projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: Start btree concurrency work.
[deliverable/linux.git] / fs / btrfs / ctree.c
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
19 #include <linux/sched.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "transaction.h"
23 #include "print-tree.h"
24 #include "locking.h"
25
26 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_path *path, int level);
28 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_key *ins_key,
30 struct btrfs_path *path, int data_size, int extend);
31 static int push_node_left(struct btrfs_trans_handle *trans,
32 struct btrfs_root *root, struct extent_buffer *dst,
33 struct extent_buffer *src, int empty);
34 static int balance_node_right(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root,
36 struct extent_buffer *dst_buf,
37 struct extent_buffer *src_buf);
38 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
39 struct btrfs_path *path, int level, int slot);
40
41 inline void btrfs_init_path(struct btrfs_path *p)
42 {
43 memset(p, 0, sizeof(*p));
44 }
45
46 struct btrfs_path *btrfs_alloc_path(void)
47 {
48 struct btrfs_path *path;
49 path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
50 if (path) {
51 btrfs_init_path(path);
52 path->reada = 1;
53 }
54 return path;
55 }
56
57 void btrfs_free_path(struct btrfs_path *p)
58 {
59 btrfs_release_path(NULL, p);
60 kmem_cache_free(btrfs_path_cachep, p);
61 }
62
63 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
64 {
65 int i;
66 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
67 if (!p->nodes[i])
68 continue;
69 if (p->locks[i]) {
70 btrfs_tree_unlock(p->nodes[i]);
71 p->locks[i] = 0;
72 }
73 free_extent_buffer(p->nodes[i]);
74 }
75 memset(p, 0, sizeof(*p));
76 }
77
78 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
79 {
80 struct extent_buffer *eb;
81 spin_lock(&root->node_lock);
82 eb = root->node;
83 extent_buffer_get(eb);
84 spin_unlock(&root->node_lock);
85 return eb;
86 }
87
88 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
89 {
90 struct extent_buffer *eb;
91
92 while(1) {
93 eb = btrfs_root_node(root);
94 btrfs_tree_lock(eb);
95
96 spin_lock(&root->node_lock);
97 if (eb == root->node) {
98 spin_unlock(&root->node_lock);
99 break;
100 }
101 spin_unlock(&root->node_lock);
102
103 btrfs_tree_unlock(eb);
104 free_extent_buffer(eb);
105 }
106 return eb;
107 }
108
109 static void add_root_to_dirty_list(struct btrfs_root *root)
110 {
111 if (root->track_dirty && list_empty(&root->dirty_list)) {
112 list_add(&root->dirty_list,
113 &root->fs_info->dirty_cowonly_roots);
114 }
115 }
116
117 int btrfs_copy_root(struct btrfs_trans_handle *trans,
118 struct btrfs_root *root,
119 struct extent_buffer *buf,
120 struct extent_buffer **cow_ret, u64 new_root_objectid)
121 {
122 struct extent_buffer *cow;
123 u32 nritems;
124 int ret = 0;
125 int level;
126 struct btrfs_key first_key;
127 struct btrfs_root *new_root;
128
129 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
130 if (!new_root)
131 return -ENOMEM;
132
133 memcpy(new_root, root, sizeof(*new_root));
134 new_root->root_key.objectid = new_root_objectid;
135
136 WARN_ON(root->ref_cows && trans->transid !=
137 root->fs_info->running_transaction->transid);
138 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
139
140 level = btrfs_header_level(buf);
141 nritems = btrfs_header_nritems(buf);
142 if (nritems) {
143 if (level == 0)
144 btrfs_item_key_to_cpu(buf, &first_key, 0);
145 else
146 btrfs_node_key_to_cpu(buf, &first_key, 0);
147 } else {
148 first_key.objectid = 0;
149 }
150 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
151 new_root_objectid,
152 trans->transid, first_key.objectid,
153 level, buf->start, 0);
154 if (IS_ERR(cow)) {
155 kfree(new_root);
156 return PTR_ERR(cow);
157 }
158
159 copy_extent_buffer(cow, buf, 0, 0, cow->len);
160 btrfs_set_header_bytenr(cow, cow->start);
161 btrfs_set_header_generation(cow, trans->transid);
162 btrfs_set_header_owner(cow, new_root_objectid);
163 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
164
165 WARN_ON(btrfs_header_generation(buf) > trans->transid);
166 ret = btrfs_inc_ref(trans, new_root, buf);
167 kfree(new_root);
168
169 if (ret)
170 return ret;
171
172 btrfs_mark_buffer_dirty(cow);
173 *cow_ret = cow;
174 return 0;
175 }
176
177 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
178 struct btrfs_root *root,
179 struct extent_buffer *buf,
180 struct extent_buffer *parent, int parent_slot,
181 struct extent_buffer **cow_ret,
182 u64 search_start, u64 empty_size)
183 {
184 u64 root_gen;
185 struct extent_buffer *cow;
186 u32 nritems;
187 int ret = 0;
188 int different_trans = 0;
189 int level;
190 int unlock_orig = 0;
191 struct btrfs_key first_key;
192
193 if (*cow_ret == buf)
194 unlock_orig = 1;
195
196 WARN_ON(!btrfs_tree_locked(buf));
197
198 if (root->ref_cows) {
199 root_gen = trans->transid;
200 } else {
201 root_gen = 0;
202 }
203 WARN_ON(root->ref_cows && trans->transid !=
204 root->fs_info->running_transaction->transid);
205 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
206
207 level = btrfs_header_level(buf);
208 nritems = btrfs_header_nritems(buf);
209 if (nritems) {
210 if (level == 0)
211 btrfs_item_key_to_cpu(buf, &first_key, 0);
212 else
213 btrfs_node_key_to_cpu(buf, &first_key, 0);
214 } else {
215 first_key.objectid = 0;
216 }
217 cow = btrfs_alloc_free_block(trans, root, buf->len,
218 root->root_key.objectid,
219 root_gen, first_key.objectid, level,
220 search_start, empty_size);
221 if (IS_ERR(cow))
222 return PTR_ERR(cow);
223
224 copy_extent_buffer(cow, buf, 0, 0, cow->len);
225 btrfs_set_header_bytenr(cow, cow->start);
226 btrfs_set_header_generation(cow, trans->transid);
227 btrfs_set_header_owner(cow, root->root_key.objectid);
228 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
229
230 WARN_ON(btrfs_header_generation(buf) > trans->transid);
231 if (btrfs_header_generation(buf) != trans->transid) {
232 different_trans = 1;
233 ret = btrfs_inc_ref(trans, root, buf);
234 if (ret)
235 return ret;
236 } else {
237 clean_tree_block(trans, root, buf);
238 }
239
240 if (buf == root->node) {
241 WARN_ON(parent && parent != buf);
242 root_gen = btrfs_header_generation(buf);
243
244 spin_lock(&root->node_lock);
245 root->node = cow;
246 extent_buffer_get(cow);
247 spin_unlock(&root->node_lock);
248
249 if (buf != root->commit_root) {
250 btrfs_free_extent(trans, root, buf->start,
251 buf->len, root->root_key.objectid,
252 root_gen, 0, 0, 1);
253 }
254 free_extent_buffer(buf);
255 add_root_to_dirty_list(root);
256 } else {
257 root_gen = btrfs_header_generation(parent);
258 btrfs_set_node_blockptr(parent, parent_slot,
259 cow->start);
260 WARN_ON(trans->transid == 0);
261 btrfs_set_node_ptr_generation(parent, parent_slot,
262 trans->transid);
263 btrfs_mark_buffer_dirty(parent);
264 WARN_ON(btrfs_header_generation(parent) != trans->transid);
265 btrfs_free_extent(trans, root, buf->start, buf->len,
266 btrfs_header_owner(parent), root_gen,
267 0, 0, 1);
268 }
269 if (unlock_orig)
270 btrfs_tree_unlock(buf);
271 free_extent_buffer(buf);
272 btrfs_mark_buffer_dirty(cow);
273 *cow_ret = cow;
274 return 0;
275 }
276
277 int btrfs_cow_block(struct btrfs_trans_handle *trans,
278 struct btrfs_root *root, struct extent_buffer *buf,
279 struct extent_buffer *parent, int parent_slot,
280 struct extent_buffer **cow_ret)
281 {
282 u64 search_start;
283 u64 header_trans;
284 int ret;
285
286 if (trans->transaction != root->fs_info->running_transaction) {
287 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
288 root->fs_info->running_transaction->transid);
289 WARN_ON(1);
290 }
291 if (trans->transid != root->fs_info->generation) {
292 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
293 root->fs_info->generation);
294 WARN_ON(1);
295 }
296
297 header_trans = btrfs_header_generation(buf);
298 spin_lock(&root->fs_info->hash_lock);
299 if (header_trans == trans->transid &&
300 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
301 *cow_ret = buf;
302 spin_unlock(&root->fs_info->hash_lock);
303 return 0;
304 }
305 spin_unlock(&root->fs_info->hash_lock);
306 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
307 ret = __btrfs_cow_block(trans, root, buf, parent,
308 parent_slot, cow_ret, search_start, 0);
309 return ret;
310 }
311
312 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
313 {
314 if (blocknr < other && other - (blocknr + blocksize) < 32768)
315 return 1;
316 if (blocknr > other && blocknr - (other + blocksize) < 32768)
317 return 1;
318 return 0;
319 }
320
321 /*
322 * compare two keys in a memcmp fashion
323 */
324 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
325 {
326 struct btrfs_key k1;
327
328 btrfs_disk_key_to_cpu(&k1, disk);
329
330 if (k1.objectid > k2->objectid)
331 return 1;
332 if (k1.objectid < k2->objectid)
333 return -1;
334 if (k1.type > k2->type)
335 return 1;
336 if (k1.type < k2->type)
337 return -1;
338 if (k1.offset > k2->offset)
339 return 1;
340 if (k1.offset < k2->offset)
341 return -1;
342 return 0;
343 }
344
345
346 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
347 struct btrfs_root *root, struct extent_buffer *parent,
348 int start_slot, int cache_only, u64 *last_ret,
349 struct btrfs_key *progress)
350 {
351 struct extent_buffer *cur;
352 struct extent_buffer *tmp;
353 u64 blocknr;
354 u64 gen;
355 u64 search_start = *last_ret;
356 u64 last_block = 0;
357 u64 other;
358 u32 parent_nritems;
359 int end_slot;
360 int i;
361 int err = 0;
362 int parent_level;
363 int uptodate;
364 u32 blocksize;
365 int progress_passed = 0;
366 struct btrfs_disk_key disk_key;
367
368 /* FIXME this code needs locking */
369 return 0;
370
371 parent_level = btrfs_header_level(parent);
372 if (cache_only && parent_level != 1)
373 return 0;
374
375 if (trans->transaction != root->fs_info->running_transaction) {
376 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
377 root->fs_info->running_transaction->transid);
378 WARN_ON(1);
379 }
380 if (trans->transid != root->fs_info->generation) {
381 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
382 root->fs_info->generation);
383 WARN_ON(1);
384 }
385
386 parent_nritems = btrfs_header_nritems(parent);
387 blocksize = btrfs_level_size(root, parent_level - 1);
388 end_slot = parent_nritems;
389
390 if (parent_nritems == 1)
391 return 0;
392
393 for (i = start_slot; i < end_slot; i++) {
394 int close = 1;
395
396 if (!parent->map_token) {
397 map_extent_buffer(parent,
398 btrfs_node_key_ptr_offset(i),
399 sizeof(struct btrfs_key_ptr),
400 &parent->map_token, &parent->kaddr,
401 &parent->map_start, &parent->map_len,
402 KM_USER1);
403 }
404 btrfs_node_key(parent, &disk_key, i);
405 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
406 continue;
407
408 progress_passed = 1;
409 blocknr = btrfs_node_blockptr(parent, i);
410 gen = btrfs_node_ptr_generation(parent, i);
411 if (last_block == 0)
412 last_block = blocknr;
413
414 if (i > 0) {
415 other = btrfs_node_blockptr(parent, i - 1);
416 close = close_blocks(blocknr, other, blocksize);
417 }
418 if (!close && i < end_slot - 2) {
419 other = btrfs_node_blockptr(parent, i + 1);
420 close = close_blocks(blocknr, other, blocksize);
421 }
422 if (close) {
423 last_block = blocknr;
424 continue;
425 }
426 if (parent->map_token) {
427 unmap_extent_buffer(parent, parent->map_token,
428 KM_USER1);
429 parent->map_token = NULL;
430 }
431
432 cur = btrfs_find_tree_block(root, blocknr, blocksize);
433 if (cur)
434 uptodate = btrfs_buffer_uptodate(cur, gen);
435 else
436 uptodate = 0;
437 if (!cur || !uptodate) {
438 if (cache_only) {
439 free_extent_buffer(cur);
440 continue;
441 }
442 if (!cur) {
443 cur = read_tree_block(root, blocknr,
444 blocksize, gen);
445 } else if (!uptodate) {
446 btrfs_read_buffer(cur, gen);
447 }
448 }
449 if (search_start == 0)
450 search_start = last_block;
451
452 err = __btrfs_cow_block(trans, root, cur, parent, i,
453 &tmp, search_start,
454 min(16 * blocksize,
455 (end_slot - i) * blocksize));
456 if (err) {
457 free_extent_buffer(cur);
458 break;
459 }
460 search_start = tmp->start;
461 last_block = tmp->start;
462 *last_ret = search_start;
463 if (parent_level == 1)
464 btrfs_clear_buffer_defrag(tmp);
465 free_extent_buffer(tmp);
466 }
467 if (parent->map_token) {
468 unmap_extent_buffer(parent, parent->map_token,
469 KM_USER1);
470 parent->map_token = NULL;
471 }
472 return err;
473 }
474
475 /*
476 * The leaf data grows from end-to-front in the node.
