projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: Reduce contention on the root node
[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
67 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
68 p->slots[i] = 0;
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 p->nodes[i] = NULL;
77 }
78 }
79
80 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
81 {
82 struct extent_buffer *eb;
83 spin_lock(&root->node_lock);
84 eb = root->node;
85 extent_buffer_get(eb);
86 spin_unlock(&root->node_lock);
87 return eb;
88 }
89
90 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
91 {
92 struct extent_buffer *eb;
93
94 while(1) {
95 eb = btrfs_root_node(root);
96 btrfs_tree_lock(eb);
97
98 spin_lock(&root->node_lock);
99 if (eb == root->node) {
100 spin_unlock(&root->node_lock);
101 break;
102 }
103 spin_unlock(&root->node_lock);
104
105 btrfs_tree_unlock(eb);
106 free_extent_buffer(eb);
107 }
108 return eb;
109 }
110
111 static void add_root_to_dirty_list(struct btrfs_root *root)
112 {
113 if (root->track_dirty && list_empty(&root->dirty_list)) {
114 list_add(&root->dirty_list,
115 &root->fs_info->dirty_cowonly_roots);
116 }
117 }
118
119 int btrfs_copy_root(struct btrfs_trans_handle *trans,
120 struct btrfs_root *root,
121 struct extent_buffer *buf,
122 struct extent_buffer **cow_ret, u64 new_root_objectid)
123 {
124 struct extent_buffer *cow;
125 u32 nritems;
126 int ret = 0;
127 int level;
128 struct btrfs_key first_key;
129 struct btrfs_root *new_root;
130
131 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
132 if (!new_root)
133 return -ENOMEM;
134
135 memcpy(new_root, root, sizeof(*new_root));
136 new_root->root_key.objectid = new_root_objectid;
137
138 WARN_ON(root->ref_cows && trans->transid !=
139 root->fs_info->running_transaction->transid);
140 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
141
142 level = btrfs_header_level(buf);
143 nritems = btrfs_header_nritems(buf);
144 if (nritems) {
145 if (level == 0)
146 btrfs_item_key_to_cpu(buf, &first_key, 0);
147 else
148 btrfs_node_key_to_cpu(buf, &first_key, 0);
149 } else {
150 first_key.objectid = 0;
151 }
152 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
153 new_root_objectid,
154 trans->transid, first_key.objectid,
155 level, buf->start, 0);
156 if (IS_ERR(cow)) {
157 kfree(new_root);
158 return PTR_ERR(cow);
159 }
160
161 copy_extent_buffer(cow, buf, 0, 0, cow->len);
162 btrfs_set_header_bytenr(cow, cow->start);
163 btrfs_set_header_generation(cow, trans->transid);
164 btrfs_set_header_owner(cow, new_root_objectid);
165 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
166
167 WARN_ON(btrfs_header_generation(buf) > trans->transid);
168 ret = btrfs_inc_ref(trans, new_root, buf);
169 kfree(new_root);
170
171 if (ret)
172 return ret;
173
174 btrfs_mark_buffer_dirty(cow);
175 *cow_ret = cow;
176 return 0;
177 }
178
179 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
180 struct btrfs_root *root,
181 struct extent_buffer *buf,
182 struct extent_buffer *parent, int parent_slot,
183 struct extent_buffer **cow_ret,
184 u64 search_start, u64 empty_size)
185 {
186 u64 root_gen;
187 struct extent_buffer *cow;
188 u32 nritems;
189 int ret = 0;
190 int different_trans = 0;
191 int level;
192 int unlock_orig = 0;
193 struct btrfs_key first_key;
194
195 if (*cow_ret == buf)
196 unlock_orig = 1;
197
198 WARN_ON(!btrfs_tree_locked(buf));
199
200 if (root->ref_cows) {
201 root_gen = trans->transid;
202 } else {
203 root_gen = 0;
204 }
205 WARN_ON(root->ref_cows && trans->transid !=
206 root->fs_info->running_transaction->transid);
207 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
208
209 level = btrfs_header_level(buf);
210 nritems = btrfs_header_nritems(buf);
211 if (nritems) {
212 if (level == 0)
213 btrfs_item_key_to_cpu(buf, &first_key, 0);
214 else
215 btrfs_node_key_to_cpu(buf, &first_key, 0);
216 } else {
217 first_key.objectid = 0;
218 }
219 cow = btrfs_alloc_free_block(trans, root, buf->len,
220 root->root_key.objectid,
221 root_gen, first_key.objectid, level,
222 search_start, empty_size);
223 if (IS_ERR(cow))
224 return PTR_ERR(cow);
225
226 copy_extent_buffer(cow, buf, 0, 0, cow->len);
227 btrfs_set_header_bytenr(cow, cow->start);
228 btrfs_set_header_generation(cow, trans->transid);
229 btrfs_set_header_owner(cow, root->root_key.objectid);
230 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
231
232 WARN_ON(btrfs_header_generation(buf) > trans->transid);
233 if (btrfs_header_generation(buf) != trans->transid) {
234 different_trans = 1;
235 ret = btrfs_inc_ref(trans, root, buf);
236 if (ret)
237 return ret;
238 } else {
239 clean_tree_block(trans, root, buf);
240 }
241
242 if (buf == root->node) {
243 WARN_ON(parent && parent != buf);
244 root_gen = btrfs_header_generation(buf);
245
246 spin_lock(&root->node_lock);
247 root->node = cow;
248 extent_buffer_get(cow);
249 spin_unlock(&root->node_lock);
250
251 if (buf != root->commit_root) {
252 btrfs_free_extent(trans, root, buf->start,
253 buf->len, root->root_key.objectid,
254 root_gen, 0, 0, 1);
255 }
256 free_extent_buffer(buf);
257 add_root_to_dirty_list(root);
258 } else {
259 root_gen = btrfs_header_generation(parent);
260 btrfs_set_node_blockptr(parent, parent_slot,
261 cow->start);
262 WARN_ON(trans->transid == 0);
263 btrfs_set_node_ptr_generation(parent, parent_slot,
264 trans->transid);
265 btrfs_mark_buffer_dirty(parent);
266 WARN_ON(btrfs_header_generation(parent) != trans->transid);
267 btrfs_free_extent(trans, root, buf->start, buf->len,
268 btrfs_header_owner(parent), root_gen,
269 0, 0, 1);
270 }
271 if (unlock_orig)
272 btrfs_tree_unlock(buf);
273 free_extent_buffer(buf);
274 btrfs_mark_buffer_dirty(cow);
275 *cow_ret = cow;
276 return 0;
277 }
278
279 int btrfs_cow_block(struct btrfs_trans_handle *trans,
280 struct btrfs_root *root, struct extent_buffer *buf,
281 struct extent_buffer *parent, int parent_slot,
282 struct extent_buffer **cow_ret)
283 {
284 u64 search_start;
285 u64 header_trans;
286 int ret;
287
288 if (trans->transaction != root->fs_info->running_transaction) {
289 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
290 root->fs_info->running_transaction->transid);
291 WARN_ON(1);
292 }
293 if (trans->transid != root->fs_info->generation) {
294 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
295 root->fs_info->generation);
296 WARN_ON(1);
297 }
298
299 header_trans = btrfs_header_generation(buf);
300 spin_lock(&root->fs_info->hash_lock);
301 if (header_trans == trans->transid &&
302 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
303 *cow_ret = buf;
304 spin_unlock(&root->fs_info->hash_lock);
305 return 0;
306 }
307 spin_unlock(&root->fs_info->hash_lock);
308 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
309 ret = __btrfs_cow_block(trans, root, buf, parent,
310 parent_slot, cow_ret, search_start, 0);
311 return ret;
312 }
313
314 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
315 {
316 if (blocknr < other && other - (blocknr + blocksize) < 32768)
317 return 1;
318 if (blocknr > other && blocknr - (other + blocksize) < 32768)
319 return 1;
320 return 0;
321 }
322
323 /*
324 * compare two keys in a memcmp fashion
325 */
326 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
327 {
328 struct btrfs_key k1;
329
330 btrfs_disk_key_to_cpu(&k1, disk);
331
332 if (k1.objectid > k2->objectid)
333 return 1;
334 if (k1.objectid < k2->objectid)
335 return -1;
336 if (k1.type > k2->type)
337 return 1;
338 if (k1.type < k2->type)
339 return -1;
340 if (k1.offset > k2->offset)
341 return 1;
342 if (k1.offset < k2->offset)
343 return -1;
344 return 0;
345 }
346
347
348 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
349 struct btrfs_root *root, struct extent_buffer *parent,
350 int start_slot, int cache_only, u64 *last_ret,
351 struct btrfs_key *progress)
352 {
353 struct extent_buffer *cur;
354 u64 blocknr;
355 u64 gen;
356 u64 search_start = *last_ret;
357 u64 last_block = 0;
358 u64 other;
359 u32 parent_nritems;
360 int end_slot;
361 int i;
362 int err = 0;
363 int parent_level;
364 int uptodate;
365 u32 blocksize;
366 int progress_passed = 0;
367 struct btrfs_disk_key disk_key;
368
369 parent_level = btrfs_header_level(parent);
370 if (cache_only && parent_level != 1)
371 return 0;
372
373 if (trans->transaction != root->fs_info->running_transaction) {
374 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
375 root->fs_info->running_transaction->transid);
376 WARN_ON(1);
377 }
378 if (trans->transid != root->fs_info->generation) {
379 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
380 root->fs_info->generation);
381 WARN_ON(1);
382 }
383
384 parent_nritems = btrfs_header_nritems(parent);
385 blocksize = btrfs_level_size(root, parent_level - 1);
386 end_slot = parent_nritems;
387
388 if (parent_nritems == 1)
389 return 0;
390
391 for (i = start_slot; i < end_slot; i++) {
392 int close = 1;
393
394 if (!parent->map_token) {
395 map_extent_buffer(parent,
396 btrfs_node_key_ptr_offset(i),
397 sizeof(struct btrfs_key_ptr),
398 &parent->map_token, &parent->kaddr,
399 &parent->map_start, &parent->map_len,
400 KM_USER1);
401 }
402 btrfs_node_key(parent, &disk_key, i);
403 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
404 continue;
405
406 progress_passed = 1;
407 blocknr = btrfs_node_blockptr(parent, i);
408 gen = btrfs_node_ptr_generation(parent, i);
409 if (last_block == 0)
410 last_block = blocknr;
411
412 if (i > 0) {
413 other = btrfs_node_blockptr(parent, i - 1);
414 close = close_blocks(blocknr, other, blocksize);
415 }
416 if (!close && i < end_slot - 2) {
417 other = btrfs_node_blockptr(parent, i + 1);
418 close = close_blocks(blocknr, other, blocksize);
419 }
420 if (close) {
421 last_block = blocknr;
422 continue;
423 }
424 if (parent->map_token) {
425 unmap_extent_buffer(parent, parent->map_token,
426 KM_USER1);
427 parent->map_token = NULL;
428 }
429
430 cur = btrfs_find_tree_block(root, blocknr, blocksize);
431 if (cur)
432 uptodate = btrfs_buffer_uptodate(cur, gen);
433 else
434 uptodate = 0;
435 if (!cur || !uptodate) {
436 if (cache_only) {
437 free_extent_buffer(cur);
438 continue;
439 }
440 if (!cur) {
441 cur = read_tree_block(root, blocknr,
442 blocksize, gen);
443 } else if (!uptodate) {
444 btrfs_read_buffer(cur, gen);
445 }
446 }
447 if (search_start == 0)
448 search_start = last_block;
449
450 btrfs_tree_lock(cur);
451 err = __btrfs_cow_block(trans, root, cur, parent, i,
452 &cur, search_start,
453 min(16 * blocksize,
454 (end_slot - i) * blocksize));
455 if (err) {
456 btrfs_tree_unlock(cur);
457 free_extent_buffer(cur);
458 break;
459 }
460 search_start = cur->start;
461 last_block = cur->start;
462 *last_ret = search_start;
463 btrfs_tree_unlock(cur);
464 free_extent_buffer(cur);
465 }
466 if (parent->map_token) {
467 unmap_extent_buffer(parent, parent->map_token,
468 KM_USER1);
469 parent->map_token = NULL;
470 }
471 return err;
472 }
473
474 /*
475 * The leaf data grows from end-to-front in the node.
