]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/ctree.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: free space accounting redo
[mirror_ubuntu-artful-kernel.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 noinline 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, 0);
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 noinline __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 u64 prealloc_dest)
186 {
187 u64 root_gen;
188 struct extent_buffer *cow;
189 u32 nritems;
190 int ret = 0;
191 int different_trans = 0;
192 int level;
193 int unlock_orig = 0;
194 struct btrfs_key first_key;
195
196 if (*cow_ret == buf)
197 unlock_orig = 1;
198
199 WARN_ON(!btrfs_tree_locked(buf));
200
201 if (root->ref_cows) {
202 root_gen = trans->transid;
203 } else {
204 root_gen = 0;
205 }
206 WARN_ON(root->ref_cows && trans->transid !=
207 root->fs_info->running_transaction->transid);
208 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
209
210 level = btrfs_header_level(buf);
211 nritems = btrfs_header_nritems(buf);
212 if (nritems) {
213 if (level == 0)
214 btrfs_item_key_to_cpu(buf, &first_key, 0);
215 else
216 btrfs_node_key_to_cpu(buf, &first_key, 0);
217 } else {
218 first_key.objectid = 0;
219 }
220 if (prealloc_dest) {
221 struct btrfs_key ins;
222
223 ins.objectid = prealloc_dest;
224 ins.offset = buf->len;
225 ins.type = BTRFS_EXTENT_ITEM_KEY;
226
227 ret = btrfs_alloc_reserved_extent(trans, root,
228 root->root_key.objectid,
229 root_gen, level,
230 first_key.objectid,
231 &ins);
232 BUG_ON(ret);
233 cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
234 buf->len);
235 } else {
236 cow = btrfs_alloc_free_block(trans, root, buf->len,
237 root->root_key.objectid,
238 root_gen, first_key.objectid,
239 level, search_start, empty_size);
240 }
241 if (IS_ERR(cow))
242 return PTR_ERR(cow);
243
244 copy_extent_buffer(cow, buf, 0, 0, cow->len);
245 btrfs_set_header_bytenr(cow, cow->start);
246 btrfs_set_header_generation(cow, trans->transid);
247 btrfs_set_header_owner(cow, root->root_key.objectid);
248 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
249
250 WARN_ON(btrfs_header_generation(buf) > trans->transid);
251 if (btrfs_header_generation(buf) != trans->transid) {
252 different_trans = 1;
253 ret = btrfs_inc_ref(trans, root, buf, 1);
254 if (ret)
255 return ret;
256 } else {
257 clean_tree_block(trans, root, buf);
258 }
259
260 if (buf == root->node) {
261 WARN_ON(parent && parent != buf);
262 root_gen = btrfs_header_generation(buf);
263
264 spin_lock(&root->node_lock);
265 root->node = cow;
266 extent_buffer_get(cow);
267 spin_unlock(&root->node_lock);
268
269 if (buf != root->commit_root) {
270 btrfs_free_extent(trans, root, buf->start,
271 buf->len, root->root_key.objectid,
272 root_gen, 0, 0, 1);
273 }
274 free_extent_buffer(buf);
275 add_root_to_dirty_list(root);
276 } else {
277 root_gen = btrfs_header_generation(parent);
278 btrfs_set_node_blockptr(parent, parent_slot,
279 cow->start);
280 WARN_ON(trans->transid == 0);
281 btrfs_set_node_ptr_generation(parent, parent_slot,
282 trans->transid);
283 btrfs_mark_buffer_dirty(parent);
284 WARN_ON(btrfs_header_generation(parent) != trans->transid);
285 btrfs_free_extent(trans, root, buf->start, buf->len,
286 btrfs_header_owner(parent), root_gen,
287 0, 0, 1);
288 }
289 if (unlock_orig)
290 btrfs_tree_unlock(buf);
291 free_extent_buffer(buf);
292 btrfs_mark_buffer_dirty(cow);
293 *cow_ret = cow;
294 return 0;
295 }
296
297 int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
298 struct btrfs_root *root, struct extent_buffer *buf,
299 struct extent_buffer *parent, int parent_slot,
300 struct extent_buffer **cow_ret, u64 prealloc_dest)
301 {
302 u64 search_start;
303 u64 header_trans;
304 int ret;
305
306 if (trans->transaction != root->fs_info->running_transaction) {
307 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
308 root->fs_info->running_transaction->transid);
309 WARN_ON(1);
310 }
311 if (trans->transid != root->fs_info->generation) {
312 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
313 root->fs_info->generation);
314 WARN_ON(1);
315 }
316
317 header_trans = btrfs_header_generation(buf);
318 spin_lock(&root->fs_info->hash_lock);
319 if (header_trans == trans->transid &&
320 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
321 *cow_ret = buf;
322 spin_unlock(&root->fs_info->hash_lock);
323 WARN_ON(prealloc_dest);
324 return 0;
325 }
326 spin_unlock(&root->fs_info->hash_lock);
327 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
328 ret = __btrfs_cow_block(trans, root, buf, parent,
329 parent_slot, cow_ret, search_start, 0,
330 prealloc_dest);
331 return ret;
332 }
333
334 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
335 {
336 if (blocknr < other && other - (blocknr + blocksize) < 32768)
337 return 1;
338 if (blocknr > other && blocknr - (other + blocksize) < 32768)
339 return 1;
340 return 0;
341 }
342
343 /*
344 * compare two keys in a memcmp fashion
345 */
346 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
347 {
348 struct btrfs_key k1;
349
350 btrfs_disk_key_to_cpu(&k1, disk);
351
352 if (k1.objectid > k2->objectid)
353 return 1;
354 if (k1.objectid < k2->objectid)
355 return -1;
356 if (k1.type > k2->type)
357 return 1;
358 if (k1.type < k2->type)
359 return -1;
360 if (k1.offset > k2->offset)
361 return 1;
362 if (k1.offset < k2->offset)
363 return -1;
364 return 0;
365 }
366
367
368 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
369 struct btrfs_root *root, struct extent_buffer *parent,
370 int start_slot, int cache_only, u64 *last_ret,
371 struct btrfs_key *progress)
372 {
373 struct extent_buffer *cur;
374 u64 blocknr;
375 u64 gen;
376 u64 search_start = *last_ret;
377 u64 last_block = 0;
378 u64 other;
379 u32 parent_nritems;
380 int end_slot;
381 int i;
382 int err = 0;
383 int parent_level;
384 int uptodate;
385 u32 blocksize;
386 int progress_passed = 0;
387 struct btrfs_disk_key disk_key;
388
389 parent_level = btrfs_header_level(parent);
390 if (cache_only && parent_level != 1)
391 return 0;
392
393 if (trans->transaction != root->fs_info->running_transaction) {
394 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
395 root->fs_info->running_transaction->transid);
396 WARN_ON(1);
397 }
398 if (trans->transid != root->fs_info->generation) {
399 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
400 root->fs_info->generation);
401 WARN_ON(1);
402 }
403
404 parent_nritems = btrfs_header_nritems(parent);
405 blocksize = btrfs_level_size(root, parent_level - 1);
406 end_slot = parent_nritems;
407
408 if (parent_nritems == 1)
409 return 0;
410
411 for (i = start_slot; i < end_slot; i++) {
412 int close = 1;
413
414 if (!parent->map_token) {
415 map_extent_buffer(parent,
416 btrfs_node_key_ptr_offset(i),
417 sizeof(struct btrfs_key_ptr),
418 &parent->map_token, &parent->kaddr,
419 &parent->map_start, &parent->map_len,
420 KM_USER1);
421 }
422 btrfs_node_key(parent, &disk_key, i);
423 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
424 continue;
425
426 progress_passed = 1;
427 blocknr = btrfs_node_blockptr(parent, i);
428 gen = btrfs_node_ptr_generation(parent, i);
429 if (last_block == 0)
430 last_block = blocknr;
431
432 if (i > 0) {
433 other = btrfs_node_blockptr(parent, i - 1);
434 close = close_blocks(blocknr, other, blocksize);
435 }
436 if (!close && i < end_slot - 2) {
437 other = btrfs_node_blockptr(parent, i + 1);
438 close = close_blocks(blocknr, other, blocksize);
439 }
440 if (close) {
441 last_block = blocknr;
442 continue;
443 }
444 if (parent->map_token) {
445 unmap_extent_buffer(parent, parent->map_token,
446 KM_USER1);
447 parent->map_token = NULL;
448 }
449
450 cur = btrfs_find_tree_block(root, blocknr, blocksize);
451 if (cur)
452 uptodate = btrfs_buffer_uptodate(cur, gen);
453 else
454 uptodate = 0;
455 if (!cur || !uptodate) {
456 if (cache_only) {
457 free_extent_buffer(cur);
458 continue;
459 }
460 if (!cur) {
461 cur = read_tree_block(root, blocknr,
462 blocksize, gen);
463 } else if (!uptodate) {
464 btrfs_read_buffer(cur, gen);
465 }
466 }
467 if (search_start == 0)
468 search_start = last_block;
469
470 btrfs_tree_lock(cur);
471 err = __btrfs_cow_block(trans, root, cur, parent, i,
472 &cur, search_start,
473 min(16 * blocksize,
474 (end_slot - i) * blocksize), 0);
475 if (err) {
476 btrfs_tree_unlock(cur);
477 free_extent_buffer(cur);
478 break;
479 }
480 search_start = cur->start;
481 last_block = cur->start;
482 *last_ret = search_start;
483 btrfs_tree_unlock(cur);
484 free_extent_buffer(cur);
485 }
486 if (parent->map_token) {
487 unmap_extent_buffer(parent, parent->map_token,
488 KM_USER1);
489 parent->map_token = NULL;
490 }
491 return err;
492 }
493
494 /*
495 * The leaf data grows from end-to-front in the node.
496 * this returns the address of the start of the last item,
497 * which is the stop of the leaf data stack
498 */
499 static inline unsigned int leaf_data_end(struct btrfs_root *root,
500 struct extent_buffer *leaf)
501 {
502 u32 nr = btrfs_header_nritems(leaf);
503 if (nr == 0)
504 return BTRFS_LEAF_DATA_SIZE(root);
505 return btrfs_item_offset_nr(leaf, nr - 1);
506 }
507
508 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
509 int level)
510 {
511 struct extent_buffer *parent = NULL;
512 struct extent_buffer *node = path->nodes[level];
513 struct btrfs_disk_key parent_key;
514 struct btrfs_disk_key node_key;
515 int parent_slot;
516 int slot;
517 struct btrfs_key cpukey;
518 u32 nritems = btrfs_header_nritems(node);
519
520 if (path->nodes[level + 1])
521 parent = path->nodes[level + 1];
522
523 slot = path->slots[level];
524 BUG_ON(nritems == 0);
525 if (parent) {
526 parent_slot = path->slots[level + 1];
527 btrfs_node_key(parent, &parent_key, parent_slot);
528 btrfs_node_key(node, &node_key, 0);
529 BUG_ON(memcmp(&parent_key, &node_key,
530 sizeof(struct btrfs_disk_key)));
531 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
532 btrfs_header_bytenr(node));
533 }
534 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
535 if (slot != 0) {
536 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
537 btrfs_node_key(node, &node_key, slot);
538 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
539 }
540 if (slot < nritems - 1) {
541 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
542 btrfs_node_key(node, &node_key, slot);
543 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
544 }
545 return 0;
546 }
547
548 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
549 int level)
550 {
551 struct extent_buffer *leaf = path->nodes[level];
552 struct extent_buffer *parent = NULL;
553 int parent_slot;
554 struct btrfs_key cpukey;
555 struct btrfs_disk_key parent_key;
556 struct btrfs_disk_key leaf_key;
557 int slot = path->slots[0];
558
559 u32 nritems = btrfs_header_nritems(leaf);
560
561 if (path->nodes[level + 1])
562 parent = path->nodes[level + 1];
563
564 if (nritems == 0)
565 return 0;
566
567 if (parent) {
568 parent_slot = path->slots[level + 1];
569 btrfs_node_key(parent, &parent_key, parent_slot);
570 btrfs_item_key(leaf, &leaf_key, 0);
571
572 BUG_ON(memcmp(&parent_key, &leaf_key,
573 sizeof(struct btrfs_disk_key)));
574 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
575 btrfs_header_bytenr(leaf));
576 }
577 #if 0
578 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
579 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
580 btrfs_item_key(leaf, &leaf_key, i);
581 if (comp_keys(&leaf_key, &cpukey) >= 0) {
582 btrfs_print_leaf(root, leaf);
583 printk("slot %d offset bad key\n", i);
584 BUG_ON(1);
585 }
586 if (btrfs_item_offset_nr(leaf, i) !=
587 btrfs_item_end_nr(leaf, i + 1)) {
588 btrfs_print_leaf(root, leaf);
589 printk("slot %d offset bad\n", i);
590 BUG_ON(1);
591 }
592 if (i == 0) {
593 if (btrfs_item_offset_nr(leaf, i) +
594 btrfs_item_size_nr(leaf, i) !=
595 BTRFS_LEAF_DATA_SIZE(root)) {
596 btrfs_print_leaf(root, leaf);
597 printk("slot %d first offset bad\n", i);
598 BUG_ON(1);
599 }
600 }
601 }
602 if (nritems > 0) {
603 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
604 btrfs_print_leaf(root, leaf);
605 printk("slot %d bad size \n", nritems - 1);
606 BUG_ON(1);
607 }
608 }
609 #endif
610 if (slot != 0 && slot < nritems - 1) {
611 btrfs_item_key(leaf, &leaf_key, slot);
612 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
613 if (comp_keys(&leaf_key, &cpukey) <= 0) {
614 btrfs_print_leaf(root, leaf);
615 printk("slot %d offset bad key\n", slot);
616 BUG_ON(1);
617 }
618 if (btrfs_item_offset_nr(leaf, slot - 1) !=
619 btrfs_item_end_nr(leaf, slot)) {
620 btrfs_print_leaf(root, leaf);
621 printk("slot %d offset bad\n", slot);
622 BUG_ON(1);
623 }
624 }
625 if (slot < nritems - 1) {
626 btrfs_item_key(leaf, &leaf_key, slot);
627 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
628 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
629 if (btrfs_item_offset_nr(leaf, slot) !=
630 btrfs_item_end_nr(leaf, slot + 1)) {
631 btrfs_print_leaf(root, leaf);
632 printk("slot %d offset bad\n", slot);
633 BUG_ON(1);
634 }
635 }
636 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
637 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
638 return 0;
639 }
640
641 static int noinline check_block(struct btrfs_root *root,
642 struct btrfs_path *path, int level)
643 {
644 u64 found_start;
645 return 0;
646 if (btrfs_header_level(path->nodes[level]) != level)
647 printk("warning: bad level %Lu wanted %d found %d\n",
648 path->nodes[level]->start, level,
649 btrfs_header_level(path->nodes[level]));
650 found_start = btrfs_header_bytenr(path->nodes[level]);
651 if (found_start != path->nodes[level]->start) {
652 printk("warning: bad bytentr %Lu found %Lu\n",
653 path->nodes[level]->start, found_start);
654 }
655 #if 0
656 struct extent_buffer *buf = path->nodes[level];
657
658 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
659 (unsigned long)btrfs_header_fsid(buf),
660 BTRFS_FSID_SIZE)) {
661 printk("warning bad block %Lu\n", buf->start);
662 return 1;
663 }
664 #endif
665 if (level == 0)
666 return check_leaf(root, path, level);
667 return check_node(root, path, level);
668 }
669
670 /*
671 * search for key in the extent_buffer. The items start at offset p,
672 * and they are item_size apart. There are 'max' items in p.