477 * this returns the address of the start of the last item,
478 * which is the stop of the leaf data stack
479 */
480 static inline unsigned int leaf_data_end(struct btrfs_root *root,
481 struct extent_buffer *leaf)
482 {
483 u32 nr = btrfs_header_nritems(leaf);
484 if (nr == 0)
485 return BTRFS_LEAF_DATA_SIZE(root);
486 return btrfs_item_offset_nr(leaf, nr - 1);
487 }
488
489 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
490 int level)
491 {
492 struct extent_buffer *parent = NULL;
493 struct extent_buffer *node = path->nodes[level];
494 struct btrfs_disk_key parent_key;
495 struct btrfs_disk_key node_key;
496 int parent_slot;
497 int slot;
498 struct btrfs_key cpukey;
499 u32 nritems = btrfs_header_nritems(node);
500
501 if (path->nodes[level + 1])
502 parent = path->nodes[level + 1];
503
504 slot = path->slots[level];
505 BUG_ON(nritems == 0);
506 if (parent) {
507 parent_slot = path->slots[level + 1];
508 btrfs_node_key(parent, &parent_key, parent_slot);
509 btrfs_node_key(node, &node_key, 0);
510 BUG_ON(memcmp(&parent_key, &node_key,
511 sizeof(struct btrfs_disk_key)));
512 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
513 btrfs_header_bytenr(node));
514 }
515 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
516 if (slot != 0) {
517 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
518 btrfs_node_key(node, &node_key, slot);
519 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
520 }
521 if (slot < nritems - 1) {
522 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
523 btrfs_node_key(node, &node_key, slot);
524 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
525 }
526 return 0;
527 }
528
529 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
530 int level)
531 {
532 struct extent_buffer *leaf = path->nodes[level];
533 struct extent_buffer *parent = NULL;
534 int parent_slot;
535 struct btrfs_key cpukey;
536 struct btrfs_disk_key parent_key;
537 struct btrfs_disk_key leaf_key;
538 int slot = path->slots[0];
539
540 u32 nritems = btrfs_header_nritems(leaf);
541
542 if (path->nodes[level + 1])
543 parent = path->nodes[level + 1];
544
545 if (nritems == 0)
546 return 0;
547
548 if (parent) {
549 parent_slot = path->slots[level + 1];
550 btrfs_node_key(parent, &parent_key, parent_slot);
551 btrfs_item_key(leaf, &leaf_key, 0);
552
553 BUG_ON(memcmp(&parent_key, &leaf_key,
554 sizeof(struct btrfs_disk_key)));
555 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
556 btrfs_header_bytenr(leaf));
557 }
558 #if 0
559 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
560 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
561 btrfs_item_key(leaf, &leaf_key, i);
562 if (comp_keys(&leaf_key, &cpukey) >= 0) {
563 btrfs_print_leaf(root, leaf);
564 printk("slot %d offset bad key\n", i);
565 BUG_ON(1);
566 }
567 if (btrfs_item_offset_nr(leaf, i) !=
568 btrfs_item_end_nr(leaf, i + 1)) {
569 btrfs_print_leaf(root, leaf);
570 printk("slot %d offset bad\n", i);
571 BUG_ON(1);
572 }
573 if (i == 0) {
574 if (btrfs_item_offset_nr(leaf, i) +
575 btrfs_item_size_nr(leaf, i) !=
576 BTRFS_LEAF_DATA_SIZE(root)) {
577 btrfs_print_leaf(root, leaf);
578 printk("slot %d first offset bad\n", i);
579 BUG_ON(1);
580 }
581 }
582 }
583 if (nritems > 0) {
584 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
585 btrfs_print_leaf(root, leaf);
586 printk("slot %d bad size \n", nritems - 1);
587 BUG_ON(1);
588 }
589 }
590 #endif
591 if (slot != 0 && slot < nritems - 1) {
592 btrfs_item_key(leaf, &leaf_key, slot);
593 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
594 if (comp_keys(&leaf_key, &cpukey) <= 0) {
595 btrfs_print_leaf(root, leaf);
596 printk("slot %d offset bad key\n", slot);
597 BUG_ON(1);
598 }
599 if (btrfs_item_offset_nr(leaf, slot - 1) !=
600 btrfs_item_end_nr(leaf, slot)) {
601 btrfs_print_leaf(root, leaf);
602 printk("slot %d offset bad\n", slot);
603 BUG_ON(1);
604 }
605 }
606 if (slot < nritems - 1) {
607 btrfs_item_key(leaf, &leaf_key, slot);
608 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
609 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
610 if (btrfs_item_offset_nr(leaf, slot) !=
611 btrfs_item_end_nr(leaf, slot + 1)) {
612 btrfs_print_leaf(root, leaf);
613 printk("slot %d offset bad\n", slot);
614 BUG_ON(1);
615 }
616 }
617 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
618 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
619 return 0;
620 }
621
622 static int noinline check_block(struct btrfs_root *root,
623 struct btrfs_path *path, int level)
624 {
625 u64 found_start;
626 return 0;
627 if (btrfs_header_level(path->nodes[level]) != level)
628 printk("warning: bad level %Lu wanted %d found %d\n",
629 path->nodes[level]->start, level,
630 btrfs_header_level(path->nodes[level]));
631 found_start = btrfs_header_bytenr(path->nodes[level]);
632 if (found_start != path->nodes[level]->start) {
633 printk("warning: bad bytentr %Lu found %Lu\n",
634 path->nodes[level]->start, found_start);
635 }
636 #if 0
637 struct extent_buffer *buf = path->nodes[level];
638
639 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
640 (unsigned long)btrfs_header_fsid(buf),
641 BTRFS_FSID_SIZE)) {
642 printk("warning bad block %Lu\n", buf->start);
643 return 1;
644 }
645 #endif
646 if (level == 0)
647 return check_leaf(root, path, level);
648 return check_node(root, path, level);
649 }
650
651 /*
652 * search for key in the extent_buffer. The items start at offset p,
653 * and they are item_size apart. There are 'max' items in p.
654 *
655 * the slot in the array is returned via slot, and it points to
656 * the place where you would insert key if it is not found in
657 * the array.
658 *
659 * slot may point to max if the key is bigger than all of the keys
660 */
661 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
662 int item_size, struct btrfs_key *key,
663 int max, int *slot)
664 {
665 int low = 0;
666 int high = max;
667 int mid;
668 int ret;
669 struct btrfs_disk_key *tmp = NULL;
670 struct btrfs_disk_key unaligned;
671 unsigned long offset;
672 char *map_token = NULL;
673 char *kaddr = NULL;
674 unsigned long map_start = 0;
675 unsigned long map_len = 0;
676 int err;
677
678 while(low < high) {
679 mid = (low + high) / 2;
680 offset = p + mid * item_size;
681
682 if (!map_token || offset < map_start ||
683 (offset + sizeof(struct btrfs_disk_key)) >
684 map_start + map_len) {
685 if (map_token) {
686 unmap_extent_buffer(eb, map_token, KM_USER0);
687 map_token = NULL;
688 }
689 err = map_extent_buffer(eb, offset,
690 sizeof(struct btrfs_disk_key),
691 &map_token, &kaddr,
692 &map_start, &map_len, KM_USER0);
693
694 if (!err) {
695 tmp = (struct btrfs_disk_key *)(kaddr + offset -
696 map_start);
697 } else {
698 read_extent_buffer(eb, &unaligned,
699 offset, sizeof(unaligned));
700 tmp = &unaligned;
701 }
702
703 } else {
704 tmp = (struct btrfs_disk_key *)(kaddr + offset -
705 map_start);
706 }
707 ret = comp_keys(tmp, key);
708
709 if (ret < 0)
710 low = mid + 1;
711 else if (ret > 0)
712 high = mid;
713 else {
714 *slot = mid;
715 if (map_token)
716 unmap_extent_buffer(eb, map_token, KM_USER0);
717 return 0;
718 }
719 }
720 *slot = low;
721 if (map_token)
722 unmap_extent_buffer(eb, map_token, KM_USER0);
723 return 1;
724 }
725
726 /*
727 * simple bin_search frontend that does the right thing for
728 * leaves vs nodes
729 */
730 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
731 int level, int *slot)
732 {
733 if (level == 0) {
734 return generic_bin_search(eb,
735 offsetof(struct btrfs_leaf, items),
736 sizeof(struct btrfs_item),
737 key, btrfs_header_nritems(eb),
738 slot);
739 } else {
740 return generic_bin_search(eb,
741 offsetof(struct btrfs_node, ptrs),
742 sizeof(struct btrfs_key_ptr),
743 key, btrfs_header_nritems(eb),
744 slot);
745 }
746 return -1;
747 }
748
749 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
750 struct extent_buffer *parent, int slot)
751 {
752 int level = btrfs_header_level(parent);
753 if (slot < 0)
754 return NULL;
755 if (slot >= btrfs_header_nritems(parent))
756 return NULL;
757
758 BUG_ON(level == 0);
759
760 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
761 btrfs_level_size(root, level - 1),
762 btrfs_node_ptr_generation(parent, slot));
763 }
764
765 static int balance_level(struct btrfs_trans_handle *trans,
766 struct btrfs_root *root,
767 struct btrfs_path *path, int level)
768 {
769 struct extent_buffer *right = NULL;
770 struct extent_buffer *mid;
771 struct extent_buffer *left = NULL;
772 struct extent_buffer *parent = NULL;
773 int ret = 0;
774 int wret;
775 int pslot;
776 int orig_slot = path->slots[level];
777 int err_on_enospc = 0;
778 u64 orig_ptr;
779
780 if (level == 0)
781 return 0;
782
783 mid = path->nodes[level];
784 WARN_ON(!path->locks[level]);
785 WARN_ON(btrfs_header_generation(mid) != trans->transid);
786
787 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
788
789 if (level < BTRFS_MAX_LEVEL - 1)
790 parent = path->nodes[level + 1];
791 pslot = path->slots[level + 1];
792
793 /*
794 * deal with the case where there is only one pointer in the root
795 * by promoting the node below to a root
796 */
797 if (!parent) {
798 struct extent_buffer *child;
799
800 if (btrfs_header_nritems(mid) != 1)
801 return 0;
802
803 /* promote the child to a root */
804 child = read_node_slot(root, mid, 0);
805 btrfs_tree_lock(child);
806 BUG_ON(!child);
807 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
808 BUG_ON(ret);
809
810 spin_lock(&root->node_lock);
811 root->node = child;
812 spin_unlock(&root->node_lock);
813
814 add_root_to_dirty_list(root);
815 btrfs_tree_unlock(child);
816 path->locks[level] = 0;
817 path->nodes[level] = NULL;
818 clean_tree_block(trans, root, mid);
819 btrfs_tree_unlock(mid);
820 /* once for the path */
821 free_extent_buffer(mid);
822 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
823 root->root_key.objectid,
824 btrfs_header_generation(mid), 0, 0, 1);
825 /* once for the root ptr */
826 free_extent_buffer(mid);
827 return ret;
828 }
829 if (btrfs_header_nritems(mid) >
830 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
831 return 0;
832
833 if (btrfs_header_nritems(mid) < 2)
834 err_on_enospc = 1;
835
836 left = read_node_slot(root, parent, pslot - 1);
837 if (left) {
838 btrfs_tree_lock(left);
839 wret = btrfs_cow_block(trans, root, left,
840 parent, pslot - 1, &left);
841 if (wret) {
842 ret = wret;
843 goto enospc;
844 }
845 }
846 right = read_node_slot(root, parent, pslot + 1);
847 if (right) {
848 btrfs_tree_lock(right);
849 wret = btrfs_cow_block(trans, root, right,
850 parent, pslot + 1, &right);
851 if (wret) {
852 ret = wret;
853 goto enospc;
854 }
855 }
856
857 /* first, try to make some room in the middle buffer */
858 if (left) {
859 orig_slot += btrfs_header_nritems(left);
860 wret = push_node_left(trans, root, left, mid, 1);
861 if (wret < 0)
862 ret = wret;
863 if (btrfs_header_nritems(mid) < 2)
864 err_on_enospc = 1;
865 }
866
867 /*
868 * then try to empty the right most buffer into the middle
869 */
870 if (right) {
871 wret = push_node_left(trans, root, mid, right, 1);
872 if (wret < 0 && wret != -ENOSPC)
873 ret = wret;
874 if (btrfs_header_nritems(right) == 0) {
875 u64 bytenr = right->start;
876 u64 generation = btrfs_header_generation(parent);
877 u32 blocksize = right->len;
878
879 clean_tree_block(trans, root, right);
880 btrfs_tree_unlock(right);
881 free_extent_buffer(right);
882 right = NULL;
883 wret = del_ptr(trans, root, path, level + 1, pslot +
884 1);
885 if (wret)
886 ret = wret;
887 wret = btrfs_free_extent(trans, root, bytenr,
888 blocksize,
889 btrfs_header_owner(parent),
890 generation, 0, 0, 1);
891 if (wret)
892 ret = wret;
893 } else {
894 struct btrfs_disk_key right_key;
895 btrfs_node_key(right, &right_key, 0);
896 btrfs_set_node_key(parent, &right_key, pslot + 1);
897 btrfs_mark_buffer_dirty(parent);
898 }
899 }
900 if (btrfs_header_nritems(mid) == 1) {
901 /*
902 * we're not allowed to leave a node with one item in the
903 * tree during a delete. A deletion from lower in the tree
904 * could try to delete the only pointer in this node.
