projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Fix btrfs_next_leaf to check for new items after dropping locks
[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 int keep = p->keep_locks;
67
68 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
69 if (!p->nodes[i])
70 continue;
71 if (p->locks[i]) {
72 btrfs_tree_unlock(p->nodes[i]);
73 p->locks[i] = 0;
74 }
75 free_extent_buffer(p->nodes[i]);
76 }
77 memset(p, 0, sizeof(*p));
78 p->keep_locks = keep;
79 }
80
81 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
82 {
83 struct extent_buffer *eb;
84 spin_lock(&root->node_lock);
85 eb = root->node;
86 extent_buffer_get(eb);
87 spin_unlock(&root->node_lock);
88 return eb;
89 }
90
91 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
92 {
93 struct extent_buffer *eb;
94
95 while(1) {
96 eb = btrfs_root_node(root);
97 btrfs_tree_lock(eb);
98
99 spin_lock(&root->node_lock);
100 if (eb == root->node) {
101 spin_unlock(&root->node_lock);
102 break;
103 }
104 spin_unlock(&root->node_lock);
105
106 btrfs_tree_unlock(eb);
107 free_extent_buffer(eb);
108 }
109 return eb;
110 }
111
112 static void add_root_to_dirty_list(struct btrfs_root *root)
113 {
114 if (root->track_dirty && list_empty(&root->dirty_list)) {
115 list_add(&root->dirty_list,
116 &root->fs_info->dirty_cowonly_roots);
117 }
118 }
119
120 int btrfs_copy_root(struct btrfs_trans_handle *trans,
121 struct btrfs_root *root,
122 struct extent_buffer *buf,
123 struct extent_buffer **cow_ret, u64 new_root_objectid)
124 {
125 struct extent_buffer *cow;
126 u32 nritems;
127 int ret = 0;
128 int level;
129 struct btrfs_key first_key;
130 struct btrfs_root *new_root;
131
132 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
133 if (!new_root)
134 return -ENOMEM;
135
136 memcpy(new_root, root, sizeof(*new_root));
137 new_root->root_key.objectid = new_root_objectid;
138
139 WARN_ON(root->ref_cows && trans->transid !=
140 root->fs_info->running_transaction->transid);
141 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
142
143 level = btrfs_header_level(buf);
144 nritems = btrfs_header_nritems(buf);
145 if (nritems) {
146 if (level == 0)
147 btrfs_item_key_to_cpu(buf, &first_key, 0);
148 else
149 btrfs_node_key_to_cpu(buf, &first_key, 0);
150 } else {
151 first_key.objectid = 0;
152 }
153 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
154 new_root_objectid,
155 trans->transid, first_key.objectid,
156 level, buf->start, 0);
157 if (IS_ERR(cow)) {
158 kfree(new_root);
159 return PTR_ERR(cow);
160 }
161
162 copy_extent_buffer(cow, buf, 0, 0, cow->len);
163 btrfs_set_header_bytenr(cow, cow->start);
164 btrfs_set_header_generation(cow, trans->transid);
165 btrfs_set_header_owner(cow, new_root_objectid);
166 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
167
168 WARN_ON(btrfs_header_generation(buf) > trans->transid);
169 ret = btrfs_inc_ref(trans, new_root, buf);
170 kfree(new_root);
171
172 if (ret)
173 return ret;
174
175 btrfs_mark_buffer_dirty(cow);
176 *cow_ret = cow;
177 return 0;
178 }
179
180 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
181 struct btrfs_root *root,
182 struct extent_buffer *buf,
183 struct extent_buffer *parent, int parent_slot,
184 struct extent_buffer **cow_ret,
185 u64 search_start, u64 empty_size)
186 {
187 u64 root_gen;
188 struct extent_buffer *cow;
189 u32 nritems;
190 int ret = 0;
191 int different_trans = 0;
192 int level;
193 int unlock_orig = 0;
194 struct btrfs_key first_key;
195
196 if (*cow_ret == buf)
197 unlock_orig = 1;
198
199 WARN_ON(!btrfs_tree_locked(buf));
200
201 if (root->ref_cows) {
202 root_gen = trans->transid;
203 } else {
204 root_gen = 0;
205 }
206 WARN_ON(root->ref_cows && trans->transid !=
207 root->fs_info->running_transaction->transid);
208 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
209
210 level = btrfs_header_level(buf);
211 nritems = btrfs_header_nritems(buf);
212 if (nritems) {
213 if (level == 0)
214 btrfs_item_key_to_cpu(buf, &first_key, 0);
215 else
216 btrfs_node_key_to_cpu(buf, &first_key, 0);
217 } else {
218 first_key.objectid = 0;
219 }
220 cow = btrfs_alloc_free_block(trans, root, buf->len,
221 root->root_key.objectid,
222 root_gen, first_key.objectid, level,
223 search_start, empty_size);
224 if (IS_ERR(cow))
225 return PTR_ERR(cow);
226
227 copy_extent_buffer(cow, buf, 0, 0, cow->len);
228 btrfs_set_header_bytenr(cow, cow->start);
229 btrfs_set_header_generation(cow, trans->transid);
230 btrfs_set_header_owner(cow, root->root_key.objectid);
231 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
232
233 WARN_ON(btrfs_header_generation(buf) > trans->transid);
234 if (btrfs_header_generation(buf) != trans->transid) {
235 different_trans = 1;
236 ret = btrfs_inc_ref(trans, root, buf);
237 if (ret)
238 return ret;
239 } else {
240 clean_tree_block(trans, root, buf);
241 }
242
243 if (buf == root->node) {
244 WARN_ON(parent && parent != buf);
245 root_gen = btrfs_header_generation(buf);
246
247 spin_lock(&root->node_lock);
248 root->node = cow;
249 extent_buffer_get(cow);
250 spin_unlock(&root->node_lock);
251
252 if (buf != root->commit_root) {
253 btrfs_free_extent(trans, root, buf->start,
254 buf->len, root->root_key.objectid,
255 root_gen, 0, 0, 1);
256 }
257 free_extent_buffer(buf);
258 add_root_to_dirty_list(root);
259 } else {
260 root_gen = btrfs_header_generation(parent);
261 btrfs_set_node_blockptr(parent, parent_slot,
262 cow->start);
263 WARN_ON(trans->transid == 0);
264 btrfs_set_node_ptr_generation(parent, parent_slot,
265 trans->transid);
266 btrfs_mark_buffer_dirty(parent);
267 WARN_ON(btrfs_header_generation(parent) != trans->transid);
268 btrfs_free_extent(trans, root, buf->start, buf->len,
269 btrfs_header_owner(parent), root_gen,
270 0, 0, 1);
271 }
272 if (unlock_orig)
273 btrfs_tree_unlock(buf);
274 free_extent_buffer(buf);
275 btrfs_mark_buffer_dirty(cow);
276 *cow_ret = cow;
277 return 0;
278 }
279
280 int btrfs_cow_block(struct btrfs_trans_handle *trans,
281 struct btrfs_root *root, struct extent_buffer *buf,
282 struct extent_buffer *parent, int parent_slot,
283 struct extent_buffer **cow_ret)
284 {
285 u64 search_start;
286 u64 header_trans;
287 int ret;
288
289 if (trans->transaction != root->fs_info->running_transaction) {
290 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
291 root->fs_info->running_transaction->transid);
292 WARN_ON(1);
293 }
294 if (trans->transid != root->fs_info->generation) {
295 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
296 root->fs_info->generation);
297 WARN_ON(1);
298 }
299
300 header_trans = btrfs_header_generation(buf);
301 spin_lock(&root->fs_info->hash_lock);
302 if (header_trans == trans->transid &&
303 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
304 *cow_ret = buf;
305 spin_unlock(&root->fs_info->hash_lock);
306 return 0;
307 }
308 spin_unlock(&root->fs_info->hash_lock);
309 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
310 ret = __btrfs_cow_block(trans, root, buf, parent,
311 parent_slot, cow_ret, search_start, 0);
312 return ret;
313 }
314
315 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
316 {
317 if (blocknr < other && other - (blocknr + blocksize) < 32768)
318 return 1;
319 if (blocknr > other && blocknr - (other + blocksize) < 32768)
320 return 1;
321 return 0;
322 }
323
324 /*
325 * compare two keys in a memcmp fashion
326 */
327 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
328 {
329 struct btrfs_key k1;
330
331 btrfs_disk_key_to_cpu(&k1, disk);
332
333 if (k1.objectid > k2->objectid)
334 return 1;
335 if (k1.objectid < k2->objectid)
336 return -1;
337 if (k1.type > k2->type)
338 return 1;
339 if (k1.type < k2->type)
340 return -1;
341 if (k1.offset > k2->offset)
342 return 1;
343 if (k1.offset < k2->offset)
344 return -1;
345 return 0;
346 }
347
348
349 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
350 struct btrfs_root *root, struct extent_buffer *parent,
351 int start_slot, int cache_only, u64 *last_ret,
352 struct btrfs_key *progress)
353 {
354 struct extent_buffer *cur;
355 struct extent_buffer *tmp;
356 u64 blocknr;
357 u64 gen;
358 u64 search_start = *last_ret;
359 u64 last_block = 0;
360 u64 other;
361 u32 parent_nritems;
362 int end_slot;
363 int i;
364 int err = 0;
365 int parent_level;
366 int uptodate;
367 u32 blocksize;
368 int progress_passed = 0;
369 struct btrfs_disk_key disk_key;
370
371 /* FIXME this code needs locking */
372 return 0;
373
374 parent_level = btrfs_header_level(parent);
375 if (cache_only && parent_level != 1)
376 return 0;
377
378 if (trans->transaction != root->fs_info->running_transaction) {
379 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
380 root->fs_info->running_transaction->transid);
381 WARN_ON(1);
382 }
383 if (trans->transid != root->fs_info->generation) {
384 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
385 root->fs_info->generation);
386 WARN_ON(1);
387 }
388
389 parent_nritems = btrfs_header_nritems(parent);
390 blocksize = btrfs_level_size(root, parent_level - 1);
391 end_slot = parent_nritems;
392
393 if (parent_nritems == 1)
394 return 0;
395
396 for (i = start_slot; i < end_slot; i++) {
397 int close = 1;
398
399 if (!parent->map_token) {
400 map_extent_buffer(parent,
401 btrfs_node_key_ptr_offset(i),
402 sizeof(struct btrfs_key_ptr),
403 &parent->map_token, &parent->kaddr,
404 &parent->map_start, &parent->map_len,
405 KM_USER1);
406 }
407 btrfs_node_key(parent, &disk_key, i);
408 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
409 continue;
410
411 progress_passed = 1;
412 blocknr = btrfs_node_blockptr(parent, i);
413 gen = btrfs_node_ptr_generation(parent, i);
414 if (last_block == 0)
415 last_block = blocknr;
416
417 if (i > 0) {
418 other = btrfs_node_blockptr(parent, i - 1);
419 close = close_blocks(blocknr, other, blocksize);
420 }
421 if (!close && i < end_slot - 2) {
422 other = btrfs_node_blockptr(parent, i + 1);
423 close = close_blocks(blocknr, other, blocksize);
424 }
425 if (close) {
426 last_block = blocknr;
427 continue;
428 }
429 if (parent->map_token) {
430 unmap_extent_buffer(parent, parent->map_token,
431 KM_USER1);
432 parent->map_token = NULL;
433 }
434
435 cur = btrfs_find_tree_block(root, blocknr, blocksize);
436 if (cur)
437 uptodate = btrfs_buffer_uptodate(cur, gen);
438 else
439 uptodate = 0;
440 if (!cur || !uptodate) {
441 if (cache_only) {
442 free_extent_buffer(cur);
443 continue;
444 }
445 if (!cur) {
446 cur = read_tree_block(root, blocknr,
447 blocksize, gen);
448 } else if (!uptodate) {
449 btrfs_read_buffer(cur, gen);
450 }
451 }
452 if (search_start == 0)
453 search_start = last_block;
454
455 err = __btrfs_cow_block(trans, root, cur, parent, i,
456 &tmp, search_start,
457 min(16 * blocksize,
458 (end_slot - i) * blocksize));
459 if (err) {
460 free_extent_buffer(cur);
461 break;
462 }
463 search_start = tmp->start;
464 last_block = tmp->start;
465 *last_ret = search_start;
466 if (parent_level == 1)
467 btrfs_clear_buffer_defrag(tmp);
468 free_extent_buffer(tmp);
469 }
470 if (parent->map_token) {
471 unmap_extent_buffer(parent, parent->map_token,
472 KM_USER1);
473 parent->map_token = NULL;
474 }
475 return err;
476 }
477
478 /*
479 * The leaf data grows from end-to-front in the node.