673 *
674 * the slot in the array is returned via slot, and it points to
675 * the place where you would insert key if it is not found in
676 * the array.
677 *
678 * slot may point to max if the key is bigger than all of the keys
679 */
680 static noinline int generic_bin_search(struct extent_buffer *eb,
681 unsigned long p,
682 int item_size, struct btrfs_key *key,
683 int max, int *slot)
684 {
685 int low = 0;
686 int high = max;
687 int mid;
688 int ret;
689 struct btrfs_disk_key *tmp = NULL;
690 struct btrfs_disk_key unaligned;
691 unsigned long offset;
692 char *map_token = NULL;
693 char *kaddr = NULL;
694 unsigned long map_start = 0;
695 unsigned long map_len = 0;
696 int err;
697
698 while(low < high) {
699 mid = (low + high) / 2;
700 offset = p + mid * item_size;
701
702 if (!map_token || offset < map_start ||
703 (offset + sizeof(struct btrfs_disk_key)) >
704 map_start + map_len) {
705 if (map_token) {
706 unmap_extent_buffer(eb, map_token, KM_USER0);
707 map_token = NULL;
708 }
709 err = map_extent_buffer(eb, offset,
710 sizeof(struct btrfs_disk_key),
711 &map_token, &kaddr,
712 &map_start, &map_len, KM_USER0);
713
714 if (!err) {
715 tmp = (struct btrfs_disk_key *)(kaddr + offset -
716 map_start);
717 } else {
718 read_extent_buffer(eb, &unaligned,
719 offset, sizeof(unaligned));
720 tmp = &unaligned;
721 }
722
723 } else {
724 tmp = (struct btrfs_disk_key *)(kaddr + offset -
725 map_start);
726 }
727 ret = comp_keys(tmp, key);
728
729 if (ret < 0)
730 low = mid + 1;
731 else if (ret > 0)
732 high = mid;
733 else {
734 *slot = mid;
735 if (map_token)
736 unmap_extent_buffer(eb, map_token, KM_USER0);
737 return 0;
738 }
739 }
740 *slot = low;
741 if (map_token)
742 unmap_extent_buffer(eb, map_token, KM_USER0);
743 return 1;
744 }
745
746 /*
747 * simple bin_search frontend that does the right thing for
748 * leaves vs nodes
749 */
750 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
751 int level, int *slot)
752 {
753 if (level == 0) {
754 return generic_bin_search(eb,
755 offsetof(struct btrfs_leaf, items),
756 sizeof(struct btrfs_item),
757 key, btrfs_header_nritems(eb),
758 slot);
759 } else {
760 return generic_bin_search(eb,
761 offsetof(struct btrfs_node, ptrs),
762 sizeof(struct btrfs_key_ptr),
763 key, btrfs_header_nritems(eb),
764 slot);
765 }
766 return -1;
767 }
768
769 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
770 struct extent_buffer *parent, int slot)
771 {
772 int level = btrfs_header_level(parent);
773 if (slot < 0)
774 return NULL;
775 if (slot >= btrfs_header_nritems(parent))
776 return NULL;
777
778 BUG_ON(level == 0);
779
780 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
781 btrfs_level_size(root, level - 1),
782 btrfs_node_ptr_generation(parent, slot));
783 }
784
785 static noinline int balance_level(struct btrfs_trans_handle *trans,
786 struct btrfs_root *root,
787 struct btrfs_path *path, int level)
788 {
789 struct extent_buffer *right = NULL;
790 struct extent_buffer *mid;
791 struct extent_buffer *left = NULL;
792 struct extent_buffer *parent = NULL;
793 int ret = 0;
794 int wret;
795 int pslot;
796 int orig_slot = path->slots[level];
797 int err_on_enospc = 0;
798 u64 orig_ptr;
799
800 if (level == 0)
801 return 0;
802
803 mid = path->nodes[level];
804 WARN_ON(!path->locks[level]);
805 WARN_ON(btrfs_header_generation(mid) != trans->transid);
806
807 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
808
809 if (level < BTRFS_MAX_LEVEL - 1)
810 parent = path->nodes[level + 1];
811 pslot = path->slots[level + 1];
812
813 /*
814 * deal with the case where there is only one pointer in the root
815 * by promoting the node below to a root
816 */
817 if (!parent) {
818 struct extent_buffer *child;
819
820 if (btrfs_header_nritems(mid) != 1)
821 return 0;
822
823 /* promote the child to a root */
824 child = read_node_slot(root, mid, 0);
825 btrfs_tree_lock(child);
826 BUG_ON(!child);
827 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
828 BUG_ON(ret);
829
830 spin_lock(&root->node_lock);
831 root->node = child;
832 spin_unlock(&root->node_lock);
833
834 add_root_to_dirty_list(root);
835 btrfs_tree_unlock(child);
836 path->locks[level] = 0;
837 path->nodes[level] = NULL;
838 clean_tree_block(trans, root, mid);
839 btrfs_tree_unlock(mid);
840 /* once for the path */
841 free_extent_buffer(mid);
842 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
843 root->root_key.objectid,
844 btrfs_header_generation(mid), 0, 0, 1);
845 /* once for the root ptr */
846 free_extent_buffer(mid);
847 return ret;
848 }
849 if (btrfs_header_nritems(mid) >
850 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
851 return 0;
852
853 if (btrfs_header_nritems(mid) < 2)
854 err_on_enospc = 1;
855
856 left = read_node_slot(root, parent, pslot - 1);
857 if (left) {
858 btrfs_tree_lock(left);
859 wret = btrfs_cow_block(trans, root, left,
860 parent, pslot - 1, &left, 0);
861 if (wret) {
862 ret = wret;
863 goto enospc;
864 }
865 }
866 right = read_node_slot(root, parent, pslot + 1);
867 if (right) {
868 btrfs_tree_lock(right);
869 wret = btrfs_cow_block(trans, root, right,
870 parent, pslot + 1, &right, 0);
871 if (wret) {
872 ret = wret;
873 goto enospc;
874 }
875 }
876
877 /* first, try to make some room in the middle buffer */
878 if (left) {
879 orig_slot += btrfs_header_nritems(left);
880 wret = push_node_left(trans, root, left, mid, 1);
881 if (wret < 0)
882 ret = wret;
883 if (btrfs_header_nritems(mid) < 2)
884 err_on_enospc = 1;
885 }
886
887 /*
888 * then try to empty the right most buffer into the middle
889 */
890 if (right) {
891 wret = push_node_left(trans, root, mid, right, 1);
892 if (wret < 0 && wret != -ENOSPC)
893 ret = wret;
894 if (btrfs_header_nritems(right) == 0) {
895 u64 bytenr = right->start;
896 u64 generation = btrfs_header_generation(parent);
897 u32 blocksize = right->len;
898
899 clean_tree_block(trans, root, right);
900 btrfs_tree_unlock(right);
901 free_extent_buffer(right);
902 right = NULL;
903 wret = del_ptr(trans, root, path, level + 1, pslot +
904 1);
905 if (wret)
906 ret = wret;
907 wret = btrfs_free_extent(trans, root, bytenr,
908 blocksize,
909 btrfs_header_owner(parent),
910 generation, 0, 0, 1);
911 if (wret)
912 ret = wret;
913 } else {
914 struct btrfs_disk_key right_key;
915 btrfs_node_key(right, &right_key, 0);
916 btrfs_set_node_key(parent, &right_key, pslot + 1);
917 btrfs_mark_buffer_dirty(parent);
918 }
919 }
920 if (btrfs_header_nritems(mid) == 1) {
921 /*
922 * we're not allowed to leave a node with one item in the
923 * tree during a delete. A deletion from lower in the tree
924 * could try to delete the only pointer in this node.
925 * So, pull some keys from the left.