476 * this returns the address of the start of the last item,
477 * which is the stop of the leaf data stack
478 */
479 static inline unsigned int leaf_data_end(struct btrfs_root *root,
480 struct extent_buffer *leaf)
481 {
482 u32 nr = btrfs_header_nritems(leaf);
483 if (nr == 0)
484 return BTRFS_LEAF_DATA_SIZE(root);
485 return btrfs_item_offset_nr(leaf, nr - 1);
486 }
487
488 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
489 int level)
490 {
491 struct extent_buffer *parent = NULL;
492 struct extent_buffer *node = path->nodes[level];
493 struct btrfs_disk_key parent_key;
494 struct btrfs_disk_key node_key;
495 int parent_slot;
496 int slot;
497 struct btrfs_key cpukey;
498 u32 nritems = btrfs_header_nritems(node);
499
500 if (path->nodes[level + 1])
501 parent = path->nodes[level + 1];
502
503 slot = path->slots[level];
504 BUG_ON(nritems == 0);
505 if (parent) {
506 parent_slot = path->slots[level + 1];
507 btrfs_node_key(parent, &parent_key, parent_slot);
508 btrfs_node_key(node, &node_key, 0);
509 BUG_ON(memcmp(&parent_key, &node_key,
510 sizeof(struct btrfs_disk_key)));
511 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
512 btrfs_header_bytenr(node));
513 }
514 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
515 if (slot != 0) {
516 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
517 btrfs_node_key(node, &node_key, slot);
518 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
519 }
520 if (slot < nritems - 1) {
521 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
522 btrfs_node_key(node, &node_key, slot);
523 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
524 }
525 return 0;
526 }
527
528 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
529 int level)
530 {
531 struct extent_buffer *leaf = path->nodes[level];
532 struct extent_buffer *parent = NULL;
533 int parent_slot;
534 struct btrfs_key cpukey;
535 struct btrfs_disk_key parent_key;
536 struct btrfs_disk_key leaf_key;
537 int slot = path->slots[0];
538
539 u32 nritems = btrfs_header_nritems(leaf);
540
541 if (path->nodes[level + 1])
542 parent = path->nodes[level + 1];
543
544 if (nritems == 0)
545 return 0;
546
547 if (parent) {
548 parent_slot = path->slots[level + 1];
549 btrfs_node_key(parent, &parent_key, parent_slot);
550 btrfs_item_key(leaf, &leaf_key, 0);
551
552 BUG_ON(memcmp(&parent_key, &leaf_key,
553 sizeof(struct btrfs_disk_key)));
554 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
555 btrfs_header_bytenr(leaf));
556 }
557 #if 0
558 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
559 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
560 btrfs_item_key(leaf, &leaf_key, i);
561 if (comp_keys(&leaf_key, &cpukey) >= 0) {
562 btrfs_print_leaf(root, leaf);
563 printk("slot %d offset bad key\n", i);
564 BUG_ON(1);
565 }
566 if (btrfs_item_offset_nr(leaf, i) !=
567 btrfs_item_end_nr(leaf, i + 1)) {
568 btrfs_print_leaf(root, leaf);
569 printk("slot %d offset bad\n", i);
570 BUG_ON(1);
571 }
572 if (i == 0) {
573 if (btrfs_item_offset_nr(leaf, i) +
574 btrfs_item_size_nr(leaf, i) !=
575 BTRFS_LEAF_DATA_SIZE(root)) {
576 btrfs_print_leaf(root, leaf);
577 printk("slot %d first offset bad\n", i);
578 BUG_ON(1);
579 }
580 }
581 }
582 if (nritems > 0) {
583 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
584 btrfs_print_leaf(root, leaf);
585 printk("slot %d bad size \n", nritems - 1);
586 BUG_ON(1);
587 }
588 }
589 #endif
590 if (slot != 0 && slot < nritems - 1) {
591 btrfs_item_key(leaf, &leaf_key, slot);
592 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
593 if (comp_keys(&leaf_key, &cpukey) <= 0) {
594 btrfs_print_leaf(root, leaf);
595 printk("slot %d offset bad key\n", slot);
596 BUG_ON(1);
597 }
598 if (btrfs_item_offset_nr(leaf, slot - 1) !=
599 btrfs_item_end_nr(leaf, slot)) {
600 btrfs_print_leaf(root, leaf);
601 printk("slot %d offset bad\n", slot);
602 BUG_ON(1);
603 }
604 }
605 if (slot < nritems - 1) {
606 btrfs_item_key(leaf, &leaf_key, slot);
607 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
608 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
609 if (btrfs_item_offset_nr(leaf, slot) !=
610 btrfs_item_end_nr(leaf, slot + 1)) {
611 btrfs_print_leaf(root, leaf);
612 printk("slot %d offset bad\n", slot);
613 BUG_ON(1);
614 }
615 }
616 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
617 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
618 return 0;
619 }
620
621 static int noinline check_block(struct btrfs_root *root,
622 struct btrfs_path *path, int level)
623 {
624 u64 found_start;
625 return 0;
626 if (btrfs_header_level(path->nodes[level]) != level)
627 printk("warning: bad level %Lu wanted %d found %d\n",
628 path->nodes[level]->start, level,
629 btrfs_header_level(path->nodes[level]));
630 found_start = btrfs_header_bytenr(path->nodes[level]);
631 if (found_start != path->nodes[level]->start) {
632 printk("warning: bad bytentr %Lu found %Lu\n",
633 path->nodes[level]->start, found_start);
634 }
635 #if 0
636 struct extent_buffer *buf = path->nodes[level];
637
638 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
639 (unsigned long)btrfs_header_fsid(buf),
640 BTRFS_FSID_SIZE)) {
641 printk("warning bad block %Lu\n", buf->start);
642 return 1;
643 }
644 #endif
645 if (level == 0)
646 return check_leaf(root, path, level);
647 return check_node(root, path, level);
648 }
649
650 /*
651 * search for key in the extent_buffer. The items start at offset p,
652 * and they are item_size apart. There are 'max' items in p.
653 *
654 * the slot in the array is returned via slot, and it points to
655 * the place where you would insert key if it is not found in
656 * the array.
657 *
658 * slot may point to max if the key is bigger than all of the keys
659 */
660 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
661 int item_size, struct btrfs_key *key,
662 int max, int *slot)
663 {
664 int low = 0;
665 int high = max;
666 int mid;
667 int ret;
668 struct btrfs_disk_key *tmp = NULL;
669 struct btrfs_disk_key unaligned;
670 unsigned long offset;
671 char *map_token = NULL;
672 char *kaddr = NULL;
673 unsigned long map_start = 0;
674 unsigned long map_len = 0;
675 int err;
676
677 while(low < high) {
678 mid = (low + high) / 2;
679 offset = p + mid * item_size;
680
681 if (!map_token || offset < map_start ||
682 (offset + sizeof(struct btrfs_disk_key)) >
683 map_start + map_len) {
684 if (map_token) {
685 unmap_extent_buffer(eb, map_token, KM_USER0);
686 map_token = NULL;
687 }
688 err = map_extent_buffer(eb, offset,
689 sizeof(struct btrfs_disk_key),
690 &map_token, &kaddr,
691 &map_start, &map_len, KM_USER0);
692
693 if (!err) {
694 tmp = (struct btrfs_disk_key *)(kaddr + offset -
695 map_start);
696 } else {
697 read_extent_buffer(eb, &unaligned,
698 offset, sizeof(unaligned));
699 tmp = &unaligned;
700 }
701
702 } else {
703 tmp = (struct btrfs_disk_key *)(kaddr + offset -
704 map_start);
705 }
706 ret = comp_keys(tmp, key);
707
708 if (ret < 0)
709 low = mid + 1;
710 else if (ret > 0)
711 high = mid;
712 else {
713 *slot = mid;
714 if (map_token)
715 unmap_extent_buffer(eb, map_token, KM_USER0);
716 return 0;
717 }
718 }
719 *slot = low;
720 if (map_token)
721 unmap_extent_buffer(eb, map_token, KM_USER0);
722 return 1;
723 }
724
725 /*
726 * simple bin_search frontend that does the right thing for
727 * leaves vs nodes
728 */
729 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
730 int level, int *slot)
731 {
732 if (level == 0) {
733 return generic_bin_search(eb,
734 offsetof(struct btrfs_leaf, items),
735 sizeof(struct btrfs_item),
736 key, btrfs_header_nritems(eb),
737 slot);
738 } else {
739 return generic_bin_search(eb,
740 offsetof(struct btrfs_node, ptrs),
741 sizeof(struct btrfs_key_ptr),
742 key, btrfs_header_nritems(eb),
743 slot);
744 }
745 return -1;
746 }
747
748 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
749 struct extent_buffer *parent, int slot)
750 {
751 int level = btrfs_header_level(parent);
752 if (slot < 0)
753 return NULL;
754 if (slot >= btrfs_header_nritems(parent))
755 return NULL;
756
757 BUG_ON(level == 0);
758
759 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
760 btrfs_level_size(root, level - 1),
761 btrfs_node_ptr_generation(parent, slot));
762 }
763
764 static int balance_level(struct btrfs_trans_handle *trans,
765 struct btrfs_root *root,
766 struct btrfs_path *path, int level)
767 {
768 struct extent_buffer *right = NULL;
769 struct extent_buffer *mid;
770 struct extent_buffer *left = NULL;
771 struct extent_buffer *parent = NULL;
772 int ret = 0;
773 int wret;
774 int pslot;
775 int orig_slot = path->slots[level];
776 int err_on_enospc = 0;
777 u64 orig_ptr;
778
779 if (level == 0)
780 return 0;
781
782 mid = path->nodes[level];
783 WARN_ON(!path->locks[level]);
784 WARN_ON(btrfs_header_generation(mid) != trans->transid);
785
786 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
787
788 if (level < BTRFS_MAX_LEVEL - 1)
789 parent = path->nodes[level + 1];
790 pslot = path->slots[level + 1];
791
792 /*
793 * deal with the case where there is only one pointer in the root
794 * by promoting the node below to a root
795 */
796 if (!parent) {
797 struct extent_buffer *child;
798
799 if (btrfs_header_nritems(mid) != 1)
800 return 0;
801
802 /* promote the child to a root */
803 child = read_node_slot(root, mid, 0);
804 btrfs_tree_lock(child);
805 BUG_ON(!child);
806 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
807 BUG_ON(ret);
808
809 spin_lock(&root->node_lock);
810 root->node = child;
811 spin_unlock(&root->node_lock);
812
813 add_root_to_dirty_list(root);
814 btrfs_tree_unlock(child);
815 path->locks[level] = 0;
816 path->nodes[level] = NULL;
817 clean_tree_block(trans, root, mid);
818 btrfs_tree_unlock(mid);
819 /* once for the path */
820 free_extent_buffer(mid);
821 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
822 root->root_key.objectid,
823 btrfs_header_generation(mid), 0, 0, 1);
824 /* once for the root ptr */
825 free_extent_buffer(mid);
826 return ret;
827 }
828 if (btrfs_header_nritems(mid) >
829 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
830 return 0;
831
832 if (btrfs_header_nritems(mid) < 2)
833 err_on_enospc = 1;
834
835 left = read_node_slot(root, parent, pslot - 1);
836 if (left) {
837 btrfs_tree_lock(left);
838 wret = btrfs_cow_block(trans, root, left,
839 parent, pslot - 1, &left);
840 if (wret) {
841 ret = wret;
842 goto enospc;
843 }
844 }
845 right = read_node_slot(root, parent, pslot + 1);
846 if (right) {
847 btrfs_tree_lock(right);
848 wret = btrfs_cow_block(trans, root, right,
849 parent, pslot + 1, &right);
850 if (wret) {
851 ret = wret;
852 goto enospc;
853 }
854 }
855
856 /* first, try to make some room in the middle buffer */
857 if (left) {
858 orig_slot += btrfs_header_nritems(left);
859 wret = push_node_left(trans, root, left, mid, 1);
860 if (wret < 0)
861 ret = wret;
862 if (btrfs_header_nritems(mid) < 2)
863 err_on_enospc = 1;
864 }
865
866 /*
867 * then try to empty the right most buffer into the middle
868 */
869 if (right) {
870 wret = push_node_left(trans, root, mid, right, 1);
871 if (wret < 0 && wret != -ENOSPC)
872 ret = wret;
873 if (btrfs_header_nritems(right) == 0) {
874 u64 bytenr = right->start;
875 u64 generation = btrfs_header_generation(parent);
876 u32 blocksize = right->len;
877
878 clean_tree_block(trans, root, right);
879 btrfs_tree_unlock(right);
880 free_extent_buffer(right);
881 right = NULL;
882 wret = del_ptr(trans, root, path, level + 1, pslot +
883 1);
884 if (wret)
885 ret = wret;
886 wret = btrfs_free_extent(trans, root, bytenr,
887 blocksize,
888 btrfs_header_owner(parent),
889 generation, 0, 0, 1);
890 if (wret)
891 ret = wret;
892 } else {
893 struct btrfs_disk_key right_key;
894 btrfs_node_key(right, &right_key, 0);
895 btrfs_set_node_key(parent, &right_key, pslot + 1);
896 btrfs_mark_buffer_dirty(parent);
897 }
898 }
899 if (btrfs_header_nritems(mid) == 1) {
900 /*
901 * we're not allowed to leave a node with one item in the
902 * tree during a delete. A deletion from lower in the tree
903 * could try to delete the only pointer in this node.