905 * So, pull some keys from the left.
906 * There has to be a left pointer at this point because
907 * otherwise we would have pulled some pointers from the
908 * right
909 */
910 BUG_ON(!left);
911 wret = balance_node_right(trans, root, mid, left);
912 if (wret < 0) {
913 ret = wret;
914 goto enospc;
915 }
916 if (wret == 1) {
917 wret = push_node_left(trans, root, left, mid, 1);
918 if (wret < 0)
919 ret = wret;
920 }
921 BUG_ON(wret == 1);
922 }
923 if (btrfs_header_nritems(mid) == 0) {
924 /* we've managed to empty the middle node, drop it */
925 u64 root_gen = btrfs_header_generation(parent);
926 u64 bytenr = mid->start;
927 u32 blocksize = mid->len;
928
929 clean_tree_block(trans, root, mid);
930 btrfs_tree_unlock(mid);
931 free_extent_buffer(mid);
932 mid = NULL;
933 wret = del_ptr(trans, root, path, level + 1, pslot);
934 if (wret)
935 ret = wret;
936 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
937 btrfs_header_owner(parent),
938 root_gen, 0, 0, 1);
939 if (wret)
940 ret = wret;
941 } else {
942 /* update the parent key to reflect our changes */
943 struct btrfs_disk_key mid_key;
944 btrfs_node_key(mid, &mid_key, 0);
945 btrfs_set_node_key(parent, &mid_key, pslot);
946 btrfs_mark_buffer_dirty(parent);
947 }
948
949 /* update the path */
950 if (left) {
951 if (btrfs_header_nritems(left) > orig_slot) {
952 extent_buffer_get(left);
953 /* left was locked after cow */
954 path->nodes[level] = left;
955 path->slots[level + 1] -= 1;
956 path->slots[level] = orig_slot;
957 if (mid) {
958 btrfs_tree_unlock(mid);
959 free_extent_buffer(mid);
960 }
961 } else {
962 orig_slot -= btrfs_header_nritems(left);
963 path->slots[level] = orig_slot;
964 }
965 }
966 /* double check we haven't messed things up */
967 check_block(root, path, level);
968 if (orig_ptr !=
969 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
970 BUG();
971 enospc:
972 if (right) {
973 btrfs_tree_unlock(right);
974 free_extent_buffer(right);
975 }
976 if (left) {
977 if (path->nodes[level] != left)
978 btrfs_tree_unlock(left);
979 free_extent_buffer(left);
980 }
981 return ret;
982 }
983
984 /* returns zero if the push worked, non-zero otherwise */
985 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
986 struct btrfs_root *root,
987 struct btrfs_path *path, int level)
988 {
989 struct extent_buffer *right = NULL;
990 struct extent_buffer *mid;
991 struct extent_buffer *left = NULL;
992 struct extent_buffer *parent = NULL;
993 int ret = 0;
994 int wret;
995 int pslot;
996 int orig_slot = path->slots[level];
997 u64 orig_ptr;
998
999 if (level == 0)
1000 return 1;
1001
1002 mid = path->nodes[level];
1003 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1004 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1005
1006 if (level < BTRFS_MAX_LEVEL - 1)
1007 parent = path->nodes[level + 1];
1008 pslot = path->slots[level + 1];
1009
1010 if (!parent)
1011 return 1;
1012
1013 left = read_node_slot(root, parent, pslot - 1);
1014
1015 /* first, try to make some room in the middle buffer */
1016 if (left) {
1017 u32 left_nr;
1018
1019 btrfs_tree_lock(left);
1020 left_nr = btrfs_header_nritems(left);
1021 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1022 wret = 1;
1023 } else {
1024 ret = btrfs_cow_block(trans, root, left, parent,
1025 pslot - 1, &left);
1026 if (ret)
1027 wret = 1;
1028 else {
1029 wret = push_node_left(trans, root,
1030 left, mid, 0);
1031 }
1032 }
1033 if (wret < 0)
1034 ret = wret;
1035 if (wret == 0) {
1036 struct btrfs_disk_key disk_key;
1037 orig_slot += left_nr;
1038 btrfs_node_key(mid, &disk_key, 0);
1039 btrfs_set_node_key(parent, &disk_key, pslot);
1040 btrfs_mark_buffer_dirty(parent);
1041 if (btrfs_header_nritems(left) > orig_slot) {
1042 path->nodes[level] = left;
1043 path->slots[level + 1] -= 1;
1044 path->slots[level] = orig_slot;
1045 btrfs_tree_unlock(mid);
1046 free_extent_buffer(mid);
1047 } else {
1048 orig_slot -=
1049 btrfs_header_nritems(left);
1050 path->slots[level] = orig_slot;
1051 btrfs_tree_unlock(left);
1052 free_extent_buffer(left);
1053 }
1054 return 0;
1055 }
1056 btrfs_tree_unlock(left);
1057 free_extent_buffer(left);
1058 }
1059 right = read_node_slot(root, parent, pslot + 1);
1060
1061 /*
1062 * then try to empty the right most buffer into the middle
1063 */
1064 if (right) {
1065 u32 right_nr;
1066 btrfs_tree_lock(right);
1067 right_nr = btrfs_header_nritems(right);
1068 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1069 wret = 1;
1070 } else {
1071 ret = btrfs_cow_block(trans, root, right,
1072 parent, pslot + 1,
1073 &right);
1074 if (ret)
1075 wret = 1;
1076 else {
1077 wret = balance_node_right(trans, root,
1078 right, mid);
1079 }
1080 }
1081 if (wret < 0)
1082 ret = wret;
1083 if (wret == 0) {
1084 struct btrfs_disk_key disk_key;
1085
1086 btrfs_node_key(right, &disk_key, 0);
1087 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1088 btrfs_mark_buffer_dirty(parent);
1089
1090 if (btrfs_header_nritems(mid) <= orig_slot) {
1091 path->nodes[level] = right;
1092 path->slots[level + 1] += 1;
1093 path->slots[level] = orig_slot -
1094 btrfs_header_nritems(mid);
1095 btrfs_tree_unlock(mid);
1096 free_extent_buffer(mid);
1097 } else {
1098 btrfs_tree_unlock(right);
1099 free_extent_buffer(right);
1100 }
1101 return 0;
1102 }
1103 btrfs_tree_unlock(right);
1104 free_extent_buffer(right);
1105 }
1106 return 1;
1107 }
1108
1109 /*
1110 * readahead one full node of leaves
1111 */
1112 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
1113 int level, int slot, u64 objectid)
1114 {
1115 struct extent_buffer *node;
1116 struct btrfs_disk_key disk_key;
1117 u32 nritems;
1118 u64 search;
1119 u64 lowest_read;
1120 u64 highest_read;
1121 u64 nread = 0;
1122 int direction = path->reada;
1123 struct extent_buffer *eb;
1124 u32 nr;
1125 u32 blocksize;
1126 u32 nscan = 0;
1127
1128 if (level != 1)
1129 return;
1130
1131 if (!path->nodes[level])
1132 return;
1133
1134 node = path->nodes[level];
1135 WARN_ON(!path->skip_locking && !btrfs_tree_locked(node));
1136
1137 search = btrfs_node_blockptr(node, slot);
1138 blocksize = btrfs_level_size(root, level - 1);
1139 eb = btrfs_find_tree_block(root, search, blocksize);
1140 if (eb) {
1141 free_extent_buffer(eb);
1142 return;
1143 }
1144
1145 highest_read = search;
1146 lowest_read = search;
1147
1148 nritems = btrfs_header_nritems(node);
1149 nr = slot;
1150 while(1) {
1151 if (direction < 0) {
1152 if (nr == 0)
1153 break;
1154 nr--;
1155 } else if (direction > 0) {
1156 nr++;
1157 if (nr >= nritems)
1158 break;
1159 }
1160 if (path->reada < 0 && objectid) {
1161 btrfs_node_key(node, &disk_key, nr);
1162 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1163 break;
1164 }
1165 search = btrfs_node_blockptr(node, nr);
1166 if ((search >= lowest_read && search <= highest_read) ||
1167 (search < lowest_read && lowest_read - search <= 32768) ||
1168 (search > highest_read && search - highest_read <= 32768)) {
1169 readahead_tree_block(root, search, blocksize,
1170 btrfs_node_ptr_generation(node, nr));
1171 nread += blocksize;
1172 }
1173 nscan++;
1174 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1175 break;
1176 if(nread > (1024 * 1024) || nscan > 128)
1177 break;
1178
1179 if (search < lowest_read)
1180 lowest_read = search;
1181 if (search > highest_read)
1182 highest_read = search;
1183 }
1184 }
1185
1186 static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock)
1187 {
1188 int i;
1189 int skip_level = level;
1190 struct extent_buffer *t;
1191
1192 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1193 if (!path->nodes[i])
1194 break;
1195 if (!path->locks[i])
1196 break;
1197 if (path->slots[i] == 0) {
1198 skip_level = i + 1;
1199 continue;
1200 }
1201 if (path->keep_locks) {
1202 u32 nritems;
1203 t = path->nodes[i];
1204 nritems = btrfs_header_nritems(t);
1205 if (path->slots[i] >= nritems - 1) {
1206 skip_level = i + 1;
1207 continue;
1208 }
1209 }
1210 t = path->nodes[i];
1211 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1212 btrfs_tree_unlock(t);
1213 path->locks[i] = 0;
1214 }
1215 }
1216 }
1217
1218 /*
1219 * look for key in the tree. path is filled in with nodes along the way
1220 * if key is found, we return zero and you can find the item in the leaf
1221 * level of the path (level 0)
1222 *
1223 * If the key isn't found, the path points to the slot where it should
1224 * be inserted, and 1 is returned. If there are other errors during the
1225 * search a negative error number is returned.