926 * There has to be a left pointer at this point because
927 * otherwise we would have pulled some pointers from the
928 * right
929 */
930 BUG_ON(!left);
931 wret = balance_node_right(trans, root, mid, left);
932 if (wret < 0) {
933 ret = wret;
934 goto enospc;
935 }
936 if (wret == 1) {
937 wret = push_node_left(trans, root, left, mid, 1);
938 if (wret < 0)
939 ret = wret;
940 }
941 BUG_ON(wret == 1);
942 }
943 if (btrfs_header_nritems(mid) == 0) {
944 /* we've managed to empty the middle node, drop it */
945 u64 root_gen = btrfs_header_generation(parent);
946 u64 bytenr = mid->start;
947 u32 blocksize = mid->len;
948
949 clean_tree_block(trans, root, mid);
950 btrfs_tree_unlock(mid);
951 free_extent_buffer(mid);
952 mid = NULL;
953 wret = del_ptr(trans, root, path, level + 1, pslot);
954 if (wret)
955 ret = wret;
956 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
957 btrfs_header_owner(parent),
958 root_gen, 0, 0, 1);
959 if (wret)
960 ret = wret;
961 } else {
962 /* update the parent key to reflect our changes */
963 struct btrfs_disk_key mid_key;
964 btrfs_node_key(mid, &mid_key, 0);
965 btrfs_set_node_key(parent, &mid_key, pslot);
966 btrfs_mark_buffer_dirty(parent);
967 }
968
969 /* update the path */
970 if (left) {
971 if (btrfs_header_nritems(left) > orig_slot) {
972 extent_buffer_get(left);
973 /* left was locked after cow */
974 path->nodes[level] = left;
975 path->slots[level + 1] -= 1;
976 path->slots[level] = orig_slot;
977 if (mid) {
978 btrfs_tree_unlock(mid);
979 free_extent_buffer(mid);
980 }
981 } else {
982 orig_slot -= btrfs_header_nritems(left);
983 path->slots[level] = orig_slot;
984 }
985 }
986 /* double check we haven't messed things up */
987 check_block(root, path, level);
988 if (orig_ptr !=
989 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
990 BUG();
991 enospc:
992 if (right) {
993 btrfs_tree_unlock(right);
994 free_extent_buffer(right);
995 }
996 if (left) {
997 if (path->nodes[level] != left)
998 btrfs_tree_unlock(left);
999 free_extent_buffer(left);
1000 }
1001 return ret;
1002 }
1003
1004 /* returns zero if the push worked, non-zero otherwise */
1005 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
1006 struct btrfs_root *root,
1007 struct btrfs_path *path, int level)
1008 {
1009 struct extent_buffer *right = NULL;
1010 struct extent_buffer *mid;
1011 struct extent_buffer *left = NULL;
1012 struct extent_buffer *parent = NULL;
1013 int ret = 0;
1014 int wret;
1015 int pslot;
1016 int orig_slot = path->slots[level];
1017 u64 orig_ptr;
1018
1019 if (level == 0)
1020 return 1;
1021
1022 mid = path->nodes[level];
1023 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1024 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1025
1026 if (level < BTRFS_MAX_LEVEL - 1)
1027 parent = path->nodes[level + 1];
1028 pslot = path->slots[level + 1];
1029
1030 if (!parent)
1031 return 1;
1032
1033 left = read_node_slot(root, parent, pslot - 1);
1034
1035 /* first, try to make some room in the middle buffer */
1036 if (left) {
1037 u32 left_nr;
1038
1039 btrfs_tree_lock(left);
1040 left_nr = btrfs_header_nritems(left);
1041 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1042 wret = 1;
1043 } else {
1044 ret = btrfs_cow_block(trans, root, left, parent,
1045 pslot - 1, &left, 0);
1046 if (ret)
1047 wret = 1;
1048 else {
1049 wret = push_node_left(trans, root,
1050 left, mid, 0);
1051 }
1052 }
1053 if (wret < 0)
1054 ret = wret;
1055 if (wret == 0) {
1056 struct btrfs_disk_key disk_key;
1057 orig_slot += left_nr;
1058 btrfs_node_key(mid, &disk_key, 0);
1059 btrfs_set_node_key(parent, &disk_key, pslot);
1060 btrfs_mark_buffer_dirty(parent);
1061 if (btrfs_header_nritems(left) > orig_slot) {
1062 path->nodes[level] = left;
1063 path->slots[level + 1] -= 1;
1064 path->slots[level] = orig_slot;
1065 btrfs_tree_unlock(mid);
1066 free_extent_buffer(mid);
1067 } else {
1068 orig_slot -=
1069 btrfs_header_nritems(left);
1070 path->slots[level] = orig_slot;
1071 btrfs_tree_unlock(left);
1072 free_extent_buffer(left);
1073 }
1074 return 0;
1075 }
1076 btrfs_tree_unlock(left);
1077 free_extent_buffer(left);
1078 }
1079 right = read_node_slot(root, parent, pslot + 1);
1080
1081 /*
1082 * then try to empty the right most buffer into the middle
1083 */
1084 if (right) {
1085 u32 right_nr;
1086 btrfs_tree_lock(right);
1087 right_nr = btrfs_header_nritems(right);
1088 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1089 wret = 1;
1090 } else {
1091 ret = btrfs_cow_block(trans, root, right,
1092 parent, pslot + 1,
1093 &right, 0);
1094 if (ret)
1095 wret = 1;
1096 else {
1097 wret = balance_node_right(trans, root,
1098 right, mid);
1099 }
1100 }
1101 if (wret < 0)
1102 ret = wret;
1103 if (wret == 0) {
1104 struct btrfs_disk_key disk_key;
1105
1106 btrfs_node_key(right, &disk_key, 0);
1107 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1108 btrfs_mark_buffer_dirty(parent);
1109
1110 if (btrfs_header_nritems(mid) <= orig_slot) {
1111 path->nodes[level] = right;
1112 path->slots[level + 1] += 1;
1113 path->slots[level] = orig_slot -
1114 btrfs_header_nritems(mid);
1115 btrfs_tree_unlock(mid);
1116 free_extent_buffer(mid);
1117 } else {
1118 btrfs_tree_unlock(right);
1119 free_extent_buffer(right);
1120 }
1121 return 0;
1122 }
1123 btrfs_tree_unlock(right);
1124 free_extent_buffer(right);
1125 }
1126 return 1;
1127 }
1128
1129 /*
1130 * readahead one full node of leaves
1131 */
1132 static noinline void reada_for_search(struct btrfs_root *root,
1133 struct btrfs_path *path,
1134 int level, int slot, u64 objectid)
1135 {
1136 struct extent_buffer *node;
1137 struct btrfs_disk_key disk_key;
1138 u32 nritems;
1139 u64 search;
1140 u64 lowest_read;
1141 u64 highest_read;
1142 u64 nread = 0;
1143 int direction = path->reada;
1144 struct extent_buffer *eb;
1145 u32 nr;
1146 u32 blocksize;
1147 u32 nscan = 0;
1148
1149 if (level != 1)
1150 return;
1151
1152 if (!path->nodes[level])
1153 return;
1154
1155 node = path->nodes[level];
1156
1157 search = btrfs_node_blockptr(node, slot);
1158 blocksize = btrfs_level_size(root, level - 1);
1159 eb = btrfs_find_tree_block(root, search, blocksize);
1160 if (eb) {
1161 free_extent_buffer(eb);
1162 return;
1163 }
1164
1165 highest_read = search;
1166 lowest_read = search;
1167
1168 nritems = btrfs_header_nritems(node);
1169 nr = slot;
1170 while(1) {
1171 if (direction < 0) {
1172 if (nr == 0)
1173 break;
1174 nr--;
1175 } else if (direction > 0) {
1176 nr++;
1177 if (nr >= nritems)
1178 break;
1179 }
1180 if (path->reada < 0 && objectid) {
1181 btrfs_node_key(node, &disk_key, nr);
1182 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1183 break;
1184 }
1185 search = btrfs_node_blockptr(node, nr);
1186 if ((search >= lowest_read && search <= highest_read) ||
1187 (search < lowest_read && lowest_read - search <= 32768) ||
1188 (search > highest_read && search - highest_read <= 32768)) {
1189 readahead_tree_block(root, search, blocksize,
1190 btrfs_node_ptr_generation(node, nr));
1191 nread += blocksize;
1192 }
1193 nscan++;
1194 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1195 break;
1196 if(nread > (1024 * 1024) || nscan > 128)
1197 break;
1198
1199 if (search < lowest_read)
1200 lowest_read = search;
1201 if (search > highest_read)
1202 highest_read = search;
1203 }
1204 }
1205
1206 static noinline void unlock_up(struct btrfs_path *path, int level,
1207 int lowest_unlock)
1208 {
1209 int i;
1210 int skip_level = level;
1211 int no_skips = 0;
1212 struct extent_buffer *t;
1213
1214 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1215 if (!path->nodes[i])
1216 break;
1217 if (!path->locks[i])
1218 break;
1219 if (!no_skips && path->slots[i] == 0) {
1220 skip_level = i + 1;
1221 continue;
1222 }
1223 if (!no_skips && path->keep_locks) {
1224 u32 nritems;
1225 t = path->nodes[i];
1226 nritems = btrfs_header_nritems(t);
1227 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1228 skip_level = i + 1;
1229 continue;
1230 }
1231 }
1232 if (skip_level < i && i >= lowest_unlock)
1233 no_skips = 1;
1234
1235 t = path->nodes[i];
1236 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1237 btrfs_tree_unlock(t);
1238 path->locks[i] = 0;
1239 }
1240 }
1241 }
1242
1243 /*
1244 * look for key in the tree. path is filled in with nodes along the way
1245 * if key is found, we return zero and you can find the item in the leaf
1246 * level of the path (level 0)
1247 *
1248 * If the key isn't found, the path points to the slot where it should
1249 * be inserted, and 1 is returned. If there are other errors during the
1250 * search a negative error number is returned.
1251 *
1252 * if ins_len > 0, nodes and leaves will be split as we walk down the
1253 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1254 * possible)
1255 */
1256 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1257 *root, struct btrfs_key *key, struct btrfs_path *p, int
1258 ins_len, int cow)
1259 {
1260 struct extent_buffer *b;
1261 struct extent_buffer *tmp;
1262 int slot;
1263 int ret;
1264 int level;
1265 int should_reada = p->reada;
1266 int lowest_unlock = 1;
1267 int blocksize;
1268 u8 lowest_level = 0;
1269 u64 blocknr;
1270 u64 gen;
1271 struct btrfs_key prealloc_block;
1272
1273 lowest_level = p->lowest_level;
1274 WARN_ON(lowest_level && ins_len);
1275 WARN_ON(p->nodes[0] != NULL);
1276 WARN_ON(cow && root == root->fs_info->extent_root &&
1277 !mutex_is_locked(&root->fs_info->alloc_mutex));
1278 if (ins_len < 0)
1279 lowest_unlock = 2;
1280
1281 prealloc_block.objectid = 0;
1282
1283 again:
1284 if (p->skip_locking)
1285 b = btrfs_root_node(root);
1286 else
1287 b = btrfs_lock_root_node(root);
1288
1289 while (b) {
1290 level = btrfs_header_level(b);
1291
1292 /*
1293 * setup the path here so we can release it under lock
1294 * contention with the cow code
1295 */
1296 p->nodes[level] = b;
1297 if (!p->skip_locking)
1298 p->locks[level] = 1;
1299
1300 if (cow) {
1301 int wret;
1302
1303 /* is a cow on this block not required */
1304 spin_lock(&root->fs_info->hash_lock);
1305 if (btrfs_header_generation(b) == trans->transid &&
1306 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1307 spin_unlock(&root->fs_info->hash_lock);
1308 goto cow_done;
1309 }
1310 spin_unlock(&root->fs_info->hash_lock);
1311
1312 /* ok, we have to cow, is our old prealloc the right
1313 * size?
1314 */
1315 if (prealloc_block.objectid &&
1316 prealloc_block.offset != b->len) {
1317 btrfs_free_reserved_extent(root,
1318 prealloc_block.objectid,
1319 prealloc_block.offset);
1320 prealloc_block.objectid = 0;
1321 }
1322
1323 /*
1324 * for higher level blocks, try not to allocate blocks
1325 * with the block and the parent locks held.
1326 */
1327 if (level > 1 && !prealloc_block.objectid &&
1328 btrfs_path_lock_waiting(p, level)) {
1329 u32 size = b->len;
1330 u64 hint = b->start;
1331
1332 btrfs_release_path(root, p);
1333 ret = btrfs_reserve_extent(trans, root,
1334 size, size, 0,
1335 hint, (u64)-1,
1336 &prealloc_block, 0);
1337 BUG_ON(ret);
1338 goto again;
1339 }
1340
1341 wret = btrfs_cow_block(trans, root, b,
1342 p->nodes[level + 1],
1343 p->slots[level + 1],
1344 &b, prealloc_block.objectid);
1345 prealloc_block.objectid = 0;
1346 if (wret) {
1347 free_extent_buffer(b);
1348 ret = wret;
1349 goto done;
1350 }
1351 }
1352 cow_done:
1353 BUG_ON(!cow && ins_len);
1354 if (level != btrfs_header_level(b))
1355 WARN_ON(1);
1356 level = btrfs_header_level(b);
1357
1358 p->nodes[level] = b;
1359 if (!p->skip_locking)
1360 p->locks[level] = 1;
1361
1362 ret = check_block(root, p, level);
1363 if (ret) {
1364 ret = -1;
1365 goto done;
1366 }
1367
1368 ret = bin_search(b, key, level, &slot);
1369 if (level != 0) {
1370 if (ret && slot > 0)
1371 slot -= 1;
1372 p->slots[level] = slot;
1373 if (ins_len > 0 && btrfs_header_nritems(b) >=
1374 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1375 int sret = split_node(trans, root, p, level);
1376 BUG_ON(sret > 0);
1377 if (sret) {
1378 ret = sret;
1379 goto done;
1380 }
1381 b = p->nodes[level];
1382 slot = p->slots[level];
1383 } else if (ins_len < 0) {
1384 int sret = balance_level(trans, root, p,
1385 level);
1386 if (sret) {
1387 ret = sret;
1388 goto done;
1389 }
1390 b = p->nodes[level];
1391 if (!b) {
1392 btrfs_release_path(NULL, p);
1393 goto again;
1394 }
1395 slot = p->slots[level];
1396 BUG_ON(btrfs_header_nritems(b) == 1);
1397 }
1398 unlock_up(p, level, lowest_unlock);
1399
1400 /* this is only true while dropping a snapshot */
1401 if (level == lowest_level) {
1402 break;
1403 }
1404
1405 blocknr = btrfs_node_blockptr(b, slot);
1406 gen = btrfs_node_ptr_generation(b, slot);
1407 blocksize = btrfs_level_size(root, level - 1);
1408
1409 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1410 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1411 b = tmp;
1412 } else {
1413 /*
1414 * reduce lock contention at high levels
1415 * of the btree by dropping locks before
1416 * we read.