904 * So, pull some keys from the left.
905 * There has to be a left pointer at this point because
906 * otherwise we would have pulled some pointers from the
907 * right
908 */
909 BUG_ON(!left);
910 wret = balance_node_right(trans, root, mid, left);
911 if (wret < 0) {
912 ret = wret;
913 goto enospc;
914 }
915 if (wret == 1) {
916 wret = push_node_left(trans, root, left, mid, 1);
917 if (wret < 0)
918 ret = wret;
919 }
920 BUG_ON(wret == 1);
921 }
922 if (btrfs_header_nritems(mid) == 0) {
923 /* we've managed to empty the middle node, drop it */
924 u64 root_gen = btrfs_header_generation(parent);
925 u64 bytenr = mid->start;
926 u32 blocksize = mid->len;
927
928 clean_tree_block(trans, root, mid);
929 btrfs_tree_unlock(mid);
930 free_extent_buffer(mid);
931 mid = NULL;
932 wret = del_ptr(trans, root, path, level + 1, pslot);
933 if (wret)
934 ret = wret;
935 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
936 btrfs_header_owner(parent),
937 root_gen, 0, 0, 1);
938 if (wret)
939 ret = wret;
940 } else {
941 /* update the parent key to reflect our changes */
942 struct btrfs_disk_key mid_key;
943 btrfs_node_key(mid, &mid_key, 0);
944 btrfs_set_node_key(parent, &mid_key, pslot);
945 btrfs_mark_buffer_dirty(parent);
946 }
947
948 /* update the path */
949 if (left) {
950 if (btrfs_header_nritems(left) > orig_slot) {
951 extent_buffer_get(left);
952 /* left was locked after cow */
953 path->nodes[level] = left;
954 path->slots[level + 1] -= 1;
955 path->slots[level] = orig_slot;
956 if (mid) {
957 btrfs_tree_unlock(mid);
958 free_extent_buffer(mid);
959 }
960 } else {
961 orig_slot -= btrfs_header_nritems(left);
962 path->slots[level] = orig_slot;
963 }
964 }
965 /* double check we haven't messed things up */
966 check_block(root, path, level);
967 if (orig_ptr !=
968 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
969 BUG();
970 enospc:
971 if (right) {
972 btrfs_tree_unlock(right);
973 free_extent_buffer(right);
974 }
975 if (left) {
976 if (path->nodes[level] != left)
977 btrfs_tree_unlock(left);
978 free_extent_buffer(left);
979 }
980 return ret;
981 }
982
983 /* returns zero if the push worked, non-zero otherwise */
984 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
985 struct btrfs_root *root,
986 struct btrfs_path *path, int level)
987 {
988 struct extent_buffer *right = NULL;
989 struct extent_buffer *mid;
990 struct extent_buffer *left = NULL;
991 struct extent_buffer *parent = NULL;
992 int ret = 0;
993 int wret;
994 int pslot;
995 int orig_slot = path->slots[level];
996 u64 orig_ptr;
997
998 if (level == 0)
999 return 1;
1000
1001 mid = path->nodes[level];
1002 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1003 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1004
1005 if (level < BTRFS_MAX_LEVEL - 1)
1006 parent = path->nodes[level + 1];
1007 pslot = path->slots[level + 1];
1008
1009 if (!parent)
1010 return 1;
1011
1012 left = read_node_slot(root, parent, pslot - 1);
1013
1014 /* first, try to make some room in the middle buffer */
1015 if (left) {
1016 u32 left_nr;
1017
1018 btrfs_tree_lock(left);
1019 left_nr = btrfs_header_nritems(left);
1020 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1021 wret = 1;
1022 } else {
1023 ret = btrfs_cow_block(trans, root, left, parent,
1024 pslot - 1, &left);
1025 if (ret)
1026 wret = 1;
1027 else {
1028 wret = push_node_left(trans, root,
1029 left, mid, 0);
1030 }
1031 }
1032 if (wret < 0)
1033 ret = wret;
1034 if (wret == 0) {
1035 struct btrfs_disk_key disk_key;
1036 orig_slot += left_nr;
1037 btrfs_node_key(mid, &disk_key, 0);
1038 btrfs_set_node_key(parent, &disk_key, pslot);
1039 btrfs_mark_buffer_dirty(parent);
1040 if (btrfs_header_nritems(left) > orig_slot) {
1041 path->nodes[level] = left;
1042 path->slots[level + 1] -= 1;
1043 path->slots[level] = orig_slot;
1044 btrfs_tree_unlock(mid);
1045 free_extent_buffer(mid);
1046 } else {
1047 orig_slot -=
1048 btrfs_header_nritems(left);
1049 path->slots[level] = orig_slot;
1050 btrfs_tree_unlock(left);
1051 free_extent_buffer(left);
1052 }
1053 return 0;
1054 }
1055 btrfs_tree_unlock(left);
1056 free_extent_buffer(left);
1057 }
1058 right = read_node_slot(root, parent, pslot + 1);
1059
1060 /*
1061 * then try to empty the right most buffer into the middle
1062 */
1063 if (right) {
1064 u32 right_nr;
1065 btrfs_tree_lock(right);
1066 right_nr = btrfs_header_nritems(right);
1067 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1068 wret = 1;
1069 } else {
1070 ret = btrfs_cow_block(trans, root, right,
1071 parent, pslot + 1,
1072 &right);
1073 if (ret)
1074 wret = 1;
1075 else {
1076 wret = balance_node_right(trans, root,
1077 right, mid);
1078 }
1079 }
1080 if (wret < 0)
1081 ret = wret;
1082 if (wret == 0) {
1083 struct btrfs_disk_key disk_key;
1084
1085 btrfs_node_key(right, &disk_key, 0);
1086 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1087 btrfs_mark_buffer_dirty(parent);
1088
1089 if (btrfs_header_nritems(mid) <= orig_slot) {
1090 path->nodes[level] = right;
1091 path->slots[level + 1] += 1;
1092 path->slots[level] = orig_slot -
1093 btrfs_header_nritems(mid);
1094 btrfs_tree_unlock(mid);
1095 free_extent_buffer(mid);
1096 } else {
1097 btrfs_tree_unlock(right);
1098 free_extent_buffer(right);
1099 }
1100 return 0;
1101 }
1102 btrfs_tree_unlock(right);
1103 free_extent_buffer(right);
1104 }
1105 return 1;
1106 }
1107
1108 /*
1109 * readahead one full node of leaves
1110 */
1111 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
1112 int level, int slot, u64 objectid)
1113 {
1114 struct extent_buffer *node;
1115 struct btrfs_disk_key disk_key;
1116 u32 nritems;
1117 u64 search;
1118 u64 lowest_read;
1119 u64 highest_read;
1120 u64 nread = 0;
1121 int direction = path->reada;
1122 struct extent_buffer *eb;
1123 u32 nr;
1124 u32 blocksize;
1125 u32 nscan = 0;
1126
1127 if (level != 1)
1128 return;
1129
1130 if (!path->nodes[level])
1131 return;
1132
1133 node = path->nodes[level];
1134
1135 search = btrfs_node_blockptr(node, slot);
1136 blocksize = btrfs_level_size(root, level - 1);
1137 eb = btrfs_find_tree_block(root, search, blocksize);
1138 if (eb) {
1139 free_extent_buffer(eb);
1140 return;
1141 }
1142
1143 highest_read = search;
1144 lowest_read = search;
1145
1146 nritems = btrfs_header_nritems(node);
1147 nr = slot;
1148 while(1) {
1149 if (direction < 0) {
1150 if (nr == 0)
1151 break;
1152 nr--;
1153 } else if (direction > 0) {
1154 nr++;
1155 if (nr >= nritems)
1156 break;
1157 }
1158 if (path->reada < 0 && objectid) {
1159 btrfs_node_key(node, &disk_key, nr);
1160 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1161 break;
1162 }
1163 search = btrfs_node_blockptr(node, nr);
1164 if ((search >= lowest_read && search <= highest_read) ||
1165 (search < lowest_read && lowest_read - search <= 32768) ||
1166 (search > highest_read && search - highest_read <= 32768)) {
1167 readahead_tree_block(root, search, blocksize,
1168 btrfs_node_ptr_generation(node, nr));
1169 nread += blocksize;
1170 }
1171 nscan++;
1172 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1173 break;
1174 if(nread > (1024 * 1024) || nscan > 128)
1175 break;
1176
1177 if (search < lowest_read)
1178 lowest_read = search;
1179 if (search > highest_read)
1180 highest_read = search;
1181 }
1182 }
1183
1184 static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock)
1185 {
1186 int i;
1187 int skip_level = level;
1188 int no_skips = 0;
1189 struct extent_buffer *t;
1190
1191 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1192 if (!path->nodes[i])
1193 break;
1194 if (!path->locks[i])
1195 break;
1196 if (!no_skips && path->slots[i] == 0) {
1197 skip_level = i + 1;
1198 continue;
1199 }
1200 if (!no_skips && path->keep_locks) {
1201 u32 nritems;
1202 t = path->nodes[i];
1203 nritems = btrfs_header_nritems(t);
1204 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1205 skip_level = i + 1;
1206 continue;
1207 }
1208 }
1209 if (skip_level < i && i >= lowest_unlock)
1210 no_skips = 1;
1211
1212 t = path->nodes[i];
1213 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1214 btrfs_tree_unlock(t);
1215 path->locks[i] = 0;
1216 }
1217 }
1218 }
1219
1220 /*
1221 * look for key in the tree. path is filled in with nodes along the way
1222 * if key is found, we return zero and you can find the item in the leaf
1223 * level of the path (level 0)
1224 *
1225 * If the key isn't found, the path points to the slot where it should
1226 * be inserted, and 1 is returned. If there are other errors during the
1227 * search a negative error number is returned.
1228 *
1229 * if ins_len > 0, nodes and leaves will be split as we walk down the
1230 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1231 * possible)
1232 */
1233 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1234 *root, struct btrfs_key *key, struct btrfs_path *p, int
1235 ins_len, int cow)
1236 {
1237 struct extent_buffer *b;
1238 struct extent_buffer *tmp;
1239 int slot;
1240 int ret;
1241 int level;
1242 int should_reada = p->reada;
1243 int lowest_unlock = 1;
1244 int blocksize;
1245 u8 lowest_level = 0;
1246 u64 blocknr;
1247 u64 gen;
1248
1249 lowest_level = p->lowest_level;
1250 WARN_ON(lowest_level && ins_len);
1251 WARN_ON(p->nodes[0] != NULL);
1252 WARN_ON(cow && root == root->fs_info->extent_root &&
1253 !mutex_is_locked(&root->fs_info->alloc_mutex));
1254 WARN_ON(root == root->fs_info->chunk_root &&
1255 !mutex_is_locked(&root->fs_info->chunk_mutex));
1256 WARN_ON(root == root->fs_info->dev_root &&
1257 !mutex_is_locked(&root->fs_info->chunk_mutex));
1258 if (ins_len < 0)
1259 lowest_unlock = 2;
1260 again:
1261 if (p->skip_locking)
1262 b = btrfs_root_node(root);
1263 else
1264 b = btrfs_lock_root_node(root);
1265
1266 while (b) {
1267 level = btrfs_header_level(b);
1268 if (cow) {
1269 int wret;
1270 wret = btrfs_cow_block(trans, root, b,
1271 p->nodes[level + 1],
1272 p->slots[level + 1],
1273 &b);
1274 if (wret) {
1275 free_extent_buffer(b);
1276 return wret;
1277 }
1278 }
1279 BUG_ON(!cow && ins_len);
1280 if (level != btrfs_header_level(b))
1281 WARN_ON(1);
1282 level = btrfs_header_level(b);
1283 p->nodes[level] = b;
1284 if (!p->skip_locking)
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 unlock_up(p, level, lowest_unlock);
1317
1318 /* this is only true while dropping a snapshot */
1319 if (level == lowest_level) {
1320 break;
1321 }
1322
1323 blocknr = btrfs_node_blockptr(b, slot);
1324 gen = btrfs_node_ptr_generation(b, slot);
1325 blocksize = btrfs_level_size(root, level - 1);
1326
1327 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1328 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1329 b = tmp;
1330 } else {
1331 /*
1332 * reduce lock contention at high levels
1333 * of the btree by dropping locks before
1334 * we read.
