?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add sync and transaction ioctl defs
[btrfs-progs-unstable.git] / ctree.c
blob881c115bfd7e98ba6a5eb1fda7e83f69193c66e4
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 gen;
305 u64 search_start = *last_ret;
306 u64 last_block = 0;
307 u64 other;
308 u32 parent_nritems;
309 int end_slot;
310 int i;
311 int err = 0;
312 int parent_level;
313 int uptodate;
314 u32 blocksize;
315 int progress_passed = 0;
316 struct btrfs_disk_key disk_key;
317
318 parent_level = btrfs_header_level(parent);
319 if (cache_only && parent_level != 1)
320 return 0;
321
322 if (trans->transaction != root->fs_info->running_transaction) {
323 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
324 root->fs_info->running_transaction->transid);
325 WARN_ON(1);
326 }
327 if (trans->transid != root->fs_info->generation) {
328 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
329 root->fs_info->generation);
330 WARN_ON(1);
331 }
332
333 parent_nritems = btrfs_header_nritems(parent);
334 blocksize = btrfs_level_size(root, parent_level - 1);
335 end_slot = parent_nritems;
336
337 if (parent_nritems == 1)
338 return 0;
339
340 for (i = start_slot; i < end_slot; i++) {
341 int close = 1;
342
343 if (!parent->map_token) {
344 map_extent_buffer(parent,
345 btrfs_node_key_ptr_offset(i),
346 sizeof(struct btrfs_key_ptr),
347 &parent->map_token, &parent->kaddr,
348 &parent->map_start, &parent->map_len,
349 KM_USER1);
350 }
351 btrfs_node_key(parent, &disk_key, i);
352 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
353 continue;
354
355 progress_passed = 1;
356 blocknr = btrfs_node_blockptr(parent, i);
357 gen = btrfs_node_ptr_generation(parent, i);
358 if (last_block == 0)
359 last_block = blocknr;
360
361 if (i > 0) {
362 other = btrfs_node_blockptr(parent, i - 1);
363 close = close_blocks(blocknr, other, blocksize);
364 }
365 if (close && i < end_slot - 2) {
366 other = btrfs_node_blockptr(parent, i + 1);
367 close = close_blocks(blocknr, other, blocksize);
368 }
369 if (close) {
370 last_block = blocknr;
371 continue;
372 }
373 if (parent->map_token) {
374 unmap_extent_buffer(parent, parent->map_token,
375 KM_USER1);
376 parent->map_token = NULL;
377 }
378
379 cur = btrfs_find_tree_block(root, blocknr, blocksize);
380 if (cur)
381 uptodate = btrfs_buffer_uptodate(cur, gen);
382 else
383 uptodate = 0;
384 if (!cur || !uptodate) {
385 if (cache_only) {
386 free_extent_buffer(cur);
387 continue;
388 }
389 if (!cur) {
390 cur = read_tree_block(root, blocknr,
391 blocksize, gen);
392 } else if (!uptodate) {
393 btrfs_read_buffer(cur, gen);
394 }
395 }
396 if (search_start == 0)
397 search_start = last_block;
398
399 err = __btrfs_cow_block(trans, root, cur, parent, i,
400 &tmp, search_start,
401 min(16 * blocksize,
402 (end_slot - i) * blocksize));
403 if (err) {
404 free_extent_buffer(cur);
405 break;
406 }
407 search_start = tmp->start;
408 last_block = tmp->start;
409 *last_ret = search_start;
410 if (parent_level == 1)
411 btrfs_clear_buffer_defrag(tmp);
412 free_extent_buffer(tmp);
413 }
414 if (parent->map_token) {
415 unmap_extent_buffer(parent, parent->map_token,
416 KM_USER1);
417 parent->map_token = NULL;
418 }
419 return err;
420 }
421 #endif
422
423 /*
424 * The leaf data grows from end-to-front in the node.
425 * this returns the address of the start of the last item,
426 * which is the stop of the leaf data stack
427 */
428 static inline unsigned int leaf_data_end(struct btrfs_root *root,
429 struct extent_buffer *leaf)
430 {
431 u32 nr = btrfs_header_nritems(leaf);
432 if (nr == 0)
433 return BTRFS_LEAF_DATA_SIZE(root);
434 return btrfs_item_offset_nr(leaf, nr - 1);
435 }
436
437 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
438 int level)
439 {
440 struct extent_buffer *parent = NULL;
441 struct extent_buffer *node = path->nodes[level];
442 struct btrfs_disk_key parent_key;
443 struct btrfs_disk_key node_key;
444 int parent_slot;
445 int slot;
446 struct btrfs_key cpukey;
447 u32 nritems = btrfs_header_nritems(node);
448
449 if (path->nodes[level + 1])
450 parent = path->nodes[level + 1];
451
452 slot = path->slots[level];
453 BUG_ON(nritems == 0);
454 if (parent) {
455 parent_slot = path->slots[level + 1];
456 btrfs_node_key(parent, &parent_key, parent_slot);
457 btrfs_node_key(node, &node_key, 0);
458 BUG_ON(memcmp(&parent_key, &node_key,
459 sizeof(struct btrfs_disk_key)));
460 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
461 btrfs_header_bytenr(node));
462 }
463 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
464 if (slot != 0) {
465 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
466 btrfs_node_key(node, &node_key, slot);
467 BUG_ON(btrfs_comp_keys(&node_key, &cpukey) <= 0);
468 }
469 if (slot < nritems - 1) {
470 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
471 btrfs_node_key(node, &node_key, slot);
472 BUG_ON(btrfs_comp_keys(&node_key, &cpukey) >= 0);
473 }
474 return 0;
475 }
476
477 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
478 int level)
479 {
480 struct extent_buffer *leaf = path->nodes[level];
481 struct extent_buffer *parent = NULL;
482 int parent_slot;
483 struct btrfs_key cpukey;
484 struct btrfs_disk_key parent_key;
485 struct btrfs_disk_key leaf_key;
486 int slot = path->slots[0];
487
488 u32 nritems = btrfs_header_nritems(leaf);
489
490 if (path->nodes[level + 1])
491 parent = path->nodes[level + 1];
492
493 if (nritems == 0)
494 return 0;
495
496 if (parent) {
497 parent_slot = path->slots[level + 1];
498 btrfs_node_key(parent, &parent_key, parent_slot);
499 btrfs_item_key(leaf, &leaf_key, 0);
500
501 BUG_ON(memcmp(&parent_key, &leaf_key,
502 sizeof(struct btrfs_disk_key)));
503 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
504 btrfs_header_bytenr(leaf));
505 }
506 #if 0
507 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
508 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
509 btrfs_item_key(leaf, &leaf_key, i);
510 if (comp_keys(&leaf_key, &cpukey) >= 0) {
511 btrfs_print_leaf(root, leaf);
512 printk("slot %d offset bad key\n", i);
513 BUG_ON(1);
514 }
515 if (btrfs_item_offset_nr(leaf, i) !=
516 btrfs_item_end_nr(leaf, i + 1)) {
517 btrfs_print_leaf(root, leaf);
518 printk("slot %d offset bad\n", i);
519 BUG_ON(1);
520 }
521 if (i == 0) {
522 if (btrfs_item_offset_nr(leaf, i) +
523 btrfs_item_size_nr(leaf, i) !=
524 BTRFS_LEAF_DATA_SIZE(root)) {
525 btrfs_print_leaf(root, leaf);
526 printk("slot %d first offset bad\n", i);
527 BUG_ON(1);
528 }
529 }
530 }
531 if (nritems > 0) {
532 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
533 btrfs_print_leaf(root, leaf);
534 printk("slot %d bad size \n", nritems - 1);
535 BUG_ON(1);
536 }
537 }
538 #endif
539 if (slot != 0 && slot < nritems - 1) {
540 btrfs_item_key(leaf, &leaf_key, slot);
541 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
542 if (btrfs_comp_keys(&leaf_key, &cpukey) <= 0) {
543 btrfs_print_leaf(root, leaf);
544 printk("slot %d offset bad key\n", slot);
545 BUG_ON(1);
546 }
547 if (btrfs_item_offset_nr(leaf, slot - 1) !=
548 btrfs_item_end_nr(leaf, slot)) {
549 btrfs_print_leaf(root, leaf);
550 printk("slot %d offset bad\n", slot);
551 BUG_ON(1);
552 }
553 }
554 if (slot < nritems - 1) {
555 btrfs_item_key(leaf, &leaf_key, slot);
556 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
557 BUG_ON(btrfs_comp_keys(&leaf_key, &cpukey) >= 0);
558 if (btrfs_item_offset_nr(leaf, slot) !=
559 btrfs_item_end_nr(leaf, slot + 1)) {
560 btrfs_print_leaf(root, leaf);
561 printk("slot %d offset bad\n", slot);
562 BUG_ON(1);
563 }
564 }
565 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
566 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
567 return 0;
568 }
569
570 static int noinline check_block(struct btrfs_root *root,
571 struct btrfs_path *path, int level)
572 {
573 return 0;
574 #if 0
575 struct extent_buffer *buf = path->nodes[level];
576
577 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
578 (unsigned long)btrfs_header_fsid(buf),
579 BTRFS_FSID_SIZE)) {
580 printk("warning bad block %Lu\n", buf->start);
581 return 1;
582 }
583 #endif
584 if (level == 0)
585 return check_leaf(root, path, level);
586 return check_node(root, path, level);
587 }
588
589 /*
590 * search for key in the extent_buffer. The items start at offset p,
591 * and they are item_size apart. There are 'max' items in p.