480 * this returns the address of the start of the last item,
481 * which is the stop of the leaf data stack
482 */
483 static inline unsigned int leaf_data_end(struct btrfs_root *root,
484 struct extent_buffer *leaf)
485 {
486 u32 nr = btrfs_header_nritems(leaf);
487 if (nr == 0)
488 return BTRFS_LEAF_DATA_SIZE(root);
489 return btrfs_item_offset_nr(leaf, nr - 1);
490 }
491
492 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
493 int level)
494 {
495 struct extent_buffer *parent = NULL;
496 struct extent_buffer *node = path->nodes[level];
497 struct btrfs_disk_key parent_key;
498 struct btrfs_disk_key node_key;
499 int parent_slot;
500 int slot;
501 struct btrfs_key cpukey;
502 u32 nritems = btrfs_header_nritems(node);
503
504 if (path->nodes[level + 1])
505 parent = path->nodes[level + 1];
506
507 slot = path->slots[level];
508 BUG_ON(nritems == 0);
509 if (parent) {
510 parent_slot = path->slots[level + 1];
511 btrfs_node_key(parent, &parent_key, parent_slot);
512 btrfs_node_key(node, &node_key, 0);
513 BUG_ON(memcmp(&parent_key, &node_key,
514 sizeof(struct btrfs_disk_key)));
515 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
516 btrfs_header_bytenr(node));
517 }
518 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
519 if (slot != 0) {
520 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
521 btrfs_node_key(node, &node_key, slot);
522 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
523 }
524 if (slot < nritems - 1) {
525 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
526 btrfs_node_key(node, &node_key, slot);
527 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
528 }
529 return 0;
530 }
531
532 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
533 int level)
534 {
535 struct extent_buffer *leaf = path->nodes[level];
536 struct extent_buffer *parent = NULL;
537 int parent_slot;
538 struct btrfs_key cpukey;
539 struct btrfs_disk_key parent_key;
540 struct btrfs_disk_key leaf_key;
541 int slot = path->slots[0];
542
543 u32 nritems = btrfs_header_nritems(leaf);
544
545 if (path->nodes[level + 1])
546 parent = path->nodes[level + 1];
547
548 if (nritems == 0)
549 return 0;
550
551 if (parent) {
552 parent_slot = path->slots[level + 1];
553 btrfs_node_key(parent, &parent_key, parent_slot);
554 btrfs_item_key(leaf, &leaf_key, 0);
555
556 BUG_ON(memcmp(&parent_key, &leaf_key,
557 sizeof(struct btrfs_disk_key)));
558 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
559 btrfs_header_bytenr(leaf));
560 }
561 #if 0
562 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
563 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
564 btrfs_item_key(leaf, &leaf_key, i);
565 if (comp_keys(&leaf_key, &cpukey) >= 0) {
566 btrfs_print_leaf(root, leaf);
567 printk("slot %d offset bad key\n", i);
568 BUG_ON(1);
569 }
570 if (btrfs_item_offset_nr(leaf, i) !=
571 btrfs_item_end_nr(leaf, i + 1)) {
572 btrfs_print_leaf(root, leaf);
573 printk("slot %d offset bad\n", i);
574 BUG_ON(1);
575 }
576 if (i == 0) {
577 if (btrfs_item_offset_nr(leaf, i) +
578 btrfs_item_size_nr(leaf, i) !=
579 BTRFS_LEAF_DATA_SIZE(root)) {
580 btrfs_print_leaf(root, leaf);
581 printk("slot %d first offset bad\n", i);
582 BUG_ON(1);
583 }
584 }
585 }
586 if (nritems > 0) {
587 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
588 btrfs_print_leaf(root, leaf);
589 printk("slot %d bad size \n", nritems - 1);
590 BUG_ON(1);
591 }
592 }
593 #endif
594 if (slot != 0 && slot < nritems - 1) {
595 btrfs_item_key(leaf, &leaf_key, slot);
596 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
597 if (comp_keys(&leaf_key, &cpukey) <= 0) {
598 btrfs_print_leaf(root, leaf);
599 printk("slot %d offset bad key\n", slot);
600 BUG_ON(1);
601 }
602 if (btrfs_item_offset_nr(leaf, slot - 1) !=
603 btrfs_item_end_nr(leaf, slot)) {
604 btrfs_print_leaf(root, leaf);
605 printk("slot %d offset bad\n", slot);
606 BUG_ON(1);
607 }
608 }
609 if (slot < nritems - 1) {
610 btrfs_item_key(leaf, &leaf_key, slot);
611 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
612 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
613 if (btrfs_item_offset_nr(leaf, slot) !=
614 btrfs_item_end_nr(leaf, slot + 1)) {
615 btrfs_print_leaf(root, leaf);
616 printk("slot %d offset bad\n", slot);
617 BUG_ON(1);
618 }
619 }
620 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
621 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
622 return 0;
623 }
624
625 static int noinline check_block(struct btrfs_root *root,
626 struct btrfs_path *path, int level)
627 {
628 u64 found_start;
629 return 0;
630 if (btrfs_header_level(path->nodes[level]) != level)
631 printk("warning: bad level %Lu wanted %d found %d\n",
632 path->nodes[level]->start, level,
633 btrfs_header_level(path->nodes[level]));
634 found_start = btrfs_header_bytenr(path->nodes[level]);
635 if (found_start != path->nodes[level]->start) {
636 printk("warning: bad bytentr %Lu found %Lu\n",
637 path->nodes[level]->start, found_start);
638 }
639 #if 0
640 struct extent_buffer *buf = path->nodes[level];
641
642 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
643 (unsigned long)btrfs_header_fsid(buf),
644 BTRFS_FSID_SIZE)) {
645 printk("warning bad block %Lu\n", buf->start);
646 return 1;
647 }
648 #endif
649 if (level == 0)
650 return check_leaf(root, path, level);
651 return check_node(root, path, level);
652 }
653
654 /*
655 * search for key in the extent_buffer. The items start at offset p,
656 * and they are item_size apart. There are 'max' items in p.
657 *
658 * the slot in the array is returned via slot, and it points to
659 * the place where you would insert key if it is not found in
660 * the array.
661 *
662 * slot may point to max if the key is bigger than all of the keys
663 */
664 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
665 int item_size, struct btrfs_key *key,
666 int max, int *slot)
667 {
668 int low = 0;
669 int high = max;
670 int mid;
671 int ret;
672 struct btrfs_disk_key *tmp = NULL;
673 struct btrfs_disk_key unaligned;
674 unsigned long offset;
675 char *map_token = NULL;
676 char *kaddr = NULL;
677 unsigned long map_start = 0;
678 unsigned long map_len = 0;
679 int err;
680
681 while(low < high) {
682 mid = (low + high) / 2;
683 offset = p + mid * item_size;
684
685 if (!map_token || offset < map_start ||
686 (offset + sizeof(struct btrfs_disk_key)) >
687 map_start + map_len) {
688 if (map_token) {
689 unmap_extent_buffer(eb, map_token, KM_USER0);
690 map_token = NULL;
691 }
692 err = map_extent_buffer(eb, offset,
693 sizeof(struct btrfs_disk_key),
694 &map_token, &kaddr,
695 &map_start, &map_len, KM_USER0);
696
697 if (!err) {
698 tmp = (struct btrfs_disk_key *)(kaddr + offset -
699 map_start);
700 } else {
701 read_extent_buffer(eb, &unaligned,
702 offset, sizeof(unaligned));
703 tmp = &unaligned;
704 }
705
706 } else {
707 tmp = (struct btrfs_disk_key *)(kaddr + offset -
708 map_start);
709 }
710 ret = comp_keys(tmp, key);
711
712 if (ret < 0)
713 low = mid + 1;
714 else if (ret > 0)
715 high = mid;
716 else {
717 *slot = mid;
718 if (map_token)
719 unmap_extent_buffer(eb, map_token, KM_USER0);
720 return 0;
721 }
722 }
723 *slot = low;
724 if (map_token)
725 unmap_extent_buffer(eb, map_token, KM_USER0);
726 return 1;
727 }
728
729 /*
730 * simple bin_search frontend that does the right thing for
731 * leaves vs nodes
732 */
733 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
734 int level, int *slot)
735 {
736 if (level == 0) {
737 return generic_bin_search(eb,
738 offsetof(struct btrfs_leaf, items),
739 sizeof(struct btrfs_item),
740 key, btrfs_header_nritems(eb),
741 slot);
742 } else {
743 return generic_bin_search(eb,
744 offsetof(struct btrfs_node, ptrs),
745 sizeof(struct btrfs_key_ptr),
746 key, btrfs_header_nritems(eb),
747 slot);
748 }
749 return -1;
750 }
751
752 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
753 struct extent_buffer *parent, int slot)
754 {
755 int level = btrfs_header_level(parent);
756 if (slot < 0)
757 return NULL;
758 if (slot >= btrfs_header_nritems(parent))
759 return NULL;
760
761 BUG_ON(level == 0);
762
763 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
764 btrfs_level_size(root, level - 1),
765 btrfs_node_ptr_generation(parent, slot));
766 }
767
768 static int balance_level(struct btrfs_trans_handle *trans,
769 struct btrfs_root *root,
770 struct btrfs_path *path, int level)
771 {
772 struct extent_buffer *right = NULL;
773 struct extent_buffer *mid;
774 struct extent_buffer *left = NULL;
775 struct extent_buffer *parent = NULL;
776 int ret = 0;
777 int wret;
778 int pslot;
779 int orig_slot = path->slots[level];
780 int err_on_enospc = 0;
781 u64 orig_ptr;
782
783 if (level == 0)
784 return 0;
785
786 mid = path->nodes[level];
787 WARN_ON(!path->locks[level]);
788 WARN_ON(btrfs_header_generation(mid) != trans->transid);
789
790 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
791
792 if (level < BTRFS_MAX_LEVEL - 1)
793 parent = path->nodes[level + 1];
794 pslot = path->slots[level + 1];
795
796 /*
797 * deal with the case where there is only one pointer in the root
798 * by promoting the node below to a root
799 */
800 if (!parent) {
801 struct extent_buffer *child;
802
803 if (btrfs_header_nritems(mid) != 1)
804 return 0;
805
806 /* promote the child to a root */
807 child = read_node_slot(root, mid, 0);
808 btrfs_tree_lock(child);
809 BUG_ON(!child);
810 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
811 BUG_ON(ret);
812
813 spin_lock(&root->node_lock);
814 root->node = child;
815 spin_unlock(&root->node_lock);
816
817 add_root_to_dirty_list(root);
818 btrfs_tree_unlock(child);
819 path->locks[level] = 0;
820 path->nodes[level] = NULL;
821 clean_tree_block(trans, root, mid);
822 btrfs_tree_unlock(mid);
823 /* once for the path */
824 free_extent_buffer(mid);
825 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
826 root->root_key.objectid,
827 btrfs_header_generation(mid), 0, 0, 1);
828 /* once for the root ptr */
829 free_extent_buffer(mid);
830 return ret;
831 }
832 if (btrfs_header_nritems(mid) >
833 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
834 return 0;
835
836 if (btrfs_header_nritems(mid) < 2)
837 err_on_enospc = 1;
838
839 left = read_node_slot(root, parent, pslot - 1);
840 if (left) {
841 btrfs_tree_lock(left);
842 wret = btrfs_cow_block(trans, root, left,
843 parent, pslot - 1, &left);
844 if (wret) {
845 ret = wret;
846 goto enospc;
847 }
848 }
849 right = read_node_slot(root, parent, pslot + 1);
850 if (right) {
851 btrfs_tree_lock(right);
852 wret = btrfs_cow_block(trans, root, right,
853 parent, pslot + 1, &right);
854 if (wret) {
855 ret = wret;
856 goto enospc;
857 }
858 }
859
860 /* first, try to make some room in the middle buffer */
861 if (left) {
862 orig_slot += btrfs_header_nritems(left);
863 wret = push_node_left(trans, root, left, mid, 1);
864 if (wret < 0)
865 ret = wret;
866 if (btrfs_header_nritems(mid) < 2)
867 err_on_enospc = 1;
868 }
869
870 /*
871 * then try to empty the right most buffer into the middle
872 */
873 if (right) {
874 wret = push_node_left(trans, root, mid, right, 1);
875 if (wret < 0 && wret != -ENOSPC)
876 ret = wret;
877 if (btrfs_header_nritems(right) == 0) {
878 u64 bytenr = right->start;
879 u64 generation = btrfs_header_generation(parent);
880 u32 blocksize = right->len;
881
882 clean_tree_block(trans, root, right);
883 btrfs_tree_unlock(right);
884 free_extent_buffer(right);
885 right = NULL;
886 wret = del_ptr(trans, root, path, level + 1, pslot +
887 1);
888 if (wret)
889 ret = wret;
890 wret = btrfs_free_extent(trans, root, bytenr,
891 blocksize,
892 btrfs_header_owner(parent),
893 generation, 0, 0, 1);
894 if (wret)
895 ret = wret;
896 } else {
897 struct btrfs_disk_key right_key;
898 btrfs_node_key(right, &right_key, 0);
899 btrfs_set_node_key(parent, &right_key, pslot + 1);
900 btrfs_mark_buffer_dirty(parent);
901 }
902 }
903 if (btrfs_header_nritems(mid) == 1) {
904 /*
905 * we're not allowed to leave a node with one item in the
906 * tree during a delete. A deletion from lower in the tree
907 * could try to delete the only pointer in this node.