1417 */
1418 if (level > 1) {
1419 btrfs_release_path(NULL, p);
1420 if (tmp)
1421 free_extent_buffer(tmp);
1422 if (should_reada)
1423 reada_for_search(root, p,
1424 level, slot,
1425 key->objectid);
1426
1427 tmp = read_tree_block(root, blocknr,
1428 blocksize, gen);
1429 if (tmp)
1430 free_extent_buffer(tmp);
1431 goto again;
1432 } else {
1433 if (tmp)
1434 free_extent_buffer(tmp);
1435 if (should_reada)
1436 reada_for_search(root, p,
1437 level, slot,
1438 key->objectid);
1439 b = read_node_slot(root, b, slot);
1440 }
1441 }
1442 if (!p->skip_locking)
1443 btrfs_tree_lock(b);
1444 } else {
1445 p->slots[level] = slot;
1446 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1447 sizeof(struct btrfs_item) + ins_len) {
1448 int sret = split_leaf(trans, root, key,
1449 p, ins_len, ret == 0);
1450 BUG_ON(sret > 0);
1451 if (sret) {
1452 ret = sret;
1453 goto done;
1454 }
1455 }
1456 unlock_up(p, level, lowest_unlock);
1457 goto done;
1458 }
1459 }
1460 ret = 1;
1461 done:
1462 if (prealloc_block.objectid) {
1463 btrfs_free_reserved_extent(root,
1464 prealloc_block.objectid,
1465 prealloc_block.offset);
1466 }
1467
1468 return ret;
1469 }
1470
1471 /*
1472 * adjust the pointers going up the tree, starting at level
1473 * making sure the right key of each node is points to 'key'.
1474 * This is used after shifting pointers to the left, so it stops
1475 * fixing up pointers when a given leaf/node is not in slot 0 of the
1476 * higher levels
1477 *
1478 * If this fails to write a tree block, it returns -1, but continues
1479 * fixing up the blocks in ram so the tree is consistent.
1480 */
1481 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1482 struct btrfs_root *root, struct btrfs_path *path,
1483 struct btrfs_disk_key *key, int level)
1484 {
1485 int i;
1486 int ret = 0;
1487 struct extent_buffer *t;
1488
1489 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1490 int tslot = path->slots[i];
1491 if (!path->nodes[i])
1492 break;
1493 t = path->nodes[i];
1494 btrfs_set_node_key(t, key, tslot);
1495 btrfs_mark_buffer_dirty(path->nodes[i]);
1496 if (tslot != 0)
1497 break;
1498 }
1499 return ret;
1500 }
1501
1502 /*
1503 * try to push data from one node into the next node left in the
1504 * tree.
1505 *
1506 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1507 * error, and > 0 if there was no room in the left hand block.
1508 */
1509 static int push_node_left(struct btrfs_trans_handle *trans,
1510 struct btrfs_root *root, struct extent_buffer *dst,
1511 struct extent_buffer *src, int empty)
1512 {
1513 int push_items = 0;
1514 int src_nritems;
1515 int dst_nritems;
1516 int ret = 0;
1517
1518 src_nritems = btrfs_header_nritems(src);
1519 dst_nritems = btrfs_header_nritems(dst);
1520 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1521 WARN_ON(btrfs_header_generation(src) != trans->transid);
1522 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1523
1524 if (!empty && src_nritems <= 8)
1525 return 1;
1526
1527 if (push_items <= 0) {
1528 return 1;
1529 }
1530
1531 if (empty) {
1532 push_items = min(src_nritems, push_items);
1533 if (push_items < src_nritems) {
1534 /* leave at least 8 pointers in the node if
1535 * we aren't going to empty it
1536 */
1537 if (src_nritems - push_items < 8) {
1538 if (push_items <= 8)
1539 return 1;
1540 push_items -= 8;
1541 }
1542 }
1543 } else
1544 push_items = min(src_nritems - 8, push_items);
1545
1546 copy_extent_buffer(dst, src,
1547 btrfs_node_key_ptr_offset(dst_nritems),
1548 btrfs_node_key_ptr_offset(0),
1549 push_items * sizeof(struct btrfs_key_ptr));
1550
1551 if (push_items < src_nritems) {
1552 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1553 btrfs_node_key_ptr_offset(push_items),
1554 (src_nritems - push_items) *
1555 sizeof(struct btrfs_key_ptr));
1556 }
1557 btrfs_set_header_nritems(src, src_nritems - push_items);
1558 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1559 btrfs_mark_buffer_dirty(src);
1560 btrfs_mark_buffer_dirty(dst);
1561 return ret;
1562 }
1563
1564 /*
1565 * try to push data from one node into the next node right in the
1566 * tree.
1567 *
1568 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1569 * error, and > 0 if there was no room in the right hand block.
1570 *
1571 * this will only push up to 1/2 the contents of the left node over
1572 */
1573 static int balance_node_right(struct btrfs_trans_handle *trans,
1574 struct btrfs_root *root,
1575 struct extent_buffer *dst,
1576 struct extent_buffer *src)
1577 {
1578 int push_items = 0;
1579 int max_push;
1580 int src_nritems;
1581 int dst_nritems;
1582 int ret = 0;
1583
1584 WARN_ON(btrfs_header_generation(src) != trans->transid);
1585 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1586
1587 src_nritems = btrfs_header_nritems(src);
1588 dst_nritems = btrfs_header_nritems(dst);
1589 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1590 if (push_items <= 0) {
1591 return 1;
1592 }
1593
1594 if (src_nritems < 4) {
1595 return 1;
1596 }
1597
1598 max_push = src_nritems / 2 + 1;
1599 /* don't try to empty the node */
1600 if (max_push >= src_nritems) {
1601 return 1;
1602 }
1603
1604 if (max_push < push_items)
1605 push_items = max_push;
1606
1607 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1608 btrfs_node_key_ptr_offset(0),
1609 (dst_nritems) *
1610 sizeof(struct btrfs_key_ptr));
1611
1612 copy_extent_buffer(dst, src,
1613 btrfs_node_key_ptr_offset(0),
1614 btrfs_node_key_ptr_offset(src_nritems - push_items),
1615 push_items * sizeof(struct btrfs_key_ptr));
1616
1617 btrfs_set_header_nritems(src, src_nritems - push_items);
1618 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1619
1620 btrfs_mark_buffer_dirty(src);
1621 btrfs_mark_buffer_dirty(dst);
1622 return ret;
1623 }
1624
1625 /*
1626 * helper function to insert a new root level in the tree.
1627 * A new node is allocated, and a single item is inserted to
1628 * point to the existing root
1629 *
1630 * returns zero on success or < 0 on failure.
1631 */
1632 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1633 struct btrfs_root *root,
1634 struct btrfs_path *path, int level)
1635 {
1636 u64 root_gen;
1637 u64 lower_gen;
1638 struct extent_buffer *lower;
1639 struct extent_buffer *c;
1640 struct extent_buffer *old;
1641 struct btrfs_disk_key lower_key;
1642
1643 BUG_ON(path->nodes[level]);
1644 BUG_ON(path->nodes[level-1] != root->node);
1645
1646 if (root->ref_cows)
1647 root_gen = trans->transid;
1648 else
1649 root_gen = 0;
1650
1651 lower = path->nodes[level-1];
1652 if (level == 1)
1653 btrfs_item_key(lower, &lower_key, 0);
1654 else
1655 btrfs_node_key(lower, &lower_key, 0);
1656
1657 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1658 root->root_key.objectid,
1659 root_gen, le64_to_cpu(lower_key.objectid),
1660 level, root->node->start, 0);
1661 if (IS_ERR(c))
1662 return PTR_ERR(c);
1663
1664 memset_extent_buffer(c, 0, 0, root->nodesize);
1665 btrfs_set_header_nritems(c, 1);
1666 btrfs_set_header_level(c, level);
1667 btrfs_set_header_bytenr(c, c->start);
1668 btrfs_set_header_generation(c, trans->transid);
1669 btrfs_set_header_owner(c, root->root_key.objectid);
1670
1671 write_extent_buffer(c, root->fs_info->fsid,
1672 (unsigned long)btrfs_header_fsid(c),
1673 BTRFS_FSID_SIZE);
1674
1675 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1676 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1677 BTRFS_UUID_SIZE);
1678
1679 btrfs_set_node_key(c, &lower_key, 0);
1680 btrfs_set_node_blockptr(c, 0, lower->start);
1681 lower_gen = btrfs_header_generation(lower);
1682 WARN_ON(lower_gen == 0);
1683
1684 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1685
1686 btrfs_mark_buffer_dirty(c);
1687
1688 spin_lock(&root->node_lock);
1689 old = root->node;
1690 root->node = c;
1691 spin_unlock(&root->node_lock);
1692
1693 /* the super has an extra ref to root->node */
1694 free_extent_buffer(old);
1695
1696 add_root_to_dirty_list(root);
1697 extent_buffer_get(c);
1698 path->nodes[level] = c;
1699 path->locks[level] = 1;
1700 path->slots[level] = 0;
1701
1702 if (root->ref_cows && lower_gen != trans->transid) {
1703 struct btrfs_path *back_path = btrfs_alloc_path();
1704 int ret;
1705 mutex_lock(&root->fs_info->alloc_mutex);
1706 ret = btrfs_insert_extent_backref(trans,
1707 root->fs_info->extent_root,
1708 path, lower->start,
1709 root->root_key.objectid,
1710 trans->transid, 0, 0);
1711 BUG_ON(ret);
1712 mutex_unlock(&root->fs_info->alloc_mutex);
1713 btrfs_free_path(back_path);
1714 }
1715 return 0;
1716 }
1717
1718 /*
1719 * worker function to insert a single pointer in a node.
1720 * the node should have enough room for the pointer already
1721 *
1722 * slot and level indicate where you want the key to go, and
1723 * blocknr is the block the key points to.
1724 *
1725 * returns zero on success and < 0 on any error
1726 */
1727 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1728 *root, struct btrfs_path *path, struct btrfs_disk_key
1729 *key, u64 bytenr, int slot, int level)
1730 {
1731 struct extent_buffer *lower;
1732 int nritems;
1733
1734 BUG_ON(!path->nodes[level]);
1735 lower = path->nodes[level];
1736 nritems = btrfs_header_nritems(lower);
1737 if (slot > nritems)
1738 BUG();
1739 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1740 BUG();
1741 if (slot != nritems) {
1742 memmove_extent_buffer(lower,
1743 btrfs_node_key_ptr_offset(slot + 1),
1744 btrfs_node_key_ptr_offset(slot),
1745 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1746 }
1747 btrfs_set_node_key(lower, key, slot);
1748 btrfs_set_node_blockptr(lower, slot, bytenr);
1749 WARN_ON(trans->transid == 0);
1750 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1751 btrfs_set_header_nritems(lower, nritems + 1);
1752 btrfs_mark_buffer_dirty(lower);
1753 return 0;
1754 }
1755
1756 /*
1757 * split the node at the specified level in path in two.
1758 * The path is corrected to point to the appropriate node after the split
1759 *
1760 * Before splitting this tries to make some room in the node by pushing
1761 * left and right, if either one works, it returns right away.
