nritems %d\n", slot, nritems);
2179 BUG_ON(1);
2180 }
2181
2182 /*
2183 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2184 */
2185 /* first correct the data pointers */
2186 for (i = slot; i < nritems; i++) {
2187 u32 ioff;
2188 item = btrfs_item_nr(leaf, i);
2189 ioff = btrfs_item_offset(leaf, item);
2190 btrfs_set_item_offset(leaf, item, ioff - data_size);
2191 }
2192
2193 /* shift the data */
2194 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2195 data_end - data_size, btrfs_leaf_data(leaf) +
2196 data_end, old_data - data_end);
2197
2198 data_end = old_data;
2199 old_size = btrfs_item_size_nr(leaf, slot);
2200 item = btrfs_item_nr(leaf, slot);
2201 btrfs_set_item_size(leaf, item, old_size + data_size);
2202 btrfs_mark_buffer_dirty(leaf);
2203
2204 ret = 0;
2205 if (btrfs_leaf_free_space(root, leaf) < 0) {
2206 btrfs_print_leaf(root, leaf);
2207 BUG();
2208 }
2209 return ret;
2210 }
2211
2212 /*
2213 * Given a key and some data, insert an item into the tree.
2214 * This does all the path init required, making room in the tree if needed.
2215 */
2216 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2217 struct btrfs_root *root,
2218 struct btrfs_path *path,
2219 struct btrfs_key *cpu_key, u32 *data_size,
2220 int nr)
2221 {
2222 struct extent_buffer *leaf;
2223 struct btrfs_item *item;
2224 int ret = 0;
2225 int slot;
2226 int slot_orig;
2227 int i;
2228 u32 nritems;
2229 u32 total_size = 0;
2230 u32 total_data = 0;
2231 unsigned int data_end;
2232 struct btrfs_disk_key disk_key;
2233
2234 for (i = 0; i < nr; i++) {
2235 total_data += data_size[i];
2236 }
2237
2238 /* create a root if there isn't one */
2239 if (!root->node)
2240 BUG();
2241
2242 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2243 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2244 if (ret == 0) {
2245 return -EEXIST;
2246 }
2247 if (ret < 0)
2248 goto out;
2249
2250 slot_orig = path->slots[0];
2251 leaf = path->nodes[0];
2252
2253 nritems = btrfs_header_nritems(leaf);
2254 data_end = leaf_data_end(root, leaf);
2255
2256 if (btrfs_leaf_free_space(root, leaf) <
2257 sizeof(struct btrfs_item) + total_size) {
2258 btrfs_print_leaf(root, leaf);
2259 printk("not enough freespace need %u have %d\n",
2260 total_size, btrfs_leaf_free_space(root, leaf));
2261 BUG();
2262 }
2263
2264 slot = path->slots[0];
2265 BUG_ON(slot < 0);
2266
2267 if (slot != nritems) {
2268 int i;
2269 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2270
2271 if (old_data < data_end) {
2272 btrfs_print_leaf(root, leaf);
2273 printk("slot %d old_data %d data_end %d\n",
2274 slot, old_data, data_end);
2275 BUG_ON(1);
2276 }
2277 /*
2278 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2279 */
2280 /* first correct the data pointers */
2281 for (i = slot; i < nritems; i++) {
2282 u32 ioff;
2283
2284 item = btrfs_item_nr(leaf, i);
2285 ioff = btrfs_item_offset(leaf, item);
2286 btrfs_set_item_offset(leaf, item, ioff - total_data);
2287 }
2288
2289 /* shift the items */
2290 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2291 btrfs_item_nr_offset(slot),
2292 (nritems - slot) * sizeof(struct btrfs_item));
2293
2294 /* shift the data */
2295 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2296 data_end - total_data, btrfs_leaf_data(leaf) +
2297 data_end, old_data - data_end);
2298 data_end = old_data;
2299 }
2300
2301 /* setup the item for the new data */
2302 for (i = 0; i < nr; i++) {
2303 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2304 btrfs_set_item_key(leaf, &disk_key, slot + i);
2305 item = btrfs_item_nr(leaf, slot + i);
2306 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2307 data_end -= data_size[i];
2308 btrfs_set_item_size(leaf, item, data_size[i]);
2309 }
2310 btrfs_set_header_nritems(leaf, nritems + nr);
2311 btrfs_mark_buffer_dirty(leaf);
2312
2313 ret = 0;
2314 if (slot == 0) {
2315 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2316 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2317 }
2318
2319 if (btrfs_leaf_free_space(root, leaf) < 0) {
2320 btrfs_print_leaf(root, leaf);
2321 BUG();
2322 }
2323
2324 out:
2325 return ret;
2326 }
2327
2328 /*
2329 * Given a key and some data, insert an item into the tree.
2330 * This does all the path init required, making room in the tree if needed.