1226 *
1227 * if ins_len > 0, nodes and leaves will be split as we walk down the
1228 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1229 * possible)
1230 */
1231 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1232 *root, struct btrfs_key *key, struct btrfs_path *p, int
1233 ins_len, int cow)
1234 {
1235 struct extent_buffer *b;
1236 int slot;
1237 int ret;
1238 int level;
1239 int should_reada = p->reada;
1240 int lowest_unlock = 1;
1241 u8 lowest_level = 0;
1242
1243 lowest_level = p->lowest_level;
1244 WARN_ON(lowest_level && ins_len);
1245 WARN_ON(p->nodes[0] != NULL);
1246 // WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1247 WARN_ON(root == root->fs_info->extent_root &&
1248 !mutex_is_locked(&root->fs_info->alloc_mutex));
1249 WARN_ON(root == root->fs_info->chunk_root &&
1250 !mutex_is_locked(&root->fs_info->chunk_mutex));
1251 WARN_ON(root == root->fs_info->dev_root &&
1252 !mutex_is_locked(&root->fs_info->chunk_mutex));
1253 if (ins_len < 0)
1254 lowest_unlock = 2;
1255 again:
1256 if (!p->skip_locking)
1257 b = btrfs_lock_root_node(root);
1258 else
1259 b = btrfs_root_node(root);
1260
1261 while (b) {
1262 level = btrfs_header_level(b);
1263 if (cow) {
1264 int wret;
1265 wret = btrfs_cow_block(trans, root, b,
1266 p->nodes[level + 1],
1267 p->slots[level + 1],
1268 &b);
1269 if (wret) {
1270 free_extent_buffer(b);
1271 return wret;
1272 }
1273 }
1274 BUG_ON(!cow && ins_len);
1275 if (level != btrfs_header_level(b))
1276 WARN_ON(1);
1277 level = btrfs_header_level(b);
1278 p->nodes[level] = b;
1279 if (!p->skip_locking)
1280 p->locks[level] = 1;
1281 ret = check_block(root, p, level);
1282 if (ret)
1283 return -1;
1284
1285 ret = bin_search(b, key, level, &slot);
1286 if (level != 0) {
1287 if (ret && slot > 0)
1288 slot -= 1;
1289 p->slots[level] = slot;
1290 if (ins_len > 0 && btrfs_header_nritems(b) >=
1291 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1292 int sret = split_node(trans, root, p, level);
1293 BUG_ON(sret > 0);
1294 if (sret)
1295 return sret;
1296 b = p->nodes[level];
1297 slot = p->slots[level];
1298 } else if (ins_len < 0) {
1299 int sret = balance_level(trans, root, p,
1300 level);
1301 if (sret)
1302 return sret;
1303 b = p->nodes[level];
1304 if (!b) {
1305 btrfs_release_path(NULL, p);
1306 goto again;
1307 }
1308 slot = p->slots[level];
1309 BUG_ON(btrfs_header_nritems(b) == 1);
1310 }
1311 /* this is only true while dropping a snapshot */
1312 if (level == lowest_level) {
1313 unlock_up(p, level, lowest_unlock);
1314 break;
1315 }
1316
1317 if (should_reada)
1318 reada_for_search(root, p, level, slot,
1319 key->objectid);
1320
1321 b = read_node_slot(root, b, slot);
1322 if (!p->skip_locking)
1323 btrfs_tree_lock(b);
1324 unlock_up(p, level, lowest_unlock);
1325 } else {
1326 p->slots[level] = slot;
1327 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1328 sizeof(struct btrfs_item) + ins_len) {
1329 int sret = split_leaf(trans, root, key,
1330 p, ins_len, ret == 0);
1331 BUG_ON(sret > 0);
1332 if (sret)
1333 return sret;
1334 }
1335 unlock_up(p, level, lowest_unlock);
1336 return ret;
1337 }
1338 }
1339 return 1;
1340 }
1341
1342 /*
1343 * adjust the pointers going up the tree, starting at level
1344 * making sure the right key of each node is points to 'key'.
1345 * This is used after shifting pointers to the left, so it stops
1346 * fixing up pointers when a given leaf/node is not in slot 0 of the
1347 * higher levels
1348 *
1349 * If this fails to write a tree block, it returns -1, but continues
1350 * fixing up the blocks in ram so the tree is consistent.
1351 */
1352 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1353 struct btrfs_root *root, struct btrfs_path *path,
1354 struct btrfs_disk_key *key, int level)
1355 {
1356 int i;
1357 int ret = 0;
1358 struct extent_buffer *t;
1359
1360 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1361 int tslot = path->slots[i];
1362 if (!path->nodes[i])
1363 break;
1364 t = path->nodes[i];
1365 btrfs_set_node_key(t, key, tslot);
1366 if (!btrfs_tree_locked(path->nodes[i])) {
1367 int ii;
1368 printk("fixup without lock on level %d\n", btrfs_header_level(path->nodes[i]));
1369 for (ii = 0; ii < BTRFS_MAX_LEVEL; ii++) {
1370 printk("level %d slot %d\n", ii, path->slots[ii]);
1371 }
1372 }
1373 btrfs_mark_buffer_dirty(path->nodes[i]);
1374 if (tslot != 0)
1375 break;
1376 }
1377 return ret;
1378 }
1379
1380 /*
1381 * try to push data from one node into the next node left in the
1382 * tree.
1383 *
1384 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1385 * error, and > 0 if there was no room in the left hand block.
1386 */
1387 static int push_node_left(struct btrfs_trans_handle *trans,
1388 struct btrfs_root *root, struct extent_buffer *dst,
1389 struct extent_buffer *src, int empty)
1390 {
1391 int push_items = 0;
1392 int src_nritems;
1393 int dst_nritems;
1394 int ret = 0;
1395
1396 src_nritems = btrfs_header_nritems(src);
1397 dst_nritems = btrfs_header_nritems(dst);
1398 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1399 WARN_ON(btrfs_header_generation(src) != trans->transid);
1400 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1401
1402 if (!empty && src_nritems <= 8)
1403 return 1;
1404
1405 if (push_items <= 0) {
1406 return 1;
1407 }
1408
1409 if (empty) {
1410 push_items = min(src_nritems, push_items);
1411 if (push_items < src_nritems) {
1412 /* leave at least 8 pointers in the node if
1413 * we aren't going to empty it
1414 */
1415 if (src_nritems - push_items < 8) {
1416 if (push_items <= 8)
1417 return 1;
1418 push_items -= 8;
1419 }
1420 }
1421 } else
1422 push_items = min(src_nritems - 8, push_items);
1423
1424 copy_extent_buffer(dst, src,
1425 btrfs_node_key_ptr_offset(dst_nritems),
1426 btrfs_node_key_ptr_offset(0),
1427 push_items * sizeof(struct btrfs_key_ptr));
1428
1429 if (push_items < src_nritems) {
1430 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1431 btrfs_node_key_ptr_offset(push_items),
1432 (src_nritems - push_items) *
1433 sizeof(struct btrfs_key_ptr));
1434 }
1435 btrfs_set_header_nritems(src, src_nritems - push_items);
1436 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1437 btrfs_mark_buffer_dirty(src);
1438 btrfs_mark_buffer_dirty(dst);
1439 return ret;
1440 }
1441
1442 /*
1443 * try to push data from one node into the next node right in the
1444 * tree.
1445 *
1446 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1447 * error, and > 0 if there was no room in the right hand block.
1448 *
1449 * this will only push up to 1/2 the contents of the left node over
1450 */
1451 static int balance_node_right(struct btrfs_trans_handle *trans,
1452 struct btrfs_root *root,
1453 struct extent_buffer *dst,
1454 struct extent_buffer *src)
1455 {
1456 int push_items = 0;
1457 int max_push;
1458 int src_nritems;
1459 int dst_nritems;
1460 int ret = 0;
1461
1462 WARN_ON(btrfs_header_generation(src) != trans->transid);
1463 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1464
1465 src_nritems = btrfs_header_nritems(src);
1466 dst_nritems = btrfs_header_nritems(dst);
1467 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1468 if (push_items <= 0) {
1469 return 1;
1470 }
1471
1472 if (src_nritems < 4) {
1473 return 1;
1474 }
1475
1476 max_push = src_nritems / 2 + 1;
1477 /* don't try to empty the node */
1478 if (max_push >= src_nritems) {
1479 return 1;
1480 }
1481
1482 if (max_push < push_items)
1483 push_items = max_push;
1484
1485 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1486 btrfs_node_key_ptr_offset(0),
1487 (dst_nritems) *
1488 sizeof(struct btrfs_key_ptr));
1489
1490 copy_extent_buffer(dst, src,
1491 btrfs_node_key_ptr_offset(0),
1492 btrfs_node_key_ptr_offset(src_nritems - push_items),
1493 push_items * sizeof(struct btrfs_key_ptr));
1494
1495 btrfs_set_header_nritems(src, src_nritems - push_items);
1496 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1497
1498 btrfs_mark_buffer_dirty(src);
1499 btrfs_mark_buffer_dirty(dst);
1500 return ret;
1501 }
1502
1503 /*
1504 * helper function to insert a new root level in the tree.
1505 * A new node is allocated, and a single item is inserted to
1506 * point to the existing root
1507 *
1508 * returns zero on success or < 0 on failure.
1509 */
1510 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1511 struct btrfs_root *root,
1512 struct btrfs_path *path, int level)
1513 {
1514 u64 root_gen;
1515 u64 lower_gen;
1516 struct extent_buffer *lower;
1517 struct extent_buffer *c;
1518 struct extent_buffer *old;
1519 struct btrfs_disk_key lower_key;
1520
1521 BUG_ON(path->nodes[level]);
1522 BUG_ON(path->nodes[level-1] != root->node);
1523
1524 if (root->ref_cows)
1525 root_gen = trans->transid;
1526 else
1527 root_gen = 0;
1528
1529 lower = path->nodes[level-1];
1530 if (level == 1)
1531 btrfs_item_key(lower, &lower_key, 0);
1532 else
1533 btrfs_node_key(lower, &lower_key, 0);
1534
1535 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1536 root->root_key.objectid,
1537 root_gen, lower_key.objectid, level,
1538 root->node->start, 0);
1539 if (IS_ERR(c))
1540 return PTR_ERR(c);
1541
1542 memset_extent_buffer(c, 0, 0, root->nodesize);
1543 btrfs_set_header_nritems(c, 1);
1544 btrfs_set_header_level(c, level);
1545 btrfs_set_header_bytenr(c, c->start);
1546 btrfs_set_header_generation(c, trans->transid);
1547 btrfs_set_header_owner(c, root->root_key.objectid);
1548
1549 write_extent_buffer(c, root->fs_info->fsid,
1550 (unsigned long)btrfs_header_fsid(c),
1551 BTRFS_FSID_SIZE);
1552
1553 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1554 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1555 BTRFS_UUID_SIZE);
1556
1557 btrfs_set_node_key(c, &lower_key, 0);
1558 btrfs_set_node_blockptr(c, 0, lower->start);
1559 lower_gen = btrfs_header_generation(lower);
1560 WARN_ON(lower_gen == 0);
1561
1562 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1563
1564 btrfs_mark_buffer_dirty(c);
1565
1566 spin_lock(&root->node_lock);
1567 old = root->node;
1568 root->node = c;
1569 spin_unlock(&root->node_lock);
1570
1571 /* the super has an extra ref to root->node */
1572 free_extent_buffer(old);
1573
1574 add_root_to_dirty_list(root);
1575 extent_buffer_get(c);
1576 path->nodes[level] = c;
1577 path->locks[level] = 1;
1578 path->slots[level] = 0;
1579
1580 if (root->ref_cows && lower_gen != trans->transid) {
1581 struct btrfs_path *back_path = btrfs_alloc_path();
1582 int ret;
1583 mutex_lock(&root->fs_info->alloc_mutex);
1584 ret = btrfs_insert_extent_backref(trans,
1585 root->fs_info->extent_root,
1586 path, lower->start,
1587 root->root_key.objectid,
1588 trans->transid, 0, 0);
1589 BUG_ON(ret);
1590 mutex_unlock(&root->fs_info->alloc_mutex);
1591 btrfs_free_path(back_path);
1592 }
1593 return 0;
1594 }
1595
1596 /*
1597 * worker function to insert a single pointer in a node.