908 * So, pull some keys from the left.
909 * There has to be a left pointer at this point because
910 * otherwise we would have pulled some pointers from the
911 * right
912 */
913 BUG_ON(!left);
914 wret = balance_node_right(trans, root, mid, left);
915 if (wret < 0) {
916 ret = wret;
917 goto enospc;
918 }
919 if (wret == 1) {
920 wret = push_node_left(trans, root, left, mid, 1);
921 if (wret < 0)
922 ret = wret;
923 }
924 BUG_ON(wret == 1);
925 }
926 if (btrfs_header_nritems(mid) == 0) {
927 /* we've managed to empty the middle node, drop it */
928 u64 root_gen = btrfs_header_generation(parent);
929 u64 bytenr = mid->start;
930 u32 blocksize = mid->len;
931
932 clean_tree_block(trans, root, mid);
933 btrfs_tree_unlock(mid);
934 free_extent_buffer(mid);
935 mid = NULL;
936 wret = del_ptr(trans, root, path, level + 1, pslot);
937 if (wret)
938 ret = wret;
939 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
940 btrfs_header_owner(parent),
941 root_gen, 0, 0, 1);
942 if (wret)
943 ret = wret;
944 } else {
945 /* update the parent key to reflect our changes */
946 struct btrfs_disk_key mid_key;
947 btrfs_node_key(mid, &mid_key, 0);
948 btrfs_set_node_key(parent, &mid_key, pslot);
949 btrfs_mark_buffer_dirty(parent);
950 }
951
952 /* update the path */
953 if (left) {
954 if (btrfs_header_nritems(left) > orig_slot) {
955 extent_buffer_get(left);
956 /* left was locked after cow */
957 path->nodes[level] = left;
958 path->slots[level + 1] -= 1;
959 path->slots[level] = orig_slot;
960 if (mid) {
961 btrfs_tree_unlock(mid);
962 free_extent_buffer(mid);
963 }
964 } else {
965 orig_slot -= btrfs_header_nritems(left);
966 path->slots[level] = orig_slot;
967 }
968 }
969 /* double check we haven't messed things up */
970 check_block(root, path, level);
971 if (orig_ptr !=
972 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
973 BUG();
974 enospc:
975 if (right) {
976 btrfs_tree_unlock(right);
977 free_extent_buffer(right);
978 }
979 if (left) {
980 if (path->nodes[level] != left)
981 btrfs_tree_unlock(left);
982 free_extent_buffer(left);
983 }
984 return ret;
985 }
986
987 /* returns zero if the push worked, non-zero otherwise */
988 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
989 struct btrfs_root *root,
990 struct btrfs_path *path, int level)
991 {
992 struct extent_buffer *right = NULL;
993 struct extent_buffer *mid;
994 struct extent_buffer *left = NULL;
995 struct extent_buffer *parent = NULL;
996 int ret = 0;
997 int wret;
998 int pslot;
999 int orig_slot = path->slots[level];
1000 u64 orig_ptr;
1001
1002 if (level == 0)
1003 return 1;
1004
1005 mid = path->nodes[level];
1006 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1007 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1008
1009 if (level < BTRFS_MAX_LEVEL - 1)
1010 parent = path->nodes[level + 1];
1011 pslot = path->slots[level + 1];
1012
1013 if (!parent)
1014 return 1;
1015
1016 left = read_node_slot(root, parent, pslot - 1);
1017
1018 /* first, try to make some room in the middle buffer */
1019 if (left) {
1020 u32 left_nr;
1021
1022 btrfs_tree_lock(left);
1023 left_nr = btrfs_header_nritems(left);
1024 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1025 wret = 1;
1026 } else {
1027 ret = btrfs_cow_block(trans, root, left, parent,
1028 pslot - 1, &left);
1029 if (ret)
1030 wret = 1;
1031 else {
1032 wret = push_node_left(trans, root,
1033 left, mid, 0);
1034 }
1035 }
1036 if (wret < 0)
1037 ret = wret;
1038 if (wret == 0) {
1039 struct btrfs_disk_key disk_key;
1040 orig_slot += left_nr;
1041 btrfs_node_key(mid, &disk_key, 0);
1042 btrfs_set_node_key(parent, &disk_key, pslot);
1043 btrfs_mark_buffer_dirty(parent);
1044 if (btrfs_header_nritems(left) > orig_slot) {
1045 path->nodes[level] = left;
1046 path->slots[level + 1] -= 1;
1047 path->slots[level] = orig_slot;
1048 btrfs_tree_unlock(mid);
1049 free_extent_buffer(mid);
1050 } else {
1051 orig_slot -=
1052 btrfs_header_nritems(left);
1053 path->slots[level] = orig_slot;
1054 btrfs_tree_unlock(left);
1055 free_extent_buffer(left);
1056 }
1057 return 0;
1058 }
1059 btrfs_tree_unlock(left);
1060 free_extent_buffer(left);
1061 }
1062 right = read_node_slot(root, parent, pslot + 1);
1063
1064 /*
1065 * then try to empty the right most buffer into the middle
1066 */
1067 if (right) {
1068 u32 right_nr;
1069 btrfs_tree_lock(right);
1070 right_nr = btrfs_header_nritems(right);
1071 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1072 wret = 1;
1073 } else {
1074 ret = btrfs_cow_block(trans, root, right,
1075 parent, pslot + 1,
1076 &right);
1077 if (ret)
1078 wret = 1;
1079 else {
1080 wret = balance_node_right(trans, root,
1081 right, mid);
1082 }
1083 }
1084 if (wret < 0)
1085 ret = wret;
1086 if (wret == 0) {
1087 struct btrfs_disk_key disk_key;
1088
1089 btrfs_node_key(right, &disk_key, 0);
1090 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1091 btrfs_mark_buffer_dirty(parent);
1092
1093 if (btrfs_header_nritems(mid) <= orig_slot) {
1094 path->nodes[level] = right;
1095 path->slots[level + 1] += 1;
1096 path->slots[level] = orig_slot -
1097 btrfs_header_nritems(mid);
1098 btrfs_tree_unlock(mid);
1099 free_extent_buffer(mid);
1100 } else {
1101 btrfs_tree_unlock(right);
1102 free_extent_buffer(right);
1103 }
1104 return 0;
1105 }
1106 btrfs_tree_unlock(right);
1107 free_extent_buffer(right);
1108 }
1109 return 1;
1110 }
1111
1112 /*
1113 * readahead one full node of leaves
1114 */
1115 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
1116 int level, int slot, u64 objectid)
1117 {
1118 struct extent_buffer *node;
1119 struct btrfs_disk_key disk_key;
1120 u32 nritems;
1121 u64 search;
1122 u64 lowest_read;
1123 u64 highest_read;
1124 u64 nread = 0;
1125 int direction = path->reada;
1126 struct extent_buffer *eb;
1127 u32 nr;
1128 u32 blocksize;
1129 u32 nscan = 0;
1130
1131 if (level != 1)
1132 return;
1133
1134 if (!path->nodes[level])
1135 return;
1136
1137 node = path->nodes[level];
1138
1139 search = btrfs_node_blockptr(node, slot);
1140 blocksize = btrfs_level_size(root, level - 1);
1141 eb = btrfs_find_tree_block(root, search, blocksize);
1142 if (eb) {
1143 free_extent_buffer(eb);
1144 return;
1145 }
1146
1147 highest_read = search;
1148 lowest_read = search;
1149
1150 nritems = btrfs_header_nritems(node);
1151 nr = slot;
1152 while(1) {
1153 if (direction < 0) {
1154 if (nr == 0)
1155 break;
1156 nr--;
1157 } else if (direction > 0) {
1158 nr++;
1159 if (nr >= nritems)
1160 break;
1161 }
1162 if (path->reada < 0 && objectid) {
1163 btrfs_node_key(node, &disk_key, nr);
1164 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1165 break;
1166 }
1167 search = btrfs_node_blockptr(node, nr);
1168 if ((search >= lowest_read && search <= highest_read) ||
1169 (search < lowest_read && lowest_read - search <= 32768) ||
1170 (search > highest_read && search - highest_read <= 32768)) {
1171 readahead_tree_block(root, search, blocksize,
1172 btrfs_node_ptr_generation(node, nr));
1173 nread += blocksize;
1174 }
1175 nscan++;
1176 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1177 break;
1178 if(nread > (1024 * 1024) || nscan > 128)
1179 break;
1180
1181 if (search < lowest_read)
1182 lowest_read = search;
1183 if (search > highest_read)
1184 highest_read = search;
1185 }
1186 }
1187
1188 static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock)
1189 {
1190 int i;
1191 int skip_level = level;
1192 int no_skips = 0;
1193 struct extent_buffer *t;
1194
1195 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1196 if (!path->nodes[i])
1197 break;
1198 if (!path->locks[i])
1199 break;
1200 if (!no_skips && path->slots[i] == 0) {
1201 skip_level = i + 1;
1202 continue;
1203 }
1204 if (!no_skips && path->keep_locks) {
1205 u32 nritems;
1206 t = path->nodes[i];
1207 nritems = btrfs_header_nritems(t);
1208 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1209 skip_level = i + 1;
1210 continue;
1211 }
1212 }
1213 if (skip_level < i && i >= lowest_unlock)
1214 no_skips = 1;
1215
1216 t = path->nodes[i];
1217 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1218 btrfs_tree_unlock(t);
1219 path->locks[i] = 0;
1220 }
1221 }
1222 }
1223
1224 /*
1225 * look for key in the tree. path is filled in with nodes along the way
1226 * if key is found, we return zero and you can find the item in the leaf
1227 * level of the path (level 0)
1228 *
1229 * If the key isn't found, the path points to the slot where it should
1230 * be inserted, and 1 is returned. If there are other errors during the
1231 * search a negative error number is returned.