1762 *
1763 * returns 0 on success and < 0 on failure
1764 */
1765 static noinline int split_node(struct btrfs_trans_handle *trans,
1766 struct btrfs_root *root,
1767 struct btrfs_path *path, int level)
1768 {
1769 u64 root_gen;
1770 struct extent_buffer *c;
1771 struct extent_buffer *split;
1772 struct btrfs_disk_key disk_key;
1773 int mid;
1774 int ret;
1775 int wret;
1776 u32 c_nritems;
1777
1778 c = path->nodes[level];
1779 WARN_ON(btrfs_header_generation(c) != trans->transid);
1780 if (c == root->node) {
1781 /* trying to split the root, lets make a new one */
1782 ret = insert_new_root(trans, root, path, level + 1);
1783 if (ret)
1784 return ret;
1785 } else {
1786 ret = push_nodes_for_insert(trans, root, path, level);
1787 c = path->nodes[level];
1788 if (!ret && btrfs_header_nritems(c) <
1789 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1790 return 0;
1791 if (ret < 0)
1792 return ret;
1793 }
1794
1795 c_nritems = btrfs_header_nritems(c);
1796 if (root->ref_cows)
1797 root_gen = trans->transid;
1798 else
1799 root_gen = 0;
1800
1801 btrfs_node_key(c, &disk_key, 0);
1802 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1803 root->root_key.objectid,
1804 root_gen,
1805 btrfs_disk_key_objectid(&disk_key),
1806 level, c->start, 0);
1807 if (IS_ERR(split))
1808 return PTR_ERR(split);
1809
1810 btrfs_set_header_flags(split, btrfs_header_flags(c));
1811 btrfs_set_header_level(split, btrfs_header_level(c));
1812 btrfs_set_header_bytenr(split, split->start);
1813 btrfs_set_header_generation(split, trans->transid);
1814 btrfs_set_header_owner(split, root->root_key.objectid);
1815 btrfs_set_header_flags(split, 0);
1816 write_extent_buffer(split, root->fs_info->fsid,
1817 (unsigned long)btrfs_header_fsid(split),
1818 BTRFS_FSID_SIZE);
1819 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1820 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1821 BTRFS_UUID_SIZE);
1822
1823 mid = (c_nritems + 1) / 2;
1824
1825 copy_extent_buffer(split, c,
1826 btrfs_node_key_ptr_offset(0),
1827 btrfs_node_key_ptr_offset(mid),
1828 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1829 btrfs_set_header_nritems(split, c_nritems - mid);
1830 btrfs_set_header_nritems(c, mid);
1831 ret = 0;
1832
1833 btrfs_mark_buffer_dirty(c);
1834 btrfs_mark_buffer_dirty(split);
1835
1836 btrfs_node_key(split, &disk_key, 0);
1837 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1838 path->slots[level + 1] + 1,
1839 level + 1);
1840 if (wret)
1841 ret = wret;
1842
1843 if (path->slots[level] >= mid) {
1844 path->slots[level] -= mid;
1845 btrfs_tree_unlock(c);
1846 free_extent_buffer(c);
1847 path->nodes[level] = split;
1848 path->slots[level + 1] += 1;
1849 } else {
1850 btrfs_tree_unlock(split);
1851 free_extent_buffer(split);
1852 }
1853 return ret;
1854 }
1855
1856 /*
1857 * how many bytes are required to store the items in a leaf. start
1858 * and nr indicate which items in the leaf to check. This totals up the
1859 * space used both by the item structs and the item data
1860 */
1861 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1862 {
1863 int data_len;
1864 int nritems = btrfs_header_nritems(l);
1865 int end = min(nritems, start + nr) - 1;
1866
1867 if (!nr)
1868 return 0;
1869 data_len = btrfs_item_end_nr(l, start);
1870 data_len = data_len - btrfs_item_offset_nr(l, end);
1871 data_len += sizeof(struct btrfs_item) * nr;
1872 WARN_ON(data_len < 0);
1873 return data_len;
1874 }
1875
1876 /*
1877 * The space between the end of the leaf items and
1878 * the start of the leaf data. IOW, how much room
1879 * the leaf has left for both items and data
1880 */
1881 int noinline btrfs_leaf_free_space(struct btrfs_root *root,
1882 struct extent_buffer *leaf)
1883 {
1884 int nritems = btrfs_header_nritems(leaf);
1885 int ret;
1886 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1887 if (ret < 0) {
1888 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1889 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1890 leaf_space_used(leaf, 0, nritems), nritems);
1891 }
1892 return ret;
1893 }
1894
1895 /*
1896 * push some data in the path leaf to the right, trying to free up at
1897 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1898 *
1899 * returns 1 if the push failed because the other node didn't have enough
1900 * room, 0 if everything worked out and < 0 if there were major errors.
1901 */
1902 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1903 *root, struct btrfs_path *path, int data_size,
1904 int empty)
1905 {
1906 struct extent_buffer *left = path->nodes[0];
1907 struct extent_buffer *right;
1908 struct extent_buffer *upper;
1909 struct btrfs_disk_key disk_key;
1910 int slot;
1911 u32 i;
1912 int free_space;
1913 int push_space = 0;
1914 int push_items = 0;
1915 struct btrfs_item *item;
1916 u32 left_nritems;
1917 u32 nr;
1918 u32 right_nritems;
1919 u32 data_end;
1920 u32 this_item_size;
1921 int ret;
1922
1923 slot = path->slots[1];
1924 if (!path->nodes[1]) {
1925 return 1;
1926 }
1927 upper = path->nodes[1];
1928 if (slot >= btrfs_header_nritems(upper) - 1)
1929 return 1;
1930
1931 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
1932
1933 right = read_node_slot(root, upper, slot + 1);
1934 btrfs_tree_lock(right);
1935 free_space = btrfs_leaf_free_space(root, right);
1936 if (free_space < data_size + sizeof(struct btrfs_item))
1937 goto out_unlock;
1938
1939 /* cow and double check */
1940 ret = btrfs_cow_block(trans, root, right, upper,
1941 slot + 1, &right, 0);
1942 if (ret)
1943 goto out_unlock;
1944
1945 free_space = btrfs_leaf_free_space(root, right);
1946 if (free_space < data_size + sizeof(struct btrfs_item))
1947 goto out_unlock;
1948
1949 left_nritems = btrfs_header_nritems(left);
1950 if (left_nritems == 0)
1951 goto out_unlock;
1952
1953 if (empty)
1954 nr = 0;
1955 else
1956 nr = 1;
1957
1958 i = left_nritems - 1;
1959 while (i >= nr) {
1960 item = btrfs_item_nr(left, i);
1961
1962 if (path->slots[0] == i)
1963 push_space += data_size + sizeof(*item);
1964
1965 if (!left->map_token) {
1966 map_extent_buffer(left, (unsigned long)item,
1967 sizeof(struct btrfs_item),
1968 &left->map_token, &left->kaddr,
1969 &left->map_start, &left->map_len,
1970 KM_USER1);
1971 }
1972
1973 this_item_size = btrfs_item_size(left, item);
1974 if (this_item_size + sizeof(*item) + push_space > free_space)
1975 break;
1976 push_items++;
1977 push_space += this_item_size + sizeof(*item);
1978 if (i == 0)
1979 break;
1980 i--;
1981 }
1982 if (left->map_token) {
1983 unmap_extent_buffer(left, left->map_token, KM_USER1);
1984 left->map_token = NULL;
1985 }
1986
1987 if (push_items == 0)
1988 goto out_unlock;
1989
1990 if (!empty && push_items == left_nritems)
1991 WARN_ON(1);
1992
1993 /* push left to right */
1994 right_nritems = btrfs_header_nritems(right);
1995
1996 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1997 push_space -= leaf_data_end(root, left);
1998
1999 /* make room in the right data area */
2000 data_end = leaf_data_end(root, right);
2001 memmove_extent_buffer(right,
2002 btrfs_leaf_data(right) + data_end - push_space,
2003 btrfs_leaf_data(right) + data_end,
2004 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2005
2006 /* copy from the left data area */
2007 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2008 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2009 btrfs_leaf_data(left) + leaf_data_end(root, left),
2010 push_space);
2011
2012 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2013 btrfs_item_nr_offset(0),
2014 right_nritems * sizeof(struct btrfs_item));
2015
2016 /* copy the items from left to right */
2017 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2018 btrfs_item_nr_offset(left_nritems - push_items),
2019 push_items * sizeof(struct btrfs_item));
2020
2021 /* update the item pointers */
2022 right_nritems += push_items;
2023 btrfs_set_header_nritems(right, right_nritems);
2024 push_space = BTRFS_LEAF_DATA_SIZE(root);
2025 for (i = 0; i < right_nritems; i++) {
2026 item = btrfs_item_nr(right, i);
2027 if (!right->map_token) {
2028 map_extent_buffer(right, (unsigned long)item,
2029 sizeof(struct btrfs_item),
2030 &right->map_token, &right->kaddr,
2031 &right->map_start, &right->map_len,
2032 KM_USER1);
2033 }
2034 push_space -= btrfs_item_size(right, item);
2035 btrfs_set_item_offset(right, item, push_space);
2036 }
2037
2038 if (right->map_token) {
2039 unmap_extent_buffer(right, right->map_token, KM_USER1);
2040 right->map_token = NULL;
2041 }
2042 left_nritems -= push_items;
2043 btrfs_set_header_nritems(left, left_nritems);
2044
2045 if (left_nritems)
2046 btrfs_mark_buffer_dirty(left);
2047 btrfs_mark_buffer_dirty(right);
2048
2049 btrfs_item_key(right, &disk_key, 0);
2050 btrfs_set_node_key(upper, &disk_key, slot + 1);
2051 btrfs_mark_buffer_dirty(upper);
2052
2053 /* then fixup the leaf pointer in the path */
2054 if (path->slots[0] >= left_nritems) {
2055 path->slots[0] -= left_nritems;
2056 if (btrfs_header_nritems(path->nodes[0]) == 0)
2057 clean_tree_block(trans, root, path->nodes[0]);
2058 btrfs_tree_unlock(path->nodes[0]);
2059 free_extent_buffer(path->nodes[0]);
2060 path->nodes[0] = right;
2061 path->slots[1] += 1;
2062 } else {
2063 btrfs_tree_unlock(right);
2064 free_extent_buffer(right);
2065 }
2066 return 0;
2067
2068 out_unlock:
2069 btrfs_tree_unlock(right);
2070 free_extent_buffer(right);
2071 return 1;
2072 }
2073
2074 /*
2075 * push some data in the path leaf to the left, trying to free up at
2076 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2077 */
2078 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2079 *root, struct btrfs_path *path, int data_size,
2080 int empty)
2081 {
2082 struct btrfs_disk_key disk_key;
2083 struct extent_buffer *right = path->nodes[0];
2084 struct extent_buffer *left;
2085 int slot;
2086 int i;
2087 int free_space;
2088 int push_space = 0;
2089 int push_items = 0;
2090 struct btrfs_item *item;
2091 u32 old_left_nritems;
2092 u32 right_nritems;
2093 u32 nr;
2094 int ret = 0;
2095 int wret;
2096 u32 this_item_size;
2097 u32 old_left_item_size;
2098
2099 slot = path->slots[1];
2100 if (slot == 0)
2101 return 1;
2102 if (!path->nodes[1])
2103 return 1;
2104
2105 right_nritems = btrfs_header_nritems(right);
2106 if (right_nritems == 0) {
2107 return 1;
2108 }
2109
2110 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2111
2112 left = read_node_slot(root, path->nodes[1], slot - 1);
2113 btrfs_tree_lock(left);
2114 free_space = btrfs_leaf_free_space(root, left);
2115 if (free_space < data_size + sizeof(struct btrfs_item)) {
2116 ret = 1;
2117 goto out;
2118 }
2119
2120 /* cow and double check */
2121 ret = btrfs_cow_block(trans, root, left,
2122 path->nodes[1], slot - 1, &left, 0);
2123 if (ret) {
2124 /* we hit -ENOSPC, but it isn't fatal here */
2125 ret = 1;
2126 goto out;
2127 }
2128
2129 free_space = btrfs_leaf_free_space(root, left);
2130 if (free_space < data_size + sizeof(struct btrfs_item)) {
2131 ret = 1;
2132 goto out;
2133 }
2134
2135 if (empty)
2136 nr = right_nritems;
2137 else
2138 nr = right_nritems - 1;
2139
2140 for (i = 0; i < nr; i++) {
2141 item = btrfs_item_nr(right, i);
2142 if (!right->map_token) {
2143 map_extent_buffer(right, (unsigned long)item,
2144 sizeof(struct btrfs_item),
2145 &right->map_token, &right->kaddr,
2146 &right->map_start, &right->map_len,
2147 KM_USER1);
2148 }
2149
2150 if (path->slots[0] == i)
2151 push_space += data_size + sizeof(*item);
2152
2153 this_item_size = btrfs_item_size(right, item);
2154 if (this_item_size + sizeof(*item) + push_space > free_space)
2155 break;
2156
2157 push_items++;
2158 push_space += this_item_size + sizeof(*item);
2159 }
2160
2161 if (right->map_token) {
2162 unmap_extent_buffer(right, right->map_token, KM_USER1);
2163 right->map_token = NULL;
2164 }
2165
2166 if (push_items == 0) {
2167 ret = 1;
2168 goto out;
2169 }
2170 if (!empty && push_items == btrfs_header_nritems(right))
2171 WARN_ON(1);
2172
2173 /* push data from right to left */
2174 copy_extent_buffer(left, right,
2175 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2176 btrfs_item_nr_offset(0),
2177 push_items * sizeof(struct btrfs_item));
2178
2179 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2180 btrfs_item_offset_nr(right, push_items -1);
2181
2182 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2183 leaf_data_end(root, left) - push_space,
2184 btrfs_leaf_data(right) +
2185 btrfs_item_offset_nr(right, push_items - 1),
2186 push_space);
2187 old_left_nritems = btrfs_header_nritems(left);
2188 BUG_ON(old_left_nritems < 0);
2189
2190 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2191 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2192 u32 ioff;
2193
2194 item = btrfs_item_nr(left, i);
2195 if (!left->map_token) {
2196 map_extent_buffer(left, (unsigned long)item,
2197 sizeof(struct btrfs_item),
2198 &left->map_token, &left->kaddr,
2199 &left->map_start, &left->map_len,
2200 KM_USER1);
2201 }
2202
2203 ioff = btrfs_item_offset(left, item);
2204 btrfs_set_item_offset(left, item,
2205 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2206 }
2207 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2208 if (left->map_token) {
2209 unmap_extent_buffer(left, left->map_token, KM_USER1);
2210 left->map_token = NULL;
2211 }
2212
2213 /* fixup right node */
2214 if (push_items > right_nritems) {
2215 printk("push items %d nr %u\n", push_items, right_nritems);
2216 WARN_ON(1);
2217 }
2218
2219 if (push_items < right_nritems) {
2220 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2221 leaf_data_end(root, right);
2222 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2223 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2224 btrfs_leaf_data(right) +
2225 leaf_data_end(root, right), push_space);
2226
2227 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2228 btrfs_item_nr_offset(push_items),
2229 (btrfs_header_nritems(right) - push_items) *
2230 sizeof(struct btrfs_item));
2231 }
2232 right_nritems -= push_items;
2233 btrfs_set_header_nritems(right, right_nritems);
2234 push_space = BTRFS_LEAF_DATA_SIZE(root);
2235 for (i = 0; i < right_nritems; i++) {
2236 item = btrfs_item_nr(right, i);
2237
2238 if (!right->map_token) {
2239 map_extent_buffer(right, (unsigned long)item,
2240 sizeof(struct btrfs_item),
2241 &right->map_token, &right->kaddr,
2242 &right->map_start, &right->map_len,
2243 KM_USER1);
2244 }
2245
2246 push_space = push_space - btrfs_item_size(right, item);
2247 btrfs_set_item_offset(right, item, push_space);
2248 }
2249 if (right->map_token) {
2250 unmap_extent_buffer(right, right->map_token, KM_USER1);
2251 right->map_token = NULL;
2252 }
2253
2254 btrfs_mark_buffer_dirty(left);
2255 if (right_nritems)
2256 btrfs_mark_buffer_dirty(right);
2257
2258 btrfs_item_key(right, &disk_key, 0);
2259 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2260 if (wret)
2261 ret = wret;
2262
2263 /* then fixup the leaf pointer in the path */
2264 if (path->slots[0] < push_items) {
2265 path->slots[0] += old_left_nritems;
2266 if (btrfs_header_nritems(path->nodes[0]) == 0)
2267 clean_tree_block(trans, root, path->nodes[0]);
2268 btrfs_tree_unlock(path->nodes[0]);
2269 free_extent_buffer(path->nodes[0]);
2270 path->nodes[0] = left;
2271 path->slots[1] -= 1;
2272 } else {
2273 btrfs_tree_unlock(left);
2274 free_extent_buffer(left);
2275 path->slots[0] -= push_items;
2276 }
2277 BUG_ON(path->slots[0] < 0);
2278 return ret;
2279 out:
2280 btrfs_tree_unlock(left);
2281 free_extent_buffer(left);
2282 return ret;
2283 }
2284
2285 /*
2286 * split the path's leaf in two, making sure there is at least data_size
2287 * available for the resulting leaf level of the path.