2331 */
2332 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2333 *root, struct btrfs_key *cpu_key, void *data, u32
2334 data_size)
2335 {
2336 int ret = 0;
2337 struct btrfs_path *path;
2338 struct extent_buffer *leaf;
2339 unsigned long ptr;
2340
2341 path = btrfs_alloc_path();
2342 BUG_ON(!path);
2343 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2344 if (!ret) {
2345 leaf = path->nodes[0];
2346 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2347 write_extent_buffer(leaf, data, ptr, data_size);
2348 btrfs_mark_buffer_dirty(leaf);
2349 }
2350 btrfs_free_path(path);
2351 return ret;
2352 }
2353
2354 /*
2355 * delete the pointer from a given node.
2356 *
2357 * If the delete empties a node, the node is removed from the tree,
2358 * continuing all the way the root if required. The root is converted into
2359 * a leaf if all the nodes are emptied.
2360 */
2361 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2362 struct btrfs_path *path, int level, int slot)
2363 {
2364 struct extent_buffer *parent = path->nodes[level];
2365 u32 nritems;
2366 int ret = 0;
2367 int wret;
2368
2369 nritems = btrfs_header_nritems(parent);
2370 if (slot != nritems -1) {
2371 memmove_extent_buffer(parent,
2372 btrfs_node_key_ptr_offset(slot),
2373 btrfs_node_key_ptr_offset(slot + 1),
2374 sizeof(struct btrfs_key_ptr) *
2375 (nritems - slot - 1));
2376 }
2377 nritems--;
2378 btrfs_set_header_nritems(parent, nritems);
2379 if (nritems == 0 && parent == root->node) {
2380 BUG_ON(btrfs_header_level(root->node) != 1);
2381 /* just turn the root into a leaf and break */
2382 btrfs_set_header_level(root->node, 0);
2383 } else if (slot == 0) {
2384 struct btrfs_disk_key disk_key;
2385
2386 btrfs_node_key(parent, &disk_key, 0);
2387 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2388 if (wret)
2389 ret = wret;
2390 }
2391 btrfs_mark_buffer_dirty(parent);
2392 return ret;
2393 }
2394
2395 /*
2396 * delete the item at the leaf level in path. If that empties
2397 * the leaf, remove it from the tree
2398 */
2399 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2400 struct btrfs_path *path, int slot, int nr)
2401 {
2402 struct extent_buffer *leaf;
2403 struct btrfs_item *item;
2404 int last_off;
2405 int dsize = 0;
2406 int ret = 0;
2407 int wret;
2408 int i;
2409 u32 nritems;
2410
2411 leaf = path->nodes[0];
2412 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2413
2414 for (i = 0; i < nr; i++)
2415 dsize += btrfs_item_size_nr(leaf, slot + i);
2416
2417 nritems = btrfs_header_nritems(leaf);
2418
2419 if (slot + nr != nritems) {
2420 int i;
2421 int data_end = leaf_data_end(root, leaf);
2422
2423 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2424 data_end + dsize,
2425 btrfs_leaf_data(leaf) + data_end,
2426 last_off - data_end);
2427
2428 for (i = slot + nr; i < nritems; i++) {
2429 u32 ioff;
2430
2431 item = btrfs_item_nr(leaf, i);
2432 ioff = btrfs_item_offset(leaf, item);
2433 btrfs_set_item_offset(leaf, item, ioff + dsize);
2434 }
2435
2436 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2437 btrfs_item_nr_offset(slot + nr),
2438 sizeof(struct btrfs_item) *
2439 (nritems - slot - nr));
2440 }
2441 btrfs_set_header_nritems(leaf, nritems - nr);
2442 nritems -= nr;
2443
2444 /* delete the leaf if we've emptied it */
2445 if (nritems == 0) {
2446 if (leaf == root->node) {
2447 btrfs_set_header_level(leaf, 0);
2448 } else {
2449 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2450 clean_tree_block(trans, root, leaf);
2451 wait_on_tree_block_writeback(root, leaf);
2452 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2453 if (wret)
2454 ret = wret;
2455 wret = btrfs_free_extent(trans, root,
2456 leaf->start, leaf->len,
2457 btrfs_header_owner(path->nodes[1]),
2458 root_gen, 0, 0, 1);
2459 if (wret)
2460 ret = wret;
2461 }
2462 } else {
2463 int used = leaf_space_used(leaf, 0, nritems);
2464 if (slot == 0) {
2465 struct btrfs_disk_key disk_key;
2466
2467 btrfs_item_key(leaf, &disk_key, 0);
2468 wret = fixup_low_keys(trans, root, path,
2469 &disk_key, 1);
2470 if (wret)
2471 ret = wret;
2472 }
2473
2474 /* delete the leaf if it is mostly empty */
2475 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2476 /* push_leaf_left fixes the path.