592 *
593 * the slot in the array is returned via slot, and it points to
594 * the place where you would insert key if it is not found in
595 * the array.
596 *
597 * slot may point to max if the key is bigger than all of the keys
598 */
599 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
600 int item_size, struct btrfs_key *key,
601 int max, int *slot)
602 {
603 int low = 0;
604 int high = max;
605 int mid;
606 int ret;
607 unsigned long offset;
608 struct btrfs_disk_key *tmp;
609
610 while(low < high) {
611 mid = (low + high) / 2;
612 offset = p + mid * item_size;
613
614 tmp = (struct btrfs_disk_key *)(eb->data + offset);
615 ret = btrfs_comp_keys(tmp, key);
616
617 if (ret < 0)
618 low = mid + 1;
619 else if (ret > 0)
620 high = mid;
621 else {
622 *slot = mid;
623 return 0;
624 }
625 }
626 *slot = low;
627 return 1;
628 }
629
630 /*
631 * simple bin_search frontend that does the right thing for
632 * leaves vs nodes
633 */
634 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
635 int level, int *slot)
636 {
637 if (level == 0) {
638 return generic_bin_search(eb,
639 offsetof(struct btrfs_leaf, items),
640 sizeof(struct btrfs_item),
641 key, btrfs_header_nritems(eb),
642 slot);
643 } else {
644 return generic_bin_search(eb,
645 offsetof(struct btrfs_node, ptrs),
646 sizeof(struct btrfs_key_ptr),
647 key, btrfs_header_nritems(eb),
648 slot);
649 }
650 return -1;
651 }
652
653 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
654 struct extent_buffer *parent, int slot)
655 {
656 int level = btrfs_header_level(parent);
657 if (slot < 0)
658 return NULL;
659 if (slot >= btrfs_header_nritems(parent))
660 return NULL;
661
662 BUG_ON(level == 0);
663
664 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
665 btrfs_level_size(root, level - 1),
666 btrfs_node_ptr_generation(parent, slot));
667 }
668
669 static int balance_level(struct btrfs_trans_handle *trans,
670 struct btrfs_root *root,
671 struct btrfs_path *path, int level)
672 {
673 struct extent_buffer *right = NULL;
674 struct extent_buffer *mid;
675 struct extent_buffer *left = NULL;
676 struct extent_buffer *parent = NULL;
677 int ret = 0;
678 int wret;
679 int pslot;
680 int orig_slot = path->slots[level];
681 int err_on_enospc = 0;
682 u64 orig_ptr;
683
684 if (level == 0)
685 return 0;
686
687 mid = path->nodes[level];
688 WARN_ON(btrfs_header_generation(mid) != trans->transid);
689
690 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
691
692 if (level < BTRFS_MAX_LEVEL - 1)
693 parent = path->nodes[level + 1];
694 pslot = path->slots[level + 1];
695
696 /*
697 * deal with the case where there is only one pointer in the root
698 * by promoting the node below to a root
699 */
700 if (!parent) {
701 struct extent_buffer *child;
702
703 if (btrfs_header_nritems(mid) != 1)
704 return 0;
705
706 /* promote the child to a root */
707 child = read_node_slot(root, mid, 0);
708 BUG_ON(!child);
709 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
710 BUG_ON(ret);
711
712 root->node = child;
713 add_root_to_dirty_list(root);
714 path->nodes[level] = NULL;
715 clean_tree_block(trans, root, mid);
716 wait_on_tree_block_writeback(root, mid);
717 /* once for the path */
718 free_extent_buffer(mid);
719 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
720 root->root_key.objectid,
721 btrfs_header_generation(mid), 0, 0, 1);
722 /* once for the root ptr */
723 free_extent_buffer(mid);
724 return ret;
725 }
726 if (btrfs_header_nritems(mid) >
727 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
728 return 0;
729
730 if (btrfs_header_nritems(mid) < 2)
731 err_on_enospc = 1;
732
733 left = read_node_slot(root, parent, pslot - 1);
734 if (left) {
735 wret = btrfs_cow_block(trans, root, left,
736 parent, pslot - 1, &left);
737 if (wret) {
738 ret = wret;
739 goto enospc;
740 }
741 }
742 right = read_node_slot(root, parent, pslot + 1);
743 if (right) {
744 wret = btrfs_cow_block(trans, root, right,
745 parent, pslot + 1, &right);
746 if (wret) {
747 ret = wret;
748 goto enospc;
749 }
750 }
751
752 /* first, try to make some room in the middle buffer */
753 if (left) {
754 orig_slot += btrfs_header_nritems(left);
755 wret = push_node_left(trans, root, left, mid, 1);
756 if (wret < 0)
757 ret = wret;
758 if (btrfs_header_nritems(mid) < 2)
759 err_on_enospc = 1;
760 }
761
762 /*
763 * then try to empty the right most buffer into the middle
764 */
765 if (right) {
766 wret = push_node_left(trans, root, mid, right, 1);
767 if (wret < 0 && wret != -ENOSPC)
768 ret = wret;
769 if (btrfs_header_nritems(right) == 0) {
770 u64 bytenr = right->start;
771 u64 generation = btrfs_header_generation(parent);
772 u32 blocksize = right->len;
773
774 clean_tree_block(trans, root, right);
775 wait_on_tree_block_writeback(root, right);
776 free_extent_buffer(right);
777 right = NULL;
778 wret = del_ptr(trans, root, path, level + 1, pslot +
779 1);
780 if (wret)
781 ret = wret;
782 wret = btrfs_free_extent(trans, root, bytenr,
783 blocksize,
784 btrfs_header_owner(parent),
785 generation, 0, 0, 1);
786 if (wret)
787 ret = wret;
788 } else {
789 struct btrfs_disk_key right_key;
790 btrfs_node_key(right, &right_key, 0);
791 btrfs_set_node_key(parent, &right_key, pslot + 1);
792 btrfs_mark_buffer_dirty(parent);
793 }
794 }
795 if (btrfs_header_nritems(mid) == 1) {
796 /*
797 * we're not allowed to leave a node with one item in the
798 * tree during a delete. A deletion from lower in the tree
799 * could try to delete the only pointer in this node.
800 * So, pull some keys from the left.
801 * There has to be a left pointer at this point because
802 * otherwise we would have pulled some pointers from the
803 * right
804 */
805 BUG_ON(!left);
806 wret = balance_node_right(trans, root, mid, left);
807 if (wret < 0) {
808 ret = wret;
809 goto enospc;
810 }
811 if (wret == 1) {
812 wret = push_node_left(trans, root, left, mid, 1);
813 if (wret < 0)
814 ret = wret;
815 }
816 BUG_ON(wret == 1);
817 }
818 if (btrfs_header_nritems(mid) == 0) {
819 /* we've managed to empty the middle node, drop it */
820 u64 root_gen = btrfs_header_generation(parent);
821 u64 bytenr = mid->start;
822 u32 blocksize = mid->len;
823 clean_tree_block(trans, root, mid);
824 wait_on_tree_block_writeback(root, mid);
825 free_extent_buffer(mid);
826 mid = NULL;
827 wret = del_ptr(trans, root, path, level + 1, pslot);
828 if (wret)
829 ret = wret;
830 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
831 btrfs_header_owner(parent),
832 root_gen, 0, 0, 1);
833 if (wret)
834 ret = wret;
835 } else {
836 /* update the parent key to reflect our changes */
837 struct btrfs_disk_key mid_key;
838 btrfs_node_key(mid, &mid_key, 0);
839 btrfs_set_node_key(parent, &mid_key, pslot);
840 btrfs_mark_buffer_dirty(parent);
841 }
842
843 /* update the path */
844 if (left) {
845 if (btrfs_header_nritems(left) > orig_slot) {
846 extent_buffer_get(left);
847 path->nodes[level] = left;
848 path->slots[level + 1] -= 1;
849 path->slots[level] = orig_slot;
850 if (mid)
851 free_extent_buffer(mid);
852 } else {
853 orig_slot -= btrfs_header_nritems(left);
854 path->slots[level] = orig_slot;
855 }
856 }
857 /* double check we haven't messed things up */
858 check_block(root, path, level);
859 if (orig_ptr !=
860 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
861 BUG();
862 enospc:
863 if (right)
864 free_extent_buffer(right);
865 if (left)
866 free_extent_buffer(left);
867 return ret;
868 }
869
870 /* returns zero if the push worked, non-zero otherwise */
871 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
872 struct btrfs_root *root,
873 struct btrfs_path *path, int level)
874 {
875 struct extent_buffer *right = NULL;
876 struct extent_buffer *mid;
877 struct extent_buffer *left = NULL;
878 struct extent_buffer *parent = NULL;
879 int ret = 0;
880 int wret;
881 int pslot;
882 int orig_slot = path->slots[level];
883 u64 orig_ptr;
884
885 if (level == 0)
886 return 1;
887
888 mid = path->nodes[level];
889 WARN_ON(btrfs_header_generation(mid) != trans->transid);
890 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
891
892 if (level < BTRFS_MAX_LEVEL - 1)
893 parent = path->nodes[level + 1];