2288 *
2289 * returns 0 if all went well and < 0 on failure.
2290 */
2291 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2292 struct btrfs_root *root,
2293 struct btrfs_key *ins_key,
2294 struct btrfs_path *path, int data_size,
2295 int extend)
2296 {
2297 u64 root_gen;
2298 struct extent_buffer *l;
2299 u32 nritems;
2300 int mid;
2301 int slot;
2302 struct extent_buffer *right;
2303 int space_needed = data_size + sizeof(struct btrfs_item);
2304 int data_copy_size;
2305 int rt_data_off;
2306 int i;
2307 int ret = 0;
2308 int wret;
2309 int double_split;
2310 int num_doubles = 0;
2311 struct btrfs_disk_key disk_key;
2312
2313 if (extend)
2314 space_needed = data_size;
2315
2316 if (root->ref_cows)
2317 root_gen = trans->transid;
2318 else
2319 root_gen = 0;
2320
2321 /* first try to make some room by pushing left and right */
2322 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2323 wret = push_leaf_right(trans, root, path, data_size, 0);
2324 if (wret < 0) {
2325 return wret;
2326 }
2327 if (wret) {
2328 wret = push_leaf_left(trans, root, path, data_size, 0);
2329 if (wret < 0)
2330 return wret;
2331 }
2332 l = path->nodes[0];
2333
2334 /* did the pushes work? */
2335 if (btrfs_leaf_free_space(root, l) >= space_needed)
2336 return 0;
2337 }
2338
2339 if (!path->nodes[1]) {
2340 ret = insert_new_root(trans, root, path, 1);
2341 if (ret)
2342 return ret;
2343 }
2344 again:
2345 double_split = 0;
2346 l = path->nodes[0];
2347 slot = path->slots[0];
2348 nritems = btrfs_header_nritems(l);
2349 mid = (nritems + 1)/ 2;
2350
2351 btrfs_item_key(l, &disk_key, 0);
2352
2353 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2354 root->root_key.objectid,
2355 root_gen,
2356 le64_to_cpu(disk_key.objectid),
2357 0, l->start, 0);
2358 if (IS_ERR(right)) {
2359 BUG_ON(1);
2360 return PTR_ERR(right);
2361 }
2362
2363 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2364 btrfs_set_header_bytenr(right, right->start);
2365 btrfs_set_header_generation(right, trans->transid);
2366 btrfs_set_header_owner(right, root->root_key.objectid);
2367 btrfs_set_header_level(right, 0);
2368 write_extent_buffer(right, root->fs_info->fsid,
2369 (unsigned long)btrfs_header_fsid(right),
2370 BTRFS_FSID_SIZE);
2371
2372 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2373 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2374 BTRFS_UUID_SIZE);
2375 if (mid <= slot) {
2376 if (nritems == 1 ||
2377 leaf_space_used(l, mid, nritems - mid) + space_needed >
2378 BTRFS_LEAF_DATA_SIZE(root)) {
2379 if (slot >= nritems) {
2380 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2381 btrfs_set_header_nritems(right, 0);
2382 wret = insert_ptr(trans, root, path,
2383 &disk_key, right->start,
2384 path->slots[1] + 1, 1);
2385 if (wret)
2386 ret = wret;
2387
2388 btrfs_tree_unlock(path->nodes[0]);
2389 free_extent_buffer(path->nodes[0]);
2390 path->nodes[0] = right;
2391 path->slots[0] = 0;
2392 path->slots[1] += 1;
2393 btrfs_mark_buffer_dirty(right);
2394 return ret;
2395 }
2396 mid = slot;
2397 if (mid != nritems &&
2398 leaf_space_used(l, mid, nritems - mid) +
2399 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2400 double_split = 1;
2401 }
2402 }
2403 } else {
2404 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2405 BTRFS_LEAF_DATA_SIZE(root)) {
2406 if (!extend && slot == 0) {
2407 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2408 btrfs_set_header_nritems(right, 0);
2409 wret = insert_ptr(trans, root, path,
2410 &disk_key,
2411 right->start,
2412 path->slots[1], 1);
2413 if (wret)
2414 ret = wret;
2415 btrfs_tree_unlock(path->nodes[0]);
2416 free_extent_buffer(path->nodes[0]);
2417 path->nodes[0] = right;
2418 path->slots[0] = 0;
2419 if (path->slots[1] == 0) {
2420 wret = fixup_low_keys(trans, root,
2421 path, &disk_key, 1);
2422 if (wret)
2423 ret = wret;
2424 }
2425 btrfs_mark_buffer_dirty(right);
2426 return ret;
2427 } else if (extend && slot == 0) {
2428 mid = 1;
2429 } else {
2430 mid = slot;
2431 if (mid != nritems &&
2432 leaf_space_used(l, mid, nritems - mid) +
2433 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2434 double_split = 1;
2435 }
2436 }
2437 }
2438 }
2439 nritems = nritems - mid;
2440 btrfs_set_header_nritems(right, nritems);
2441 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2442
2443 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2444 btrfs_item_nr_offset(mid),
2445 nritems * sizeof(struct btrfs_item));
2446
2447 copy_extent_buffer(right, l,
2448 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2449 data_copy_size, btrfs_leaf_data(l) +
2450 leaf_data_end(root, l), data_copy_size);
2451
2452 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2453 btrfs_item_end_nr(l, mid);
2454
2455 for (i = 0; i < nritems; i++) {
2456 struct btrfs_item *item = btrfs_item_nr(right, i);
2457 u32 ioff;
2458
2459 if (!right->map_token) {
2460 map_extent_buffer(right, (unsigned long)item,
2461 sizeof(struct btrfs_item),
2462 &right->map_token, &right->kaddr,
2463 &right->map_start, &right->map_len,
2464 KM_USER1);
2465 }
2466
2467 ioff = btrfs_item_offset(right, item);
2468 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2469 }
2470
2471 if (right->map_token) {
2472 unmap_extent_buffer(right, right->map_token, KM_USER1);
2473 right->map_token = NULL;
2474 }
2475
2476 btrfs_set_header_nritems(l, mid);
2477 ret = 0;
2478 btrfs_item_key(right, &disk_key, 0);
2479 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2480 path->slots[1] + 1, 1);
2481 if (wret)
2482 ret = wret;
2483
2484 btrfs_mark_buffer_dirty(right);
2485 btrfs_mark_buffer_dirty(l);
2486 BUG_ON(path->slots[0] != slot);
2487
2488 if (mid <= slot) {
2489 btrfs_tree_unlock(path->nodes[0]);
2490 free_extent_buffer(path->nodes[0]);
2491 path->nodes[0] = right;
2492 path->slots[0] -= mid;
2493 path->slots[1] += 1;
2494 } else {
2495 btrfs_tree_unlock(right);
2496 free_extent_buffer(right);
2497 }
2498
2499 BUG_ON(path->slots[0] < 0);
2500
2501 if (double_split) {
2502 BUG_ON(num_doubles != 0);
2503 num_doubles++;
2504 goto again;
2505 }
2506 return ret;
2507 }
2508
2509 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2510 struct btrfs_root *root,
2511 struct btrfs_path *path,
2512 u32 new_size, int from_end)
2513 {
2514 int ret = 0;
2515 int slot;
2516 int slot_orig;
2517 struct extent_buffer *leaf;
2518 struct btrfs_item *item;
2519 u32 nritems;
2520 unsigned int data_end;
2521 unsigned int old_data_start;
2522 unsigned int old_size;
2523 unsigned int size_diff;
2524 int i;
2525
2526 slot_orig = path->slots[0];
2527 leaf = path->nodes[0];
2528 slot = path->slots[0];
2529
2530 old_size = btrfs_item_size_nr(leaf, slot);
2531 if (old_size == new_size)
2532 return 0;
2533
2534 nritems = btrfs_header_nritems(leaf);
2535 data_end = leaf_data_end(root, leaf);
2536
2537 old_data_start = btrfs_item_offset_nr(leaf, slot);
2538
2539 size_diff = old_size - new_size;
2540
2541 BUG_ON(slot < 0);
2542 BUG_ON(slot >= nritems);
2543
2544 /*
2545 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2546 */
2547 /* first correct the data pointers */
2548 for (i = slot; i < nritems; i++) {
2549 u32 ioff;
2550 item = btrfs_item_nr(leaf, i);
2551
2552 if (!leaf->map_token) {
2553 map_extent_buffer(leaf, (unsigned long)item,
2554 sizeof(struct btrfs_item),
2555 &leaf->map_token, &leaf->kaddr,
2556 &leaf->map_start, &leaf->map_len,
2557 KM_USER1);
2558 }
2559
2560 ioff = btrfs_item_offset(leaf, item);
2561 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2562 }
2563
2564 if (leaf->map_token) {
2565 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2566 leaf->map_token = NULL;
2567 }
2568
2569 /* shift the data */
2570 if (from_end) {
2571 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2572 data_end + size_diff, btrfs_leaf_data(leaf) +
2573 data_end, old_data_start + new_size - data_end);
2574 } else {
2575 struct btrfs_disk_key disk_key;
2576 u64 offset;
2577
2578 btrfs_item_key(leaf, &disk_key, slot);
2579
2580 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2581 unsigned long ptr;
2582 struct btrfs_file_extent_item *fi;
2583
2584 fi = btrfs_item_ptr(leaf, slot,
2585 struct btrfs_file_extent_item);
2586 fi = (struct btrfs_file_extent_item *)(
2587 (unsigned long)fi - size_diff);
2588
2589 if (btrfs_file_extent_type(leaf, fi) ==
2590 BTRFS_FILE_EXTENT_INLINE) {
2591 ptr = btrfs_item_ptr_offset(leaf, slot);
2592 memmove_extent_buffer(leaf, ptr,
2593 (unsigned long)fi,
2594 offsetof(struct btrfs_file_extent_item,
2595 disk_bytenr));
2596 }
2597 }
2598
2599 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2600 data_end + size_diff, btrfs_leaf_data(leaf) +
2601 data_end, old_data_start - data_end);
2602
2603 offset = btrfs_disk_key_offset(&disk_key);
2604 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2605 btrfs_set_item_key(leaf, &disk_key, slot);
2606 if (slot == 0)
2607 fixup_low_keys(trans, root, path, &disk_key, 1);
2608 }
2609
2610 item = btrfs_item_nr(leaf, slot);
2611 btrfs_set_item_size(leaf, item, new_size);
2612 btrfs_mark_buffer_dirty(leaf);
2613
2614 ret = 0;
2615 if (btrfs_leaf_free_space(root, leaf) < 0) {
2616 btrfs_print_leaf(root, leaf);
2617 BUG();
2618 }
2619 return ret;
2620 }
2621
2622 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2623 struct btrfs_root *root, struct btrfs_path *path,
2624 u32 data_size)
2625 {
2626 int ret = 0;
2627 int slot;
2628 int slot_orig;
2629 struct extent_buffer *leaf;
2630 struct btrfs_item *item;
2631 u32 nritems;
2632 unsigned int data_end;
2633 unsigned int old_data;
2634 unsigned int old_size;
2635 int i;
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) < data_size) {
2644 btrfs_print_leaf(root, leaf);
2645 BUG();
2646 }
2647 slot = path->slots[0];
2648 old_data = btrfs_item_end_nr(leaf, slot);
2649
2650 BUG_ON(slot < 0);
2651 if (slot >= nritems) {
2652 btrfs_print_leaf(root, leaf);
2653 printk("slot %d too large, nritems %d\n", slot, nritems);
2654 BUG_ON(1);
2655 }
2656
2657 /*
2658 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2659 */
2660 /* first correct the data pointers */
2661 for (i = slot; i < nritems; i++) {
2662 u32 ioff;
2663 item = btrfs_item_nr(leaf, i);
2664
2665 if (!leaf->map_token) {
2666 map_extent_buffer(leaf, (unsigned long)item,
2667 sizeof(struct btrfs_item),
2668 &leaf->map_token, &leaf->kaddr,
2669 &leaf->map_start, &leaf->map_len,
2670 KM_USER1);
2671 }
2672 ioff = btrfs_item_offset(leaf, item);
2673 btrfs_set_item_offset(leaf, item, ioff - data_size);
2674 }
2675
2676 if (leaf->map_token) {
2677 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2678 leaf->map_token = NULL;
2679 }
2680
2681 /* shift the data */
2682 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2683 data_end - data_size, btrfs_leaf_data(leaf) +
2684 data_end, old_data - data_end);
2685
2686 data_end = old_data;
2687 old_size = btrfs_item_size_nr(leaf, slot);
2688 item = btrfs_item_nr(leaf, slot);
2689 btrfs_set_item_size(leaf, item, old_size + data_size);
2690 btrfs_mark_buffer_dirty(leaf);
2691
2692 ret = 0;
2693 if (btrfs_leaf_free_space(root, leaf) < 0) {
2694 btrfs_print_leaf(root, leaf);
2695 BUG();
2696 }
2697 return ret;
2698 }
2699
2700 /*
2701 * Given a key and some data, insert an item into the tree.