1335 */
1336 if (level > 1) {
1337 btrfs_release_path(NULL, p);
1338 if (tmp)
1339 free_extent_buffer(tmp);
1340 if (should_reada)
1341 reada_for_search(root, p,
1342 level, slot,
1343 key->objectid);
1344
1345 tmp = read_tree_block(root, blocknr,
1346 blocksize, gen);
1347 if (tmp)
1348 free_extent_buffer(tmp);
1349 goto again;
1350 } else {
1351 if (tmp)
1352 free_extent_buffer(tmp);
1353 if (should_reada)
1354 reada_for_search(root, p,
1355 level, slot,
1356 key->objectid);
1357 b = read_node_slot(root, b, slot);
1358 }
1359 }
1360 if (!p->skip_locking)
1361 btrfs_tree_lock(b);
1362 } else {
1363 p->slots[level] = slot;
1364 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1365 sizeof(struct btrfs_item) + ins_len) {
1366 int sret = split_leaf(trans, root, key,
1367 p, ins_len, ret == 0);
1368 BUG_ON(sret > 0);
1369 if (sret)
1370 return sret;
1371 }
1372 unlock_up(p, level, lowest_unlock);
1373 return ret;
1374 }
1375 }
1376 return 1;
1377 }
1378
1379 /*
1380 * adjust the pointers going up the tree, starting at level
1381 * making sure the right key of each node is points to 'key'.
1382 * This is used after shifting pointers to the left, so it stops
1383 * fixing up pointers when a given leaf/node is not in slot 0 of the
1384 * higher levels
1385 *
1386 * If this fails to write a tree block, it returns -1, but continues
1387 * fixing up the blocks in ram so the tree is consistent.
1388 */
1389 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1390 struct btrfs_root *root, struct btrfs_path *path,
1391 struct btrfs_disk_key *key, int level)
1392 {
1393 int i;
1394 int ret = 0;
1395 struct extent_buffer *t;
1396
1397 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1398 int tslot = path->slots[i];
1399 if (!path->nodes[i])
1400 break;
1401 t = path->nodes[i];
1402 btrfs_set_node_key(t, key, tslot);
1403 btrfs_mark_buffer_dirty(path->nodes[i]);
1404 if (tslot != 0)
1405 break;
1406 }
1407 return ret;
1408 }
1409
1410 /*
1411 * try to push data from one node into the next node left in the
1412 * tree.
1413 *
1414 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1415 * error, and > 0 if there was no room in the left hand block.
1416 */
1417 static int push_node_left(struct btrfs_trans_handle *trans,
1418 struct btrfs_root *root, struct extent_buffer *dst,
1419 struct extent_buffer *src, int empty)
1420 {
1421 int push_items = 0;
1422 int src_nritems;
1423 int dst_nritems;
1424 int ret = 0;
1425
1426 src_nritems = btrfs_header_nritems(src);
1427 dst_nritems = btrfs_header_nritems(dst);
1428 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1429 WARN_ON(btrfs_header_generation(src) != trans->transid);
1430 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1431
1432 if (!empty && src_nritems <= 8)
1433 return 1;
1434
1435 if (push_items <= 0) {
1436 return 1;
1437 }
1438
1439 if (empty) {
1440 push_items = min(src_nritems, push_items);
1441 if (push_items < src_nritems) {
1442 /* leave at least 8 pointers in the node if
1443 * we aren't going to empty it
1444 */
1445 if (src_nritems - push_items < 8) {
1446 if (push_items <= 8)
1447 return 1;
1448 push_items -= 8;
1449 }
1450 }
1451 } else
1452 push_items = min(src_nritems - 8, push_items);
1453
1454 copy_extent_buffer(dst, src,
1455 btrfs_node_key_ptr_offset(dst_nritems),
1456 btrfs_node_key_ptr_offset(0),
1457 push_items * sizeof(struct btrfs_key_ptr));
1458
1459 if (push_items < src_nritems) {
1460 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1461 btrfs_node_key_ptr_offset(push_items),
1462 (src_nritems - push_items) *
1463 sizeof(struct btrfs_key_ptr));
1464 }
1465 btrfs_set_header_nritems(src, src_nritems - push_items);
1466 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1467 btrfs_mark_buffer_dirty(src);
1468 btrfs_mark_buffer_dirty(dst);
1469 return ret;
1470 }
1471
1472 /*
1473 * try to push data from one node into the next node right in the
1474 * tree.
1475 *
1476 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1477 * error, and > 0 if there was no room in the right hand block.
1478 *
1479 * this will only push up to 1/2 the contents of the left node over
1480 */
1481 static int balance_node_right(struct btrfs_trans_handle *trans,
1482 struct btrfs_root *root,
1483 struct extent_buffer *dst,
1484 struct extent_buffer *src)
1485 {
1486 int push_items = 0;
1487 int max_push;
1488 int src_nritems;
1489 int dst_nritems;
1490 int ret = 0;
1491
1492 WARN_ON(btrfs_header_generation(src) != trans->transid);
1493 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1494
1495 src_nritems = btrfs_header_nritems(src);
1496 dst_nritems = btrfs_header_nritems(dst);
1497 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1498 if (push_items <= 0) {
1499 return 1;
1500 }
1501
1502 if (src_nritems < 4) {
1503 return 1;
1504 }
1505
1506 max_push = src_nritems / 2 + 1;
1507 /* don't try to empty the node */
1508 if (max_push >= src_nritems) {
1509 return 1;
1510 }
1511
1512 if (max_push < push_items)
1513 push_items = max_push;
1514
1515 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1516 btrfs_node_key_ptr_offset(0),
1517 (dst_nritems) *
1518 sizeof(struct btrfs_key_ptr));
1519
1520 copy_extent_buffer(dst, src,
1521 btrfs_node_key_ptr_offset(0),
1522 btrfs_node_key_ptr_offset(src_nritems - push_items),
1523 push_items * sizeof(struct btrfs_key_ptr));
1524
1525 btrfs_set_header_nritems(src, src_nritems - push_items);
1526 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1527
1528 btrfs_mark_buffer_dirty(src);
1529 btrfs_mark_buffer_dirty(dst);
1530 return ret;
1531 }
1532
1533 /*
1534 * helper function to insert a new root level in the tree.
1535 * A new node is allocated, and a single item is inserted to
1536 * point to the existing root
1537 *
1538 * returns zero on success or < 0 on failure.
1539 */
1540 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1541 struct btrfs_root *root,
1542 struct btrfs_path *path, int level)
1543 {
1544 u64 root_gen;
1545 u64 lower_gen;
1546 struct extent_buffer *lower;
1547 struct extent_buffer *c;
1548 struct extent_buffer *old;
1549 struct btrfs_disk_key lower_key;
1550
1551 BUG_ON(path->nodes[level]);
1552 BUG_ON(path->nodes[level-1] != root->node);
1553
1554 if (root->ref_cows)
1555 root_gen = trans->transid;
1556 else
1557 root_gen = 0;
1558
1559 lower = path->nodes[level-1];
1560 if (level == 1)
1561 btrfs_item_key(lower, &lower_key, 0);
1562 else
1563 btrfs_node_key(lower, &lower_key, 0);
1564
1565 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1566 root->root_key.objectid,
1567 root_gen, lower_key.objectid, level,
1568 root->node->start, 0);
1569 if (IS_ERR(c))
1570 return PTR_ERR(c);
1571
1572 memset_extent_buffer(c, 0, 0, root->nodesize);
1573 btrfs_set_header_nritems(c, 1);
1574 btrfs_set_header_level(c, level);
1575 btrfs_set_header_bytenr(c, c->start);
1576 btrfs_set_header_generation(c, trans->transid);
1577 btrfs_set_header_owner(c, root->root_key.objectid);
1578
1579 write_extent_buffer(c, root->fs_info->fsid,
1580 (unsigned long)btrfs_header_fsid(c),
1581 BTRFS_FSID_SIZE);
1582
1583 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1584 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1585 BTRFS_UUID_SIZE);
1586
1587 btrfs_set_node_key(c, &lower_key, 0);
1588 btrfs_set_node_blockptr(c, 0, lower->start);
1589 lower_gen = btrfs_header_generation(lower);
1590 WARN_ON(lower_gen == 0);
1591
1592 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1593
1594 btrfs_mark_buffer_dirty(c);
1595
1596 spin_lock(&root->node_lock);
1597 old = root->node;
1598 root->node = c;
1599 spin_unlock(&root->node_lock);
1600
1601 /* the super has an extra ref to root->node */
1602 free_extent_buffer(old);
1603
1604 add_root_to_dirty_list(root);
1605 extent_buffer_get(c);
1606 path->nodes[level] = c;
1607 path->locks[level] = 1;
1608 path->slots[level] = 0;
1609
1610 if (root->ref_cows && lower_gen != trans->transid) {
1611 struct btrfs_path *back_path = btrfs_alloc_path();
1612 int ret;
1613 mutex_lock(&root->fs_info->alloc_mutex);
1614 ret = btrfs_insert_extent_backref(trans,
1615 root->fs_info->extent_root,
1616 path, lower->start,
1617 root->root_key.objectid,
1618 trans->transid, 0, 0);
1619 BUG_ON(ret);
1620 mutex_unlock(&root->fs_info->alloc_mutex);
1621 btrfs_free_path(back_path);
1622 }
1623 return 0;
1624 }
1625
1626 /*
1627 * worker function to insert a single pointer in a node.
1628 * the node should have enough room for the pointer already
1629 *
1630 * slot and level indicate where you want the key to go, and
1631 * blocknr is the block the key points to.
1632 *
1633 * returns zero on success and < 0 on any error
1634 */
1635 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1636 *root, struct btrfs_path *path, struct btrfs_disk_key
1637 *key, u64 bytenr, int slot, int level)
1638 {
1639 struct extent_buffer *lower;
1640 int nritems;
1641
1642 BUG_ON(!path->nodes[level]);
1643 lower = path->nodes[level];
1644 nritems = btrfs_header_nritems(lower);
1645 if (slot > nritems)
1646 BUG();
1647 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1648 BUG();
1649 if (slot != nritems) {
1650 memmove_extent_buffer(lower,
1651 btrfs_node_key_ptr_offset(slot + 1),
1652 btrfs_node_key_ptr_offset(slot),
1653 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1654 }
1655 btrfs_set_node_key(lower, key, slot);
1656 btrfs_set_node_blockptr(lower, slot, bytenr);
1657 WARN_ON(trans->transid == 0);
1658 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1659 btrfs_set_header_nritems(lower, nritems + 1);
1660 btrfs_mark_buffer_dirty(lower);
1661 return 0;
1662 }
1663
1664 /*
1665 * split the node at the specified level in path in two.
1666 * The path is corrected to point to the appropriate node after the split
1667 *
1668 * Before splitting this tries to make some room in the node by pushing
1669 * left and right, if either one works, it returns right away.