1232 *
1233 * if ins_len > 0, nodes and leaves will be split as we walk down the
1234 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1235 * possible)
1236 */
1237 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1238 *root, struct btrfs_key *key, struct btrfs_path *p, int
1239 ins_len, int cow)
1240 {
1241 struct extent_buffer *b;
1242 struct extent_buffer *tmp;
1243 int slot;
1244 int ret;
1245 int level;
1246 int should_reada = p->reada;
1247 int lowest_unlock = 1;
1248 int blocksize;
1249 u8 lowest_level = 0;
1250 u64 blocknr;
1251 u64 gen;
1252
1253 lowest_level = p->lowest_level;
1254 WARN_ON(lowest_level && ins_len);
1255 WARN_ON(p->nodes[0] != NULL);
1256 WARN_ON(root == root->fs_info->extent_root &&
1257 !mutex_is_locked(&root->fs_info->alloc_mutex));
1258 WARN_ON(root == root->fs_info->chunk_root &&
1259 !mutex_is_locked(&root->fs_info->chunk_mutex));
1260 WARN_ON(root == root->fs_info->dev_root &&
1261 !mutex_is_locked(&root->fs_info->chunk_mutex));
1262 if (ins_len < 0)
1263 lowest_unlock = 2;
1264 again:
1265 b = btrfs_lock_root_node(root);
1266
1267 while (b) {
1268 level = btrfs_header_level(b);
1269 if (cow) {
1270 int wret;
1271 wret = btrfs_cow_block(trans, root, b,
1272 p->nodes[level + 1],
1273 p->slots[level + 1],
1274 &b);
1275 if (wret) {
1276 free_extent_buffer(b);
1277 return wret;
1278 }
1279 }
1280 BUG_ON(!cow && ins_len);
1281 if (level != btrfs_header_level(b))
1282 WARN_ON(1);
1283 level = btrfs_header_level(b);
1284 p->nodes[level] = b;
1285 p->locks[level] = 1;
1286 ret = check_block(root, p, level);
1287 if (ret)
1288 return -1;
1289
1290 ret = bin_search(b, key, level, &slot);
1291 if (level != 0) {
1292 if (ret && slot > 0)
1293 slot -= 1;
1294 p->slots[level] = slot;
1295 if (ins_len > 0 && btrfs_header_nritems(b) >=
1296 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1297 int sret = split_node(trans, root, p, level);
1298 BUG_ON(sret > 0);
1299 if (sret)
1300 return sret;
1301 b = p->nodes[level];
1302 slot = p->slots[level];
1303 } else if (ins_len < 0) {
1304 int sret = balance_level(trans, root, p,
1305 level);
1306 if (sret)
1307 return sret;
1308 b = p->nodes[level];
1309 if (!b) {
1310 btrfs_release_path(NULL, p);
1311 goto again;
1312 }
1313 slot = p->slots[level];
1314 BUG_ON(btrfs_header_nritems(b) == 1);
1315 }
1316 /* this is only true while dropping a snapshot */
1317 if (level == lowest_level) {
1318 unlock_up(p, level, lowest_unlock);
1319 break;
1320 }
1321
1322 if (should_reada)
1323 reada_for_search(root, p, level, slot,
1324 key->objectid);
1325
1326 blocknr = btrfs_node_blockptr(b, slot);
1327 gen = btrfs_node_ptr_generation(b, slot);
1328 blocksize = btrfs_level_size(root, level - 1);
1329
1330 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1331 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1332 b = tmp;
1333 } else {
1334 /*
1335 * reduce lock contention at high levels
1336 * of the btree by dropping locks before
1337 * we read.
1338 */
1339 if (level > 1) {
1340 btrfs_release_path(NULL, p);
1341 if (tmp)
1342 free_extent_buffer(tmp);
1343 tmp = read_tree_block(root, blocknr,
1344 blocksize, gen);
1345 if (tmp)
1346 free_extent_buffer(tmp);
1347 goto again;
1348 } else {
1349 b = read_node_slot(root, b, slot);
1350 }
1351 }
1352 btrfs_tree_lock(b);
1353 unlock_up(p, level, lowest_unlock);
1354 } else {
1355 p->slots[level] = slot;
1356 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1357 sizeof(struct btrfs_item) + ins_len) {
1358 int sret = split_leaf(trans, root, key,
1359 p, ins_len, ret == 0);
1360 BUG_ON(sret > 0);
1361 if (sret)
1362 return sret;
1363 }
1364 unlock_up(p, level, lowest_unlock);
1365 return ret;
1366 }
1367 }
1368 return 1;
1369 }
1370
1371 /*
1372 * adjust the pointers going up the tree, starting at level
1373 * making sure the right key of each node is points to 'key'.
1374 * This is used after shifting pointers to the left, so it stops
1375 * fixing up pointers when a given leaf/node is not in slot 0 of the
1376 * higher levels
1377 *
1378 * If this fails to write a tree block, it returns -1, but continues
1379 * fixing up the blocks in ram so the tree is consistent.
1380 */
1381 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1382 struct btrfs_root *root, struct btrfs_path *path,
1383 struct btrfs_disk_key *key, int level)
1384 {
1385 int i;
1386 int ret = 0;
1387 struct extent_buffer *t;
1388
1389 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1390 int tslot = path->slots[i];
1391 if (!path->nodes[i])
1392 break;
1393 t = path->nodes[i];
1394 btrfs_set_node_key(t, key, tslot);
1395 if (!btrfs_tree_locked(path->nodes[i])) {
1396 int ii;
1397 printk("fixup without lock on level %d\n", btrfs_header_level(path->nodes[i]));
1398 for (ii = 0; ii < BTRFS_MAX_LEVEL; ii++) {
1399 printk("level %d slot %d\n", ii, path->slots[ii]);
1400 }
1401 }
1402 btrfs_mark_buffer_dirty(path->nodes[i]);
1403 if (tslot != 0)
1404 break;
1405 }
1406 return ret;
1407 }
1408
1409 /*
1410 * try to push data from one node into the next node left in the
1411 * tree.
1412 *
1413 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1414 * error, and > 0 if there was no room in the left hand block.
1415 */
1416 static int push_node_left(struct btrfs_trans_handle *trans,
1417 struct btrfs_root *root, struct extent_buffer *dst,
1418 struct extent_buffer *src, int empty)
1419 {
1420 int push_items = 0;
1421 int src_nritems;
1422 int dst_nritems;
1423 int ret = 0;
1424
1425 src_nritems = btrfs_header_nritems(src);
1426 dst_nritems = btrfs_header_nritems(dst);
1427 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1428 WARN_ON(btrfs_header_generation(src) != trans->transid);
1429 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1430
1431 if (!empty && src_nritems <= 8)
1432 return 1;
1433
1434 if (push_items <= 0) {
1435 return 1;
1436 }
1437
1438 if (empty) {
1439 push_items = min(src_nritems, push_items);
1440 if (push_items < src_nritems) {
1441 /* leave at least 8 pointers in the node if
1442 * we aren't going to empty it
1443 */
1444 if (src_nritems - push_items < 8) {
1445 if (push_items <= 8)
1446 return 1;
1447 push_items -= 8;
1448 }
1449 }
1450 } else
1451 push_items = min(src_nritems - 8, push_items);
1452
1453 copy_extent_buffer(dst, src,
1454 btrfs_node_key_ptr_offset(dst_nritems),
1455 btrfs_node_key_ptr_offset(0),
1456 push_items * sizeof(struct btrfs_key_ptr));
1457
1458 if (push_items < src_nritems) {
1459 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1460 btrfs_node_key_ptr_offset(push_items),
1461 (src_nritems - push_items) *
1462 sizeof(struct btrfs_key_ptr));
1463 }
1464 btrfs_set_header_nritems(src, src_nritems - push_items);
1465 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1466 btrfs_mark_buffer_dirty(src);
1467 btrfs_mark_buffer_dirty(dst);
1468 return ret;
1469 }
1470
1471 /*
1472 * try to push data from one node into the next node right in the
1473 * tree.
1474 *
1475 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1476 * error, and > 0 if there was no room in the right hand block.
1477 *
1478 * this will only push up to 1/2 the contents of the left node over
1479 */
1480 static int balance_node_right(struct btrfs_trans_handle *trans,
1481 struct btrfs_root *root,
1482 struct extent_buffer *dst,
1483 struct extent_buffer *src)
1484 {
1485 int push_items = 0;
1486 int max_push;
1487 int src_nritems;
1488 int dst_nritems;
1489 int ret = 0;
1490
1491 WARN_ON(btrfs_header_generation(src) != trans->transid);
1492 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1493
1494 src_nritems = btrfs_header_nritems(src);
1495 dst_nritems = btrfs_header_nritems(dst);
1496 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1497 if (push_items <= 0) {
1498 return 1;
1499 }
1500
1501 if (src_nritems < 4) {
1502 return 1;
1503 }
1504
1505 max_push = src_nritems / 2 + 1;
1506 /* don't try to empty the node */
1507 if (max_push >= src_nritems) {
1508 return 1;
1509 }
1510
1511 if (max_push < push_items)
1512 push_items = max_push;
1513
1514 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1515 btrfs_node_key_ptr_offset(0),
1516 (dst_nritems) *
1517 sizeof(struct btrfs_key_ptr));
1518
1519 copy_extent_buffer(dst, src,
1520 btrfs_node_key_ptr_offset(0),
1521 btrfs_node_key_ptr_offset(src_nritems - push_items),
1522 push_items * sizeof(struct btrfs_key_ptr));
1523
1524 btrfs_set_header_nritems(src, src_nritems - push_items);
1525 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1526
1527 btrfs_mark_buffer_dirty(src);
1528 btrfs_mark_buffer_dirty(dst);
1529 return ret;
1530 }
1531
1532 /*
1533 * helper function to insert a new root level in the tree.
1534 * A new node is allocated, and a single item is inserted to
1535 * point to the existing root
1536 *
1537 * returns zero on success or < 0 on failure.
1538 */
1539 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1540 struct btrfs_root *root,
1541 struct btrfs_path *path, int level)
1542 {
1543 u64 root_gen;
1544 u64 lower_gen;
1545 struct extent_buffer *lower;
1546 struct extent_buffer *c;
1547 struct extent_buffer *old;
1548 struct btrfs_disk_key lower_key;
1549
1550 BUG_ON(path->nodes[level]);
1551 BUG_ON(path->nodes[level-1] != root->node);
1552
1553 if (root->ref_cows)
1554 root_gen = trans->transid;
1555 else
1556 root_gen = 0;
1557
1558 lower = path->nodes[level-1];
1559 if (level == 1)
1560 btrfs_item_key(lower, &lower_key, 0);
1561 else
1562 btrfs_node_key(lower, &lower_key, 0);
1563
1564 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1565 root->root_key.objectid,
1566 root_gen, lower_key.objectid, level,
1567 root->node->start, 0);
1568 if (IS_ERR(c))
1569 return PTR_ERR(c);
1570
1571 memset_extent_buffer(c, 0, 0, root->nodesize);
1572 btrfs_set_header_nritems(c, 1);
1573 btrfs_set_header_level(c, level);
1574 btrfs_set_header_bytenr(c, c->start);
1575 btrfs_set_header_generation(c, trans->transid);
1576 btrfs_set_header_owner(c, root->root_key.objectid);
1577
1578 write_extent_buffer(c, root->fs_info->fsid,
1579 (unsigned long)btrfs_header_fsid(c),
1580 BTRFS_FSID_SIZE);
1581
1582 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1583 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1584 BTRFS_UUID_SIZE);
1585
1586 btrfs_set_node_key(c, &lower_key, 0);
1587 btrfs_set_node_blockptr(c, 0, lower->start);
1588 lower_gen = btrfs_header_generation(lower);
1589 WARN_ON(lower_gen == 0);
1590
1591 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1592
1593 btrfs_mark_buffer_dirty(c);
1594
1595 spin_lock(&root->node_lock);
1596 old = root->node;
1597 root->node = c;
1598 spin_unlock(&root->node_lock);
1599
1600 /* the super has an extra ref to root->node */
1601 free_extent_buffer(old);
1602
1603 add_root_to_dirty_list(root);
1604 extent_buffer_get(c);
1605 path->nodes[level] = c;
1606 path->locks[level] = 1;
1607 path->slots[level] = 0;
1608
1609 if (root->ref_cows && lower_gen != trans->transid) {
1610 struct btrfs_path *back_path = btrfs_alloc_path();
1611 int ret;
1612 mutex_lock(&root->fs_info->alloc_mutex);
1613 ret = btrfs_insert_extent_backref(trans,
1614 root->fs_info->extent_root,
1615 path, lower->start,
1616 root->root_key.objectid,
1617 trans->transid, 0, 0);
1618 BUG_ON(ret);
1619 mutex_unlock(&root->fs_info->alloc_mutex);
1620 btrfs_free_path(back_path);
1621 }
1622 return 0;
1623 }
1624
1625 /*
1626 * worker function to insert a single pointer in a node.