2702 * This does all the path init required, making room in the tree if needed.
2703 */
2704 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2705 struct btrfs_root *root,
2706 struct btrfs_path *path,
2707 struct btrfs_key *cpu_key, u32 *data_size,
2708 int nr)
2709 {
2710 struct extent_buffer *leaf;
2711 struct btrfs_item *item;
2712 int ret = 0;
2713 int slot;
2714 int slot_orig;
2715 int i;
2716 u32 nritems;
2717 u32 total_size = 0;
2718 u32 total_data = 0;
2719 unsigned int data_end;
2720 struct btrfs_disk_key disk_key;
2721
2722 for (i = 0; i < nr; i++) {
2723 total_data += data_size[i];
2724 }
2725
2726 total_size = total_data + (nr * sizeof(struct btrfs_item));
2727 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2728 if (ret == 0)
2729 return -EEXIST;
2730 if (ret < 0)
2731 goto out;
2732
2733 slot_orig = path->slots[0];
2734 leaf = path->nodes[0];
2735
2736 nritems = btrfs_header_nritems(leaf);
2737 data_end = leaf_data_end(root, leaf);
2738
2739 if (btrfs_leaf_free_space(root, leaf) < total_size) {
2740 btrfs_print_leaf(root, leaf);
2741 printk("not enough freespace need %u have %d\n",
2742 total_size, btrfs_leaf_free_space(root, leaf));
2743 BUG();
2744 }
2745
2746 slot = path->slots[0];
2747 BUG_ON(slot < 0);
2748
2749 if (slot != nritems) {
2750 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2751
2752 if (old_data < data_end) {
2753 btrfs_print_leaf(root, leaf);
2754 printk("slot %d old_data %d data_end %d\n",
2755 slot, old_data, data_end);
2756 BUG_ON(1);
2757 }
2758 /*
2759 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2760 */
2761 /* first correct the data pointers */
2762 WARN_ON(leaf->map_token);
2763 for (i = slot; i < nritems; i++) {
2764 u32 ioff;
2765
2766 item = btrfs_item_nr(leaf, i);
2767 if (!leaf->map_token) {
2768 map_extent_buffer(leaf, (unsigned long)item,
2769 sizeof(struct btrfs_item),
2770 &leaf->map_token, &leaf->kaddr,
2771 &leaf->map_start, &leaf->map_len,
2772 KM_USER1);
2773 }
2774
2775 ioff = btrfs_item_offset(leaf, item);
2776 btrfs_set_item_offset(leaf, item, ioff - total_data);
2777 }
2778 if (leaf->map_token) {
2779 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2780 leaf->map_token = NULL;
2781 }
2782
2783 /* shift the items */
2784 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2785 btrfs_item_nr_offset(slot),
2786 (nritems - slot) * sizeof(struct btrfs_item));
2787
2788 /* shift the data */
2789 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2790 data_end - total_data, btrfs_leaf_data(leaf) +
2791 data_end, old_data - data_end);
2792 data_end = old_data;
2793 }
2794
2795 /* setup the item for the new data */
2796 for (i = 0; i < nr; i++) {
2797 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2798 btrfs_set_item_key(leaf, &disk_key, slot + i);
2799 item = btrfs_item_nr(leaf, slot + i);
2800 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2801 data_end -= data_size[i];
2802 btrfs_set_item_size(leaf, item, data_size[i]);
2803 }
2804 btrfs_set_header_nritems(leaf, nritems + nr);
2805 btrfs_mark_buffer_dirty(leaf);
2806
2807 ret = 0;
2808 if (slot == 0) {
2809 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2810 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2811 }
2812
2813 if (btrfs_leaf_free_space(root, leaf) < 0) {
2814 btrfs_print_leaf(root, leaf);
2815 BUG();
2816 }
2817 out:
2818 return ret;
2819 }
2820
2821 /*
2822 * Given a key and some data, insert an item into the tree.
2823 * This does all the path init required, making room in the tree if needed.
2824 */
2825 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2826 *root, struct btrfs_key *cpu_key, void *data, u32
2827 data_size)
2828 {
2829 int ret = 0;
2830 struct btrfs_path *path;
2831 struct extent_buffer *leaf;
2832 unsigned long ptr;
2833
2834 path = btrfs_alloc_path();
2835 BUG_ON(!path);
2836 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2837 if (!ret) {
2838 leaf = path->nodes[0];
2839 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2840 write_extent_buffer(leaf, data, ptr, data_size);
2841 btrfs_mark_buffer_dirty(leaf);
2842 }
2843 btrfs_free_path(path);
2844 return ret;
2845 }
2846
2847 /*
2848 * delete the pointer from a given node.
2849 *
2850 * If the delete empties a node, the node is removed from the tree,
2851 * continuing all the way the root if required. The root is converted into
2852 * a leaf if all the nodes are emptied.
2853 */
2854 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2855 struct btrfs_path *path, int level, int slot)
2856 {
2857 struct extent_buffer *parent = path->nodes[level];
2858 u32 nritems;
2859 int ret = 0;
2860 int wret;
2861
2862 nritems = btrfs_header_nritems(parent);
2863 if (slot != nritems -1) {
2864 memmove_extent_buffer(parent,
2865 btrfs_node_key_ptr_offset(slot),
2866 btrfs_node_key_ptr_offset(slot + 1),
2867 sizeof(struct btrfs_key_ptr) *
2868 (nritems - slot - 1));
2869 }
2870 nritems--;
2871 btrfs_set_header_nritems(parent, nritems);
2872 if (nritems == 0 && parent == root->node) {
2873 BUG_ON(btrfs_header_level(root->node) != 1);
2874 /* just turn the root into a leaf and break */
2875 btrfs_set_header_level(root->node, 0);
2876 } else if (slot == 0) {
2877 struct btrfs_disk_key disk_key;
2878
2879 btrfs_node_key(parent, &disk_key, 0);
2880 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2881 if (wret)
2882 ret = wret;
2883 }
2884 btrfs_mark_buffer_dirty(parent);
2885 return ret;
2886 }
2887
2888 /*
2889 * delete the item at the leaf level in path. If that empties
2890 * the leaf, remove it from the tree
2891 */
2892 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2893 struct btrfs_path *path, int slot, int nr)
2894 {
2895 struct extent_buffer *leaf;
2896 struct btrfs_item *item;
2897 int last_off;
2898 int dsize = 0;
2899 int ret = 0;
2900 int wret;
2901 int i;
2902 u32 nritems;
2903
2904 leaf = path->nodes[0];
2905 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2906
2907 for (i = 0; i < nr; i++)
2908 dsize += btrfs_item_size_nr(leaf, slot + i);
2909
2910 nritems = btrfs_header_nritems(leaf);
2911
2912 if (slot + nr != nritems) {
2913 int data_end = leaf_data_end(root, leaf);
2914
2915 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2916 data_end + dsize,
2917 btrfs_leaf_data(leaf) + data_end,
2918 last_off - data_end);
2919
2920 for (i = slot + nr; i < nritems; i++) {
2921 u32 ioff;
2922
2923 item = btrfs_item_nr(leaf, i);
2924 if (!leaf->map_token) {
2925 map_extent_buffer(leaf, (unsigned long)item,
2926 sizeof(struct btrfs_item),
2927 &leaf->map_token, &leaf->kaddr,
2928 &leaf->map_start, &leaf->map_len,
2929 KM_USER1);
2930 }
2931 ioff = btrfs_item_offset(leaf, item);
2932 btrfs_set_item_offset(leaf, item, ioff + dsize);
2933 }
2934
2935 if (leaf->map_token) {
2936 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2937 leaf->map_token = NULL;
2938 }
2939
2940 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2941 btrfs_item_nr_offset(slot + nr),
2942 sizeof(struct btrfs_item) *
2943 (nritems - slot - nr));
2944 }
2945 btrfs_set_header_nritems(leaf, nritems - nr);
2946 nritems -= nr;
2947
2948 /* delete the leaf if we've emptied it */
2949 if (nritems == 0) {
2950 if (leaf == root->node) {
2951 btrfs_set_header_level(leaf, 0);
2952 } else {
2953 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2954 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2955 if (wret)
2956 ret = wret;
2957 wret = btrfs_free_extent(trans, root,
2958 leaf->start, leaf->len,
2959 btrfs_header_owner(path->nodes[1]),
2960 root_gen, 0, 0, 1);
2961 if (wret)
2962 ret = wret;
2963 }
2964 } else {
2965 int used = leaf_space_used(leaf, 0, nritems);
2966 if (slot == 0) {
2967 struct btrfs_disk_key disk_key;
2968
2969 btrfs_item_key(leaf, &disk_key, 0);
2970 wret = fixup_low_keys(trans, root, path,
2971 &disk_key, 1);
2972 if (wret)
2973 ret = wret;
2974 }
2975
2976 /* delete the leaf if it is mostly empty */
2977 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2978 /* push_leaf_left fixes the path.