894 pslot = path->slots[level + 1];
895
896 if (!parent)
897 return 1;
898
899 left = read_node_slot(root, parent, pslot - 1);
900
901 /* first, try to make some room in the middle buffer */
902 if (left) {
903 u32 left_nr;
904 left_nr = btrfs_header_nritems(left);
905 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
906 wret = 1;
907 } else {
908 ret = btrfs_cow_block(trans, root, left, parent,
909 pslot - 1, &left);
910 if (ret)
911 wret = 1;
912 else {
913 wret = push_node_left(trans, root,
914 left, mid, 0);
915 }
916 }
917 if (wret < 0)
918 ret = wret;
919 if (wret == 0) {
920 struct btrfs_disk_key disk_key;
921 orig_slot += left_nr;
922 btrfs_node_key(mid, &disk_key, 0);
923 btrfs_set_node_key(parent, &disk_key, pslot);
924 btrfs_mark_buffer_dirty(parent);
925 if (btrfs_header_nritems(left) > orig_slot) {
926 path->nodes[level] = left;
927 path->slots[level + 1] -= 1;
928 path->slots[level] = orig_slot;
929 free_extent_buffer(mid);
930 } else {
931 orig_slot -=
932 btrfs_header_nritems(left);
933 path->slots[level] = orig_slot;
934 free_extent_buffer(left);
935 }
936 return 0;
937 }
938 free_extent_buffer(left);
939 }
940 right= read_node_slot(root, parent, pslot + 1);
941
942 /*
943 * then try to empty the right most buffer into the middle
944 */
945 if (right) {
946 u32 right_nr;
947 right_nr = btrfs_header_nritems(right);
948 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
949 wret = 1;
950 } else {
951 ret = btrfs_cow_block(trans, root, right,
952 parent, pslot + 1,
953 &right);
954 if (ret)
955 wret = 1;
956 else {
957 wret = balance_node_right(trans, root,
958 right, mid);
959 }
960 }
961 if (wret < 0)
962 ret = wret;
963 if (wret == 0) {
964 struct btrfs_disk_key disk_key;
965
966 btrfs_node_key(right, &disk_key, 0);
967 btrfs_set_node_key(parent, &disk_key, pslot + 1);
968 btrfs_mark_buffer_dirty(parent);
969
970 if (btrfs_header_nritems(mid) <= orig_slot) {
971 path->nodes[level] = right;
972 path->slots[level + 1] += 1;
973 path->slots[level] = orig_slot -
974 btrfs_header_nritems(mid);
975 free_extent_buffer(mid);
976 } else {
977 free_extent_buffer(right);
978 }
979 return 0;
980 }
981 free_extent_buffer(right);
982 }
983 return 1;
984 }
985
986 /*
987 * readahead one full node of leaves
988 */
989 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
990 int level, int slot, u64 objectid)
991 {
992 struct extent_buffer *node;
993 struct btrfs_disk_key disk_key;
994 u32 nritems;
995 u64 search;
996 u64 lowest_read;
997 u64 highest_read;
998 u64 nread = 0;
999 int direction = path->reada;
1000 struct extent_buffer *eb;
1001 u32 nr;
1002 u32 blocksize;
1003 u32 nscan = 0;
1004
1005 if (level != 1)
1006 return;
1007
1008 if (!path->nodes[level])
1009 return;
1010
1011 node = path->nodes[level];
1012 search = btrfs_node_blockptr(node, slot);
1013 blocksize = btrfs_level_size(root, level - 1);
1014 eb = btrfs_find_tree_block(root, search, blocksize);
1015 if (eb) {
1016 free_extent_buffer(eb);
1017 return;
1018 }
1019
1020 highest_read = search;
1021 lowest_read = search;
1022
1023 nritems = btrfs_header_nritems(node);
1024 nr = slot;
1025 while(1) {
1026 if (direction < 0) {
1027 if (nr == 0)
1028 break;
1029 nr--;
1030 } else if (direction > 0) {
1031 nr++;
1032 if (nr >= nritems)
1033 break;
1034 }
1035 if (path->reada < 0 && objectid) {
1036 btrfs_node_key(node, &disk_key, nr);
1037 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1038 break;
1039 }
1040 search = btrfs_node_blockptr(node, nr);
1041 if ((search >= lowest_read && search <= highest_read) ||
1042 (search < lowest_read && lowest_read - search <= 32768) ||
1043 (search > highest_read && search - highest_read <= 32768)) {
1044 readahead_tree_block(root, search, blocksize,
1045 btrfs_node_ptr_generation(node, nr));
1046 nread += blocksize;
1047 }
1048 nscan++;
1049 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1050 break;
1051 if(nread > (1024 * 1024) || nscan > 128)
1052 break;
1053
1054 if (search < lowest_read)
1055 lowest_read = search;
1056 if (search > highest_read)
1057 highest_read = search;
1058 }
1059 }
1060
1061 /*
1062 * look for key in the tree. path is filled in with nodes along the way
1063 * if key is found, we return zero and you can find the item in the leaf
1064 * level of the path (level 0)
1065 *
1066 * If the key isn't found, the path points to the slot where it should
1067 * be inserted, and 1 is returned. If there are other errors during the
1068 * search a negative error number is returned.
1069 *
1070 * if ins_len > 0, nodes and leaves will be split as we walk down the
1071 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1072 * possible)
1073 */
1074 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1075 *root, struct btrfs_key *key, struct btrfs_path *p, int
1076 ins_len, int cow)
1077 {
1078 struct extent_buffer *b;
1079 int slot;
1080 int ret;
1081 int level;
1082 int should_reada = p->reada;
1083 u8 lowest_level = 0;
1084
1085 lowest_level = p->lowest_level;
1086 WARN_ON(lowest_level && ins_len);
1087 WARN_ON(p->nodes[0] != NULL);
1088 /*
1089 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1090 */
1091 again:
1092 b = root->node;
1093 extent_buffer_get(b);
1094 while (b) {
1095 level = btrfs_header_level(b);
1096 if (cow) {
1097 int wret;
1098 wret = btrfs_cow_block(trans, root, b,
1099 p->nodes[level + 1],
1100 p->slots[level + 1],
1101 &b);
1102 if (wret) {
1103 free_extent_buffer(b);
1104 return wret;
1105 }
1106 }
1107 BUG_ON(!cow && ins_len);
1108 if (level != btrfs_header_level(b))
1109 WARN_ON(1);
1110 level = btrfs_header_level(b);
1111 p->nodes[level] = b;
1112 ret = check_block(root, p, level);
1113 if (ret)
1114 return -1;
1115 ret = bin_search(b, key, level, &slot);
1116 if (level != 0) {
1117 if (ret && slot > 0)
1118 slot -= 1;
1119 p->slots[level] = slot;
1120 if (ins_len > 0 && btrfs_header_nritems(b) >=
1121 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1122 int sret = split_node(trans, root, p, level);
1123 BUG_ON(sret > 0);
1124 if (sret)
1125 return sret;
1126 b = p->nodes[level];
1127 slot = p->slots[level];
1128 } else if (ins_len < 0) {
1129 int sret = balance_level(trans, root, p,
1130 level);
1131 if (sret)
1132 return sret;
1133 b = p->nodes[level];
1134 if (!b) {
1135 btrfs_release_path(NULL, p);
1136 goto again;
1137 }
1138 slot = p->slots[level];
1139 BUG_ON(btrfs_header_nritems(b) == 1);
1140 }
1141 /* this is only true while dropping a snapshot */
1142 if (level == lowest_level)
1143 break;
1144
1145 if (should_reada)
1146 reada_for_search(root, p, level, slot,
1147 key->objectid);
1148
1149 b = read_node_slot(root, b, slot);
1150 } else {
1151 p->slots[level] = slot;
1152 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1153 sizeof(struct btrfs_item) + ins_len) {
1154 int sret = split_leaf(trans, root, key,
1155 p, ins_len, ret == 0);
1156 BUG_ON(sret > 0);
1157 if (sret)
1158 return sret;
1159 }
1160 return ret;
1161 }
1162 }
1163 return 1;
1164 }
1165
1166 /*
1167 * adjust the pointers going up the tree, starting at level
1168 * making sure the right key of each node is points to 'key'.
1169 * This is used after shifting pointers to the left, so it stops
1170 * fixing up pointers when a given leaf/node is not in slot 0 of the
1171 * higher levels
1172 *
1173 * If this fails to write a tree block, it returns -1, but continues
1174 * fixing up the blocks in ram so the tree is consistent.
1175 */
1176 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1177 struct btrfs_root *root, struct btrfs_path *path,
1178 struct btrfs_disk_key *key, int level)
1179 {
1180 int i;
1181 int ret = 0;
1182 struct extent_buffer *t;
1183
1184 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1185 int tslot = path->slots[i];
1186 if (!path->nodes[i])
1187 break;
1188 t = path->nodes[i];
1189 btrfs_set_node_key(t, key, tslot);
1190 btrfs_mark_buffer_dirty(path->nodes[i]);
1191 if (tslot != 0)
1192 break;
1193 }
1194 return ret;
1195 }
1196
1197 /*
1198 * try to push data from one node into the next node left in the
1199 * tree.
1200 *
1201 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1202 * error, and > 0 if there was no room in the left hand block.