1598 * the node should have enough room for the pointer already
1599 *
1600 * slot and level indicate where you want the key to go, and
1601 * blocknr is the block the key points to.
1602 *
1603 * returns zero on success and < 0 on any error
1604 */
1605 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1606 *root, struct btrfs_path *path, struct btrfs_disk_key
1607 *key, u64 bytenr, int slot, int level)
1608 {
1609 struct extent_buffer *lower;
1610 int nritems;
1611
1612 BUG_ON(!path->nodes[level]);
1613 lower = path->nodes[level];
1614 nritems = btrfs_header_nritems(lower);
1615 if (slot > nritems)
1616 BUG();
1617 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1618 BUG();
1619 if (slot != nritems) {
1620 memmove_extent_buffer(lower,
1621 btrfs_node_key_ptr_offset(slot + 1),
1622 btrfs_node_key_ptr_offset(slot),
1623 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1624 }
1625 btrfs_set_node_key(lower, key, slot);
1626 btrfs_set_node_blockptr(lower, slot, bytenr);
1627 WARN_ON(trans->transid == 0);
1628 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1629 btrfs_set_header_nritems(lower, nritems + 1);
1630 btrfs_mark_buffer_dirty(lower);
1631 return 0;
1632 }
1633
1634 /*
1635 * split the node at the specified level in path in two.
1636 * The path is corrected to point to the appropriate node after the split
1637 *
1638 * Before splitting this tries to make some room in the node by pushing
1639 * left and right, if either one works, it returns right away.
1640 *
1641 * returns 0 on success and < 0 on failure
1642 */
1643 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1644 *root, struct btrfs_path *path, int level)
1645 {
1646 u64 root_gen;
1647 struct extent_buffer *c;
1648 struct extent_buffer *split;
1649 struct btrfs_disk_key disk_key;
1650 int mid;
1651 int ret;
1652 int wret;
1653 u32 c_nritems;
1654
1655 c = path->nodes[level];
1656 WARN_ON(btrfs_header_generation(c) != trans->transid);
1657 if (c == root->node) {
1658 /* trying to split the root, lets make a new one */
1659 ret = insert_new_root(trans, root, path, level + 1);
1660 if (ret)
1661 return ret;
1662 } else {
1663 ret = push_nodes_for_insert(trans, root, path, level);
1664 c = path->nodes[level];
1665 if (!ret && btrfs_header_nritems(c) <
1666 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1667 return 0;
1668 if (ret < 0)
1669 return ret;
1670 }
1671
1672 c_nritems = btrfs_header_nritems(c);
1673 if (root->ref_cows)
1674 root_gen = trans->transid;
1675 else
1676 root_gen = 0;
1677
1678 btrfs_node_key(c, &disk_key, 0);
1679 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1680 root->root_key.objectid,
1681 root_gen,
1682 btrfs_disk_key_objectid(&disk_key),
1683 level, c->start, 0);
1684 if (IS_ERR(split))
1685 return PTR_ERR(split);
1686
1687 btrfs_set_header_flags(split, btrfs_header_flags(c));
1688 btrfs_set_header_level(split, btrfs_header_level(c));
1689 btrfs_set_header_bytenr(split, split->start);
1690 btrfs_set_header_generation(split, trans->transid);
1691 btrfs_set_header_owner(split, root->root_key.objectid);
1692 btrfs_set_header_flags(split, 0);
1693 write_extent_buffer(split, root->fs_info->fsid,
1694 (unsigned long)btrfs_header_fsid(split),
1695 BTRFS_FSID_SIZE);
1696 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1697 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1698 BTRFS_UUID_SIZE);
1699
1700 mid = (c_nritems + 1) / 2;
1701
1702 copy_extent_buffer(split, c,
1703 btrfs_node_key_ptr_offset(0),
1704 btrfs_node_key_ptr_offset(mid),
1705 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1706 btrfs_set_header_nritems(split, c_nritems - mid);
1707 btrfs_set_header_nritems(c, mid);
1708 ret = 0;
1709
1710 btrfs_mark_buffer_dirty(c);
1711 btrfs_mark_buffer_dirty(split);
1712
1713 btrfs_node_key(split, &disk_key, 0);
1714 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1715 path->slots[level + 1] + 1,
1716 level + 1);
1717 if (wret)
1718 ret = wret;
1719
1720 if (path->slots[level] >= mid) {
1721 path->slots[level] -= mid;
1722 btrfs_tree_unlock(c);
1723 free_extent_buffer(c);
1724 path->nodes[level] = split;
1725 path->slots[level + 1] += 1;
1726 } else {
1727 btrfs_tree_unlock(split);
1728 free_extent_buffer(split);
1729 }
1730 return ret;
1731 }
1732
1733 /*
1734 * how many bytes are required to store the items in a leaf. start
1735 * and nr indicate which items in the leaf to check. This totals up the
1736 * space used both by the item structs and the item data
1737 */
1738 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1739 {
1740 int data_len;
1741 int nritems = btrfs_header_nritems(l);
1742 int end = min(nritems, start + nr) - 1;
1743
1744 if (!nr)
1745 return 0;
1746 data_len = btrfs_item_end_nr(l, start);
1747 data_len = data_len - btrfs_item_offset_nr(l, end);
1748 data_len += sizeof(struct btrfs_item) * nr;
1749 WARN_ON(data_len < 0);
1750 return data_len;
1751 }
1752
1753 /*
1754 * The space between the end of the leaf items and
1755 * the start of the leaf data. IOW, how much room
1756 * the leaf has left for both items and data
1757 */
1758 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1759 {
1760 int nritems = btrfs_header_nritems(leaf);
1761 int ret;
1762 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1763 if (ret < 0) {
1764 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1765 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1766 leaf_space_used(leaf, 0, nritems), nritems);
1767 }
1768 return ret;
1769 }
1770
1771 /*
1772 * push some data in the path leaf to the right, trying to free up at
1773 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1774 *
1775 * returns 1 if the push failed because the other node didn't have enough
1776 * room, 0 if everything worked out and < 0 if there were major errors.
1777 */
1778 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1779 *root, struct btrfs_path *path, int data_size,
1780 int empty)
1781 {
1782 struct extent_buffer *left = path->nodes[0];
1783 struct extent_buffer *right;
1784 struct extent_buffer *upper;
1785 struct btrfs_disk_key disk_key;
1786 int slot;
1787 u32 i;
1788 int free_space;
1789 int push_space = 0;
1790 int push_items = 0;
1791 struct btrfs_item *item;
1792 u32 left_nritems;
1793 u32 nr;
1794 u32 right_nritems;
1795 u32 data_end;
1796 u32 this_item_size;
1797 int ret;
1798
1799 slot = path->slots[1];
1800 if (!path->nodes[1]) {
1801 return 1;
1802 }
1803 upper = path->nodes[1];
1804 if (slot >= btrfs_header_nritems(upper) - 1)
1805 return 1;
1806
1807 right = read_node_slot(root, upper, slot + 1);
1808 btrfs_tree_lock(right);
1809 free_space = btrfs_leaf_free_space(root, right);
1810 if (free_space < data_size + sizeof(struct btrfs_item))
1811 goto out_unlock;
1812
1813 /* cow and double check */
1814 ret = btrfs_cow_block(trans, root, right, upper,
1815 slot + 1, &right);
1816 if (ret)
1817 goto out_unlock;
1818
1819 free_space = btrfs_leaf_free_space(root, right);
1820 if (free_space < data_size + sizeof(struct btrfs_item))
1821 goto out_unlock;
1822
1823 left_nritems = btrfs_header_nritems(left);
1824 if (left_nritems == 0)
1825 goto out_unlock;
1826
1827 if (empty)
1828 nr = 0;
1829 else
1830 nr = 1;
1831
1832 i = left_nritems - 1;
1833 while (i >= nr) {
1834 item = btrfs_item_nr(left, i);
1835
1836 if (path->slots[0] == i)
1837 push_space += data_size + sizeof(*item);
1838
1839 if (!left->map_token) {
1840 map_extent_buffer(left, (unsigned long)item,
1841 sizeof(struct btrfs_item),
1842 &left->map_token, &left->kaddr,
1843 &left->map_start, &left->map_len,
1844 KM_USER1);
1845 }
1846
1847 this_item_size = btrfs_item_size(left, item);
1848 if (this_item_size + sizeof(*item) + push_space > free_space)
1849 break;
1850 push_items++;
1851 push_space += this_item_size + sizeof(*item);
1852 if (i == 0)
1853 break;
1854 i--;
1855 }
1856 if (left->map_token) {
1857 unmap_extent_buffer(left, left->map_token, KM_USER1);
1858 left->map_token = NULL;
1859 }
1860
1861 if (push_items == 0)
1862 goto out_unlock;
1863
1864 if (!empty && push_items == left_nritems)
1865 WARN_ON(1);
1866
1867 /* push left to right */
1868 right_nritems = btrfs_header_nritems(right);
1869
1870 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1871 push_space -= leaf_data_end(root, left);
1872
1873 /* make room in the right data area */
1874 data_end = leaf_data_end(root, right);
1875 memmove_extent_buffer(right,
1876 btrfs_leaf_data(right) + data_end - push_space,
1877 btrfs_leaf_data(right) + data_end,
1878 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1879
1880 /* copy from the left data area */
1881 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1882 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1883 btrfs_leaf_data(left) + leaf_data_end(root, left),
1884 push_space);
1885
1886 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1887 btrfs_item_nr_offset(0),
1888 right_nritems * sizeof(struct btrfs_item));
1889
1890 /* copy the items from left to right */
1891 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1892 btrfs_item_nr_offset(left_nritems - push_items),
1893 push_items * sizeof(struct btrfs_item));
1894
1895 /* update the item pointers */
1896 right_nritems += push_items;
1897 btrfs_set_header_nritems(right, right_nritems);
1898 push_space = BTRFS_LEAF_DATA_SIZE(root);
1899 for (i = 0; i < right_nritems; i++) {
1900 item = btrfs_item_nr(right, i);
1901 if (!right->map_token) {
1902 map_extent_buffer(right, (unsigned long)item,
1903 sizeof(struct btrfs_item),
1904 &right->map_token, &right->kaddr,
1905 &right->map_start, &right->map_len,
1906 KM_USER1);
1907 }
1908 push_space -= btrfs_item_size(right, item);
1909 btrfs_set_item_offset(right, item, push_space);
1910 }
1911
1912 if (right->map_token) {
1913 unmap_extent_buffer(right, right->map_token, KM_USER1);
1914 right->map_token = NULL;
1915 }
1916 left_nritems -= push_items;
1917 btrfs_set_header_nritems(left, left_nritems);
1918
1919 if (left_nritems)
1920 btrfs_mark_buffer_dirty(left);
1921 btrfs_mark_buffer_dirty(right);
1922
1923 btrfs_item_key(right, &disk_key, 0);
1924 btrfs_set_node_key(upper, &disk_key, slot + 1);
1925 btrfs_mark_buffer_dirty(upper);
1926
1927 /* then fixup the leaf pointer in the path */
1928 if (path->slots[0] >= left_nritems) {
1929 path->slots[0] -= left_nritems;
1930 if (btrfs_header_nritems(path->nodes[0]) == 0)
1931 clean_tree_block(trans, root, path->nodes[0]);
1932 btrfs_tree_unlock(path->nodes[0]);
1933 free_extent_buffer(path->nodes[0]);
1934 path->nodes[0] = right;
1935 path->slots[1] += 1;
1936 } else {
1937 btrfs_tree_unlock(right);