1627 * the node should have enough room for the pointer already
1628 *
1629 * slot and level indicate where you want the key to go, and
1630 * blocknr is the block the key points to.
1631 *
1632 * returns zero on success and < 0 on any error
1633 */
1634 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1635 *root, struct btrfs_path *path, struct btrfs_disk_key
1636 *key, u64 bytenr, int slot, int level)
1637 {
1638 struct extent_buffer *lower;
1639 int nritems;
1640
1641 BUG_ON(!path->nodes[level]);
1642 lower = path->nodes[level];
1643 nritems = btrfs_header_nritems(lower);
1644 if (slot > nritems)
1645 BUG();
1646 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1647 BUG();
1648 if (slot != nritems) {
1649 memmove_extent_buffer(lower,
1650 btrfs_node_key_ptr_offset(slot + 1),
1651 btrfs_node_key_ptr_offset(slot),
1652 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1653 }
1654 btrfs_set_node_key(lower, key, slot);
1655 btrfs_set_node_blockptr(lower, slot, bytenr);
1656 WARN_ON(trans->transid == 0);
1657 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1658 btrfs_set_header_nritems(lower, nritems + 1);
1659 btrfs_mark_buffer_dirty(lower);
1660 return 0;
1661 }
1662
1663 /*
1664 * split the node at the specified level in path in two.
1665 * The path is corrected to point to the appropriate node after the split
1666 *
1667 * Before splitting this tries to make some room in the node by pushing
1668 * left and right, if either one works, it returns right away.
1669 *
1670 * returns 0 on success and < 0 on failure
1671 */
1672 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1673 *root, struct btrfs_path *path, int level)
1674 {
1675 u64 root_gen;
1676 struct extent_buffer *c;
1677 struct extent_buffer *split;
1678 struct btrfs_disk_key disk_key;
1679 int mid;
1680 int ret;
1681 int wret;
1682 u32 c_nritems;
1683
1684 c = path->nodes[level];
1685 WARN_ON(btrfs_header_generation(c) != trans->transid);
1686 if (c == root->node) {
1687 /* trying to split the root, lets make a new one */
1688 ret = insert_new_root(trans, root, path, level + 1);
1689 if (ret)
1690 return ret;
1691 } else {
1692 ret = push_nodes_for_insert(trans, root, path, level);
1693 c = path->nodes[level];
1694 if (!ret && btrfs_header_nritems(c) <
1695 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1696 return 0;
1697 if (ret < 0)
1698 return ret;
1699 }
1700
1701 c_nritems = btrfs_header_nritems(c);
1702 if (root->ref_cows)
1703 root_gen = trans->transid;
1704 else
1705 root_gen = 0;
1706
1707 btrfs_node_key(c, &disk_key, 0);
1708 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1709 root->root_key.objectid,
1710 root_gen,
1711 btrfs_disk_key_objectid(&disk_key),
1712 level, c->start, 0);
1713 if (IS_ERR(split))
1714 return PTR_ERR(split);
1715
1716 btrfs_set_header_flags(split, btrfs_header_flags(c));
1717 btrfs_set_header_level(split, btrfs_header_level(c));
1718 btrfs_set_header_bytenr(split, split->start);
1719 btrfs_set_header_generation(split, trans->transid);
1720 btrfs_set_header_owner(split, root->root_key.objectid);
1721 btrfs_set_header_flags(split, 0);
1722 write_extent_buffer(split, root->fs_info->fsid,
1723 (unsigned long)btrfs_header_fsid(split),
1724 BTRFS_FSID_SIZE);
1725 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1726 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1727 BTRFS_UUID_SIZE);
1728
1729 mid = (c_nritems + 1) / 2;
1730
1731 copy_extent_buffer(split, c,
1732 btrfs_node_key_ptr_offset(0),
1733 btrfs_node_key_ptr_offset(mid),
1734 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1735 btrfs_set_header_nritems(split, c_nritems - mid);
1736 btrfs_set_header_nritems(c, mid);
1737 ret = 0;
1738
1739 btrfs_mark_buffer_dirty(c);
1740 btrfs_mark_buffer_dirty(split);
1741
1742 btrfs_node_key(split, &disk_key, 0);
1743 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1744 path->slots[level + 1] + 1,
1745 level + 1);
1746 if (wret)
1747 ret = wret;
1748
1749 if (path->slots[level] >= mid) {
1750 path->slots[level] -= mid;
1751 btrfs_tree_unlock(c);
1752 free_extent_buffer(c);
1753 path->nodes[level] = split;
1754 path->slots[level + 1] += 1;
1755 } else {
1756 btrfs_tree_unlock(split);
1757 free_extent_buffer(split);
1758 }
1759 return ret;
1760 }
1761
1762 /*
1763 * how many bytes are required to store the items in a leaf. start
1764 * and nr indicate which items in the leaf to check. This totals up the
1765 * space used both by the item structs and the item data
1766 */
1767 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1768 {
1769 int data_len;
1770 int nritems = btrfs_header_nritems(l);
1771 int end = min(nritems, start + nr) - 1;
1772
1773 if (!nr)
1774 return 0;
1775 data_len = btrfs_item_end_nr(l, start);
1776 data_len = data_len - btrfs_item_offset_nr(l, end);
1777 data_len += sizeof(struct btrfs_item) * nr;
1778 WARN_ON(data_len < 0);
1779 return data_len;
1780 }
1781
1782 /*
1783 * The space between the end of the leaf items and
1784 * the start of the leaf data. IOW, how much room
1785 * the leaf has left for both items and data
1786 */
1787 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1788 {
1789 int nritems = btrfs_header_nritems(leaf);
1790 int ret;
1791 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1792 if (ret < 0) {
1793 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1794 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1795 leaf_space_used(leaf, 0, nritems), nritems);
1796 }
1797 return ret;
1798 }
1799
1800 /*
1801 * push some data in the path leaf to the right, trying to free up at
1802 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1803 *
1804 * returns 1 if the push failed because the other node didn't have enough
1805 * room, 0 if everything worked out and < 0 if there were major errors.
1806 */
1807 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1808 *root, struct btrfs_path *path, int data_size,
1809 int empty)
1810 {
1811 struct extent_buffer *left = path->nodes[0];
1812 struct extent_buffer *right;
1813 struct extent_buffer *upper;
1814 struct btrfs_disk_key disk_key;
1815 int slot;
1816 u32 i;
1817 int free_space;
1818 int push_space = 0;
1819 int push_items = 0;
1820 struct btrfs_item *item;
1821 u32 left_nritems;
1822 u32 nr;
1823 u32 right_nritems;
1824 u32 data_end;
1825 u32 this_item_size;
1826 int ret;
1827
1828 slot = path->slots[1];
1829 if (!path->nodes[1]) {
1830 return 1;
1831 }
1832 upper = path->nodes[1];
1833 if (slot >= btrfs_header_nritems(upper) - 1)
1834 return 1;
1835
1836 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
1837
1838 right = read_node_slot(root, upper, slot + 1);
1839 btrfs_tree_lock(right);
1840 free_space = btrfs_leaf_free_space(root, right);
1841 if (free_space < data_size + sizeof(struct btrfs_item))
1842 goto out_unlock;
1843
1844 /* cow and double check */
1845 ret = btrfs_cow_block(trans, root, right, upper,
1846 slot + 1, &right);
1847 if (ret)
1848 goto out_unlock;
1849
1850 free_space = btrfs_leaf_free_space(root, right);
1851 if (free_space < data_size + sizeof(struct btrfs_item))
1852 goto out_unlock;
1853
1854 left_nritems = btrfs_header_nritems(left);
1855 if (left_nritems == 0)
1856 goto out_unlock;
1857
1858 if (empty)
1859 nr = 0;
1860 else
1861 nr = 1;
1862
1863 i = left_nritems - 1;
1864 while (i >= nr) {
1865 item = btrfs_item_nr(left, i);
1866
1867 if (path->slots[0] == i)
1868 push_space += data_size + sizeof(*item);
1869
1870 if (!left->map_token) {
1871 map_extent_buffer(left, (unsigned long)item,
1872 sizeof(struct btrfs_item),
1873 &left->map_token, &left->kaddr,
1874 &left->map_start, &left->map_len,
1875 KM_USER1);
1876 }
1877
1878 this_item_size = btrfs_item_size(left, item);
1879 if (this_item_size + sizeof(*item) + push_space > free_space)
1880 break;
1881 push_items++;
1882 push_space += this_item_size + sizeof(*item);
1883 if (i == 0)
1884 break;
1885 i--;
1886 }
1887 if (left->map_token) {
1888 unmap_extent_buffer(left, left->map_token, KM_USER1);
1889 left->map_token = NULL;
1890 }
1891
1892 if (push_items == 0)
1893 goto out_unlock;
1894
1895 if (!empty && push_items == left_nritems)
1896 WARN_ON(1);
1897
1898 /* push left to right */
1899 right_nritems = btrfs_header_nritems(right);
1900
1901 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1902 push_space -= leaf_data_end(root, left);
1903
1904 /* make room in the right data area */
1905 data_end = leaf_data_end(root, right);
1906 memmove_extent_buffer(right,
1907 btrfs_leaf_data(right) + data_end - push_space,
1908 btrfs_leaf_data(right) + data_end,
1909 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1910
1911 /* copy from the left data area */
1912 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1913 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1914 btrfs_leaf_data(left) + leaf_data_end(root, left),
1915 push_space);
1916
1917 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1918 btrfs_item_nr_offset(0),
1919 right_nritems * sizeof(struct btrfs_item));
1920
1921 /* copy the items from left to right */
1922 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1923 btrfs_item_nr_offset(left_nritems - push_items),
1924 push_items * sizeof(struct btrfs_item));
1925
1926 /* update the item pointers */
1927 right_nritems += push_items;
1928 btrfs_set_header_nritems(right, right_nritems);
1929 push_space = BTRFS_LEAF_DATA_SIZE(root);
1930 for (i = 0; i < right_nritems; i++) {
1931 item = btrfs_item_nr(right, i);
1932 if (!right->map_token) {
1933 map_extent_buffer(right, (unsigned long)item,
1934 sizeof(struct btrfs_item),
1935 &right->map_token, &right->kaddr,
1936 &right->map_start, &right->map_len,
1937 KM_USER1);
1938 }
1939 push_space -= btrfs_item_size(right, item);
1940 btrfs_set_item_offset(right, item, push_space);
1941 }
1942
1943 if (right->map_token) {
1944 unmap_extent_buffer(right, right->map_token, KM_USER1);
1945 right->map_token = NULL;
1946 }
1947 left_nritems -= push_items;
1948 btrfs_set_header_nritems(left, left_nritems);
1949
1950 if (left_nritems)
1951 btrfs_mark_buffer_dirty(left);
1952 btrfs_mark_buffer_dirty(right);
1953
1954 btrfs_item_key(right, &disk_key, 0);
1955 btrfs_set_node_key(upper, &disk_key, slot + 1);
1956 btrfs_mark_buffer_dirty(upper);
1957
1958 /* then fixup the leaf pointer in the path */
1959 if (path->slots[0] >= left_nritems) {
1960 path->slots[0] -= left_nritems;
1961 if (btrfs_header_nritems(path->nodes[0]) == 0)
1962 clean_tree_block(trans, root, path->nodes[0]);
1963 btrfs_tree_unlock(path->nodes[0]);
1964 free_extent_buffer(path->nodes[0]);
1965 path->nodes[0] = right;
1966 path->slots[1] += 1;
1967 } else {
1968 btrfs_tree_unlock(right);