1203 */
1204 static int push_node_left(struct btrfs_trans_handle *trans,
1205 struct btrfs_root *root, struct extent_buffer *dst,
1206 struct extent_buffer *src, int empty)
1207 {
1208 int push_items = 0;
1209 int src_nritems;
1210 int dst_nritems;
1211 int ret = 0;
1212
1213 src_nritems = btrfs_header_nritems(src);
1214 dst_nritems = btrfs_header_nritems(dst);
1215 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1216 WARN_ON(btrfs_header_generation(src) != trans->transid);
1217 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1218
1219 if (!empty && src_nritems <= 8)
1220 return 1;
1221
1222 if (push_items <= 0) {
1223 return 1;
1224 }
1225
1226 if (empty) {
1227 push_items = min(src_nritems, push_items);
1228 if (push_items < src_nritems) {
1229 /* leave at least 8 pointers in the node if
1230 * we aren't going to empty it
1231 */
1232 if (src_nritems - push_items < 8) {
1233 if (push_items <= 8)
1234 return 1;
1235 push_items -= 8;
1236 }
1237 }
1238 } else
1239 push_items = min(src_nritems - 8, push_items);
1240
1241 copy_extent_buffer(dst, src,
1242 btrfs_node_key_ptr_offset(dst_nritems),
1243 btrfs_node_key_ptr_offset(0),
1244 push_items * sizeof(struct btrfs_key_ptr));
1245
1246 if (push_items < src_nritems) {
1247 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1248 btrfs_node_key_ptr_offset(push_items),
1249 (src_nritems - push_items) *
1250 sizeof(struct btrfs_key_ptr));
1251 }
1252 btrfs_set_header_nritems(src, src_nritems - push_items);
1253 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1254 btrfs_mark_buffer_dirty(src);
1255 btrfs_mark_buffer_dirty(dst);
1256 return ret;
1257 }
1258
1259 /*
1260 * try to push data from one node into the next node right in the
1261 * tree.
1262 *
1263 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1264 * error, and > 0 if there was no room in the right hand block.
1265 *
1266 * this will only push up to 1/2 the contents of the left node over
1267 */
1268 static int balance_node_right(struct btrfs_trans_handle *trans,
1269 struct btrfs_root *root,
1270 struct extent_buffer *dst,
1271 struct extent_buffer *src)
1272 {
1273 int push_items = 0;
1274 int max_push;
1275 int src_nritems;
1276 int dst_nritems;
1277 int ret = 0;
1278
1279 WARN_ON(btrfs_header_generation(src) != trans->transid);
1280 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1281
1282 src_nritems = btrfs_header_nritems(src);
1283 dst_nritems = btrfs_header_nritems(dst);
1284 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1285 if (push_items <= 0) {
1286 return 1;
1287 }
1288
1289 if (src_nritems < 4) {
1290 return 1;
1291 }
1292
1293 max_push = src_nritems / 2 + 1;
1294 /* don't try to empty the node */
1295 if (max_push >= src_nritems) {
1296 return 1;
1297 }
1298
1299 if (max_push < push_items)
1300 push_items = max_push;
1301
1302 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1303 btrfs_node_key_ptr_offset(0),
1304 (dst_nritems) *
1305 sizeof(struct btrfs_key_ptr));
1306
1307 copy_extent_buffer(dst, src,
1308 btrfs_node_key_ptr_offset(0),
1309 btrfs_node_key_ptr_offset(src_nritems - push_items),
1310 push_items * sizeof(struct btrfs_key_ptr));
1311
1312 btrfs_set_header_nritems(src, src_nritems - push_items);
1313 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1314
1315 btrfs_mark_buffer_dirty(src);
1316 btrfs_mark_buffer_dirty(dst);
1317 return ret;
1318 }
1319
1320 /*
1321 * helper function to insert a new root level in the tree.
1322 * A new node is allocated, and a single item is inserted to
1323 * point to the existing root
1324 *
1325 * returns zero on success or < 0 on failure.
1326 */
1327 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1328 struct btrfs_root *root,
1329 struct btrfs_path *path, int level)
1330 {
1331 u64 root_gen;
1332 u64 lower_gen;
1333 struct extent_buffer *lower;
1334 struct extent_buffer *c;
1335 struct btrfs_disk_key lower_key;
1336
1337 BUG_ON(path->nodes[level]);
1338 BUG_ON(path->nodes[level-1] != root->node);
1339
1340 if (root->ref_cows)
1341 root_gen = trans->transid;
1342 else
1343 root_gen = 0;
1344
1345 lower = path->nodes[level-1];
1346 if (level == 1)
1347 btrfs_item_key(lower, &lower_key, 0);
1348 else
1349 btrfs_node_key(lower, &lower_key, 0);
1350
1351 c = __btrfs_alloc_free_block(trans, root, root->nodesize,
1352 root->root_key.objectid,
1353 root_gen, lower_key.objectid, level,
1354 root->node->start, 0);
1355 if (IS_ERR(c))
1356 return PTR_ERR(c);
1357 memset_extent_buffer(c, 0, 0, root->nodesize);
1358 btrfs_set_header_nritems(c, 1);
1359 btrfs_set_header_level(c, level);
1360 btrfs_set_header_bytenr(c, c->start);
1361 btrfs_set_header_generation(c, trans->transid);
1362 btrfs_set_header_owner(c, root->root_key.objectid);
1363
1364 write_extent_buffer(c, root->fs_info->fsid,
1365 (unsigned long)btrfs_header_fsid(c),
1366 BTRFS_FSID_SIZE);
1367
1368 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1369 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1370 BTRFS_UUID_SIZE);
1371
1372 btrfs_set_node_key(c, &lower_key, 0);
1373 btrfs_set_node_blockptr(c, 0, lower->start);
1374 lower_gen = btrfs_header_generation(lower);
1375 WARN_ON(lower_gen == 0);
1376
1377 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1378
1379 btrfs_mark_buffer_dirty(c);
1380
1381 /* the super has an extra ref to root->node */
1382 free_extent_buffer(root->node);
1383 root->node = c;
1384 add_root_to_dirty_list(root);
1385 extent_buffer_get(c);
1386 path->nodes[level] = c;
1387 path->slots[level] = 0;
1388
1389 if (root->ref_cows && lower_gen != trans->transid) {
1390 struct btrfs_path *back_path = btrfs_alloc_path();
1391 int ret;
1392 ret = btrfs_insert_extent_backref(trans,
1393 root->fs_info->extent_root,
1394 path, lower->start,
1395 root->root_key.objectid,
1396 trans->transid, 0, 0);
1397 BUG_ON(ret);
1398 btrfs_free_path(back_path);
1399 }
1400 return 0;
1401 }
1402
1403 /*
1404 * worker function to insert a single pointer in a node.
1405 * the node should have enough room for the pointer already
1406 *
1407 * slot and level indicate where you want the key to go, and
1408 * blocknr is the block the key points to.
1409 *
1410 * returns zero on success and < 0 on any error
1411 */
1412 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1413 *root, struct btrfs_path *path, struct btrfs_disk_key
1414 *key, u64 bytenr, int slot, int level)
1415 {
1416 struct extent_buffer *lower;
1417 int nritems;
1418
1419 BUG_ON(!path->nodes[level]);
1420 lower = path->nodes[level];
1421 nritems = btrfs_header_nritems(lower);
1422 if (slot > nritems)
1423 BUG();
1424 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1425 BUG();
1426 if (slot != nritems) {
1427 memmove_extent_buffer(lower,
1428 btrfs_node_key_ptr_offset(slot + 1),
1429 btrfs_node_key_ptr_offset(slot),
1430 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1431 }
1432 btrfs_set_node_key(lower, key, slot);
1433 btrfs_set_node_blockptr(lower, slot, bytenr);
1434 WARN_ON(trans->transid == 0);
1435 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1436 btrfs_set_header_nritems(lower, nritems + 1);
1437 btrfs_mark_buffer_dirty(lower);
1438 return 0;
1439 }
1440
1441 /*
1442 * split the node at the specified level in path in two.
1443 * The path is corrected to point to the appropriate node after the split
1444 *
1445 * Before splitting this tries to make some room in the node by pushing
1446 * left and right, if either one works, it returns right away.