1670 *
1671 * returns 0 on success and < 0 on failure
1672 */
1673 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1674 *root, struct btrfs_path *path, int level)
1675 {
1676 u64 root_gen;
1677 struct extent_buffer *c;
1678 struct extent_buffer *split;
1679 struct btrfs_disk_key disk_key;
1680 int mid;
1681 int ret;
1682 int wret;
1683 u32 c_nritems;
1684
1685 c = path->nodes[level];
1686 WARN_ON(btrfs_header_generation(c) != trans->transid);
1687 if (c == root->node) {
1688 /* trying to split the root, lets make a new one */
1689 ret = insert_new_root(trans, root, path, level + 1);
1690 if (ret)
1691 return ret;
1692 } else {
1693 ret = push_nodes_for_insert(trans, root, path, level);
1694 c = path->nodes[level];
1695 if (!ret && btrfs_header_nritems(c) <
1696 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1697 return 0;
1698 if (ret < 0)
1699 return ret;
1700 }
1701
1702 c_nritems = btrfs_header_nritems(c);
1703 if (root->ref_cows)
1704 root_gen = trans->transid;
1705 else
1706 root_gen = 0;
1707
1708 btrfs_node_key(c, &disk_key, 0);
1709 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1710 root->root_key.objectid,
1711 root_gen,
1712 btrfs_disk_key_objectid(&disk_key),
1713 level, c->start, 0);
1714 if (IS_ERR(split))
1715 return PTR_ERR(split);
1716
1717 btrfs_set_header_flags(split, btrfs_header_flags(c));
1718 btrfs_set_header_level(split, btrfs_header_level(c));
1719 btrfs_set_header_bytenr(split, split->start);
1720 btrfs_set_header_generation(split, trans->transid);
1721 btrfs_set_header_owner(split, root->root_key.objectid);
1722 btrfs_set_header_flags(split, 0);
1723 write_extent_buffer(split, root->fs_info->fsid,
1724 (unsigned long)btrfs_header_fsid(split),
1725 BTRFS_FSID_SIZE);
1726 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1727 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1728 BTRFS_UUID_SIZE);
1729
1730 mid = (c_nritems + 1) / 2;
1731
1732 copy_extent_buffer(split, c,
1733 btrfs_node_key_ptr_offset(0),
1734 btrfs_node_key_ptr_offset(mid),
1735 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1736 btrfs_set_header_nritems(split, c_nritems - mid);
1737 btrfs_set_header_nritems(c, mid);
1738 ret = 0;
1739
1740 btrfs_mark_buffer_dirty(c);
1741 btrfs_mark_buffer_dirty(split);
1742
1743 btrfs_node_key(split, &disk_key, 0);
1744 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1745 path->slots[level + 1] + 1,
1746 level + 1);
1747 if (wret)
1748 ret = wret;
1749
1750 if (path->slots[level] >= mid) {
1751 path->slots[level] -= mid;
1752 btrfs_tree_unlock(c);
1753 free_extent_buffer(c);
1754 path->nodes[level] = split;
1755 path->slots[level + 1] += 1;
1756 } else {
1757 btrfs_tree_unlock(split);
1758 free_extent_buffer(split);
1759 }
1760 return ret;
1761 }
1762
1763 /*
1764 * how many bytes are required to store the items in a leaf. start
1765 * and nr indicate which items in the leaf to check. This totals up the
1766 * space used both by the item structs and the item data
1767 */
1768 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1769 {
1770 int data_len;
1771 int nritems = btrfs_header_nritems(l);
1772 int end = min(nritems, start + nr) - 1;
1773
1774 if (!nr)
1775 return 0;
1776 data_len = btrfs_item_end_nr(l, start);
1777 data_len = data_len - btrfs_item_offset_nr(l, end);
1778 data_len += sizeof(struct btrfs_item) * nr;
1779 WARN_ON(data_len < 0);
1780 return data_len;
1781 }
1782
1783 /*
1784 * The space between the end of the leaf items and
1785 * the start of the leaf data. IOW, how much room
1786 * the leaf has left for both items and data
1787 */
1788 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1789 {
1790 int nritems = btrfs_header_nritems(leaf);
1791 int ret;
1792 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1793 if (ret < 0) {
1794 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1795 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1796 leaf_space_used(leaf, 0, nritems), nritems);
1797 }
1798 return ret;
1799 }
1800
1801 /*
1802 * push some data in the path leaf to the right, trying to free up at
1803 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1804 *
1805 * returns 1 if the push failed because the other node didn't have enough
1806 * room, 0 if everything worked out and < 0 if there were major errors.
1807 */
1808 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1809 *root, struct btrfs_path *path, int data_size,
1810 int empty)
1811 {
1812 struct extent_buffer *left = path->nodes[0];
1813 struct extent_buffer *right;
1814 struct extent_buffer *upper;
1815 struct btrfs_disk_key disk_key;
1816 int slot;
1817 u32 i;
1818 int free_space;
1819 int push_space = 0;
1820 int push_items = 0;
1821 struct btrfs_item *item;
1822 u32 left_nritems;
1823 u32 nr;
1824 u32 right_nritems;
1825 u32 data_end;
1826 u32 this_item_size;
1827 int ret;
1828
1829 slot = path->slots[1];
1830 if (!path->nodes[1]) {
1831 return 1;
1832 }
1833 upper = path->nodes[1];
1834 if (slot >= btrfs_header_nritems(upper) - 1)
1835 return 1;
1836
1837 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
1838
1839 right = read_node_slot(root, upper, slot + 1);
1840 btrfs_tree_lock(right);
1841 free_space = btrfs_leaf_free_space(root, right);
1842 if (free_space < data_size + sizeof(struct btrfs_item))
1843 goto out_unlock;
1844
1845 /* cow and double check */
1846 ret = btrfs_cow_block(trans, root, right, upper,
1847 slot + 1, &right);
1848 if (ret)
1849 goto out_unlock;
1850
1851 free_space = btrfs_leaf_free_space(root, right);
1852 if (free_space < data_size + sizeof(struct btrfs_item))
1853 goto out_unlock;
1854
1855 left_nritems = btrfs_header_nritems(left);
1856 if (left_nritems == 0)
1857 goto out_unlock;
1858
1859 if (empty)
1860 nr = 0;
1861 else
1862 nr = 1;
1863
1864 i = left_nritems - 1;
1865 while (i >= nr) {
1866 item = btrfs_item_nr(left, i);
1867
1868 if (path->slots[0] == i)
1869 push_space += data_size + sizeof(*item);
1870
1871 if (!left->map_token) {
1872 map_extent_buffer(left, (unsigned long)item,
1873 sizeof(struct btrfs_item),
1874 &left->map_token, &left->kaddr,
1875 &left->map_start, &left->map_len,
1876 KM_USER1);
1877 }
1878
1879 this_item_size = btrfs_item_size(left, item);
1880 if (this_item_size + sizeof(*item) + push_space > free_space)
1881 break;
1882 push_items++;
1883 push_space += this_item_size + sizeof(*item);
1884 if (i == 0)
1885 break;
1886 i--;
1887 }
1888 if (left->map_token) {
1889 unmap_extent_buffer(left, left->map_token, KM_USER1);
1890 left->map_token = NULL;
1891 }
1892
1893 if (push_items == 0)
1894 goto out_unlock;
1895
1896 if (!empty && push_items == left_nritems)
1897 WARN_ON(1);
1898
1899 /* push left to right */
1900 right_nritems = btrfs_header_nritems(right);
1901
1902 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1903 push_space -= leaf_data_end(root, left);
1904
1905 /* make room in the right data area */
1906 data_end = leaf_data_end(root, right);
1907 memmove_extent_buffer(right,
1908 btrfs_leaf_data(right) + data_end - push_space,
1909 btrfs_leaf_data(right) + data_end,
1910 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1911
1912 /* copy from the left data area */
1913 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1914 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1915 btrfs_leaf_data(left) + leaf_data_end(root, left),
1916 push_space);
1917
1918 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1919 btrfs_item_nr_offset(0),
1920 right_nritems * sizeof(struct btrfs_item));
1921
1922 /* copy the items from left to right */
1923 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1924 btrfs_item_nr_offset(left_nritems - push_items),
1925 push_items * sizeof(struct btrfs_item));
1926
1927 /* update the item pointers */
1928 right_nritems += push_items;
1929 btrfs_set_header_nritems(right, right_nritems);
1930 push_space = BTRFS_LEAF_DATA_SIZE(root);
1931 for (i = 0; i < right_nritems; i++) {
1932 item = btrfs_item_nr(right, i);
1933 if (!right->map_token) {
1934 map_extent_buffer(right, (unsigned long)item,
1935 sizeof(struct btrfs_item),
1936 &right->map_token, &right->kaddr,
1937 &right->map_start, &right->map_len,
1938 KM_USER1);
1939 }
1940 push_space -= btrfs_item_size(right, item);
1941 btrfs_set_item_offset(right, item, push_space);
1942 }
1943
1944 if (right->map_token) {
1945 unmap_extent_buffer(right, right->map_token, KM_USER1);
1946 right->map_token = NULL;
1947 }
1948 left_nritems -= push_items;
1949 btrfs_set_header_nritems(left, left_nritems);
1950
1951 if (left_nritems)
1952 btrfs_mark_buffer_dirty(left);
1953 btrfs_mark_buffer_dirty(right);
1954
1955 btrfs_item_key(right, &disk_key, 0);
1956 btrfs_set_node_key(upper, &disk_key, slot + 1);
1957 btrfs_mark_buffer_dirty(upper);
1958
1959 /* then fixup the leaf pointer in the path */
1960 if (path->slots[0] >= left_nritems) {
1961 path->slots[0] -= left_nritems;
1962 if (btrfs_header_nritems(path->nodes[0]) == 0)
1963 clean_tree_block(trans, root, path->nodes[0]);
1964 btrfs_tree_unlock(path->nodes[0]);
1965 free_extent_buffer(path->nodes[0]);
1966 path->nodes[0] = right;
1967 path->slots[1] += 1;
1968 } else {
1969 btrfs_tree_unlock(right);
1970 free_extent_buffer(right);
1971 }
1972 return 0;
1973
1974 out_unlock:
1975 btrfs_tree_unlock(right);
1976 free_extent_buffer(right);
1977 return 1;
1978 }
1979
1980 /*
1981 * push some data in the path leaf to the left, trying to free up at
1982 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1983 */
1984 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1985 *root, struct btrfs_path *path, int data_size,
1986 int empty)
1987 {
1988 struct btrfs_disk_key disk_key;
1989 struct extent_buffer *right = path->nodes[0];
1990 struct extent_buffer *left;
1991 int slot;
1992 int i;
1993 int free_space;
1994 int push_space = 0;
1995 int push_items = 0;
1996 struct btrfs_item *item;
1997 u32 old_left_nritems;
1998 u32 right_nritems;
1999 u32 nr;
2000 int ret = 0;
2001 int wret;
2002 u32 this_item_size;
2003 u32 old_left_item_size;
2004
2005 slot = path->slots[1];
2006 if (slot == 0)
2007 return 1;
2008 if (!path->nodes[1])
2009 return 1;
2010
2011 right_nritems = btrfs_header_nritems(right);
2012 if (right_nritems == 0) {
2013 return 1;
2014 }
2015
2016 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2017
2018 left = read_node_slot(root, path->nodes[1], slot - 1);
2019 btrfs_tree_lock(left);
2020 free_space = btrfs_leaf_free_space(root, left);
2021 if (free_space < data_size + sizeof(struct btrfs_item)) {
2022 ret = 1;
2023 goto out;
2024 }
2025
2026 /* cow and double check */
2027 ret = btrfs_cow_block(trans, root, left,
2028 path->nodes[1], slot - 1, &left);
2029 if (ret) {
2030 /* we hit -ENOSPC, but it isn't fatal here */
2031 ret = 1;
2032 goto out;
2033 }
2034
2035 free_space = btrfs_leaf_free_space(root, left);
2036 if (free_space < data_size + sizeof(struct btrfs_item)) {
2037 ret = 1;
2038 goto out;
2039 }
2040
2041 if (empty)
2042 nr = right_nritems;
2043 else
2044 nr = right_nritems - 1;
2045
2046 for (i = 0; i < nr; i++) {
2047 item = btrfs_item_nr(right, i);
2048 if (!right->map_token) {
2049 map_extent_buffer(right, (unsigned long)item,
2050 sizeof(struct btrfs_item),
2051 &right->map_token, &right->kaddr,
2052 &right->map_start, &right->map_len,
2053 KM_USER1);
2054 }
2055
2056 if (path->slots[0] == i)
2057 push_space += data_size + sizeof(*item);
2058
2059 this_item_size = btrfs_item_size(right, item);
2060 if (this_item_size + sizeof(*item) + push_space > free_space)
2061 break;
2062
2063 push_items++;
2064 push_space += this_item_size + sizeof(*item);
2065 }
2066
2067 if (right->map_token) {
2068 unmap_extent_buffer(right, right->map_token, KM_USER1);
2069 right->map_token = NULL;
2070 }
2071
2072 if (push_items == 0) {
2073 ret = 1;
2074 goto out;
2075 }
2076 if (!empty && push_items == btrfs_header_nritems(right))
2077 WARN_ON(1);
2078
2079 /* push data from right to left */
2080 copy_extent_buffer(left, right,
2081 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2082 btrfs_item_nr_offset(0),
2083 push_items * sizeof(struct btrfs_item));
2084
2085 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2086 btrfs_item_offset_nr(right, push_items -1);
2087
2088 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2089 leaf_data_end(root, left) - push_space,
2090 btrfs_leaf_data(right) +
2091 btrfs_item_offset_nr(right, push_items - 1),
2092 push_space);
2093 old_left_nritems = btrfs_header_nritems(left);
2094 BUG_ON(old_left_nritems < 0);
2095
2096 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2097 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2098 u32 ioff;
2099
2100 item = btrfs_item_nr(left, i);
2101 if (!left->map_token) {
2102 map_extent_buffer(left, (unsigned long)item,
2103 sizeof(struct btrfs_item),
2104 &left->map_token, &left->kaddr,
2105 &left->map_start, &left->map_len,
2106 KM_USER1);
2107 }
2108
2109 ioff = btrfs_item_offset(left, item);
2110 btrfs_set_item_offset(left, item,
2111 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2112 }
2113 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2114 if (left->map_token) {
2115 unmap_extent_buffer(left, left->map_token, KM_USER1);
2116 left->map_token = NULL;
2117 }
2118
2119 /* fixup right node */
2120 if (push_items > right_nritems) {
2121 printk("push items %d nr %u\n", push_items, right_nritems);
2122 WARN_ON(1);
2123 }
2124
2125 if (push_items < right_nritems) {
2126 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2127 leaf_data_end(root, right);
2128 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2129 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2130 btrfs_leaf_data(right) +
2131 leaf_data_end(root, right), push_space);
2132
2133 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2134 btrfs_item_nr_offset(push_items),
2135 (btrfs_header_nritems(right) - push_items) *
2136 sizeof(struct btrfs_item));
2137 }
2138 right_nritems -= push_items;
2139 btrfs_set_header_nritems(right, right_nritems);
2140 push_space = BTRFS_LEAF_DATA_SIZE(root);
2141 for (i = 0; i < right_nritems; i++) {
2142 item = btrfs_item_nr(right, i);
2143
2144 if (!right->map_token) {
2145 map_extent_buffer(right, (unsigned long)item,
2146 sizeof(struct btrfs_item),
2147 &right->map_token, &right->kaddr,
2148 &right->map_start, &right->map_len,
2149 KM_USER1);
2150 }
2151
2152 push_space = push_space - btrfs_item_size(right, item);
2153 btrfs_set_item_offset(right, item, push_space);
2154 }
2155 if (right->map_token) {
2156 unmap_extent_buffer(right, right->map_token, KM_USER1);
2157 right->map_token = NULL;
2158 }
2159
2160 btrfs_mark_buffer_dirty(left);
2161 if (right_nritems)
2162 btrfs_mark_buffer_dirty(right);
2163
2164 btrfs_item_key(right, &disk_key, 0);
2165 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2166 if (wret)
2167 ret = wret;
2168
2169 /* then fixup the leaf pointer in the path */
2170 if (path->slots[0] < push_items) {
2171 path->slots[0] += old_left_nritems;
2172 if (btrfs_header_nritems(path->nodes[0]) == 0)
2173 clean_tree_block(trans, root, path->nodes[0]);
2174 btrfs_tree_unlock(path->nodes[0]);
2175 free_extent_buffer(path->nodes[0]);
2176 path->nodes[0] = left;
2177 path->slots[1] -= 1;
2178 } else {
2179 btrfs_tree_unlock(left);
2180 free_extent_buffer(left);
2181 path->slots[0] -= push_items;
2182 }
2183 BUG_ON(path->slots[0] < 0);
2184 return ret;
2185 out:
2186 btrfs_tree_unlock(left);
2187 free_extent_buffer(left);
2188 return ret;
2189 }
2190
2191 /*
2192 * split the path's leaf in two, making sure there is at least data_size
2193 * available for the resulting leaf level of the path.