1969 free_extent_buffer(right);
1970 }
1971 return 0;
1972
1973 out_unlock:
1974 btrfs_tree_unlock(right);
1975 free_extent_buffer(right);
1976 return 1;
1977 }
1978
1979 /*
1980 * push some data in the path leaf to the left, trying to free up at
1981 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1982 */
1983 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1984 *root, struct btrfs_path *path, int data_size,
1985 int empty)
1986 {
1987 struct btrfs_disk_key disk_key;
1988 struct extent_buffer *right = path->nodes[0];
1989 struct extent_buffer *left;
1990 int slot;
1991 int i;
1992 int free_space;
1993 int push_space = 0;
1994 int push_items = 0;
1995 struct btrfs_item *item;
1996 u32 old_left_nritems;
1997 u32 right_nritems;
1998 u32 nr;
1999 int ret = 0;
2000 int wret;
2001 u32 this_item_size;
2002 u32 old_left_item_size;
2003
2004 slot = path->slots[1];
2005 if (slot == 0)
2006 return 1;
2007 if (!path->nodes[1])
2008 return 1;
2009
2010 right_nritems = btrfs_header_nritems(right);
2011 if (right_nritems == 0) {
2012 return 1;
2013 }
2014
2015 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2016
2017 left = read_node_slot(root, path->nodes[1], slot - 1);
2018 btrfs_tree_lock(left);
2019 free_space = btrfs_leaf_free_space(root, left);
2020 if (free_space < data_size + sizeof(struct btrfs_item)) {
2021 ret = 1;
2022 goto out;
2023 }
2024
2025 /* cow and double check */
2026 ret = btrfs_cow_block(trans, root, left,
2027 path->nodes[1], slot - 1, &left);
2028 if (ret) {
2029 /* we hit -ENOSPC, but it isn't fatal here */
2030 ret = 1;
2031 goto out;
2032 }
2033
2034 free_space = btrfs_leaf_free_space(root, left);
2035 if (free_space < data_size + sizeof(struct btrfs_item)) {
2036 ret = 1;
2037 goto out;
2038 }
2039
2040 if (empty)
2041 nr = right_nritems;
2042 else
2043 nr = right_nritems - 1;
2044
2045 for (i = 0; i < nr; i++) {
2046 item = btrfs_item_nr(right, i);
2047 if (!right->map_token) {
2048 map_extent_buffer(right, (unsigned long)item,
2049 sizeof(struct btrfs_item),
2050 &right->map_token, &right->kaddr,
2051 &right->map_start, &right->map_len,
2052 KM_USER1);
2053 }
2054
2055 if (path->slots[0] == i)
2056 push_space += data_size + sizeof(*item);
2057
2058 this_item_size = btrfs_item_size(right, item);
2059 if (this_item_size + sizeof(*item) + push_space > free_space)
2060 break;
2061
2062 push_items++;
2063 push_space += this_item_size + sizeof(*item);
2064 }
2065
2066 if (right->map_token) {
2067 unmap_extent_buffer(right, right->map_token, KM_USER1);
2068 right->map_token = NULL;
2069 }
2070
2071 if (push_items == 0) {
2072 ret = 1;
2073 goto out;
2074 }
2075 if (!empty && push_items == btrfs_header_nritems(right))
2076 WARN_ON(1);
2077
2078 /* push data from right to left */
2079 copy_extent_buffer(left, right,
2080 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2081 btrfs_item_nr_offset(0),
2082 push_items * sizeof(struct btrfs_item));
2083
2084 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2085 btrfs_item_offset_nr(right, push_items -1);
2086
2087 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2088 leaf_data_end(root, left) - push_space,
2089 btrfs_leaf_data(right) +
2090 btrfs_item_offset_nr(right, push_items - 1),
2091 push_space);
2092 old_left_nritems = btrfs_header_nritems(left);
2093 BUG_ON(old_left_nritems < 0);
2094
2095 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2096 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2097 u32 ioff;
2098
2099 item = btrfs_item_nr(left, i);
2100 if (!left->map_token) {
2101 map_extent_buffer(left, (unsigned long)item,
2102 sizeof(struct btrfs_item),
2103 &left->map_token, &left->kaddr,
2104 &left->map_start, &left->map_len,
2105 KM_USER1);
2106 }
2107
2108 ioff = btrfs_item_offset(left, item);
2109 btrfs_set_item_offset(left, item,
2110 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2111 }
2112 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2113 if (left->map_token) {
2114 unmap_extent_buffer(left, left->map_token, KM_USER1);
2115 left->map_token = NULL;
2116 }
2117
2118 /* fixup right node */
2119 if (push_items > right_nritems) {
2120 printk("push items %d nr %u\n", push_items, right_nritems);
2121 WARN_ON(1);
2122 }
2123
2124 if (push_items < right_nritems) {
2125 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2126 leaf_data_end(root, right);
2127 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2128 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2129 btrfs_leaf_data(right) +
2130 leaf_data_end(root, right), push_space);
2131
2132 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2133 btrfs_item_nr_offset(push_items),
2134 (btrfs_header_nritems(right) - push_items) *
2135 sizeof(struct btrfs_item));
2136 }
2137 right_nritems -= push_items;
2138 btrfs_set_header_nritems(right, right_nritems);
2139 push_space = BTRFS_LEAF_DATA_SIZE(root);
2140 for (i = 0; i < right_nritems; i++) {
2141 item = btrfs_item_nr(right, i);
2142
2143 if (!right->map_token) {
2144 map_extent_buffer(right, (unsigned long)item,
2145 sizeof(struct btrfs_item),
2146 &right->map_token, &right->kaddr,
2147 &right->map_start, &right->map_len,
2148 KM_USER1);
2149 }
2150
2151 push_space = push_space - btrfs_item_size(right, item);
2152 btrfs_set_item_offset(right, item, push_space);
2153 }
2154 if (right->map_token) {
2155 unmap_extent_buffer(right, right->map_token, KM_USER1);
2156 right->map_token = NULL;
2157 }
2158
2159 btrfs_mark_buffer_dirty(left);
2160 if (right_nritems)
2161 btrfs_mark_buffer_dirty(right);
2162
2163 btrfs_item_key(right, &disk_key, 0);
2164 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2165 if (wret)
2166 ret = wret;
2167
2168 /* then fixup the leaf pointer in the path */
2169 if (path->slots[0] < push_items) {
2170 path->slots[0] += old_left_nritems;
2171 if (btrfs_header_nritems(path->nodes[0]) == 0)
2172 clean_tree_block(trans, root, path->nodes[0]);
2173 btrfs_tree_unlock(path->nodes[0]);
2174 free_extent_buffer(path->nodes[0]);
2175 path->nodes[0] = left;
2176 path->slots[1] -= 1;
2177 } else {
2178 btrfs_tree_unlock(left);
2179 free_extent_buffer(left);
2180 path->slots[0] -= push_items;
2181 }
2182 BUG_ON(path->slots[0] < 0);
2183 return ret;
2184 out:
2185 btrfs_tree_unlock(left);
2186 free_extent_buffer(left);
2187 return ret;
2188 }
2189
2190 /*
2191 * split the path's leaf in two, making sure there is at least data_size
2192 * available for the resulting leaf level of the path.
2193 *
2194 * returns 0 if all went well and < 0 on failure.
2195 */
2196 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
2197 *root, struct btrfs_key *ins_key,
2198 struct btrfs_path *path, int data_size, int extend)
2199 {
2200 u64 root_gen;
2201 struct extent_buffer *l;
2202 u32 nritems;
2203 int mid;
2204 int slot;
2205 struct extent_buffer *right;
2206 int space_needed = data_size + sizeof(struct btrfs_item);
2207 int data_copy_size;
2208 int rt_data_off;
2209 int i;
2210 int ret = 0;
2211 int wret;
2212 int double_split;
2213 int num_doubles = 0;
2214 struct btrfs_disk_key disk_key;
2215
2216 if (extend)
2217 space_needed = data_size;
2218
2219 if (root->ref_cows)
2220 root_gen = trans->transid;
2221 else
2222 root_gen = 0;
2223
2224 /* first try to make some room by pushing left and right */
2225 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2226 wret = push_leaf_right(trans, root, path, data_size, 0);
2227 if (wret < 0) {
2228 return wret;
2229 }
2230 if (wret) {
2231 wret = push_leaf_left(trans, root, path, data_size, 0);
2232 if (wret < 0)
2233 return wret;
2234 }
2235 l = path->nodes[0];
2236
2237 /* did the pushes work? */
2238 if (btrfs_leaf_free_space(root, l) >= space_needed)
2239 return 0;
2240 }
2241
2242 if (!path->nodes[1]) {
2243 ret = insert_new_root(trans, root, path, 1);
2244 if (ret)
2245 return ret;
2246 }
2247 again:
2248 double_split = 0;
2249 l = path->nodes[0];
2250 slot = path->slots[0];
2251 nritems = btrfs_header_nritems(l);
2252 mid = (nritems + 1)/ 2;
2253
2254 btrfs_item_key(l, &disk_key, 0);
2255
2256 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2257 root->root_key.objectid,
2258 root_gen, disk_key.objectid, 0,
2259 l->start, 0);
2260 if (IS_ERR(right)) {
2261 BUG_ON(1);
2262 return PTR_ERR(right);
2263 }
2264
2265 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2266 btrfs_set_header_bytenr(right, right->start);
2267 btrfs_set_header_generation(right, trans->transid);
2268 btrfs_set_header_owner(right, root->root_key.objectid);
2269 btrfs_set_header_level(right, 0);
2270 write_extent_buffer(right, root->fs_info->fsid,
2271 (unsigned long)btrfs_header_fsid(right),
2272 BTRFS_FSID_SIZE);
2273
2274 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2275 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2276 BTRFS_UUID_SIZE);
2277 if (mid <= slot) {
2278 if (nritems == 1 ||
2279 leaf_space_used(l, mid, nritems - mid) + space_needed >
2280 BTRFS_LEAF_DATA_SIZE(root)) {
2281 if (slot >= nritems) {
2282 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2283 btrfs_set_header_nritems(right, 0);
2284 wret = insert_ptr(trans, root, path,
2285 &disk_key, right->start,
2286 path->slots[1] + 1, 1);
2287 if (wret)
2288 ret = wret;
2289
2290 btrfs_tree_unlock(path->nodes[0]);
2291 free_extent_buffer(path->nodes[0]);
2292 path->nodes[0] = right;
2293 path->slots[0] = 0;
2294 path->slots[1] += 1;
2295 btrfs_mark_buffer_dirty(right);
2296 return ret;
2297 }
2298 mid = slot;
2299 if (mid != nritems &&
2300 leaf_space_used(l, mid, nritems - mid) +
2301 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2302 double_split = 1;
2303 }
2304 }
2305 } else {
2306 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2307 BTRFS_LEAF_DATA_SIZE(root)) {
2308 if (!extend && slot == 0) {
2309 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2310 btrfs_set_header_nritems(right, 0);
2311 wret = insert_ptr(trans, root, path,
2312 &disk_key,
2313 right->start,
2314 path->slots[1], 1);
2315 if (wret)
2316 ret = wret;
2317 btrfs_tree_unlock(path->nodes[0]);
2318 free_extent_buffer(path->nodes[0]);
2319 path->nodes[0] = right;
2320 path->slots[0] = 0;
2321 if (path->slots[1] == 0) {
2322 wret = fixup_low_keys(trans, root,
2323 path, &disk_key, 1);
2324 if (wret)
2325 ret = wret;
2326 }
2327 btrfs_mark_buffer_dirty(right);
2328 return ret;
2329 } else if (extend && slot == 0) {
2330 mid = 1;
2331 } else {
2332 mid = slot;
2333 if (mid != nritems &&
2334 leaf_space_used(l, mid, nritems - mid) +
2335 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2336 double_split = 1;
2337 }
2338 }
2339 }
2340 }
2341 nritems = nritems - mid;
2342 btrfs_set_header_nritems(right, nritems);