1447 *
1448 * returns 0 on success and < 0 on failure
1449 */
1450 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1451 *root, struct btrfs_path *path, int level)
1452 {
1453 u64 root_gen;
1454 struct extent_buffer *c;
1455 struct extent_buffer *split;
1456 struct btrfs_disk_key disk_key;
1457 int mid;
1458 int ret;
1459 int wret;
1460 u32 c_nritems;
1461
1462 c = path->nodes[level];
1463 WARN_ON(btrfs_header_generation(c) != trans->transid);
1464 if (c == root->node) {
1465 /* trying to split the root, lets make a new one */
1466 ret = insert_new_root(trans, root, path, level + 1);
1467 if (ret)
1468 return ret;
1469 } else {
1470 ret = push_nodes_for_insert(trans, root, path, level);
1471 c = path->nodes[level];
1472 if (!ret && btrfs_header_nritems(c) <
1473 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1474 return 0;
1475 if (ret < 0)
1476 return ret;
1477 }
1478
1479 c_nritems = btrfs_header_nritems(c);
1480 if (root->ref_cows)
1481 root_gen = trans->transid;
1482 else
1483 root_gen = 0;
1484
1485 btrfs_node_key(c, &disk_key, 0);
1486 split = __btrfs_alloc_free_block(trans, root, root->nodesize,
1487 root->root_key.objectid,
1488 root_gen,
1489 btrfs_disk_key_objectid(&disk_key),
1490 level, c->start, 0);
1491 if (IS_ERR(split))
1492 return PTR_ERR(split);
1493
1494 btrfs_set_header_flags(split, btrfs_header_flags(c));
1495 btrfs_set_header_level(split, btrfs_header_level(c));
1496 btrfs_set_header_bytenr(split, split->start);
1497 btrfs_set_header_generation(split, trans->transid);
1498 btrfs_set_header_owner(split, root->root_key.objectid);
1499 btrfs_set_header_flags(split, 0);
1500 write_extent_buffer(split, root->fs_info->fsid,
1501 (unsigned long)btrfs_header_fsid(split),
1502 BTRFS_FSID_SIZE);
1503 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1504 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1505 BTRFS_UUID_SIZE);
1506
1507 mid = (c_nritems + 1) / 2;
1508
1509 copy_extent_buffer(split, c,
1510 btrfs_node_key_ptr_offset(0),
1511 btrfs_node_key_ptr_offset(mid),
1512 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1513 btrfs_set_header_nritems(split, c_nritems - mid);
1514 btrfs_set_header_nritems(c, mid);
1515 ret = 0;
1516
1517 btrfs_mark_buffer_dirty(c);
1518 btrfs_mark_buffer_dirty(split);
1519
1520 btrfs_node_key(split, &disk_key, 0);
1521 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1522 path->slots[level + 1] + 1,
1523 level + 1);
1524 if (wret)
1525 ret = wret;
1526
1527 if (path->slots[level] >= mid) {
1528 path->slots[level] -= mid;
1529 free_extent_buffer(c);
1530 path->nodes[level] = split;
1531 path->slots[level + 1] += 1;
1532 } else {
1533 free_extent_buffer(split);
1534 }
1535 return ret;
1536 }
1537
1538 /*
1539 * how many bytes are required to store the items in a leaf. start
1540 * and nr indicate which items in the leaf to check. This totals up the
1541 * space used both by the item structs and the item data
1542 */
1543 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1544 {
1545 int data_len;
1546 int nritems = btrfs_header_nritems(l);
1547 int end = min(nritems, start + nr) - 1;
1548
1549 if (!nr)
1550 return 0;
1551 data_len = btrfs_item_end_nr(l, start);
1552 data_len = data_len - btrfs_item_offset_nr(l, end);
1553 data_len += sizeof(struct btrfs_item) * nr;
1554 WARN_ON(data_len < 0);
1555 return data_len;
1556 }
1557
1558 /*
1559 * The space between the end of the leaf items and
1560 * the start of the leaf data. IOW, how much room
1561 * the leaf has left for both items and data
1562 */
1563 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1564 {
1565 int nritems = btrfs_header_nritems(leaf);
1566 int ret;
1567 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1568 if (ret < 0) {
1569 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1570 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1571 leaf_space_used(leaf, 0, nritems), nritems);
1572 }
1573 return ret;
1574 }
1575
1576 /*
1577 * push some data in the path leaf to the right, trying to free up at
1578 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1579 *
1580 * returns 1 if the push failed because the other node didn't have enough
1581 * room, 0 if everything worked out and < 0 if there were major errors.
1582 */
1583 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1584 *root, struct btrfs_path *path, int data_size,
1585 int empty)
1586 {
1587 struct extent_buffer *left = path->nodes[0];
1588 struct extent_buffer *right;
1589 struct extent_buffer *upper;
1590 struct btrfs_disk_key disk_key;
1591 int slot;
1592 u32 i;
1593 int free_space;
1594 int push_space = 0;
1595 int push_items = 0;
1596 struct btrfs_item *item;
1597 u32 left_nritems;
1598 u32 nr;
1599 u32 right_nritems;
1600 u32 data_end;
1601 u32 this_item_size;
1602 int ret;
1603
1604 slot = path->slots[1];
1605 if (!path->nodes[1]) {
1606 return 1;
1607 }
1608 upper = path->nodes[1];
1609 if (slot >= btrfs_header_nritems(upper) - 1)
1610 return 1;
1611
1612 right = read_node_slot(root, upper, slot + 1);
1613 free_space = btrfs_leaf_free_space(root, right);
1614 if (free_space < data_size + sizeof(struct btrfs_item)) {
1615 free_extent_buffer(right);
1616 return 1;
1617 }
1618
1619 /* cow and double check */
1620 ret = btrfs_cow_block(trans, root, right, upper,
1621 slot + 1, &right);
1622 if (ret) {
1623 free_extent_buffer(right);
1624 return 1;
1625 }
1626 free_space = btrfs_leaf_free_space(root, right);
1627 if (free_space < data_size + sizeof(struct btrfs_item)) {
1628 free_extent_buffer(right);
1629 return 1;
1630 }
1631
1632 left_nritems = btrfs_header_nritems(left);
1633 if (left_nritems == 0) {
1634 free_extent_buffer(right);
1635 return 1;
1636 }
1637
1638 if (empty)
1639 nr = 0;
1640 else
1641 nr = 1;
1642
1643 i = left_nritems - 1;
1644 while (i >= nr) {
1645 item = btrfs_item_nr(left, i);
1646
1647 if (path->slots[0] == i)
1648 push_space += data_size + sizeof(*item);
1649
1650 this_item_size = btrfs_item_size(left, item);
1651 if (this_item_size + sizeof(*item) + push_space > free_space)
1652 break;
1653 push_items++;
1654 push_space += this_item_size + sizeof(*item);
1655 if (i == 0)
1656 break;
1657 i--;
1658 }
1659
1660 if (push_items == 0) {
1661 free_extent_buffer(right);
1662 return 1;
1663 }
1664
1665 if (!empty && push_items == left_nritems)
1666 WARN_ON(1);
1667
1668 /* push left to right */
1669 right_nritems = btrfs_header_nritems(right);
1670
1671 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1672 push_space -= leaf_data_end(root, left);
1673
1674 /* make room in the right data area */
1675 data_end = leaf_data_end(root, right);
1676 memmove_extent_buffer(right,
1677 btrfs_leaf_data(right) + data_end - push_space,
1678 btrfs_leaf_data(right) + data_end,
1679 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1680
1681 /* copy from the left data area */
1682 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1683 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1684 btrfs_leaf_data(left) + leaf_data_end(root, left),
1685 push_space);
1686
1687 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1688 btrfs_item_nr_offset(0),
1689 right_nritems * sizeof(struct btrfs_item));
1690
1691 /* copy the items from left to right */
1692 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1693 btrfs_item_nr_offset(left_nritems - push_items),
1694 push_items * sizeof(struct btrfs_item));
1695
1696 /* update the item pointers */
1697 right_nritems += push_items;
1698 btrfs_set_header_nritems(right, right_nritems);
1699 push_space = BTRFS_LEAF_DATA_SIZE(root);
1700 for (i = 0; i < right_nritems; i++) {
1701 item = btrfs_item_nr(right, i);
1702 push_space -= btrfs_item_size(right, item);
1703 btrfs_set_item_offset(right, item, push_space);
1704 }
1705
1706 left_nritems -= push_items;
1707 btrfs_set_header_nritems(left, left_nritems);
1708
1709 if (left_nritems)
1710 btrfs_mark_buffer_dirty(left);
1711 btrfs_mark_buffer_dirty(right);
1712
1713 btrfs_item_key(right, &disk_key, 0);
1714 btrfs_set_node_key(upper, &disk_key, slot + 1);
1715 btrfs_mark_buffer_dirty(upper);
1716
1717 /* then fixup the leaf pointer in the path */
1718 if (path->slots[0] >= left_nritems) {
1719 path->slots[0] -= left_nritems;
1720 free_extent_buffer(path->nodes[0]);
1721 path->nodes[0] = right;
1722 path->slots[1] += 1;
1723 } else {
1724 free_extent_buffer(right);
1725 }
1726 return 0;
1727 }
1728 /*
1729 * push some data in the path leaf to the left, trying to free up at
1730 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1731 */
1732 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1733 *root, struct btrfs_path *path, int data_size,
1734 int empty)
1735 {
1736 struct btrfs_disk_key disk_key;
1737 struct extent_buffer *right = path->nodes[0];
1738 struct extent_buffer *left;
1739 int slot;
1740 int i;
1741 int free_space;
1742 int push_space = 0;
1743 int push_items = 0;
1744 struct btrfs_item *item;
1745 u32 old_left_nritems;
1746 u32 right_nritems;
1747 u32 nr;
1748 int ret = 0;
1749 int wret;
1750 u32 this_item_size;
1751 u32 old_left_item_size;
1752
1753 slot = path->slots[1];
1754 if (slot == 0)