2194 *
2195 * returns 0 if all went well and < 0 on failure.
2196 */
2197 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
2198 *root, struct btrfs_key *ins_key,
2199 struct btrfs_path *path, int data_size, int extend)
2200 {
2201 u64 root_gen;
2202 struct extent_buffer *l;
2203 u32 nritems;
2204 int mid;
2205 int slot;
2206 struct extent_buffer *right;
2207 int space_needed = data_size + sizeof(struct btrfs_item);
2208 int data_copy_size;
2209 int rt_data_off;
2210 int i;
2211 int ret = 0;
2212 int wret;
2213 int double_split;
2214 int num_doubles = 0;
2215 struct btrfs_disk_key disk_key;
2216
2217 if (extend)
2218 space_needed = data_size;
2219
2220 if (root->ref_cows)
2221 root_gen = trans->transid;
2222 else
2223 root_gen = 0;
2224
2225 /* first try to make some room by pushing left and right */
2226 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2227 wret = push_leaf_right(trans, root, path, data_size, 0);
2228 if (wret < 0) {
2229 return wret;
2230 }
2231 if (wret) {
2232 wret = push_leaf_left(trans, root, path, data_size, 0);
2233 if (wret < 0)
2234 return wret;
2235 }
2236 l = path->nodes[0];
2237
2238 /* did the pushes work? */
2239 if (btrfs_leaf_free_space(root, l) >= space_needed)
2240 return 0;
2241 }
2242
2243 if (!path->nodes[1]) {
2244 ret = insert_new_root(trans, root, path, 1);
2245 if (ret)
2246 return ret;
2247 }
2248 again:
2249 double_split = 0;
2250 l = path->nodes[0];
2251 slot = path->slots[0];
2252 nritems = btrfs_header_nritems(l);
2253 mid = (nritems + 1)/ 2;
2254
2255 btrfs_item_key(l, &disk_key, 0);
2256
2257 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2258 root->root_key.objectid,
2259 root_gen, disk_key.objectid, 0,
2260 l->start, 0);
2261 if (IS_ERR(right)) {
2262 BUG_ON(1);
2263 return PTR_ERR(right);
2264 }
2265
2266 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2267 btrfs_set_header_bytenr(right, right->start);
2268 btrfs_set_header_generation(right, trans->transid);
2269 btrfs_set_header_owner(right, root->root_key.objectid);
2270 btrfs_set_header_level(right, 0);
2271 write_extent_buffer(right, root->fs_info->fsid,
2272 (unsigned long)btrfs_header_fsid(right),
2273 BTRFS_FSID_SIZE);
2274
2275 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2276 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2277 BTRFS_UUID_SIZE);
2278 if (mid <= slot) {
2279 if (nritems == 1 ||
2280 leaf_space_used(l, mid, nritems - mid) + space_needed >
2281 BTRFS_LEAF_DATA_SIZE(root)) {
2282 if (slot >= nritems) {
2283 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2284 btrfs_set_header_nritems(right, 0);
2285 wret = insert_ptr(trans, root, path,
2286 &disk_key, right->start,
2287 path->slots[1] + 1, 1);
2288 if (wret)
2289 ret = wret;
2290
2291 btrfs_tree_unlock(path->nodes[0]);
2292 free_extent_buffer(path->nodes[0]);
2293 path->nodes[0] = right;
2294 path->slots[0] = 0;
2295 path->slots[1] += 1;
2296 btrfs_mark_buffer_dirty(right);
2297 return ret;
2298 }
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 } else {
2307 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2308 BTRFS_LEAF_DATA_SIZE(root)) {
2309 if (!extend && slot == 0) {
2310 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2311 btrfs_set_header_nritems(right, 0);
2312 wret = insert_ptr(trans, root, path,
2313 &disk_key,
2314 right->start,
2315 path->slots[1], 1);
2316 if (wret)
2317 ret = wret;
2318 btrfs_tree_unlock(path->nodes[0]);
2319 free_extent_buffer(path->nodes[0]);
2320 path->nodes[0] = right;
2321 path->slots[0] = 0;
2322 if (path->slots[1] == 0) {
2323 wret = fixup_low_keys(trans, root,
2324 path, &disk_key, 1);
2325 if (wret)
2326 ret = wret;
2327 }
2328 btrfs_mark_buffer_dirty(right);
2329 return ret;
2330 } else if (extend && slot == 0) {
2331 mid = 1;
2332 } else {
2333 mid = slot;
2334 if (mid != nritems &&
2335 leaf_space_used(l, mid, nritems - mid) +
2336 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2337 double_split = 1;
2338 }
2339 }
2340 }
2341 }
2342 nritems = nritems - mid;
2343 btrfs_set_header_nritems(right, nritems);
2344 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2345
2346 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2347 btrfs_item_nr_offset(mid),
2348 nritems * sizeof(struct btrfs_item));
2349
2350 copy_extent_buffer(right, l,
2351 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2352 data_copy_size, btrfs_leaf_data(l) +
2353 leaf_data_end(root, l), data_copy_size);
2354
2355 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2356 btrfs_item_end_nr(l, mid);
2357
2358 for (i = 0; i < nritems; i++) {
2359 struct btrfs_item *item = btrfs_item_nr(right, i);
2360 u32 ioff;
2361
2362 if (!right->map_token) {
2363 map_extent_buffer(right, (unsigned long)item,
2364 sizeof(struct btrfs_item),
2365 &right->map_token, &right->kaddr,
2366 &right->map_start, &right->map_len,
2367 KM_USER1);
2368 }
2369
2370 ioff = btrfs_item_offset(right, item);
2371 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2372 }
2373
2374 if (right->map_token) {
2375 unmap_extent_buffer(right, right->map_token, KM_USER1);
2376 right->map_token = NULL;
2377 }
2378
2379 btrfs_set_header_nritems(l, mid);
2380 ret = 0;
2381 btrfs_item_key(right, &disk_key, 0);
2382 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2383 path->slots[1] + 1, 1);
2384 if (wret)
2385 ret = wret;
2386
2387 btrfs_mark_buffer_dirty(right);
2388 btrfs_mark_buffer_dirty(l);
2389 BUG_ON(path->slots[0] != slot);
2390
2391 if (mid <= slot) {
2392 btrfs_tree_unlock(path->nodes[0]);
2393 free_extent_buffer(path->nodes[0]);
2394 path->nodes[0] = right;
2395 path->slots[0] -= mid;
2396 path->slots[1] += 1;
2397 } else {
2398 btrfs_tree_unlock(right);
2399 free_extent_buffer(right);
2400 }
2401
2402 BUG_ON(path->slots[0] < 0);
2403
2404 if (double_split) {
2405 BUG_ON(num_doubles != 0);
2406 num_doubles++;
2407 goto again;
2408 }
2409 return ret;
2410 }
2411
2412 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2413 struct btrfs_root *root,
2414 struct btrfs_path *path,
2415 u32 new_size, int from_end)
2416 {
2417 int ret = 0;
2418 int slot;
2419 int slot_orig;
2420 struct extent_buffer *leaf;
2421 struct btrfs_item *item;
2422 u32 nritems;
2423 unsigned int data_end;
2424 unsigned int old_data_start;
2425 unsigned int old_size;
2426 unsigned int size_diff;
2427 int i;
2428
2429 slot_orig = path->slots[0];
2430 leaf = path->nodes[0];
2431 slot = path->slots[0];
2432
2433 old_size = btrfs_item_size_nr(leaf, slot);
2434 if (old_size == new_size)
2435 return 0;
2436
2437 nritems = btrfs_header_nritems(leaf);
2438 data_end = leaf_data_end(root, leaf);
2439
2440 old_data_start = btrfs_item_offset_nr(leaf, slot);
2441
2442 size_diff = old_size - new_size;
2443
2444 BUG_ON(slot < 0);
2445 BUG_ON(slot >= nritems);
2446
2447 /*
2448 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2449 */
2450 /* first correct the data pointers */
2451 for (i = slot; i < nritems; i++) {
2452 u32 ioff;
2453 item = btrfs_item_nr(leaf, i);
2454
2455 if (!leaf->map_token) {
2456 map_extent_buffer(leaf, (unsigned long)item,
2457 sizeof(struct btrfs_item),
2458 &leaf->map_token, &leaf->kaddr,
2459 &leaf->map_start, &leaf->map_len,
2460 KM_USER1);
2461 }
2462
2463 ioff = btrfs_item_offset(leaf, item);
2464 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2465 }
2466
2467 if (leaf->map_token) {
2468 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2469 leaf->map_token = NULL;
2470 }
2471
2472 /* shift the data */
2473 if (from_end) {
2474 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2475 data_end + size_diff, btrfs_leaf_data(leaf) +
2476 data_end, old_data_start + new_size - data_end);
2477 } else {
2478 struct btrfs_disk_key disk_key;
2479 u64 offset;
2480
2481 btrfs_item_key(leaf, &disk_key, slot);
2482
2483 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2484 unsigned long ptr;
2485 struct btrfs_file_extent_item *fi;
2486
2487 fi = btrfs_item_ptr(leaf, slot,
2488 struct btrfs_file_extent_item);
2489 fi = (struct btrfs_file_extent_item *)(
2490 (unsigned long)fi - size_diff);
2491
2492 if (btrfs_file_extent_type(leaf, fi) ==
2493 BTRFS_FILE_EXTENT_INLINE) {
2494 ptr = btrfs_item_ptr_offset(leaf, slot);
2495 memmove_extent_buffer(leaf, ptr,
2496 (unsigned long)fi,
2497 offsetof(struct btrfs_file_extent_item,
2498 disk_bytenr));
2499 }
2500 }
2501
2502 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2503 data_end + size_diff, btrfs_leaf_data(leaf) +
2504 data_end, old_data_start - data_end);
2505
2506 offset = btrfs_disk_key_offset(&disk_key);
2507 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2508 btrfs_set_item_key(leaf, &disk_key, slot);
2509 if (slot == 0)
2510 fixup_low_keys(trans, root, path, &disk_key, 1);
2511 }
2512
2513 item = btrfs_item_nr(leaf, slot);
2514 btrfs_set_item_size(leaf, item, new_size);
2515 btrfs_mark_buffer_dirty(leaf);
2516
2517 ret = 0;
2518 if (btrfs_leaf_free_space(root, leaf) < 0) {
2519 btrfs_print_leaf(root, leaf);
2520 BUG();
2521 }
2522 return ret;
2523 }
2524
2525 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2526 struct btrfs_root *root, struct btrfs_path *path,
2527 u32 data_size)
2528 {
2529 int ret = 0;
2530 int slot;
2531 int slot_orig;
2532 struct extent_buffer *leaf;
2533 struct btrfs_item *item;
2534 u32 nritems;
2535 unsigned int data_end;
2536 unsigned int old_data;
2537 unsigned int old_size;
2538 int i;
2539
2540 slot_orig = path->slots[0];
2541 leaf = path->nodes[0];
2542
2543 nritems = btrfs_header_nritems(leaf);
2544 data_end = leaf_data_end(root, leaf);
2545
2546 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2547 btrfs_print_leaf(root, leaf);