2343 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2344
2345 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2346 btrfs_item_nr_offset(mid),
2347 nritems * sizeof(struct btrfs_item));
2348
2349 copy_extent_buffer(right, l,
2350 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2351 data_copy_size, btrfs_leaf_data(l) +
2352 leaf_data_end(root, l), data_copy_size);
2353
2354 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2355 btrfs_item_end_nr(l, mid);
2356
2357 for (i = 0; i < nritems; i++) {
2358 struct btrfs_item *item = btrfs_item_nr(right, i);
2359 u32 ioff;
2360
2361 if (!right->map_token) {
2362 map_extent_buffer(right, (unsigned long)item,
2363 sizeof(struct btrfs_item),
2364 &right->map_token, &right->kaddr,
2365 &right->map_start, &right->map_len,
2366 KM_USER1);
2367 }
2368
2369 ioff = btrfs_item_offset(right, item);
2370 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2371 }
2372
2373 if (right->map_token) {
2374 unmap_extent_buffer(right, right->map_token, KM_USER1);
2375 right->map_token = NULL;
2376 }
2377
2378 btrfs_set_header_nritems(l, mid);
2379 ret = 0;
2380 btrfs_item_key(right, &disk_key, 0);
2381 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2382 path->slots[1] + 1, 1);
2383 if (wret)
2384 ret = wret;
2385
2386 btrfs_mark_buffer_dirty(right);
2387 btrfs_mark_buffer_dirty(l);
2388 BUG_ON(path->slots[0] != slot);
2389
2390 if (mid <= slot) {
2391 btrfs_tree_unlock(path->nodes[0]);
2392 free_extent_buffer(path->nodes[0]);
2393 path->nodes[0] = right;
2394 path->slots[0] -= mid;
2395 path->slots[1] += 1;
2396 } else {
2397 btrfs_tree_unlock(right);
2398 free_extent_buffer(right);
2399 }
2400
2401 BUG_ON(path->slots[0] < 0);
2402
2403 if (double_split) {
2404 BUG_ON(num_doubles != 0);
2405 num_doubles++;
2406 goto again;
2407 }
2408 return ret;
2409 }
2410
2411 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2412 struct btrfs_root *root,
2413 struct btrfs_path *path,
2414 u32 new_size, int from_end)
2415 {
2416 int ret = 0;
2417 int slot;
2418 int slot_orig;
2419 struct extent_buffer *leaf;
2420 struct btrfs_item *item;
2421 u32 nritems;
2422 unsigned int data_end;
2423 unsigned int old_data_start;
2424 unsigned int old_size;
2425 unsigned int size_diff;
2426 int i;
2427
2428 slot_orig = path->slots[0];
2429 leaf = path->nodes[0];
2430 slot = path->slots[0];
2431
2432 old_size = btrfs_item_size_nr(leaf, slot);
2433 if (old_size == new_size)
2434 return 0;
2435
2436 nritems = btrfs_header_nritems(leaf);
2437 data_end = leaf_data_end(root, leaf);
2438
2439 old_data_start = btrfs_item_offset_nr(leaf, slot);
2440
2441 size_diff = old_size - new_size;
2442
2443 BUG_ON(slot < 0);
2444 BUG_ON(slot >= nritems);
2445
2446 /*
2447 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2448 */
2449 /* first correct the data pointers */
2450 for (i = slot; i < nritems; i++) {
2451 u32 ioff;
2452 item = btrfs_item_nr(leaf, i);
2453
2454 if (!leaf->map_token) {
2455 map_extent_buffer(leaf, (unsigned long)item,
2456 sizeof(struct btrfs_item),
2457 &leaf->map_token, &leaf->kaddr,
2458 &leaf->map_start, &leaf->map_len,
2459 KM_USER1);
2460 }
2461
2462 ioff = btrfs_item_offset(leaf, item);
2463 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2464 }
2465
2466 if (leaf->map_token) {
2467 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2468 leaf->map_token = NULL;
2469 }
2470
2471 /* shift the data */
2472 if (from_end) {
2473 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2474 data_end + size_diff, btrfs_leaf_data(leaf) +
2475 data_end, old_data_start + new_size - data_end);
2476 } else {
2477 struct btrfs_disk_key disk_key;
2478 u64 offset;
2479
2480 btrfs_item_key(leaf, &disk_key, slot);
2481
2482 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2483 unsigned long ptr;
2484 struct btrfs_file_extent_item *fi;
2485
2486 fi = btrfs_item_ptr(leaf, slot,
2487 struct btrfs_file_extent_item);
2488 fi = (struct btrfs_file_extent_item *)(
2489 (unsigned long)fi - size_diff);
2490
2491 if (btrfs_file_extent_type(leaf, fi) ==
2492 BTRFS_FILE_EXTENT_INLINE) {
2493 ptr = btrfs_item_ptr_offset(leaf, slot);
2494 memmove_extent_buffer(leaf, ptr,
2495 (unsigned long)fi,
2496 offsetof(struct btrfs_file_extent_item,
2497 disk_bytenr));
2498 }
2499 }
2500
2501 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2502 data_end + size_diff, btrfs_leaf_data(leaf) +
2503 data_end, old_data_start - data_end);
2504
2505 offset = btrfs_disk_key_offset(&disk_key);
2506 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2507 btrfs_set_item_key(leaf, &disk_key, slot);
2508 if (slot == 0)
2509 fixup_low_keys(trans, root, path, &disk_key, 1);
2510 }
2511
2512 item = btrfs_item_nr(leaf, slot);
2513 btrfs_set_item_size(leaf, item, new_size);
2514 btrfs_mark_buffer_dirty(leaf);
2515
2516 ret = 0;
2517 if (btrfs_leaf_free_space(root, leaf) < 0) {
2518 btrfs_print_leaf(root, leaf);
2519 BUG();
2520 }
2521 return ret;
2522 }
2523
2524 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2525 struct btrfs_root *root, struct btrfs_path *path,
2526 u32 data_size)
2527 {
2528 int ret = 0;
2529 int slot;
2530 int slot_orig;
2531 struct extent_buffer *leaf;
2532 struct btrfs_item *item;
2533 u32 nritems;
2534 unsigned int data_end;
2535 unsigned int old_data;
2536 unsigned int old_size;
2537 int i;
2538
2539 slot_orig = path->slots[0];
2540 leaf = path->nodes[0];
2541
2542 nritems = btrfs_header_nritems(leaf);
2543 data_end = leaf_data_end(root, leaf);
2544
2545 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2546 btrfs_print_leaf(root, leaf);
2547 BUG();
2548 }
2549 slot = path->slots[0];
2550 old_data = btrfs_item_end_nr(leaf, slot);
2551
2552 BUG_ON(slot < 0);
2553 if (slot >= nritems) {
2554 btrfs_print_leaf(root, leaf);
2555 printk("slot %d too large, nritems %d\n", slot, nritems);
2556 BUG_ON(1);
2557 }
2558
2559 /*
2560 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2561 */
2562 /* first correct the data pointers */
2563 for (i = slot; i < nritems; i++) {
2564 u32 ioff;
2565 item = btrfs_item_nr(leaf, i);
2566
2567 if (!leaf->map_token) {
2568 map_extent_buffer(leaf, (unsigned long)item,
2569 sizeof(struct btrfs_item),
2570 &leaf->map_token, &leaf->kaddr,
2571 &leaf->map_start, &leaf->map_len,
2572 KM_USER1);
2573 }
2574 ioff = btrfs_item_offset(leaf, item);
2575 btrfs_set_item_offset(leaf, item, ioff - data_size);
2576 }
2577
2578 if (leaf->map_token) {
2579 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2580 leaf->map_token = NULL;
2581 }
2582
2583 /* shift the data */
2584 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2585 data_end - data_size, btrfs_leaf_data(leaf) +
2586 data_end, old_data - data_end);
2587
2588 data_end = old_data;
2589 old_size = btrfs_item_size_nr(leaf, slot);
2590 item = btrfs_item_nr(leaf, slot);
2591 btrfs_set_item_size(leaf, item, old_size + data_size);
2592 btrfs_mark_buffer_dirty(leaf);
2593
2594 ret = 0;
2595 if (btrfs_leaf_free_space(root, leaf) < 0) {
2596 btrfs_print_leaf(root, leaf);
2597 BUG();
2598 }
2599 return ret;
2600 }
2601
2602 /*
2603 * Given a key and some data, insert an item into the tree.
2604 * This does all the path init required, making room in the tree if needed.
2605 */
2606 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2607 struct btrfs_root *root,
2608 struct btrfs_path *path,
2609 struct btrfs_key *cpu_key, u32 *data_size,
2610 int nr)
2611 {
2612 struct extent_buffer *leaf;
2613 struct btrfs_item *item;
2614 int ret = 0;
2615 int slot;
2616 int slot_orig;
2617 int i;
2618 u32 nritems;
2619 u32 total_size = 0;
2620 u32 total_data = 0;
2621 unsigned int data_end;
2622 struct btrfs_disk_key disk_key;
2623
2624 for (i = 0; i < nr; i++) {
2625 total_data += data_size[i];
2626 }
2627
2628 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2629 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2630 if (ret == 0) {
2631 return -EEXIST;
2632 }
2633 if (ret < 0)
2634 goto out;
2635
2636 slot_orig = path->slots[0];
2637 leaf = path->nodes[0];
2638
2639 nritems = btrfs_header_nritems(leaf);
2640 data_end = leaf_data_end(root, leaf);
2641
2642 if (btrfs_leaf_free_space(root, leaf) <
2643 sizeof(struct btrfs_item) + total_size) {
2644 btrfs_print_leaf(root, leaf);
2645 printk("not enough freespace need %u have %d\n",
2646 total_size, btrfs_leaf_free_space(root, leaf));
2647 BUG();
2648 }
2649
2650 slot = path->slots[0];
2651 BUG_ON(slot < 0);
2652
2653 if (slot != nritems) {
2654 int i;
2655 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2656
2657 if (old_data < data_end) {
2658 btrfs_print_leaf(root, leaf);
2659 printk("slot %d old_data %d data_end %d\n",
2660 slot, old_data, data_end);
2661 BUG_ON(1);
2662 }
2663 /*
2664 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2665 */
2666 /* first correct the data pointers */
2667 WARN_ON(leaf->map_token);
2668 for (i = slot; i < nritems; i++) {
2669 u32 ioff;
2670
2671 item = btrfs_item_nr(leaf, i);
2672 if (!leaf->map_token) {
2673 map_extent_buffer(leaf, (unsigned long)item,
2674 sizeof(struct btrfs_item),
2675 &leaf->map_token, &leaf->kaddr,
2676 &leaf->map_start, &leaf->map_len,
2677 KM_USER1);
2678 }
2679
2680 ioff = btrfs_item_offset(leaf, item);
2681 btrfs_set_item_offset(leaf, item, ioff - total_data);
2682 }
2683 if (leaf->map_token) {
2684 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2685 leaf->map_token = NULL;
2686 }
2687
2688 /* shift the items */
2689 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2690 btrfs_item_nr_offset(slot),
2691 (nritems - slot) * sizeof(struct btrfs_item));
2692
2693 /* shift the data */
2694 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2695 data_end - total_data, btrfs_leaf_data(leaf) +
2696 data_end, old_data - data_end);
2697 data_end = old_data;
2698 }
2699
2700 /* setup the item for the new data */
2701 for (i = 0; i < nr; i++) {
2702 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2703 btrfs_set_item_key(leaf, &disk_key, slot + i);
2704 item = btrfs_item_nr(leaf, slot + i);
2705 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2706 data_end -= data_size[i];
2707 btrfs_set_item_size(leaf, item, data_size[i]);
2708 }
2709 btrfs_set_header_nritems(leaf, nritems + nr);
2710 btrfs_mark_buffer_dirty(leaf);
2711
2712 ret = 0;
2713 if (slot == 0) {
2714 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2715 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2716 }
2717
2718 if (btrfs_leaf_free_space(root, leaf) < 0) {
2719 btrfs_print_leaf(root, leaf);
2720 BUG();
2721 }
2722 out:
2723 return ret;
2724 }
2725
2726 /*
2727 * Given a key and some data, insert an item into the tree.