1755 return 1;
1756 if (!path->nodes[1])
1757 return 1;
1758
1759 right_nritems = btrfs_header_nritems(right);
1760 if (right_nritems == 0) {
1761 return 1;
1762 }
1763
1764 left = read_node_slot(root, path->nodes[1], slot - 1);
1765 free_space = btrfs_leaf_free_space(root, left);
1766 if (free_space < data_size + sizeof(struct btrfs_item)) {
1767 free_extent_buffer(left);
1768 return 1;
1769 }
1770
1771 /* cow and double check */
1772 ret = btrfs_cow_block(trans, root, left,
1773 path->nodes[1], slot - 1, &left);
1774 if (ret) {
1775 /* we hit -ENOSPC, but it isn't fatal here */
1776 free_extent_buffer(left);
1777 return 1;
1778 }
1779
1780 free_space = btrfs_leaf_free_space(root, left);
1781 if (free_space < data_size + sizeof(struct btrfs_item)) {
1782 free_extent_buffer(left);
1783 return 1;
1784 }
1785
1786 if (empty)
1787 nr = right_nritems;
1788 else
1789 nr = right_nritems - 1;
1790
1791 for (i = 0; i < nr; i++) {
1792 item = btrfs_item_nr(right, i);
1793
1794 if (path->slots[0] == i)
1795 push_space += data_size + sizeof(*item);
1796
1797 this_item_size = btrfs_item_size(right, item);
1798 if (this_item_size + sizeof(*item) + push_space > free_space)
1799 break;
1800
1801 push_items++;
1802 push_space += this_item_size + sizeof(*item);
1803 }
1804
1805 if (push_items == 0) {
1806 free_extent_buffer(left);
1807 return 1;
1808 }
1809 if (!empty && push_items == btrfs_header_nritems(right))
1810 WARN_ON(1);
1811
1812 /* push data from right to left */
1813 copy_extent_buffer(left, right,
1814 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1815 btrfs_item_nr_offset(0),
1816 push_items * sizeof(struct btrfs_item));
1817
1818 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1819 btrfs_item_offset_nr(right, push_items -1);
1820
1821 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1822 leaf_data_end(root, left) - push_space,
1823 btrfs_leaf_data(right) +
1824 btrfs_item_offset_nr(right, push_items - 1),
1825 push_space);
1826 old_left_nritems = btrfs_header_nritems(left);
1827 BUG_ON(old_left_nritems < 0);
1828
1829 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1830 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1831 u32 ioff;
1832
1833 item = btrfs_item_nr(left, i);
1834 ioff = btrfs_item_offset(left, item);
1835 btrfs_set_item_offset(left, item,
1836 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1837 }
1838 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1839
1840 /* fixup right node */
1841 if (push_items > right_nritems) {
1842 printk("push items %d nr %u\n", push_items, right_nritems);
1843 WARN_ON(1);
1844 }
1845
1846 if (push_items < right_nritems) {
1847 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1848 leaf_data_end(root, right);
1849 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1850 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1851 btrfs_leaf_data(right) +
1852 leaf_data_end(root, right), push_space);
1853
1854 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1855 btrfs_item_nr_offset(push_items),
1856 (btrfs_header_nritems(right) - push_items) *
1857 sizeof(struct btrfs_item));
1858 }
1859 right_nritems -= push_items;
1860 btrfs_set_header_nritems(right, right_nritems);
1861 push_space = BTRFS_LEAF_DATA_SIZE(root);
1862 for (i = 0; i < right_nritems; i++) {
1863 item = btrfs_item_nr(right, i);
1864 push_space = push_space - btrfs_item_size(right, item);
1865 btrfs_set_item_offset(right, item, push_space);
1866 }
1867
1868 btrfs_mark_buffer_dirty(left);
1869 if (right_nritems)
1870 btrfs_mark_buffer_dirty(right);
1871
1872 btrfs_item_key(right, &disk_key, 0);
1873 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
1874 if (wret)
1875 ret = wret;
1876
1877 /* then fixup the leaf pointer in the path */
1878 if (path->slots[0] < push_items) {
1879 path->slots[0] += old_left_nritems;
1880 free_extent_buffer(path->nodes[0]);
1881 path->nodes[0] = left;
1882 path->slots[1] -= 1;
1883 } else {
1884 free_extent_buffer(left);
1885 path->slots[0] -= push_items;
1886 }
1887 BUG_ON(path->slots[0] < 0);
1888 return ret;
1889 }
1890
1891 /*
1892 * split the path's leaf in two, making sure there is at least data_size
1893 * available for the resulting leaf level of the path.
1894 *
1895 * returns 0 if all went well and < 0 on failure.
1896 */
1897 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
1898 *root, struct btrfs_key *ins_key,
1899 struct btrfs_path *path, int data_size, int extend)
1900 {
1901 u64 root_gen;
1902 struct extent_buffer *l;
1903 u32 nritems;
1904 int mid;
1905 int slot;
1906 struct extent_buffer *right;
1907 int space_needed = data_size + sizeof(struct btrfs_item);
1908 int data_copy_size;
1909 int rt_data_off;
1910 int i;
1911 int ret = 0;
1912 int wret;
1913 int double_split;
1914 int num_doubles = 0;
1915 struct btrfs_disk_key disk_key;
1916
1917 if (extend)
1918 space_needed = data_size;
1919
1920 if (root->ref_cows)
1921 root_gen = trans->transid;
1922 else
1923 root_gen = 0;
1924
1925 /* first try to make some room by pushing left and right */
1926 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
1927 wret = push_leaf_right(trans, root, path, data_size, 0);
1928 if (wret < 0) {
1929 return wret;
1930 }
1931 if (wret) {
1932 wret = push_leaf_left(trans, root, path, data_size, 0);
1933 if (wret < 0)
1934 return wret;
1935 }
1936 l = path->nodes[0];
1937
1938 /* did the pushes work? */
1939 if (btrfs_leaf_free_space(root, l) >= space_needed)
1940 return 0;
1941 }
1942
1943 if (!path->nodes[1]) {
1944 ret = insert_new_root(trans, root, path, 1);
1945 if (ret)
1946 return ret;
1947 }
1948 again:
1949 double_split = 0;
1950 l = path->nodes[0];
1951 slot = path->slots[0];
1952 nritems = btrfs_header_nritems(l);
1953 mid = (nritems + 1)/ 2;
1954
1955 btrfs_item_key(l, &disk_key, 0);
1956
1957 right = __btrfs_alloc_free_block(trans, root, root->leafsize,
1958 root->root_key.objectid,
1959 root_gen, disk_key.objectid, 0,
1960 l->start, 0);
1961 if (IS_ERR(right)) {
1962 BUG_ON(1);
1963 return PTR_ERR(right);
1964 }
1965
1966 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
1967 btrfs_set_header_bytenr(right, right->start);
1968 btrfs_set_header_generation(right, trans->transid);
1969 btrfs_set_header_owner(right, root->root_key.objectid);
1970 btrfs_set_header_level(right, 0);
1971 write_extent_buffer(right, root->fs_info->fsid,
1972 (unsigned long)btrfs_header_fsid(right),
1973 BTRFS_FSID_SIZE);
1974
1975 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
1976 (unsigned long)btrfs_header_chunk_tree_uuid(right),
1977 BTRFS_UUID_SIZE);
1978 if (mid <= slot) {
1979 if (nritems == 1 ||
1980 leaf_space_used(l, mid, nritems - mid) + space_needed >
1981 BTRFS_LEAF_DATA_SIZE(root)) {
1982 if (slot >= nritems) {
1983 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1984 btrfs_set_header_nritems(right, 0);
1985 wret = insert_ptr(trans, root, path,
1986 &disk_key, right->start,
1987 path->slots[1] + 1, 1);
1988 if (wret)
1989 ret = wret;
1990 free_extent_buffer(path->nodes[0]);
1991 path->nodes[0] = right;
1992 path->slots[0] = 0;
1993 path->slots[1] += 1;
1994 return ret;
1995 }
1996 mid = slot;
1997 if (mid != nritems &&
1998 leaf_space_used(l, mid, nritems - mid) +
1999 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2000 double_split = 1;
2001 }
2002 }
2003 } else {
2004 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2005 BTRFS_LEAF_DATA_SIZE(root)) {
2006 if (!extend && slot == 0) {
2007 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2008 btrfs_set_header_nritems(right, 0);
2009 wret = insert_ptr(trans, root, path,
2010 &disk_key,
2011 right->start,
2012 path->slots[1], 1);
2013 if (wret)
2014 ret = wret;
2015 free_extent_buffer(path->nodes[0]);
2016 path->nodes[0] = right;
2017 path->slots[0] = 0;
2018 if (path->slots[1] == 0) {
2019 wret = fixup_low_keys(trans, root,
2020 path, &disk_key, 1);
2021 if (wret)
2022 ret = wret;
2023 }
2024 return ret;
2025 } else if (extend && slot == 0) {
2026 mid = 1;
2027 } else {
2028 mid = slot;
2029 if (mid != nritems &&
2030 leaf_space_used(l, mid, nritems - mid) +
2031 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2032 double_split = 1;
2033 }
2034 }
2035 }
2036 }
2037 nritems = nritems - mid;
2038 btrfs_set_header_nritems(right, nritems);
2039 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2040
2041 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2042 btrfs_item_nr_offset(mid),
2043 nritems * sizeof(struct btrfs_item));
2044
2045 copy_extent_buffer(right, l,
2046 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2047 data_copy_size, btrfs_leaf_data(l) +
2048 leaf_data_end(root, l), data_copy_size);
2049
2050 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2051 btrfs_item_end_nr(l, mid);
2052
2053 for (i = 0; i < nritems; i++) {
2054 struct btrfs_item *item = btrfs_item_nr(right, i);
2055 u32 ioff = btrfs_item_offset(right, item);
2056 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2057 }
2058
2059 btrfs_set_header_nritems(l, mid);
2060 ret = 0;
2061 btrfs_item_key(right, &disk_key, 0);
2062 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2063 path->slots[1] + 1, 1);