1938 free_extent_buffer(right);
1939 }
1940 return 0;
1941
1942 out_unlock:
1943 btrfs_tree_unlock(right);
1944 free_extent_buffer(right);
1945 return 1;
1946 }
1947
1948 /*
1949 * push some data in the path leaf to the left, trying to free up at
1950 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1951 */
1952 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1953 *root, struct btrfs_path *path, int data_size,
1954 int empty)
1955 {
1956 struct btrfs_disk_key disk_key;
1957 struct extent_buffer *right = path->nodes[0];
1958 struct extent_buffer *left;
1959 int slot;
1960 int i;
1961 int free_space;
1962 int push_space = 0;
1963 int push_items = 0;
1964 struct btrfs_item *item;
1965 u32 old_left_nritems;
1966 u32 right_nritems;
1967 u32 nr;
1968 int ret = 0;
1969 int wret;
1970 u32 this_item_size;
1971 u32 old_left_item_size;
1972
1973 slot = path->slots[1];
1974 if (slot == 0)
1975 return 1;
1976 if (!path->nodes[1])
1977 return 1;
1978
1979 right_nritems = btrfs_header_nritems(right);
1980 if (right_nritems == 0) {
1981 return 1;
1982 }
1983
1984 left = read_node_slot(root, path->nodes[1], slot - 1);
1985 btrfs_tree_lock(left);
1986 free_space = btrfs_leaf_free_space(root, left);
1987 if (free_space < data_size + sizeof(struct btrfs_item)) {
1988 ret = 1;
1989 goto out;
1990 }
1991
1992 /* cow and double check */
1993 ret = btrfs_cow_block(trans, root, left,
1994 path->nodes[1], slot - 1, &left);
1995 if (ret) {
1996 /* we hit -ENOSPC, but it isn't fatal here */
1997 ret = 1;
1998 goto out;
1999 }
2000
2001 free_space = btrfs_leaf_free_space(root, left);
2002 if (free_space < data_size + sizeof(struct btrfs_item)) {
2003 ret = 1;
2004 goto out;
2005 }
2006
2007 if (empty)
2008 nr = right_nritems;
2009 else
2010 nr = right_nritems - 1;
2011
2012 for (i = 0; i < nr; i++) {
2013 item = btrfs_item_nr(right, i);
2014 if (!right->map_token) {
2015 map_extent_buffer(right, (unsigned long)item,
2016 sizeof(struct btrfs_item),
2017 &right->map_token, &right->kaddr,
2018 &right->map_start, &right->map_len,
2019 KM_USER1);
2020 }
2021
2022 if (path->slots[0] == i)
2023 push_space += data_size + sizeof(*item);
2024
2025 this_item_size = btrfs_item_size(right, item);
2026 if (this_item_size + sizeof(*item) + push_space > free_space)
2027 break;
2028
2029 push_items++;
2030 push_space += this_item_size + sizeof(*item);
2031 }
2032
2033 if (right->map_token) {
2034 unmap_extent_buffer(right, right->map_token, KM_USER1);
2035 right->map_token = NULL;
2036 }
2037
2038 if (push_items == 0) {
2039 ret = 1;
2040 goto out;
2041 }
2042 if (!empty && push_items == btrfs_header_nritems(right))
2043 WARN_ON(1);
2044
2045 /* push data from right to left */
2046 copy_extent_buffer(left, right,
2047 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2048 btrfs_item_nr_offset(0),
2049 push_items * sizeof(struct btrfs_item));
2050
2051 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2052 btrfs_item_offset_nr(right, push_items -1);
2053
2054 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2055 leaf_data_end(root, left) - push_space,
2056 btrfs_leaf_data(right) +
2057 btrfs_item_offset_nr(right, push_items - 1),
2058 push_space);
2059 old_left_nritems = btrfs_header_nritems(left);
2060 BUG_ON(old_left_nritems < 0);
2061
2062 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2063 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2064 u32 ioff;
2065
2066 item = btrfs_item_nr(left, i);
2067 if (!left->map_token) {
2068 map_extent_buffer(left, (unsigned long)item,
2069 sizeof(struct btrfs_item),
2070 &left->map_token, &left->kaddr,
2071 &left->map_start, &left->map_len,
2072 KM_USER1);
2073 }
2074
2075 ioff = btrfs_item_offset(left, item);
2076 btrfs_set_item_offset(left, item,
2077 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2078 }
2079 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2080 if (left->map_token) {
2081 unmap_extent_buffer(left, left->map_token, KM_USER1);
2082 left->map_token = NULL;
2083 }
2084
2085 /* fixup right node */
2086 if (push_items > right_nritems) {
2087 printk("push items %d nr %u\n", push_items, right_nritems);
2088 WARN_ON(1);
2089 }
2090
2091 if (push_items < right_nritems) {
2092 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2093 leaf_data_end(root, right);
2094 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2095 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2096 btrfs_leaf_data(right) +
2097 leaf_data_end(root, right), push_space);
2098
2099 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2100 btrfs_item_nr_offset(push_items),
2101 (btrfs_header_nritems(right) - push_items) *
2102 sizeof(struct btrfs_item));
2103 }
2104 right_nritems -= push_items;
2105 btrfs_set_header_nritems(right, right_nritems);
2106 push_space = BTRFS_LEAF_DATA_SIZE(root);
2107 for (i = 0; i < right_nritems; i++) {
2108 item = btrfs_item_nr(right, i);
2109
2110 if (!right->map_token) {
2111 map_extent_buffer(right, (unsigned long)item,
2112 sizeof(struct btrfs_item),
2113 &right->map_token, &right->kaddr,
2114 &right->map_start, &right->map_len,
2115 KM_USER1);
2116 }
2117
2118 push_space = push_space - btrfs_item_size(right, item);
2119 btrfs_set_item_offset(right, item, push_space);
2120 }
2121 if (right->map_token) {
2122 unmap_extent_buffer(right, right->map_token, KM_USER1);
2123 right->map_token = NULL;
2124 }
2125
2126 btrfs_mark_buffer_dirty(left);
2127 if (right_nritems)
2128 btrfs_mark_buffer_dirty(right);
2129
2130 btrfs_item_key(right, &disk_key, 0);
2131 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2132 if (wret)
2133 ret = wret;
2134
2135 /* then fixup the leaf pointer in the path */
2136 if (path->slots[0] < push_items) {
2137 path->slots[0] += old_left_nritems;
2138 if (btrfs_header_nritems(path->nodes[0]) == 0)
2139 clean_tree_block(trans, root, path->nodes[0]);
2140 btrfs_tree_unlock(path->nodes[0]);
2141 free_extent_buffer(path->nodes[0]);
2142 path->nodes[0] = left;
2143 path->slots[1] -= 1;
2144 } else {
2145 btrfs_tree_unlock(left);
2146 free_extent_buffer(left);
2147 path->slots[0] -= push_items;
2148 }
2149 BUG_ON(path->slots[0] < 0);
2150 return ret;
2151 out:
2152 btrfs_tree_unlock(left);
2153 free_extent_buffer(left);
2154 return ret;
2155 }
2156
2157 /*
2158 * split the path's leaf in two, making sure there is at least data_size
2159 * available for the resulting leaf level of the path.
2160 *
2161 * returns 0 if all went well and < 0 on failure.
2162 */
2163 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
2164 *root, struct btrfs_key *ins_key,
2165 struct btrfs_path *path, int data_size, int extend)
2166 {
2167 u64 root_gen;
2168 struct extent_buffer *l;
2169 u32 nritems;
2170 int mid;
2171 int slot;
2172 struct extent_buffer *right;
2173 int space_needed = data_size + sizeof(struct btrfs_item);
2174 int data_copy_size;
2175 int rt_data_off;
2176 int i;
2177 int ret = 0;
2178 int wret;
2179 int double_split;
2180 int num_doubles = 0;
2181 struct btrfs_disk_key disk_key;
2182
2183 if (extend)
2184 space_needed = data_size;
2185
2186 if (root->ref_cows)
2187 root_gen = trans->transid;
2188 else
2189 root_gen = 0;
2190
2191 /* first try to make some room by pushing left and right */
2192 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2193 wret = push_leaf_right(trans, root, path, data_size, 0);
2194 if (wret < 0) {
2195 return wret;
2196 }
2197 if (wret) {
2198 wret = push_leaf_left(trans, root, path, data_size, 0);
2199 if (wret < 0)
2200 return wret;
2201 }
2202 l = path->nodes[0];
2203
2204 /* did the pushes work? */
2205 if (btrfs_leaf_free_space(root, l) >= space_needed)
2206 return 0;
2207 }
2208
2209 if (!path->nodes[1]) {
2210 ret = insert_new_root(trans, root, path, 1);
2211 if (ret)
2212 return ret;
2213 }
2214 again:
2215 double_split = 0;
2216 l = path->nodes[0];
2217 slot = path->slots[0];
2218 nritems = btrfs_header_nritems(l);
2219 mid = (nritems + 1)/ 2;
2220
2221 btrfs_item_key(l, &disk_key, 0);
2222
2223 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2224 root->root_key.objectid,
2225 root_gen, disk_key.objectid, 0,
2226 l->start, 0);
2227 if (IS_ERR(right)) {
2228 BUG_ON(1);
2229 return PTR_ERR(right);
2230 }
2231
2232 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2233 btrfs_set_header_bytenr(right, right->start);
2234 btrfs_set_header_generation(right, trans->transid);
2235 btrfs_set_header_owner(right, root->root_key.objectid);
2236 btrfs_set_header_level(right, 0);
2237 write_extent_buffer(right, root->fs_info->fsid,
2238 (unsigned long)btrfs_header_fsid(right),
2239 BTRFS_FSID_SIZE);
2240
2241 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2242 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2243 BTRFS_UUID_SIZE);
2244 if (mid <= slot) {
2245 if (nritems == 1 ||
2246 leaf_space_used(l, mid, nritems - mid) + space_needed >
2247 BTRFS_LEAF_DATA_SIZE(root)) {
2248 if (slot >= nritems) {
2249 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2250 btrfs_set_header_nritems(right, 0);
2251 wret = insert_ptr(trans, root, path,
2252 &disk_key, right->start,
2253 path->slots[1] + 1, 1);
2254 if (wret)
2255 ret = wret;
2256
2257 btrfs_tree_unlock(path->nodes[0]);
2258 free_extent_buffer(path->nodes[0]);
2259 path->nodes[0] = right;
2260 path->slots[0] = 0;
2261 path->slots[1] += 1;
2262 btrfs_mark_buffer_dirty(right);
2263 return ret;
2264 }
2265 mid = slot;
2266 if (mid != nritems &&
2267 leaf_space_used(l, mid, nritems - mid) +
2268 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2269 double_split = 1;
2270 }
2271 }
2272 } else {
2273 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2274 BTRFS_LEAF_DATA_SIZE(root)) {
2275 if (!extend && slot == 0) {
2276 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2277 btrfs_set_header_nritems(right, 0);
2278 wret = insert_ptr(trans, root, path,
2279 &disk_key,
2280 right->start,
2281 path->slots[1], 1);
2282 if (wret)
2283 ret = wret;
2284 btrfs_tree_unlock(path->nodes[0]);
2285 free_extent_buffer(path->nodes[0]);
2286 path->nodes[0] = right;
2287 path->slots[0] = 0;
2288 if (path->slots[1] == 0) {
2289 wret = fixup_low_keys(trans, root,
2290 path, &disk_key, 1);
2291 if (wret)
2292 ret = wret;
2293 }
2294 btrfs_mark_buffer_dirty(right);
2295 return ret;
2296 } else if (extend && slot == 0) {
2297 mid = 1;
2298 } else {
2299 mid = slot;
2300 if (mid != nritems &&
2301 leaf_space_used(l, mid, nritems - mid) +
2302 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2303 double_split = 1;
2304 }
2305 }
2306 }
2307 }
2308 nritems = nritems - mid;
2309 btrfs_set_header_nritems(right, nritems);