2728 * This does all the path init required, making room in the tree if needed.
2729 */
2730 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2731 *root, struct btrfs_key *cpu_key, void *data, u32
2732 data_size)
2733 {
2734 int ret = 0;
2735 struct btrfs_path *path;
2736 struct extent_buffer *leaf;
2737 unsigned long ptr;
2738
2739 path = btrfs_alloc_path();
2740 BUG_ON(!path);
2741 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2742 if (!ret) {
2743 leaf = path->nodes[0];
2744 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2745 write_extent_buffer(leaf, data, ptr, data_size);
2746 btrfs_mark_buffer_dirty(leaf);
2747 }
2748 btrfs_free_path(path);
2749 return ret;
2750 }
2751
2752 /*
2753 * delete the pointer from a given node.
2754 *
2755 * If the delete empties a node, the node is removed from the tree,
2756 * continuing all the way the root if required. The root is converted into
2757 * a leaf if all the nodes are emptied.
2758 */
2759 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2760 struct btrfs_path *path, int level, int slot)
2761 {
2762 struct extent_buffer *parent = path->nodes[level];
2763 u32 nritems;
2764 int ret = 0;
2765 int wret;
2766
2767 nritems = btrfs_header_nritems(parent);
2768 if (slot != nritems -1) {
2769 memmove_extent_buffer(parent,
2770 btrfs_node_key_ptr_offset(slot),
2771 btrfs_node_key_ptr_offset(slot + 1),
2772 sizeof(struct btrfs_key_ptr) *
2773 (nritems - slot - 1));
2774 }
2775 nritems--;
2776 btrfs_set_header_nritems(parent, nritems);
2777 if (nritems == 0 && parent == root->node) {
2778 BUG_ON(btrfs_header_level(root->node) != 1);
2779 /* just turn the root into a leaf and break */
2780 btrfs_set_header_level(root->node, 0);
2781 } else if (slot == 0) {
2782 struct btrfs_disk_key disk_key;
2783
2784 btrfs_node_key(parent, &disk_key, 0);
2785 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2786 if (wret)
2787 ret = wret;
2788 }
2789 btrfs_mark_buffer_dirty(parent);
2790 return ret;
2791 }
2792
2793 /*
2794 * delete the item at the leaf level in path. If that empties
2795 * the leaf, remove it from the tree
2796 */
2797 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2798 struct btrfs_path *path, int slot, int nr)
2799 {
2800 struct extent_buffer *leaf;
2801 struct btrfs_item *item;
2802 int last_off;
2803 int dsize = 0;
2804 int ret = 0;
2805 int wret;
2806 int i;
2807 u32 nritems;
2808
2809 leaf = path->nodes[0];
2810 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2811
2812 for (i = 0; i < nr; i++)
2813 dsize += btrfs_item_size_nr(leaf, slot + i);
2814
2815 nritems = btrfs_header_nritems(leaf);
2816
2817 if (slot + nr != nritems) {
2818 int i;
2819 int data_end = leaf_data_end(root, leaf);
2820
2821 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2822 data_end + dsize,
2823 btrfs_leaf_data(leaf) + data_end,
2824 last_off - data_end);
2825
2826 for (i = slot + nr; i < nritems; i++) {
2827 u32 ioff;
2828
2829 item = btrfs_item_nr(leaf, i);
2830 if (!leaf->map_token) {
2831 map_extent_buffer(leaf, (unsigned long)item,
2832 sizeof(struct btrfs_item),
2833 &leaf->map_token, &leaf->kaddr,
2834 &leaf->map_start, &leaf->map_len,
2835 KM_USER1);
2836 }
2837 ioff = btrfs_item_offset(leaf, item);
2838 btrfs_set_item_offset(leaf, item, ioff + dsize);
2839 }
2840
2841 if (leaf->map_token) {
2842 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2843 leaf->map_token = NULL;
2844 }
2845
2846 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2847 btrfs_item_nr_offset(slot + nr),
2848 sizeof(struct btrfs_item) *
2849 (nritems - slot - nr));
2850 }
2851 btrfs_set_header_nritems(leaf, nritems - nr);
2852 nritems -= nr;
2853
2854 /* delete the leaf if we've emptied it */
2855 if (nritems == 0) {
2856 if (leaf == root->node) {
2857 btrfs_set_header_level(leaf, 0);
2858 } else {
2859 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2860 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2861 if (wret)
2862 ret = wret;
2863 wret = btrfs_free_extent(trans, root,
2864 leaf->start, leaf->len,
2865 btrfs_header_owner(path->nodes[1]),
2866 root_gen, 0, 0, 1);
2867 if (wret)
2868 ret = wret;
2869 }
2870 } else {
2871 int used = leaf_space_used(leaf, 0, nritems);
2872 if (slot == 0) {
2873 struct btrfs_disk_key disk_key;
2874
2875 btrfs_item_key(leaf, &disk_key, 0);
2876 wret = fixup_low_keys(trans, root, path,
2877 &disk_key, 1);
2878 if (wret)
2879 ret = wret;
2880 }
2881
2882 /* delete the leaf if it is mostly empty */
2883 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2884 /* push_leaf_left fixes the path.
2885 * make sure the path still points to our leaf
2886 * for possible call to del_ptr below
2887 */
2888 slot = path->slots[1];
2889 extent_buffer_get(leaf);
2890
2891 wret = push_leaf_left(trans, root, path, 1, 1);
2892 if (wret < 0 && wret != -ENOSPC)
2893 ret = wret;
2894
2895 if (path->nodes[0] == leaf &&
2896 btrfs_header_nritems(leaf)) {
2897 wret = push_leaf_right(trans, root, path, 1, 1);
2898 if (wret < 0 && wret != -ENOSPC)
2899 ret = wret;
2900 }
2901
2902 if (btrfs_header_nritems(leaf) == 0) {
2903 u64 root_gen;
2904 u64 bytenr = leaf->start;
2905 u32 blocksize = leaf->len;
2906
2907 root_gen = btrfs_header_generation(
2908 path->nodes[1]);
2909
2910 wret = del_ptr(trans, root, path, 1, slot);
2911 if (wret)
2912 ret = wret;
2913
2914 free_extent_buffer(leaf);
2915 wret = btrfs_free_extent(trans, root, bytenr,
2916 blocksize,
2917 btrfs_header_owner(path->nodes[1]),
2918 root_gen, 0, 0, 1);
2919 if (wret)
2920 ret = wret;
2921 } else {
2922 /* if we're still in the path, make sure
2923 * we're dirty. Otherwise, one of the
2924 * push_leaf functions must have already
2925 * dirtied this buffer
2926 */
2927 if (path->nodes[0] == leaf)
2928 btrfs_mark_buffer_dirty(leaf);
2929 free_extent_buffer(leaf);
2930 }
2931 } else {
2932 btrfs_mark_buffer_dirty(leaf);
2933 }
2934 }
2935 return ret;
2936 }
2937
2938 /*
2939 * search the tree again to find a leaf with lesser keys
2940 * returns 0 if it found something or 1 if there are no lesser leaves.
2941 * returns < 0 on io errors.
2942 */
2943 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2944 {
2945 struct btrfs_key key;
2946 struct btrfs_disk_key found_key;
2947 int ret;
2948
2949 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
2950
2951 if (key.offset > 0)
2952 key.offset--;
2953 else if (key.type > 0)
2954 key.type--;
2955 else if (key.objectid > 0)
2956 key.objectid--;
2957 else
2958 return 1;
2959
2960 btrfs_release_path(root, path);
2961 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2962 if (ret < 0)
2963 return ret;
2964 btrfs_item_key(path->nodes[0], &found_key, 0);
2965 ret = comp_keys(&found_key, &key);
2966 if (ret < 0)
2967 return 0;
2968 return 1;
2969 }
2970
2971 /*
2972 * search the tree again to find a leaf with greater keys
2973 * returns 0 if it found something or 1 if there are no greater leaves.
2974 * returns < 0 on io errors.
2975 */
2976 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2977 {
2978 int slot;
2979 int level = 1;
2980 struct extent_buffer *c;
2981 struct extent_buffer *next = NULL;
2982 struct btrfs_key key;
2983 u32 nritems;
2984 int ret;
2985
2986 nritems = btrfs_header_nritems(path->nodes[0]);
2987 if (nritems == 0) {
2988 return 1;
2989 }
2990
2991 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
2992
2993 btrfs_release_path(root, path);
2994 path->keep_locks = 1;
2995 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2996 path->keep_locks = 0;
2997
2998 if (ret < 0)
2999 return ret;
3000
3001 nritems = btrfs_header_nritems(path->nodes[0]);
3002 /*
3003 * by releasing the path above we dropped all our locks. A balance
3004 * could have added more items next to the key that used to be
3005 * at the very end of the block. So, check again here and
3006 * advance the path if there are now more items available.
3007 */
3008 if (nritems > 0 && path->slots[0] < nritems - 1) {
3009 path->slots[0]++;
3010 goto done;
3011 }
3012
3013 while(level < BTRFS_MAX_LEVEL) {
3014 if (!path->nodes[level])
3015 return 1;
3016
3017 slot = path->slots[level] + 1;
3018 c = path->nodes[level];
3019 if (slot >= btrfs_header_nritems(c)) {
3020 level++;
3021 if (level == BTRFS_MAX_LEVEL) {
3022 return 1;
3023 }
3024 continue;
3025 }
3026
3027 if (next) {
3028 btrfs_tree_unlock(next);
3029 free_extent_buffer(next);
3030 }
3031
3032 if (level == 1 && path->locks[1] && path->reada)
3033 reada_for_search(root, path, level, slot, 0);
3034
3035 next = read_node_slot(root, c, slot);
3036 WARN_ON(!btrfs_tree_locked(c));
3037 btrfs_tree_lock(next);
3038 break;
3039 }
3040 path->slots[level] = slot;
3041 while(1) {
3042 level--;
3043 c = path->nodes[level];
3044 if (path->locks[level])
3045 btrfs_tree_unlock(c);
3046 free_extent_buffer(c);
3047 path->nodes[level] = next;
3048 path->slots[level] = 0;
3049 path->locks[level] = 1;
3050 if (!level)
3051 break;
3052 if (level == 1 && path->locks[1] && path->reada)
3053 reada_for_search(root, path, level, slot, 0);
3054 next = read_node_slot(root, next, 0);
3055 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3056 btrfs_tree_lock(next);
3057 }
3058 done:
3059 unlock_up(path, 0, 1);
3060 return 0;
3061 }
3062
3063 int btrfs_previous_item(struct btrfs_root *root,
3064 struct btrfs_path *path, u64 min_objectid,
3065 int type)
3066 {
3067 struct btrfs_key found_key;
3068 struct extent_buffer *leaf;
3069 int ret;
3070
3071 while(1) {
3072 if (path->slots[0] == 0) {
3073 ret = btrfs_prev_leaf(root, path);
3074 if (ret != 0)
3075 return ret;
3076 } else {
3077 path->slots[0]--;
3078 }
3079 leaf = path->nodes[0];
3080 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3081 if (found_key.type == type)
3082 return 0;
3083 }
3084 return 1;
3085 }
3086
Linux kernel - Deliverables
This page took 0.096604 seconds and 6 git commands to generate.