2548 BUG();
2549 }
2550 slot = path->slots[0];
2551 old_data = btrfs_item_end_nr(leaf, slot);
2552
2553 BUG_ON(slot < 0);
2554 if (slot >= nritems) {
2555 btrfs_print_leaf(root, leaf);
2556 printk("slot %d too large, nritems %d\n", slot, nritems);
2557 BUG_ON(1);
2558 }
2559
2560 /*
2561 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2562 */
2563 /* first correct the data pointers */
2564 for (i = slot; i < nritems; i++) {
2565 u32 ioff;
2566 item = btrfs_item_nr(leaf, i);
2567
2568 if (!leaf->map_token) {
2569 map_extent_buffer(leaf, (unsigned long)item,
2570 sizeof(struct btrfs_item),
2571 &leaf->map_token, &leaf->kaddr,
2572 &leaf->map_start, &leaf->map_len,
2573 KM_USER1);
2574 }
2575 ioff = btrfs_item_offset(leaf, item);
2576 btrfs_set_item_offset(leaf, item, ioff - data_size);
2577 }
2578
2579 if (leaf->map_token) {
2580 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2581 leaf->map_token = NULL;
2582 }
2583
2584 /* shift the data */
2585 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2586 data_end - data_size, btrfs_leaf_data(leaf) +
2587 data_end, old_data - data_end);
2588
2589 data_end = old_data;
2590 old_size = btrfs_item_size_nr(leaf, slot);
2591 item = btrfs_item_nr(leaf, slot);
2592 btrfs_set_item_size(leaf, item, old_size + data_size);
2593 btrfs_mark_buffer_dirty(leaf);
2594
2595 ret = 0;
2596 if (btrfs_leaf_free_space(root, leaf) < 0) {
2597 btrfs_print_leaf(root, leaf);
2598 BUG();
2599 }
2600 return ret;
2601 }
2602
2603 /*
2604 * Given a key and some data, insert an item into the tree.
2605 * This does all the path init required, making room in the tree if needed.
2606 */
2607 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2608 struct btrfs_root *root,
2609 struct btrfs_path *path,
2610 struct btrfs_key *cpu_key, u32 *data_size,
2611 int nr)
2612 {
2613 struct extent_buffer *leaf;
2614 struct btrfs_item *item;
2615 int ret = 0;
2616 int slot;
2617 int slot_orig;
2618 int i;
2619 u32 nritems;
2620 u32 total_size = 0;
2621 u32 total_data = 0;
2622 unsigned int data_end;
2623 struct btrfs_disk_key disk_key;
2624
2625 for (i = 0; i < nr; i++) {
2626 total_data += data_size[i];
2627 }
2628
2629 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2630 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2631 if (ret == 0) {
2632 return -EEXIST;
2633 }
2634 if (ret < 0)
2635 goto out;
2636
2637 slot_orig = path->slots[0];
2638 leaf = path->nodes[0];
2639
2640 nritems = btrfs_header_nritems(leaf);
2641 data_end = leaf_data_end(root, leaf);
2642
2643 if (btrfs_leaf_free_space(root, leaf) <
2644 sizeof(struct btrfs_item) + total_size) {
2645 btrfs_print_leaf(root, leaf);
2646 printk("not enough freespace need %u have %d\n",
2647 total_size, btrfs_leaf_free_space(root, leaf));
2648 BUG();
2649 }
2650
2651 slot = path->slots[0];
2652 BUG_ON(slot < 0);
2653
2654 if (slot != nritems) {
2655 int i;
2656 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2657
2658 if (old_data < data_end) {
2659 btrfs_print_leaf(root, leaf);
2660 printk("slot %d old_data %d data_end %d\n",
2661 slot, old_data, data_end);
2662 BUG_ON(1);
2663 }
2664 /*
2665 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2666 */
2667 /* first correct the data pointers */
2668 WARN_ON(leaf->map_token);
2669 for (i = slot; i < nritems; i++) {
2670 u32 ioff;
2671
2672 item = btrfs_item_nr(leaf, i);
2673 if (!leaf->map_token) {
2674 map_extent_buffer(leaf, (unsigned long)item,
2675 sizeof(struct btrfs_item),
2676 &leaf->map_token, &leaf->kaddr,
2677 &leaf->map_start, &leaf->map_len,
2678 KM_USER1);
2679 }
2680
2681 ioff = btrfs_item_offset(leaf, item);
2682 btrfs_set_item_offset(leaf, item, ioff - total_data);
2683 }
2684 if (leaf->map_token) {
2685 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2686 leaf->map_token = NULL;
2687 }
2688
2689 /* shift the items */
2690 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2691 btrfs_item_nr_offset(slot),
2692 (nritems - slot) * sizeof(struct btrfs_item));
2693
2694 /* shift the data */
2695 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2696 data_end - total_data, btrfs_leaf_data(leaf) +
2697 data_end, old_data - data_end);
2698 data_end = old_data;
2699 }
2700
2701 /* setup the item for the new data */
2702 for (i = 0; i < nr; i++) {
2703 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2704 btrfs_set_item_key(leaf, &disk_key, slot + i);
2705 item = btrfs_item_nr(leaf, slot + i);
2706 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2707 data_end -= data_size[i];
2708 btrfs_set_item_size(leaf, item, data_size[i]);
2709 }
2710 btrfs_set_header_nritems(leaf, nritems + nr);
2711 btrfs_mark_buffer_dirty(leaf);
2712
2713 ret = 0;
2714 if (slot == 0) {
2715 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2716 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2717 }
2718
2719 if (btrfs_leaf_free_space(root, leaf) < 0) {
2720 btrfs_print_leaf(root, leaf);
2721 BUG();
2722 }
2723 out:
2724 return ret;
2725 }
2726
2727 /*
2728 * Given a key and some data, insert an item into the tree.
2729 * This does all the path init required, making room in the tree if needed.
2730 */
2731 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2732 *root, struct btrfs_key *cpu_key, void *data, u32
2733 data_size)
2734 {
2735 int ret = 0;
2736 struct btrfs_path *path;
2737 struct extent_buffer *leaf;
2738 unsigned long ptr;
2739
2740 path = btrfs_alloc_path();
2741 BUG_ON(!path);
2742 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2743 if (!ret) {
2744 leaf = path->nodes[0];
2745 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2746 write_extent_buffer(leaf, data, ptr, data_size);
2747 btrfs_mark_buffer_dirty(leaf);
2748 }
2749 btrfs_free_path(path);
2750 return ret;
2751 }
2752
2753 /*
2754 * delete the pointer from a given node.
2755 *
2756 * If the delete empties a node, the node is removed from the tree,
2757 * continuing all the way the root if required. The root is converted into
2758 * a leaf if all the nodes are emptied.
2759 */
2760 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2761 struct btrfs_path *path, int level, int slot)
2762 {
2763 struct extent_buffer *parent = path->nodes[level];
2764 u32 nritems;
2765 int ret = 0;
2766 int wret;
2767
2768 nritems = btrfs_header_nritems(parent);
2769 if (slot != nritems -1) {
2770 memmove_extent_buffer(parent,
2771 btrfs_node_key_ptr_offset(slot),
2772 btrfs_node_key_ptr_offset(slot + 1),
2773 sizeof(struct btrfs_key_ptr) *
2774 (nritems - slot - 1));
2775 }
2776 nritems--;
2777 btrfs_set_header_nritems(parent, nritems);
2778 if (nritems == 0 && parent == root->node) {
2779 BUG_ON(btrfs_header_level(root->node) != 1);
2780 /* just turn the root into a leaf and break */
2781 btrfs_set_header_level(root->node, 0);
2782 } else if (slot == 0) {
2783 struct btrfs_disk_key disk_key;
2784
2785 btrfs_node_key(parent, &disk_key, 0);
2786 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2787 if (wret)
2788 ret = wret;
2789 }
2790 btrfs_mark_buffer_dirty(parent);
2791 return ret;
2792 }
2793
2794 /*
2795 * delete the item at the leaf level in path. If that empties
2796 * the leaf, remove it from the tree
2797 */
2798 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2799 struct btrfs_path *path, int slot, int nr)
2800 {
2801 struct extent_buffer *leaf;
2802 struct btrfs_item *item;
2803 int last_off;
2804 int dsize = 0;
2805 int ret = 0;
2806 int wret;
2807 int i;
2808 u32 nritems;
2809
2810 leaf = path->nodes[0];
2811 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2812
2813 for (i = 0; i < nr; i++)
2814 dsize += btrfs_item_size_nr(leaf, slot + i);
2815
2816 nritems = btrfs_header_nritems(leaf);
2817
2818 if (slot + nr != nritems) {
2819 int i;
2820 int data_end = leaf_data_end(root, leaf);
2821
2822 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2823 data_end + dsize,
2824 btrfs_leaf_data(leaf) + data_end,
2825 last_off - data_end);
2826
2827 for (i = slot + nr; i < nritems; i++) {
2828 u32 ioff;
2829
2830 item = btrfs_item_nr(leaf, i);
2831 if (!leaf->map_token) {
2832 map_extent_buffer(leaf, (unsigned long)item,
2833 sizeof(struct btrfs_item),
2834 &leaf->map_token, &leaf->kaddr,
2835 &leaf->map_start, &leaf->map_len,
2836 KM_USER1);
2837 }
2838 ioff = btrfs_item_offset(leaf, item);
2839 btrfs_set_item_offset(leaf, item, ioff + dsize);
2840 }
2841
2842 if (leaf->map_token) {
2843 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2844 leaf->map_token = NULL;
2845 }
2846
2847 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2848 btrfs_item_nr_offset(slot + nr),
2849 sizeof(struct btrfs_item) *
2850 (nritems - slot - nr));
2851 }
2852 btrfs_set_header_nritems(leaf, nritems - nr);
2853 nritems -= nr;
2854
2855 /* delete the leaf if we've emptied it */
2856 if (nritems == 0) {
2857 if (leaf == root->node) {
2858 btrfs_set_header_level(leaf, 0);
2859 } else {
2860 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2861 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2862 if (wret)
2863 ret = wret;
2864 wret = btrfs_free_extent(trans, root,
2865 leaf->start, leaf->len,
2866 btrfs_header_owner(path->nodes[1]),
2867 root_gen, 0, 0, 1);
2868 if (wret)
2869 ret = wret;
2870 }
2871 } else {
2872 int used = leaf_space_used(leaf, 0, nritems);
2873 if (slot == 0) {
2874 struct btrfs_disk_key disk_key;
2875
2876 btrfs_item_key(leaf, &disk_key, 0);
2877 wret = fixup_low_keys(trans, root, path,
2878 &disk_key, 1);
2879 if (wret)
2880 ret = wret;
2881 }
2882
2883 /* delete the leaf if it is mostly empty */
2884 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2885 /* push_leaf_left fixes the path.