2310 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2311
2312 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2313 btrfs_item_nr_offset(mid),
2314 nritems * sizeof(struct btrfs_item));
2315
2316 copy_extent_buffer(right, l,
2317 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2318 data_copy_size, btrfs_leaf_data(l) +
2319 leaf_data_end(root, l), data_copy_size);
2320
2321 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2322 btrfs_item_end_nr(l, mid);
2323
2324 for (i = 0; i < nritems; i++) {
2325 struct btrfs_item *item = btrfs_item_nr(right, i);
2326 u32 ioff;
2327
2328 if (!right->map_token) {
2329 map_extent_buffer(right, (unsigned long)item,
2330 sizeof(struct btrfs_item),
2331 &right->map_token, &right->kaddr,
2332 &right->map_start, &right->map_len,
2333 KM_USER1);
2334 }
2335
2336 ioff = btrfs_item_offset(right, item);
2337 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2338 }
2339
2340 if (right->map_token) {
2341 unmap_extent_buffer(right, right->map_token, KM_USER1);
2342 right->map_token = NULL;
2343 }
2344
2345 btrfs_set_header_nritems(l, mid);
2346 ret = 0;
2347 btrfs_item_key(right, &disk_key, 0);
2348 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2349 path->slots[1] + 1, 1);
2350 if (wret)
2351 ret = wret;
2352
2353 btrfs_mark_buffer_dirty(right);
2354 btrfs_mark_buffer_dirty(l);
2355 BUG_ON(path->slots[0] != slot);
2356
2357 if (mid <= slot) {
2358 btrfs_tree_unlock(path->nodes[0]);
2359 free_extent_buffer(path->nodes[0]);
2360 path->nodes[0] = right;
2361 path->slots[0] -= mid;
2362 path->slots[1] += 1;
2363 } else {
2364 btrfs_tree_unlock(right);
2365 free_extent_buffer(right);
2366 }
2367
2368 BUG_ON(path->slots[0] < 0);
2369
2370 if (double_split) {
2371 BUG_ON(num_doubles != 0);
2372 num_doubles++;
2373 goto again;
2374 }
2375 return ret;
2376 }
2377
2378 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2379 struct btrfs_root *root,
2380 struct btrfs_path *path,
2381 u32 new_size, int from_end)
2382 {
2383 int ret = 0;
2384 int slot;
2385 int slot_orig;
2386 struct extent_buffer *leaf;
2387 struct btrfs_item *item;
2388 u32 nritems;
2389 unsigned int data_end;
2390 unsigned int old_data_start;
2391 unsigned int old_size;
2392 unsigned int size_diff;
2393 int i;
2394
2395 slot_orig = path->slots[0];
2396 leaf = path->nodes[0];
2397 slot = path->slots[0];
2398
2399 old_size = btrfs_item_size_nr(leaf, slot);
2400 if (old_size == new_size)
2401 return 0;
2402
2403 nritems = btrfs_header_nritems(leaf);
2404 data_end = leaf_data_end(root, leaf);
2405
2406 old_data_start = btrfs_item_offset_nr(leaf, slot);
2407
2408 size_diff = old_size - new_size;
2409
2410 BUG_ON(slot < 0);
2411 BUG_ON(slot >= nritems);
2412
2413 /*
2414 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2415 */
2416 /* first correct the data pointers */
2417 for (i = slot; i < nritems; i++) {
2418 u32 ioff;
2419 item = btrfs_item_nr(leaf, i);
2420
2421 if (!leaf->map_token) {
2422 map_extent_buffer(leaf, (unsigned long)item,
2423 sizeof(struct btrfs_item),
2424 &leaf->map_token, &leaf->kaddr,
2425 &leaf->map_start, &leaf->map_len,
2426 KM_USER1);
2427 }
2428
2429 ioff = btrfs_item_offset(leaf, item);
2430 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2431 }
2432
2433 if (leaf->map_token) {
2434 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2435 leaf->map_token = NULL;
2436 }
2437
2438 /* shift the data */
2439 if (from_end) {
2440 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2441 data_end + size_diff, btrfs_leaf_data(leaf) +
2442 data_end, old_data_start + new_size - data_end);
2443 } else {
2444 struct btrfs_disk_key disk_key;
2445 u64 offset;
2446
2447 btrfs_item_key(leaf, &disk_key, slot);
2448
2449 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2450 unsigned long ptr;
2451 struct btrfs_file_extent_item *fi;
2452
2453 fi = btrfs_item_ptr(leaf, slot,
2454 struct btrfs_file_extent_item);
2455 fi = (struct btrfs_file_extent_item *)(
2456 (unsigned long)fi - size_diff);
2457
2458 if (btrfs_file_extent_type(leaf, fi) ==
2459 BTRFS_FILE_EXTENT_INLINE) {
2460 ptr = btrfs_item_ptr_offset(leaf, slot);
2461 memmove_extent_buffer(leaf, ptr,
2462 (unsigned long)fi,
2463 offsetof(struct btrfs_file_extent_item,
2464 disk_bytenr));
2465 }
2466 }
2467
2468 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2469 data_end + size_diff, btrfs_leaf_data(leaf) +
2470 data_end, old_data_start - data_end);
2471
2472 offset = btrfs_disk_key_offset(&disk_key);
2473 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2474 btrfs_set_item_key(leaf, &disk_key, slot);
2475 if (slot == 0)
2476 fixup_low_keys(trans, root, path, &disk_key, 1);
2477 }
2478
2479 item = btrfs_item_nr(leaf, slot);
2480 btrfs_set_item_size(leaf, item, new_size);
2481 btrfs_mark_buffer_dirty(leaf);
2482
2483 ret = 0;
2484 if (btrfs_leaf_free_space(root, leaf) < 0) {
2485 btrfs_print_leaf(root, leaf);
2486 BUG();
2487 }
2488 return ret;
2489 }
2490
2491 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2492 struct btrfs_root *root, struct btrfs_path *path,
2493 u32 data_size)
2494 {
2495 int ret = 0;
2496 int slot;
2497 int slot_orig;
2498 struct extent_buffer *leaf;
2499 struct btrfs_item *item;
2500 u32 nritems;
2501 unsigned int data_end;
2502 unsigned int old_data;
2503 unsigned int old_size;
2504 int i;
2505
2506 slot_orig = path->slots[0];
2507 leaf = path->nodes[0];
2508
2509 nritems = btrfs_header_nritems(leaf);
2510 data_end = leaf_data_end(root, leaf);
2511
2512 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2513 btrfs_print_leaf(root, leaf);
2514 BUG();
2515 }
2516 slot = path->slots[0];
2517 old_data = btrfs_item_end_nr(leaf, slot);
2518
2519 BUG_ON(slot < 0);
2520 if (slot >= nritems) {
2521 btrfs_print_leaf(root, leaf);
2522 printk("slot %d too large, nritems %d\n", slot, nritems);
2523 BUG_ON(1);
2524 }
2525
2526 /*
2527 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2528 */
2529 /* first correct the data pointers */
2530 for (i = slot; i < nritems; i++) {
2531 u32 ioff;
2532 item = btrfs_item_nr(leaf, i);
2533
2534 if (!leaf->map_token) {
2535 map_extent_buffer(leaf, (unsigned long)item,
2536 sizeof(struct btrfs_item),
2537 &leaf->map_token, &leaf->kaddr,
2538 &leaf->map_start, &leaf->map_len,
2539 KM_USER1);
2540 }
2541 ioff = btrfs_item_offset(leaf, item);
2542 btrfs_set_item_offset(leaf, item, ioff - data_size);
2543 }
2544
2545 if (leaf->map_token) {
2546 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2547 leaf->map_token = NULL;
2548 }
2549
2550 /* shift the data */
2551 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2552 data_end - data_size, btrfs_leaf_data(leaf) +
2553 data_end, old_data - data_end);
2554
2555 data_end = old_data;
2556 old_size = btrfs_item_size_nr(leaf, slot);
2557 item = btrfs_item_nr(leaf, slot);
2558 btrfs_set_item_size(leaf, item, old_size + data_size);
2559 btrfs_mark_buffer_dirty(leaf);
2560
2561 ret = 0;
2562 if (btrfs_leaf_free_space(root, leaf) < 0) {
2563 btrfs_print_leaf(root, leaf);
2564 BUG();
2565 }
2566 return ret;
2567 }
2568
2569 /*
2570 * Given a key and some data, insert an item into the tree.
2571 * This does all the path init required, making room in the tree if needed.
2572 */
2573 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2574 struct btrfs_root *root,
2575 struct btrfs_path *path,
2576 struct btrfs_key *cpu_key, u32 *data_size,
2577 int nr)
2578 {
2579 struct extent_buffer *leaf;
2580 struct btrfs_item *item;
2581 int ret = 0;
2582 int slot;
2583 int slot_orig;
2584 int i;
2585 u32 nritems;
2586 u32 total_size = 0;
2587 u32 total_data = 0;
2588 unsigned int data_end;
2589 struct btrfs_disk_key disk_key;
2590
2591 for (i = 0; i < nr; i++) {
2592 total_data += data_size[i];
2593 }
2594
2595 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2596 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2597 if (ret == 0) {
2598 return -EEXIST;
2599 }
2600 if (ret < 0)
2601 goto out;
2602
2603 slot_orig = path->slots[0];
2604 leaf = path->nodes[0];
2605
2606 nritems = btrfs_header_nritems(leaf);
2607 data_end = leaf_data_end(root, leaf);
2608
2609 if (btrfs_leaf_free_space(root, leaf) <
2610 sizeof(struct btrfs_item) + total_size) {
2611 btrfs_print_leaf(root, leaf);
2612 printk("not enough freespace need %u have %d\n",
2613 total_size, btrfs_leaf_free_space(root, leaf));
2614 BUG();
2615 }
2616
2617 slot = path->slots[0];
2618 BUG_ON(slot < 0);
2619
2620 if (slot != nritems) {
2621 int i;
2622 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2623
2624 if (old_data < data_end) {
2625 btrfs_print_leaf(root, leaf);
2626 printk("slot %d old_data %d data_end %d\n",
2627 slot, old_data, data_end);
2628 BUG_ON(1);
2629 }
2630 /*
2631 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2632 */
2633 /* first correct the data pointers */
2634 WARN_ON(leaf->map_token);
2635 for (i = slot; i < nritems; i++) {
2636 u32 ioff;
2637
2638 item = btrfs_item_nr(leaf, i);
2639 if (!leaf->map_token) {
2640 map_extent_buffer(leaf, (unsigned long)item,
2641 sizeof(struct btrfs_item),
2642 &leaf->map_token, &leaf->kaddr,
2643 &leaf->map_start, &leaf->map_len,
2644 KM_USER1);
2645 }
2646
2647 ioff = btrfs_item_offset(leaf, item);
2648 btrfs_set_item_offset(leaf, item, ioff - total_data);
2649 }
2650 if (leaf->map_token) {
2651 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2652 leaf->map_token = NULL;
2653 }
2654
2655 /* shift the items */
2656 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2657 btrfs_item_nr_offset(slot),
2658 (nritems - slot) * sizeof(struct btrfs_item));
2659
2660 /* shift the data */
2661 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2662 data_end - total_data, btrfs_leaf_data(leaf) +
2663 data_end, old_data - data_end);
2664 data_end = old_data;
2665 }
2666
2667 /* setup the item for the new data */
2668 for (i = 0; i < nr; i++) {
2669 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2670 btrfs_set_item_key(leaf, &disk_key, slot + i);
2671 item = btrfs_item_nr(leaf, slot + i);
2672 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2673 data_end -= data_size[i];
2674 btrfs_set_item_size(leaf, item, data_size[i]);
2675 }
2676 btrfs_set_header_nritems(leaf, nritems + nr);
2677 btrfs_mark_buffer_dirty(leaf);
2678
2679 ret = 0;
2680 if (slot == 0) {
2681 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2682 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2683 }
2684
2685 if (btrfs_leaf_free_space(root, leaf) < 0) {
2686 btrfs_print_leaf(root, leaf);
2687 BUG();
2688 }
2689 out:
2690 return ret;
2691 }
2692
2693 /*
2694 * Given a key and some data, insert an item into the tree.