2064 if (wret)
2065 ret = wret;
2066
2067 btrfs_mark_buffer_dirty(right);
2068 btrfs_mark_buffer_dirty(l);
2069 BUG_ON(path->slots[0] != slot);
2070
2071 if (mid <= slot) {
2072 free_extent_buffer(path->nodes[0]);
2073 path->nodes[0] = right;
2074 path->slots[0] -= mid;
2075 path->slots[1] += 1;
2076 } else
2077 free_extent_buffer(right);
2078
2079 BUG_ON(path->slots[0] < 0);
2080
2081 if (double_split) {
2082 BUG_ON(num_doubles != 0);
2083 num_doubles++;
2084 goto again;
2085 }
2086 return ret;
2087 }
2088
2089 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2090 struct btrfs_root *root,
2091 struct btrfs_path *path,
2092 u32 new_size, int from_end)
2093 {
2094 int ret = 0;
2095 int slot;
2096 int slot_orig;
2097 struct extent_buffer *leaf;
2098 struct btrfs_item *item;
2099 u32 nritems;
2100 unsigned int data_end;
2101 unsigned int old_data_start;
2102 unsigned int old_size;
2103 unsigned int size_diff;
2104 int i;
2105
2106 slot_orig = path->slots[0];
2107 leaf = path->nodes[0];
2108 slot = path->slots[0];
2109
2110 old_size = btrfs_item_size_nr(leaf, slot);
2111 if (old_size == new_size)
2112 return 0;
2113
2114 nritems = btrfs_header_nritems(leaf);
2115 data_end = leaf_data_end(root, leaf);
2116
2117 old_data_start = btrfs_item_offset_nr(leaf, slot);
2118
2119 size_diff = old_size - new_size;
2120
2121 BUG_ON(slot < 0);
2122 BUG_ON(slot >= nritems);
2123
2124 /*
2125 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2126 */
2127 /* first correct the data pointers */
2128 for (i = slot; i < nritems; i++) {
2129 u32 ioff;
2130 item = btrfs_item_nr(leaf, i);
2131 ioff = btrfs_item_offset(leaf, item);
2132 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2133 }
2134
2135 /* shift the data */
2136 if (from_end) {
2137 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2138 data_end + size_diff, btrfs_leaf_data(leaf) +
2139 data_end, old_data_start + new_size - data_end);
2140 } else {
2141 struct btrfs_disk_key disk_key;
2142 u64 offset;
2143
2144 btrfs_item_key(leaf, &disk_key, slot);
2145
2146 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2147 unsigned long ptr;
2148 struct btrfs_file_extent_item *fi;
2149
2150 fi = btrfs_item_ptr(leaf, slot,
2151 struct btrfs_file_extent_item);
2152 fi = (struct btrfs_file_extent_item *)(
2153 (unsigned long)fi - size_diff);
2154
2155 if (btrfs_file_extent_type(leaf, fi) ==
2156 BTRFS_FILE_EXTENT_INLINE) {
2157 ptr = btrfs_item_ptr_offset(leaf, slot);
2158 memmove_extent_buffer(leaf, ptr,
2159 (unsigned long)fi,
2160 offsetof(struct btrfs_file_extent_item,
2161 disk_bytenr));
2162 }
2163 }
2164
2165 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2166 data_end + size_diff, btrfs_leaf_data(leaf) +
2167 data_end, old_data_start - data_end);
2168
2169 offset = btrfs_disk_key_offset(&disk_key);
2170 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2171 btrfs_set_item_key(leaf, &disk_key, slot);
2172 if (slot == 0)
2173 fixup_low_keys(trans, root, path, &disk_key, 1);
2174 }
2175
2176 item = btrfs_item_nr(leaf, slot);
2177 btrfs_set_item_size(leaf, item, new_size);
2178 btrfs_mark_buffer_dirty(leaf);
2179
2180 ret = 0;
2181 if (btrfs_leaf_free_space(root, leaf) < 0) {
2182 btrfs_print_leaf(root, leaf);
2183 BUG();
2184 }
2185 return ret;
2186 }
2187
2188 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2189 struct btrfs_root *root, struct btrfs_path *path,
2190 u32 data_size)
2191 {
2192 int ret = 0;
2193 int slot;
2194 int slot_orig;
2195 struct extent_buffer *leaf;
2196 struct btrfs_item *item;
2197 u32 nritems;
2198 unsigned int data_end;
2199 unsigned int old_data;
2200 unsigned int old_size;
2201 int i;
2202
2203 slot_orig = path->slots[0];
2204 leaf = path->nodes[0];
2205
2206 nritems = btrfs_header_nritems(leaf);
2207 data_end = leaf_data_end(root, leaf);
2208
2209 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2210 btrfs_print_leaf(root, leaf);
2211 BUG();
2212 }
2213 slot = path->slots[0];
2214 old_data = btrfs_item_end_nr(leaf, slot);
2215
2216 BUG_ON(slot < 0);
2217 if (slot >= nritems) {
2218 btrfs_print_leaf(root, leaf);
2219 printk("slot %d too large, nritems %d\n", slot, nritems);
2220 BUG_ON(1);
2221 }
2222
2223 /*
2224 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2225 */
2226 /* first correct the data pointers */
2227 for (i = slot; i < nritems; i++) {
2228 u32 ioff;
2229 item = btrfs_item_nr(leaf, i);
2230 ioff = btrfs_item_offset(leaf, item);
2231 btrfs_set_item_offset(leaf, item, ioff - data_size);
2232 }
2233
2234 /* shift the data */
2235 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2236 data_end - data_size, btrfs_leaf_data(leaf) +
2237 data_end, old_data - data_end);
2238
2239 data_end = old_data;
2240 old_size = btrfs_item_size_nr(leaf, slot);
2241 item = btrfs_item_nr(leaf, slot);
2242 btrfs_set_item_size(leaf, item, old_size + data_size);
2243 btrfs_mark_buffer_dirty(leaf);
2244
2245 ret = 0;
2246 if (btrfs_leaf_free_space(root, leaf) < 0) {
2247 btrfs_print_leaf(root, leaf);
2248 BUG();
2249 }
2250 return ret;
2251 }
2252
2253 /*
2254 * Given a key and some data, insert an item into the tree.
2255 * This does all the path init required, making room in the tree if needed.
2256 */
2257 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2258 struct btrfs_root *root,
2259 struct btrfs_path *path,
2260 struct btrfs_key *cpu_key, u32 *data_size,
2261 int nr)
2262 {
2263 struct extent_buffer *leaf;
2264 struct btrfs_item *item;
2265 int ret = 0;
2266 int slot;
2267 int slot_orig;
2268 int i;
2269 u32 nritems;
2270 u32 total_size = 0;
2271 u32 total_data = 0;
2272 unsigned int data_end;
2273 struct btrfs_disk_key disk_key;
2274
2275 for (i = 0; i < nr; i++) {
2276 total_data += data_size[i];
2277 }
2278
2279 /* create a root if there isn't one */
2280 if (!root->node)
2281 BUG();
2282
2283 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2284 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2285 if (ret == 0) {
2286 return -EEXIST;
2287 }
2288 if (ret < 0)
2289 goto out;
2290
2291 slot_orig = path->slots[0];
2292 leaf = path->nodes[0];
2293
2294 nritems = btrfs_header_nritems(leaf);
2295 data_end = leaf_data_end(root, leaf);
2296
2297 if (btrfs_leaf_free_space(root, leaf) <
2298 sizeof(struct btrfs_item) + total_size) {
2299 btrfs_print_leaf(root, leaf);
2300 printk("not enough freespace need %u have %d\n",
2301 total_size, btrfs_leaf_free_space(root, leaf));
2302 BUG();
2303 }
2304
2305 slot = path->slots[0];
2306 BUG_ON(slot < 0);
2307
2308 if (slot != nritems) {
2309 int i;
2310 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2311
2312 if (old_data < data_end) {
2313 btrfs_print_leaf(root, leaf);
2314 printk("slot %d old_data %d data_end %d\n",
2315 slot, old_data, data_end);
2316 BUG_ON(1);
2317 }
2318 /*
2319 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2320 */
2321 /* first correct the data pointers */
2322 for (i = slot; i < nritems; i++) {
2323 u32 ioff;
2324
2325 item = btrfs_item_nr(leaf, i);
2326 ioff = btrfs_item_offset(leaf, item);
2327 btrfs_set_item_offset(leaf, item, ioff - total_data);
2328 }
2329
2330 /* shift the items */
2331 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2332 btrfs_item_nr_offset(slot),
2333 (nritems - slot) * sizeof(struct btrfs_item));
2334
2335 /* shift the data */
2336 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2337 data_end - total_data, btrfs_leaf_data(leaf) +
2338 data_end, old_data - data_end);
2339 data_end = old_data;
2340 }
2341
2342 /* setup the item for the new data */
2343 for (i = 0; i < nr; i++) {
2344 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2345 btrfs_set_item_key(leaf, &disk_key, slot + i);
2346 item = btrfs_item_nr(leaf, slot + i);
2347 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2348 data_end -= data_size[i];
2349 btrfs_set_item_size(leaf, item, data_size[i]);
2350 }
2351 btrfs_set_header_nritems(leaf, nritems + nr);
2352 btrfs_mark_buffer_dirty(leaf);
2353
2354 ret = 0;
2355 if (slot == 0) {
2356 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2357 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2358 }
2359
2360 if (btrfs_leaf_free_space(root, leaf) < 0) {
2361 btrfs_print_leaf(root, leaf);
2362 BUG();
2363 }
2364
2365 out:
2366 return ret;
2367 }
2368
2369 /*
2370 * Given a key and some data, insert an item into the tree.
2371 * This does all the path init required, making room in the tree if needed.
2372 */
2373 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2374 *root, struct btrfs_key *cpu_key, void *data, u32
2375 data_size)
2376 {
2377 int ret = 0;
2378 struct btrfs_path *path;
2379 struct extent_buffer *leaf;
2380 unsigned long ptr;
2381
2382 path = btrfs_alloc_path();
2383 BUG_ON(!path);
2384 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2385 if (!ret) {
2386 leaf = path->nodes[0];
2387 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2388 write_extent_buffer(leaf, data, ptr, data_size);
2389 btrfs_mark_buffer_dirty(leaf);
2390 }
2391 btrfs_free_path(path);
2392 return ret;
2393 }
2394
2395 /*
2396 * delete the pointer from a given node.