2886 * make sure the path still points to our leaf
2887 * for possible call to del_ptr below
2888 */
2889 slot = path->slots[1];
2890 extent_buffer_get(leaf);
2891
2892 wret = push_leaf_left(trans, root, path, 1, 1);
2893 if (wret < 0 && wret != -ENOSPC)
2894 ret = wret;
2895
2896 if (path->nodes[0] == leaf &&
2897 btrfs_header_nritems(leaf)) {
2898 wret = push_leaf_right(trans, root, path, 1, 1);
2899 if (wret < 0 && wret != -ENOSPC)
2900 ret = wret;
2901 }
2902
2903 if (btrfs_header_nritems(leaf) == 0) {
2904 u64 root_gen;
2905 u64 bytenr = leaf->start;
2906 u32 blocksize = leaf->len;
2907
2908 root_gen = btrfs_header_generation(
2909 path->nodes[1]);
2910
2911 wret = del_ptr(trans, root, path, 1, slot);
2912 if (wret)
2913 ret = wret;
2914
2915 free_extent_buffer(leaf);
2916 wret = btrfs_free_extent(trans, root, bytenr,
2917 blocksize,
2918 btrfs_header_owner(path->nodes[1]),
2919 root_gen, 0, 0, 1);
2920 if (wret)
2921 ret = wret;
2922 } else {
2923 /* if we're still in the path, make sure
2924 * we're dirty. Otherwise, one of the
2925 * push_leaf functions must have already
2926 * dirtied this buffer
2927 */
2928 if (path->nodes[0] == leaf)
2929 btrfs_mark_buffer_dirty(leaf);
2930 free_extent_buffer(leaf);
2931 }
2932 } else {
2933 btrfs_mark_buffer_dirty(leaf);
2934 }
2935 }
2936 return ret;
2937 }
2938
2939 /*
2940 * search the tree again to find a leaf with lesser keys
2941 * returns 0 if it found something or 1 if there are no lesser leaves.
2942 * returns < 0 on io errors.
2943 */
2944 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2945 {
2946 struct btrfs_key key;
2947 struct btrfs_disk_key found_key;
2948 int ret;
2949
2950 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
2951
2952 if (key.offset > 0)
2953 key.offset--;
2954 else if (key.type > 0)
2955 key.type--;
2956 else if (key.objectid > 0)
2957 key.objectid--;
2958 else
2959 return 1;
2960
2961 btrfs_release_path(root, path);
2962 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2963 if (ret < 0)
2964 return ret;
2965 btrfs_item_key(path->nodes[0], &found_key, 0);
2966 ret = comp_keys(&found_key, &key);
2967 if (ret < 0)
2968 return 0;
2969 return 1;
2970 }
2971
2972 /*
2973 * A helper function to walk down the tree starting at min_key, and looking
2974 * for nodes or leaves that are either in cache or have a minimum
2975 * transaction id. This is used by the btree defrag code, but could
2976 * also be used to search for blocks that have changed since a given
2977 * transaction id.
2978 *
2979 * This does not cow, but it does stuff the starting key it finds back
2980 * into min_key, so you can call btrfs_search_slot with cow=1 on the
2981 * key and get a writable path.
2982 *
2983 * This does lock as it descends, and path->keep_locks should be set
2984 * to 1 by the caller.
2985 *
2986 * This honors path->lowest_level to prevent descent past a given level
2987 * of the tree.
2988 *
2989 * returns zero if something useful was found, < 0 on error and 1 if there
2990 * was nothing in the tree that matched the search criteria.
2991 */
2992 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
2993 struct btrfs_path *path, int cache_only,
2994 u64 min_trans)
2995 {
2996 struct extent_buffer *cur;
2997 struct btrfs_key found_key;
2998 int slot;
2999 u32 nritems;
3000 int level;
3001 int ret = 1;
3002
3003 again:
3004 cur = btrfs_lock_root_node(root);
3005 level = btrfs_header_level(cur);
3006 path->nodes[level] = cur;
3007 path->locks[level] = 1;
3008
3009 if (btrfs_header_generation(cur) < min_trans) {
3010 ret = 1;
3011 goto out;
3012 }
3013 while(1) {
3014 nritems = btrfs_header_nritems(cur);
3015 level = btrfs_header_level(cur);
3016 bin_search(cur, min_key, level, &slot);
3017
3018 /* at level = 0, we're done, setup the path and exit */
3019 if (level == 0) {
3020 ret = 0;
3021 path->slots[level] = slot;
3022 btrfs_item_key_to_cpu(cur, &found_key, slot);
3023 goto out;
3024 }
3025 /*
3026 * check this node pointer against the cache_only and
3027 * min_trans parameters. If it isn't in cache or is too
3028 * old, skip to the next one.
3029 */
3030 while(slot < nritems) {
3031 u64 blockptr;
3032 u64 gen;
3033 struct extent_buffer *tmp;
3034 blockptr = btrfs_node_blockptr(cur, slot);
3035 gen = btrfs_node_ptr_generation(cur, slot);
3036 if (gen < min_trans) {
3037 slot++;
3038 continue;
3039 }
3040 if (!cache_only)
3041 break;
3042
3043 tmp = btrfs_find_tree_block(root, blockptr,
3044 btrfs_level_size(root, level - 1));
3045
3046 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3047 free_extent_buffer(tmp);
3048 break;
3049 }
3050 if (tmp)
3051 free_extent_buffer(tmp);
3052 slot++;
3053 }
3054 /*
3055 * we didn't find a candidate key in this node, walk forward
3056 * and find another one
3057 */
3058 if (slot >= nritems) {
3059 ret = btrfs_find_next_key(root, path, min_key, level,
3060 cache_only, min_trans);
3061 if (ret == 0) {
3062 btrfs_release_path(root, path);
3063 goto again;
3064 } else {
3065 goto out;
3066 }
3067 }
3068 /* save our key for returning back */
3069 btrfs_node_key_to_cpu(cur, &found_key, slot);
3070 path->slots[level] = slot;
3071 if (level == path->lowest_level) {
3072 ret = 0;
3073 unlock_up(path, level, 1);
3074 goto out;
3075 }
3076 cur = read_node_slot(root, cur, slot);
3077
3078 btrfs_tree_lock(cur);
3079 path->locks[level - 1] = 1;
3080 path->nodes[level - 1] = cur;
3081 unlock_up(path, level, 1);
3082 }
3083 out:
3084 if (ret == 0)
3085 memcpy(min_key, &found_key, sizeof(found_key));
3086 return ret;
3087 }
3088
3089 /*
3090 * this is similar to btrfs_next_leaf, but does not try to preserve
3091 * and fixup the path. It looks for and returns the next key in the
3092 * tree based on the current path and the cache_only and min_trans
3093 * parameters.
3094 *
3095 * 0 is returned if another key is found, < 0 if there are any errors
3096 * and 1 is returned if there are no higher keys in the tree
3097 *
3098 * path->keep_locks should be set to 1 on the search made before
3099 * calling this function.
3100 */
3101 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3102 struct btrfs_key *key, int lowest_level,
3103 int cache_only, u64 min_trans)
3104 {
3105 int level = lowest_level;
3106 int slot;
3107 struct extent_buffer *c;
3108
3109 while(level < BTRFS_MAX_LEVEL) {
3110 if (!path->nodes[level])
3111 return 1;
3112
3113 slot = path->slots[level] + 1;
3114 c = path->nodes[level];
3115 next:
3116 if (slot >= btrfs_header_nritems(c)) {
3117 level++;
3118 if (level == BTRFS_MAX_LEVEL) {
3119 return 1;
3120 }
3121 continue;
3122 }
3123 if (level == 0)
3124 btrfs_item_key_to_cpu(c, key, slot);
3125 else {
3126 u64 blockptr = btrfs_node_blockptr(c, slot);
3127 u64 gen = btrfs_node_ptr_generation(c, slot);
3128
3129 if (cache_only) {
3130 struct extent_buffer *cur;
3131 cur = btrfs_find_tree_block(root, blockptr,
3132 btrfs_level_size(root, level - 1));
3133 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3134 slot++;
3135 if (cur)
3136 free_extent_buffer(cur);
3137 goto next;
3138 }
3139 free_extent_buffer(cur);
3140 }
3141 if (gen < min_trans) {
3142 slot++;
3143 goto next;
3144 }
3145 btrfs_node_key_to_cpu(c, key, slot);
3146 }
3147 return 0;
3148 }
3149 return 1;
3150 }
3151
3152 /*
3153 * search the tree again to find a leaf with greater keys
3154 * returns 0 if it found something or 1 if there are no greater leaves.
3155 * returns < 0 on io errors.
3156 */
3157 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3158 {
3159 int slot;
3160 int level = 1;
3161 struct extent_buffer *c;
3162 struct extent_buffer *next = NULL;
3163 struct btrfs_key key;
3164 u32 nritems;
3165 int ret;
3166
3167 nritems = btrfs_header_nritems(path->nodes[0]);
3168 if (nritems == 0) {
3169 return 1;
3170 }
3171
3172 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3173
3174 btrfs_release_path(root, path);
3175 path->keep_locks = 1;
3176 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3177 path->keep_locks = 0;
3178
3179 if (ret < 0)
3180 return ret;
3181
3182 nritems = btrfs_header_nritems(path->nodes[0]);
3183 /*
3184 * by releasing the path above we dropped all our locks. A balance
3185 * could have added more items next to the key that used to be
3186 * at the very end of the block. So, check again here and
3187 * advance the path if there are now more items available.
3188 */
3189 if (nritems > 0 && path->slots[0] < nritems - 1) {
3190 path->slots[0]++;
3191 goto done;
3192 }
3193
3194 while(level < BTRFS_MAX_LEVEL) {
3195 if (!path->nodes[level])
3196 return 1;
3197
3198 slot = path->slots[level] + 1;
3199 c = path->nodes[level];
3200 if (slot >= btrfs_header_nritems(c)) {
3201 level++;
3202 if (level == BTRFS_MAX_LEVEL) {
3203 return 1;
3204 }
3205 continue;
3206 }
3207
3208 if (next) {
3209 btrfs_tree_unlock(next);
3210 free_extent_buffer(next);
3211 }
3212
3213 if (level == 1 && path->locks[1] && path->reada)
3214 reada_for_search(root, path, level, slot, 0);
3215
3216 next = read_node_slot(root, c, slot);
3217 if (!path->skip_locking) {
3218 WARN_ON(!btrfs_tree_locked(c));
3219 btrfs_tree_lock(next);
3220 }
3221 break;
3222 }
3223 path->slots[level] = slot;
3224 while(1) {
3225 level--;
3226 c = path->nodes[level];
3227 if (path->locks[level])
3228 btrfs_tree_unlock(c);
3229 free_extent_buffer(c);
3230 path->nodes[level] = next;
3231 path->slots[level] = 0;
3232 if (!path->skip_locking)
3233 path->locks[level] = 1;
3234 if (!level)
3235 break;
3236 if (level == 1 && path->locks[1] && path->reada)
3237 reada_for_search(root, path, level, slot, 0);
3238 next = read_node_slot(root, next, 0);
3239 if (!path->skip_locking) {
3240 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3241 btrfs_tree_lock(next);
3242 }
3243 }
3244 done:
3245 unlock_up(path, 0, 1);
3246 return 0;
3247 }
3248
3249 /*
3250 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
3251 * searching until it gets past min_objectid or finds an item of 'type'
3252 *
3253 * returns 0 if something is found, 1 if nothing was found and < 0 on error
3254 */
3255 int btrfs_previous_item(struct btrfs_root *root,
3256 struct btrfs_path *path, u64 min_objectid,
3257 int type)
3258 {
3259 struct btrfs_key found_key;
3260 struct extent_buffer *leaf;
3261 int ret;
3262
3263 while(1) {
3264 if (path->slots[0] == 0) {
3265 ret = btrfs_prev_leaf(root, path);
3266 if (ret != 0)
3267 return ret;
3268 } else {
3269 path->slots[0]--;
3270 }
3271 leaf = path->nodes[0];
3272 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3273 if (found_key.type == type)
3274 return 0;
3275 }
3276 return 1;
3277 }
3278
Linux kernel - Deliverables
This page took 0.095491 seconds and 6 git commands to generate.