2695 * This does all the path init required, making room in the tree if needed.
2696 */
2697 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2698 *root, struct btrfs_key *cpu_key, void *data, u32
2699 data_size)
2700 {
2701 int ret = 0;
2702 struct btrfs_path *path;
2703 struct extent_buffer *leaf;
2704 unsigned long ptr;
2705
2706 path = btrfs_alloc_path();
2707 BUG_ON(!path);
2708 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2709 if (!ret) {
2710 leaf = path->nodes[0];
2711 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2712 write_extent_buffer(leaf, data, ptr, data_size);
2713 btrfs_mark_buffer_dirty(leaf);
2714 }
2715 btrfs_free_path(path);
2716 return ret;
2717 }
2718
2719 /*
2720 * delete the pointer from a given node.
2721 *
2722 * If the delete empties a node, the node is removed from the tree,
2723 * continuing all the way the root if required. The root is converted into
2724 * a leaf if all the nodes are emptied.
2725 */
2726 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2727 struct btrfs_path *path, int level, int slot)
2728 {
2729 struct extent_buffer *parent = path->nodes[level];
2730 u32 nritems;
2731 int ret = 0;
2732 int wret;
2733
2734 nritems = btrfs_header_nritems(parent);
2735 if (slot != nritems -1) {
2736 memmove_extent_buffer(parent,
2737 btrfs_node_key_ptr_offset(slot),
2738 btrfs_node_key_ptr_offset(slot + 1),
2739 sizeof(struct btrfs_key_ptr) *
2740 (nritems - slot - 1));
2741 }
2742 nritems--;
2743 btrfs_set_header_nritems(parent, nritems);
2744 if (nritems == 0 && parent == root->node) {
2745 BUG_ON(btrfs_header_level(root->node) != 1);
2746 /* just turn the root into a leaf and break */
2747 btrfs_set_header_level(root->node, 0);
2748 } else if (slot == 0) {
2749 struct btrfs_disk_key disk_key;
2750
2751 btrfs_node_key(parent, &disk_key, 0);
2752 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2753 if (wret)
2754 ret = wret;
2755 }
2756 btrfs_mark_buffer_dirty(parent);
2757 return ret;
2758 }
2759
2760 /*
2761 * delete the item at the leaf level in path. If that empties
2762 * the leaf, remove it from the tree
2763 */
2764 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2765 struct btrfs_path *path, int slot, int nr)
2766 {
2767 struct extent_buffer *leaf;
2768 struct btrfs_item *item;
2769 int last_off;
2770 int dsize = 0;
2771 int ret = 0;
2772 int wret;
2773 int i;
2774 u32 nritems;
2775
2776 leaf = path->nodes[0];
2777 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2778
2779 for (i = 0; i < nr; i++)
2780 dsize += btrfs_item_size_nr(leaf, slot + i);
2781
2782 nritems = btrfs_header_nritems(leaf);
2783
2784 if (slot + nr != nritems) {
2785 int i;
2786 int data_end = leaf_data_end(root, leaf);
2787
2788 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2789 data_end + dsize,
2790 btrfs_leaf_data(leaf) + data_end,
2791 last_off - data_end);
2792
2793 for (i = slot + nr; i < nritems; i++) {
2794 u32 ioff;
2795
2796 item = btrfs_item_nr(leaf, i);
2797 if (!leaf->map_token) {
2798 map_extent_buffer(leaf, (unsigned long)item,
2799 sizeof(struct btrfs_item),
2800 &leaf->map_token, &leaf->kaddr,
2801 &leaf->map_start, &leaf->map_len,
2802 KM_USER1);
2803 }
2804 ioff = btrfs_item_offset(leaf, item);
2805 btrfs_set_item_offset(leaf, item, ioff + dsize);
2806 }
2807
2808 if (leaf->map_token) {
2809 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2810 leaf->map_token = NULL;
2811 }
2812
2813 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2814 btrfs_item_nr_offset(slot + nr),
2815 sizeof(struct btrfs_item) *
2816 (nritems - slot - nr));
2817 }
2818 btrfs_set_header_nritems(leaf, nritems - nr);
2819 nritems -= nr;
2820
2821 /* delete the leaf if we've emptied it */
2822 if (nritems == 0) {
2823 if (leaf == root->node) {
2824 btrfs_set_header_level(leaf, 0);
2825 } else {
2826 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2827 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2828 if (wret)
2829 ret = wret;
2830 wret = btrfs_free_extent(trans, root,
2831 leaf->start, leaf->len,
2832 btrfs_header_owner(path->nodes[1]),
2833 root_gen, 0, 0, 1);
2834 if (wret)
2835 ret = wret;
2836 }
2837 } else {
2838 int used = leaf_space_used(leaf, 0, nritems);
2839 if (slot == 0) {
2840 struct btrfs_disk_key disk_key;
2841
2842 btrfs_item_key(leaf, &disk_key, 0);
2843 wret = fixup_low_keys(trans, root, path,
2844 &disk_key, 1);
2845 if (wret)
2846 ret = wret;
2847 }
2848
2849 /* delete the leaf if it is mostly empty */
2850 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2851 /* push_leaf_left fixes the path.
2852 * make sure the path still points to our leaf
2853 * for possible call to del_ptr below
2854 */
2855 slot = path->slots[1];
2856 extent_buffer_get(leaf);
2857
2858 wret = push_leaf_left(trans, root, path, 1, 1);
2859 if (wret < 0 && wret != -ENOSPC)
2860 ret = wret;
2861
2862 if (path->nodes[0] == leaf &&
2863 btrfs_header_nritems(leaf)) {
2864 wret = push_leaf_right(trans, root, path, 1, 1);
2865 if (wret < 0 && wret != -ENOSPC)
2866 ret = wret;
2867 }
2868
2869 if (btrfs_header_nritems(leaf) == 0) {
2870 u64 root_gen;
2871 u64 bytenr = leaf->start;
2872 u32 blocksize = leaf->len;
2873
2874 root_gen = btrfs_header_generation(
2875 path->nodes[1]);
2876
2877 wret = del_ptr(trans, root, path, 1, slot);
2878 if (wret)
2879 ret = wret;
2880
2881 free_extent_buffer(leaf);
2882 wret = btrfs_free_extent(trans, root, bytenr,
2883 blocksize,
2884 btrfs_header_owner(path->nodes[1]),
2885 root_gen, 0, 0, 1);
2886 if (wret)
2887 ret = wret;
2888 } else {
2889 /* if we're still in the path, make sure
2890 * we're dirty. Otherwise, one of the
2891 * push_leaf functions must have already
2892 * dirtied this buffer
2893 */
2894 if (path->nodes[0] == leaf)
2895 btrfs_mark_buffer_dirty(leaf);
2896 free_extent_buffer(leaf);
2897 }
2898 } else {
2899 btrfs_mark_buffer_dirty(leaf);
2900 }
2901 }
2902 return ret;
2903 }
2904
2905 /*
2906 * search the tree again to find a leaf with lesser keys
2907 * returns 0 if it found something or 1 if there are no lesser leaves.
2908 * returns < 0 on io errors.
2909 */
2910 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2911 {
2912 struct btrfs_key key;
2913 struct btrfs_disk_key found_key;
2914 int ret;
2915
2916 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
2917
2918 if (key.offset > 0)
2919 key.offset--;
2920 else if (key.type > 0)
2921 key.type--;
2922 else if (key.objectid > 0)
2923 key.objectid--;
2924 else
2925 return 1;
2926
2927 btrfs_release_path(root, path);
2928 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2929 if (ret < 0)
2930 return ret;
2931 btrfs_item_key(path->nodes[0], &found_key, 0);
2932 ret = comp_keys(&found_key, &key);
2933 if (ret < 0)
2934 return 0;
2935 return 1;
2936 }
2937
2938 /*
2939 * search the tree again to find a leaf with greater keys
2940 * returns 0 if it found something or 1 if there are no greater leaves.
2941 * returns < 0 on io errors.
2942 */
2943 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2944 {
2945 int slot;
2946 int level = 1;
2947 struct extent_buffer *c;
2948 struct extent_buffer *next = NULL;
2949 struct btrfs_key key;
2950 u32 nritems;
2951 int ret;
2952
2953 nritems = btrfs_header_nritems(path->nodes[0]);
2954 if (nritems == 0) {
2955 return 1;
2956 }
2957
2958 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
2959
2960 path->keep_locks = 1;
2961 btrfs_release_path(root, path);
2962 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2963 path->keep_locks = 0;
2964
2965 if (ret < 0)
2966 return ret;
2967
2968 if (path->slots[0] < nritems - 1) {
2969 goto done;
2970 }
2971
2972 while(level < BTRFS_MAX_LEVEL) {
2973 if (!path->nodes[level])
2974 return 1;
2975
2976 slot = path->slots[level] + 1;
2977 c = path->nodes[level];
2978 if (slot >= btrfs_header_nritems(c)) {
2979 level++;
2980 if (level == BTRFS_MAX_LEVEL) {
2981 return 1;
2982 }
2983 continue;
2984 }
2985
2986 if (next) {
2987 btrfs_tree_unlock(next);
2988 free_extent_buffer(next);
2989 }
2990
2991 if (level == 1 && path->locks[1] && path->reada)
2992 reada_for_search(root, path, level, slot, 0);
2993
2994 next = read_node_slot(root, c, slot);
2995 if (!path->skip_locking)
2996 btrfs_tree_lock(next);
2997 break;
2998 }
2999 path->slots[level] = slot;
3000 while(1) {
3001 level--;
3002 c = path->nodes[level];
3003 if (path->locks[level])
3004 btrfs_tree_unlock(c);
3005 free_extent_buffer(c);
3006 path->nodes[level] = next;
3007 path->slots[level] = 0;
3008 path->locks[level] = 1;
3009 if (!level)
3010 break;
3011 if (level == 1 && path->locks[1] && path->reada)
3012 reada_for_search(root, path, level, slot, 0);
3013 next = read_node_slot(root, next, 0);
3014 if (!path->skip_locking)
3015 btrfs_tree_lock(next);
3016 }
3017 done:
3018 unlock_up(path, 0, 1);
3019 return 0;
3020 }
3021
3022 int btrfs_previous_item(struct btrfs_root *root,
3023 struct btrfs_path *path, u64 min_objectid,
3024 int type)
3025 {
3026 struct btrfs_key found_key;
3027 struct extent_buffer *leaf;
3028 int ret;
3029
3030 while(1) {
3031 if (path->slots[0] == 0) {
3032 ret = btrfs_prev_leaf(root, path);
3033 if (ret != 0)
3034 return ret;
3035 } else {
3036 path->slots[0]--;
3037 }
3038 leaf = path->nodes[0];
3039 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3040 if (found_key.type == type)
3041 return 0;
3042 }
3043 return 1;
3044 }
3045
Linux kernel - Deliverables
This page took 0.089856 seconds and 6 git commands to generate.