2397 *
2398 * If the delete empties a node, the node is removed from the tree,
2399 * continuing all the way the root if required. The root is converted into
2400 * a leaf if all the nodes are emptied.
2401 */
2402 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2403 struct btrfs_path *path, int level, int slot)
2404 {
2405 struct extent_buffer *parent = path->nodes[level];
2406 u32 nritems;
2407 int ret = 0;
2408 int wret;
2409
2410 nritems = btrfs_header_nritems(parent);
2411 if (slot != nritems -1) {
2412 memmove_extent_buffer(parent,
2413 btrfs_node_key_ptr_offset(slot),
2414 btrfs_node_key_ptr_offset(slot + 1),
2415 sizeof(struct btrfs_key_ptr) *
2416 (nritems - slot - 1));
2417 }
2418 nritems--;
2419 btrfs_set_header_nritems(parent, nritems);
2420 if (nritems == 0 && parent == root->node) {
2421 BUG_ON(btrfs_header_level(root->node) != 1);
2422 /* just turn the root into a leaf and break */
2423 btrfs_set_header_level(root->node, 0);
2424 } else if (slot == 0) {
2425 struct btrfs_disk_key disk_key;
2426
2427 btrfs_node_key(parent, &disk_key, 0);
2428 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2429 if (wret)
2430 ret = wret;
2431 }
2432 btrfs_mark_buffer_dirty(parent);
2433 return ret;
2434 }
2435
2436 /*
2437 * delete the item at the leaf level in path. If that empties
2438 * the leaf, remove it from the tree
2439 */
2440 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2441 struct btrfs_path *path, int slot, int nr)
2442 {
2443 struct extent_buffer *leaf;
2444 struct btrfs_item *item;
2445 int last_off;
2446 int dsize = 0;
2447 int ret = 0;
2448 int wret;
2449 int i;
2450 u32 nritems;
2451
2452 leaf = path->nodes[0];
2453 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2454
2455 for (i = 0; i < nr; i++)
2456 dsize += btrfs_item_size_nr(leaf, slot + i);
2457
2458 nritems = btrfs_header_nritems(leaf);
2459
2460 if (slot + nr != nritems) {
2461 int i;
2462 int data_end = leaf_data_end(root, leaf);
2463
2464 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2465 data_end + dsize,
2466 btrfs_leaf_data(leaf) + data_end,
2467 last_off - data_end);
2468
2469 for (i = slot + nr; i < nritems; i++) {
2470 u32 ioff;
2471
2472 item = btrfs_item_nr(leaf, i);
2473 ioff = btrfs_item_offset(leaf, item);
2474 btrfs_set_item_offset(leaf, item, ioff + dsize);
2475 }
2476
2477 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2478 btrfs_item_nr_offset(slot + nr),
2479 sizeof(struct btrfs_item) *
2480 (nritems - slot - nr));
2481 }
2482 btrfs_set_header_nritems(leaf, nritems - nr);
2483 nritems -= nr;
2484
2485 /* delete the leaf if we've emptied it */
2486 if (nritems == 0) {
2487 if (leaf == root->node) {
2488 btrfs_set_header_level(leaf, 0);
2489 } else {
2490 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2491 clean_tree_block(trans, root, leaf);
2492 wait_on_tree_block_writeback(root, leaf);
2493 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2494 if (wret)
2495 ret = wret;
2496 wret = btrfs_free_extent(trans, root,
2497 leaf->start, leaf->len,
2498 btrfs_header_owner(path->nodes[1]),
2499 root_gen, 0, 0, 1);
2500 if (wret)
2501 ret = wret;
2502 }
2503 } else {
2504 int used = leaf_space_used(leaf, 0, nritems);
2505 if (slot == 0) {
2506 struct btrfs_disk_key disk_key;
2507
2508 btrfs_item_key(leaf, &disk_key, 0);
2509 wret = fixup_low_keys(trans, root, path,
2510 &disk_key, 1);
2511 if (wret)
2512 ret = wret;
2513 }
2514
2515 /* delete the leaf if it is mostly empty */
2516 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2517 /* push_leaf_left fixes the path.
2518 * make sure the path still points to our leaf
2519 * for possible call to del_ptr below
2520 */
2521 slot = path->slots[1];
2522 extent_buffer_get(leaf);
2523
2524 wret = push_leaf_left(trans, root, path, 1, 1);
2525 if (wret < 0 && wret != -ENOSPC)
2526 ret = wret;
2527
2528 if (path->nodes[0] == leaf &&
2529 btrfs_header_nritems(leaf)) {
2530 wret = push_leaf_right(trans, root, path, 1, 1);
2531 if (wret < 0 && wret != -ENOSPC)
2532 ret = wret;
2533 }
2534
2535 if (btrfs_header_nritems(leaf) == 0) {
2536 u64 root_gen;
2537 u64 bytenr = leaf->start;
2538 u32 blocksize = leaf->len;
2539
2540 root_gen = btrfs_header_generation(
2541 path->nodes[1]);
2542
2543 clean_tree_block(trans, root, leaf);
2544 wait_on_tree_block_writeback(root, leaf);
2545
2546 wret = del_ptr(trans, root, path, 1, slot);
2547 if (wret)
2548 ret = wret;
2549
2550 free_extent_buffer(leaf);
2551 wret = btrfs_free_extent(trans, root, bytenr,
2552 blocksize,
2553 btrfs_header_owner(path->nodes[1]),
2554 root_gen, 0, 0, 1);
2555 if (wret)
2556 ret = wret;
2557 } else {
2558 btrfs_mark_buffer_dirty(leaf);
2559 free_extent_buffer(leaf);
2560 }
2561 } else {
2562 btrfs_mark_buffer_dirty(leaf);
2563 }
2564 }
2565 return ret;
2566 }
2567
2568 /*
2569 * walk up the tree as far as required to find the previous leaf.
2570 * returns 0 if it found something or 1 if there are no lesser leaves.
2571 * returns < 0 on io errors.
2572 */
2573 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2574 {
2575 int slot;
2576 int level = 1;
2577 struct extent_buffer *c;
2578 struct extent_buffer *next = NULL;
2579
2580 while(level < BTRFS_MAX_LEVEL) {
2581 if (!path->nodes[level])
2582 return 1;
2583
2584 slot = path->slots[level];
2585 c = path->nodes[level];
2586 if (slot == 0) {
2587 level++;
2588 if (level == BTRFS_MAX_LEVEL)
2589 return 1;
2590 continue;
2591 }
2592 slot--;
2593
2594 if (next)
2595 free_extent_buffer(next);
2596
2597 next = read_node_slot(root, c, slot);
2598 break;
2599 }
2600 path->slots[level] = slot;
2601 while(1) {
2602 level--;
2603 c = path->nodes[level];
2604 free_extent_buffer(c);
2605 slot = btrfs_header_nritems(next);
2606 if (slot != 0)
2607 slot--;
2608 path->nodes[level] = next;
2609 path->slots[level] = slot;
2610 if (!level)
2611 break;
2612 next = read_node_slot(root, next, slot);
2613 }
2614 return 0;
2615 }
2616
2617 /*
2618 * walk up the tree as far as required to find the next leaf.
2619 * returns 0 if it found something or 1 if there are no greater leaves.
2620 * returns < 0 on io errors.
2621 */
2622 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2623 {
2624 int slot;
2625 int level = 1;
2626 struct extent_buffer *c;
2627 struct extent_buffer *next = NULL;
2628
2629 while(level < BTRFS_MAX_LEVEL) {
2630 if (!path->nodes[level])
2631 return 1;
2632
2633 slot = path->slots[level] + 1;
2634 c = path->nodes[level];
2635 if (slot >= btrfs_header_nritems(c)) {
2636 level++;
2637 if (level == BTRFS_MAX_LEVEL)
2638 return 1;
2639 continue;
2640 }
2641
2642 if (next)
2643 free_extent_buffer(next);
2644
2645 if (path->reada)
2646 reada_for_search(root, path, level, slot, 0);
2647
2648 next = read_node_slot(root, c, slot);
2649 break;
2650 }
2651 path->slots[level] = slot;
2652 while(1) {
2653 level--;
2654 c = path->nodes[level];
2655 free_extent_buffer(c);
2656 path->nodes[level] = next;
2657 path->slots[level] = 0;
2658 if (!level)
2659 break;
2660 if (path->reada)
2661 reada_for_search(root, path, level, 0, 0);
2662 next = read_node_slot(root, next, 0);
2663 }
2664 return 0;
2665 }
2666
2667 int btrfs_previous_item(struct btrfs_root *root,
2668 struct btrfs_path *path, u64 min_objectid,
2669 int type)
2670 {
2671 struct btrfs_key found_key;
2672 struct extent_buffer *leaf;
2673 int ret;
2674
2675 while(1) {
2676 if (path->slots[0] == 0) {
2677 ret = btrfs_prev_leaf(root, path);
2678 if (ret != 0)
2679 return ret;
2680 } else {
2681 path->slots[0]--;
2682 }
2683 leaf = path->nodes[0];
2684 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2685 if (found_key.type == type)
2686 return 0;
2687 }
2688 return 1;
2689 }
2690
Btrfs-progs-unstable mirror