2477 * make sure the path still points to our leaf
2478 * for possible call to del_ptr below
2479 */
2480 slot = path->slots[1];
2481 extent_buffer_get(leaf);
2482
2483 wret = push_leaf_left(trans, root, path, 1, 1);
2484 if (wret < 0 && wret != -ENOSPC)
2485 ret = wret;
2486
2487 if (path->nodes[0] == leaf &&
2488 btrfs_header_nritems(leaf)) {
2489 wret = push_leaf_right(trans, root, path, 1, 1);
2490 if (wret < 0 && wret != -ENOSPC)
2491 ret = wret;
2492 }
2493
2494 if (btrfs_header_nritems(leaf) == 0) {
2495 u64 root_gen;
2496 u64 bytenr = leaf->start;
2497 u32 blocksize = leaf->len;
2498
2499 root_gen = btrfs_header_generation(
2500 path->nodes[1]);
2501
2502 clean_tree_block(trans, root, leaf);
2503 wait_on_tree_block_writeback(root, leaf);
2504
2505 wret = del_ptr(trans, root, path, 1, slot);
2506 if (wret)
2507 ret = wret;
2508
2509 free_extent_buffer(leaf);
2510 wret = btrfs_free_extent(trans, root, bytenr,
2511 blocksize,
2512 btrfs_header_owner(path->nodes[1]),
2513 root_gen, 0, 0, 1);
2514 if (wret)
2515 ret = wret;
2516 } else {
2517 btrfs_mark_buffer_dirty(leaf);
2518 free_extent_buffer(leaf);
2519 }
2520 } else {
2521 btrfs_mark_buffer_dirty(leaf);
2522 }
2523 }
2524 return ret;
2525 }
2526
2527 /*
2528 * walk up the tree as far as required to find the previous leaf.
2529 * returns 0 if it found something or 1 if there are no lesser leaves.
2530 * returns < 0 on io errors.
2531 */
2532 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2533 {
2534 u64 bytenr;
2535 int slot;
2536 int level = 1;
2537 struct extent_buffer *c;
2538 struct extent_buffer *next = NULL;
2539
2540 while(level < BTRFS_MAX_LEVEL) {
2541 if (!path->nodes[level])
2542 return 1;
2543
2544 slot = path->slots[level];
2545 c = path->nodes[level];
2546 if (slot == 0) {
2547 level++;
2548 if (level == BTRFS_MAX_LEVEL)
2549 return 1;
2550 continue;
2551 }
2552 slot--;
2553
2554 bytenr = btrfs_node_blockptr(c, slot);
2555 if (next)
2556 free_extent_buffer(next);
2557
2558 next = read_tree_block(root, bytenr,
2559 btrfs_level_size(root, level - 1));
2560 break;
2561 }
2562 path->slots[level] = slot;
2563 while(1) {
2564 level--;
2565 c = path->nodes[level];
2566 free_extent_buffer(c);
2567 slot = btrfs_header_nritems(next);
2568 if (slot != 0)
2569 slot--;
2570 path->nodes[level] = next;
2571 path->slots[level] = slot;
2572 if (!level)
2573 break;
2574 next = read_tree_block(root, btrfs_node_blockptr(next, slot),
2575 btrfs_level_size(root, level - 1));
2576 }
2577 return 0;
2578 }
2579
2580 /*
2581 * walk up the tree as far as required to find the next leaf.
2582 * returns 0 if it found something or 1 if there are no greater leaves.
2583 * returns < 0 on io errors.
2584 */
2585 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2586 {
2587 int slot;
2588 int level = 1;
2589 u64 bytenr;
2590 struct extent_buffer *c;
2591 struct extent_buffer *next = NULL;
2592
2593 while(level < BTRFS_MAX_LEVEL) {
2594 if (!path->nodes[level])
2595 return 1;
2596
2597 slot = path->slots[level] + 1;
2598 c = path->nodes[level];
2599 if (slot >= btrfs_header_nritems(c)) {
2600 level++;
2601 if (level == BTRFS_MAX_LEVEL)
2602 return 1;
2603 continue;
2604 }
2605
2606 bytenr = btrfs_node_blockptr(c, slot);
2607 if (next)
2608 free_extent_buffer(next);
2609
2610 if (path->reada)
2611 reada_for_search(root, path, level, slot, 0);
2612
2613 next = read_tree_block(root, bytenr,
2614 btrfs_level_size(root, level -1));
2615 break;
2616 }
2617 path->slots[level] = slot;
2618 while(1) {
2619 level--;
2620 c = path->nodes[level];
2621 free_extent_buffer(c);
2622 path->nodes[level] = next;
2623 path->slots[level] = 0;
2624 if (!level)
2625 break;
2626 if (path->reada)
2627 reada_for_search(root, path, level, 0, 0);
2628 next = read_tree_block(root, btrfs_node_blockptr(next, 0),
2629 btrfs_level_size(root, level - 1));
2630 }
2631 return 0;
2632 }
2633
2634 int btrfs_previous_item(struct btrfs_root *root,
2635 struct btrfs_path *path, u64 min_objectid,
2636 int type)
2637 {
2638 struct btrfs_key found_key;
2639 struct extent_buffer *leaf;
2640 int ret;
2641
2642 while(1) {
2643 if (path->slots[0] == 0) {
2644 ret = btrfs_prev_leaf(root, path);
2645 if (ret != 0)
2646 return ret;
2647 } else {
2648 path->slots[0]--;
2649 }
2650 leaf = path->nodes[0];
2651 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2652 if (found_key.type == type)
2653 return 0;
2654 }
2655 return 1;
2656 }
2657
Btrfs-progs-unstable mirror