2979 * make sure the path still points to our leaf
2980 * for possible call to del_ptr below
2981 */
2982 slot = path->slots[1];
2983 extent_buffer_get(leaf);
2984
2985 wret = push_leaf_left(trans, root, path, 1, 1);
2986 if (wret < 0 && wret != -ENOSPC)
2987 ret = wret;
2988
2989 if (path->nodes[0] == leaf &&
2990 btrfs_header_nritems(leaf)) {
2991 wret = push_leaf_right(trans, root, path, 1, 1);
2992 if (wret < 0 && wret != -ENOSPC)
2993 ret = wret;
2994 }
2995
2996 if (btrfs_header_nritems(leaf) == 0) {
2997 u64 root_gen;
2998 u64 bytenr = leaf->start;
2999 u32 blocksize = leaf->len;
3000
3001 root_gen = btrfs_header_generation(
3002 path->nodes[1]);
3003
3004 wret = del_ptr(trans, root, path, 1, slot);
3005 if (wret)
3006 ret = wret;
3007
3008 free_extent_buffer(leaf);
3009 wret = btrfs_free_extent(trans, root, bytenr,
3010 blocksize,
3011 btrfs_header_owner(path->nodes[1]),
3012 root_gen, 0, 0, 1);
3013 if (wret)
3014 ret = wret;
3015 } else {
3016 /* if we're still in the path, make sure
3017 * we're dirty. Otherwise, one of the
3018 * push_leaf functions must have already
3019 * dirtied this buffer
3020 */
3021 if (path->nodes[0] == leaf)
3022 btrfs_mark_buffer_dirty(leaf);
3023 free_extent_buffer(leaf);
3024 }
3025 } else {
3026 btrfs_mark_buffer_dirty(leaf);
3027 }
3028 }
3029 return ret;
3030 }
3031
3032 /*
3033 * search the tree again to find a leaf with lesser keys
3034 * returns 0 if it found something or 1 if there are no lesser leaves.
3035 * returns < 0 on io errors.
3036 */
3037 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3038 {
3039 struct btrfs_key key;
3040 struct btrfs_disk_key found_key;
3041 int ret;
3042
3043 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3044
3045 if (key.offset > 0)
3046 key.offset--;
3047 else if (key.type > 0)
3048 key.type--;
3049 else if (key.objectid > 0)
3050 key.objectid--;
3051 else
3052 return 1;
3053
3054 btrfs_release_path(root, path);
3055 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3056 if (ret < 0)
3057 return ret;
3058 btrfs_item_key(path->nodes[0], &found_key, 0);
3059 ret = comp_keys(&found_key, &key);
3060 if (ret < 0)
3061 return 0;
3062 return 1;
3063 }
3064
3065 /*
3066 * A helper function to walk down the tree starting at min_key, and looking
3067 * for nodes or leaves that are either in cache or have a minimum
3068 * transaction id. This is used by the btree defrag code, but could
3069 * also be used to search for blocks that have changed since a given
3070 * transaction id.
3071 *
3072 * This does not cow, but it does stuff the starting key it finds back
3073 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3074 * key and get a writable path.
3075 *
3076 * This does lock as it descends, and path->keep_locks should be set
3077 * to 1 by the caller.
3078 *
3079 * This honors path->lowest_level to prevent descent past a given level
3080 * of the tree.
3081 *
3082 * returns zero if something useful was found, < 0 on error and 1 if there
3083 * was nothing in the tree that matched the search criteria.
3084 */
3085 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3086 struct btrfs_key *max_key,
3087 struct btrfs_path *path, int cache_only,
3088 u64 min_trans)
3089 {
3090 struct extent_buffer *cur;
3091 struct btrfs_key found_key;
3092 int slot;
3093 int sret;
3094 u32 nritems;
3095 int level;
3096 int ret = 1;
3097
3098 again:
3099 cur = btrfs_lock_root_node(root);
3100 level = btrfs_header_level(cur);
3101 WARN_ON(path->nodes[level]);
3102 path->nodes[level] = cur;
3103 path->locks[level] = 1;
3104
3105 if (btrfs_header_generation(cur) < min_trans) {
3106 ret = 1;
3107 goto out;
3108 }
3109 while(1) {
3110 nritems = btrfs_header_nritems(cur);
3111 level = btrfs_header_level(cur);
3112 sret = bin_search(cur, min_key, level, &slot);
3113
3114 /* at level = 0, we're done, setup the path and exit */
3115 if (level == 0) {
3116 if (slot >= nritems)
3117 goto find_next_key;
3118 ret = 0;
3119 path->slots[level] = slot;
3120 btrfs_item_key_to_cpu(cur, &found_key, slot);
3121 goto out;
3122 }
3123 if (sret && slot > 0)
3124 slot--;
3125 /*
3126 * check this node pointer against the cache_only and
3127 * min_trans parameters. If it isn't in cache or is too
3128 * old, skip to the next one.
3129 */
3130 while(slot < nritems) {
3131 u64 blockptr;
3132 u64 gen;
3133 struct extent_buffer *tmp;
3134 struct btrfs_disk_key disk_key;
3135
3136 blockptr = btrfs_node_blockptr(cur, slot);
3137 gen = btrfs_node_ptr_generation(cur, slot);
3138 if (gen < min_trans) {
3139 slot++;
3140 continue;
3141 }
3142 if (!cache_only)
3143 break;
3144
3145 if (max_key) {
3146 btrfs_node_key(cur, &disk_key, slot);
3147 if (comp_keys(&disk_key, max_key) >= 0) {
3148 ret = 1;
3149 goto out;
3150 }
3151 }
3152
3153 tmp = btrfs_find_tree_block(root, blockptr,
3154 btrfs_level_size(root, level - 1));
3155
3156 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3157 free_extent_buffer(tmp);
3158 break;
3159 }
3160 if (tmp)
3161 free_extent_buffer(tmp);
3162 slot++;
3163 }
3164 find_next_key:
3165 /*
3166 * we didn't find a candidate key in this node, walk forward
3167 * and find another one
3168 */
3169 if (slot >= nritems) {
3170 path->slots[level] = slot;
3171 sret = btrfs_find_next_key(root, path, min_key, level,
3172 cache_only, min_trans);
3173 if (sret == 0) {
3174 btrfs_release_path(root, path);
3175 goto again;
3176 } else {
3177 goto out;
3178 }
3179 }
3180 /* save our key for returning back */
3181 btrfs_node_key_to_cpu(cur, &found_key, slot);
3182 path->slots[level] = slot;
3183 if (level == path->lowest_level) {
3184 ret = 0;
3185 unlock_up(path, level, 1);
3186 goto out;
3187 }
3188 cur = read_node_slot(root, cur, slot);
3189
3190 btrfs_tree_lock(cur);
3191 path->locks[level - 1] = 1;
3192 path->nodes[level - 1] = cur;
3193 unlock_up(path, level, 1);
3194 }
3195 out:
3196 if (ret == 0)
3197 memcpy(min_key, &found_key, sizeof(found_key));
3198 return ret;
3199 }
3200
3201 /*
3202 * this is similar to btrfs_next_leaf, but does not try to preserve
3203 * and fixup the path. It looks for and returns the next key in the
3204 * tree based on the current path and the cache_only and min_trans
3205 * parameters.
3206 *
3207 * 0 is returned if another key is found, < 0 if there are any errors
3208 * and 1 is returned if there are no higher keys in the tree
3209 *
3210 * path->keep_locks should be set to 1 on the search made before
3211 * calling this function.
3212 */
3213 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3214 struct btrfs_key *key, int lowest_level,
3215 int cache_only, u64 min_trans)
3216 {
3217 int level = lowest_level;
3218 int slot;
3219 struct extent_buffer *c;
3220
3221 while(level < BTRFS_MAX_LEVEL) {
3222 if (!path->nodes[level])
3223 return 1;
3224
3225 slot = path->slots[level] + 1;
3226 c = path->nodes[level];
3227 next:
3228 if (slot >= btrfs_header_nritems(c)) {
3229 level++;
3230 if (level == BTRFS_MAX_LEVEL) {
3231 return 1;
3232 }
3233 continue;
3234 }
3235 if (level == 0)
3236 btrfs_item_key_to_cpu(c, key, slot);
3237 else {
3238 u64 blockptr = btrfs_node_blockptr(c, slot);
3239 u64 gen = btrfs_node_ptr_generation(c, slot);
3240
3241 if (cache_only) {
3242 struct extent_buffer *cur;
3243 cur = btrfs_find_tree_block(root, blockptr,
3244 btrfs_level_size(root, level - 1));
3245 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3246 slot++;
3247 if (cur)
3248 free_extent_buffer(cur);
3249 goto next;
3250 }
3251 free_extent_buffer(cur);
3252 }
3253 if (gen < min_trans) {
3254 slot++;
3255 goto next;
3256 }
3257 btrfs_node_key_to_cpu(c, key, slot);
3258 }
3259 return 0;
3260 }
3261 return 1;
3262 }
3263
3264 /*
3265 * search the tree again to find a leaf with greater keys
3266 * returns 0 if it found something or 1 if there are no greater leaves.
3267 * returns < 0 on io errors.
3268 */
3269 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3270 {
3271 int slot;
3272 int level = 1;
3273 struct extent_buffer *c;
3274 struct extent_buffer *next = NULL;
3275 struct btrfs_key key;
3276 u32 nritems;
3277 int ret;
3278
3279 nritems = btrfs_header_nritems(path->nodes[0]);
3280 if (nritems == 0) {
3281 return 1;
3282 }
3283
3284 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3285
3286 btrfs_release_path(root, path);
3287 path->keep_locks = 1;
3288 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3289 path->keep_locks = 0;
3290
3291 if (ret < 0)
3292 return ret;
3293
3294 nritems = btrfs_header_nritems(path->nodes[0]);
3295 /*
3296 * by releasing the path above we dropped all our locks. A balance
3297 * could have added more items next to the key that used to be
3298 * at the very end of the block. So, check again here and
3299 * advance the path if there are now more items available.
3300 */
3301 if (nritems > 0 && path->slots[0] < nritems - 1) {
3302 path->slots[0]++;
3303 goto done;
3304 }
3305
3306 while(level < BTRFS_MAX_LEVEL) {
3307 if (!path->nodes[level])
3308 return 1;
3309
3310 slot = path->slots[level] + 1;
3311 c = path->nodes[level];
3312 if (slot >= btrfs_header_nritems(c)) {
3313 level++;
3314 if (level == BTRFS_MAX_LEVEL) {
3315 return 1;
3316 }
3317 continue;
3318 }
3319
3320 if (next) {
3321 btrfs_tree_unlock(next);
3322 free_extent_buffer(next);
3323 }
3324
3325 if (level == 1 && (path->locks[1] || path->skip_locking) &&
3326 path->reada)
3327 reada_for_search(root, path, level, slot, 0);
3328
3329 next = read_node_slot(root, c, slot);
3330 if (!path->skip_locking) {
3331 WARN_ON(!btrfs_tree_locked(c));
3332 btrfs_tree_lock(next);
3333 }
3334 break;
3335 }
3336 path->slots[level] = slot;
3337 while(1) {
3338 level--;
3339 c = path->nodes[level];
3340 if (path->locks[level])
3341 btrfs_tree_unlock(c);
3342 free_extent_buffer(c);
3343 path->nodes[level] = next;
3344 path->slots[level] = 0;
3345 if (!path->skip_locking)
3346 path->locks[level] = 1;
3347 if (!level)
3348 break;
3349 if (level == 1 && path->locks[1] && path->reada)
3350 reada_for_search(root, path, level, slot, 0);
3351 next = read_node_slot(root, next, 0);
3352 if (!path->skip_locking) {
3353 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3354 btrfs_tree_lock(next);
3355 }
3356 }
3357 done:
3358 unlock_up(path, 0, 1);
3359 return 0;
3360 }
3361
3362 /*
3363 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
3364 * searching until it gets past min_objectid or finds an item of 'type'
3365 *
3366 * returns 0 if something is found, 1 if nothing was found and < 0 on error
3367 */
3368 int btrfs_previous_item(struct btrfs_root *root,
3369 struct btrfs_path *path, u64 min_objectid,
3370 int type)
3371 {
3372 struct btrfs_key found_key;
3373 struct extent_buffer *leaf;
3374 u32 nritems;
3375 int ret;
3376
3377 while(1) {
3378 if (path->slots[0] == 0) {
3379 ret = btrfs_prev_leaf(root, path);
3380 if (ret != 0)
3381 return ret;
3382 } else {
3383 path->slots[0]--;
3384 }
3385 leaf = path->nodes[0];
3386 nritems = btrfs_header_nritems(leaf);
3387 if (nritems == 0)
3388 return 1;
3389 if (path->slots[0] == nritems)
3390 path->slots[0]--;
3391
3392 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3393 if (found_key.type == type)
3394 return 0;
3395 if (found_key.objectid < min_objectid)
3396 break;
3397 if (found_key.objectid == min_objectid &&
3398 found_key.type < type)
3399 break;
3400 }
3401 return 1;
3402 }
Ubuntu Artful kernel mirror