projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: Replace the transaction work queue with kthreads
[deliverable/linux.git] / fs / btrfs / ctree.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/sched.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "transaction.h"
23 #include "print-tree.h"
24 #include "locking.h"
25
26 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_path *path, int level);
28 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_key *ins_key,
30 struct btrfs_path *path, int data_size, int extend);
31 static int push_node_left(struct btrfs_trans_handle *trans,
32 struct btrfs_root *root, struct extent_buffer *dst,
33 struct extent_buffer *src, int empty);
34 static int balance_node_right(struct btrfs_trans_handle *trans,
35 struct btrfs_root *root,
36 struct extent_buffer *dst_buf,
37 struct extent_buffer *src_buf);
38 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
39 struct btrfs_path *path, int level, int slot);
40
41 inline void btrfs_init_path(struct btrfs_path *p)
42 {
43 memset(p, 0, sizeof(*p));
44 }
45
46 struct btrfs_path *btrfs_alloc_path(void)
47 {
48 struct btrfs_path *path;
49 path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
50 if (path) {
51 btrfs_init_path(path);
52 path->reada = 1;
53 }
54 return path;
55 }
56
57 void btrfs_free_path(struct btrfs_path *p)
58 {
59 btrfs_release_path(NULL, p);
60 kmem_cache_free(btrfs_path_cachep, p);
61 }
62
63 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
64 {
65 int i;
66 int keep = p->keep_locks;
67 int skip = p->skip_locking;
68
69 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
70 if (!p->nodes[i])
71 continue;
72 if (p->locks[i]) {
73 btrfs_tree_unlock(p->nodes[i]);
74 p->locks[i] = 0;
75 }
76 free_extent_buffer(p->nodes[i]);
77 }
78 memset(p, 0, sizeof(*p));
79 p->keep_locks = keep;
80 p->skip_locking = skip;
81 }
82
83 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
84 {
85 struct extent_buffer *eb;
86 spin_lock(&root->node_lock);
87 eb = root->node;
88 extent_buffer_get(eb);
89 spin_unlock(&root->node_lock);
90 return eb;
91 }
92
93 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
94 {
95 struct extent_buffer *eb;
96
97 while(1) {
98 eb = btrfs_root_node(root);
99 btrfs_tree_lock(eb);
100
101 spin_lock(&root->node_lock);
102 if (eb == root->node) {
103 spin_unlock(&root->node_lock);
104 break;
105 }
106 spin_unlock(&root->node_lock);
107
108 btrfs_tree_unlock(eb);
109 free_extent_buffer(eb);
110 }
111 return eb;
112 }
113
114 static void add_root_to_dirty_list(struct btrfs_root *root)
115 {
116 if (root->track_dirty && list_empty(&root->dirty_list)) {
117 list_add(&root->dirty_list,
118 &root->fs_info->dirty_cowonly_roots);
119 }
120 }
121
122 int btrfs_copy_root(struct btrfs_trans_handle *trans,
123 struct btrfs_root *root,
124 struct extent_buffer *buf,
125 struct extent_buffer **cow_ret, u64 new_root_objectid)
126 {
127 struct extent_buffer *cow;
128 u32 nritems;
129 int ret = 0;
130 int level;
131 struct btrfs_key first_key;
132 struct btrfs_root *new_root;
133
134 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
135 if (!new_root)
136 return -ENOMEM;
137
138 memcpy(new_root, root, sizeof(*new_root));
139 new_root->root_key.objectid = new_root_objectid;
140
141 WARN_ON(root->ref_cows && trans->transid !=
142 root->fs_info->running_transaction->transid);
143 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
144
145 level = btrfs_header_level(buf);
146 nritems = btrfs_header_nritems(buf);
147 if (nritems) {
148 if (level == 0)
149 btrfs_item_key_to_cpu(buf, &first_key, 0);
150 else
151 btrfs_node_key_to_cpu(buf, &first_key, 0);
152 } else {
153 first_key.objectid = 0;
154 }
155 cow = btrfs_alloc_free_block(trans, new_root, buf->len,
156 new_root_objectid,
157 trans->transid, first_key.objectid,
158 level, buf->start, 0);
159 if (IS_ERR(cow)) {
160 kfree(new_root);
161 return PTR_ERR(cow);
162 }
163
164 copy_extent_buffer(cow, buf, 0, 0, cow->len);
165 btrfs_set_header_bytenr(cow, cow->start);
166 btrfs_set_header_generation(cow, trans->transid);
167 btrfs_set_header_owner(cow, new_root_objectid);
168 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
169
170 WARN_ON(btrfs_header_generation(buf) > trans->transid);
171 ret = btrfs_inc_ref(trans, new_root, buf);
172 kfree(new_root);
173
174 if (ret)
175 return ret;
176
177 btrfs_mark_buffer_dirty(cow);
178 *cow_ret = cow;
179 return 0;
180 }
181
182 int __btrfs_cow_block(struct btrfs_trans_handle *trans,
183 struct btrfs_root *root,
184 struct extent_buffer *buf,
185 struct extent_buffer *parent, int parent_slot,
186 struct extent_buffer **cow_ret,
187 u64 search_start, u64 empty_size)
188 {
189 u64 root_gen;
190 struct extent_buffer *cow;
191 u32 nritems;
192 int ret = 0;
193 int different_trans = 0;
194 int level;
195 int unlock_orig = 0;
196 struct btrfs_key first_key;
197
198 if (*cow_ret == buf)
199 unlock_orig = 1;
200
201 WARN_ON(!btrfs_tree_locked(buf));
202
203 if (root->ref_cows) {
204 root_gen = trans->transid;
205 } else {
206 root_gen = 0;
207 }
208 WARN_ON(root->ref_cows && trans->transid !=
209 root->fs_info->running_transaction->transid);
210 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
211
212 level = btrfs_header_level(buf);
213 nritems = btrfs_header_nritems(buf);
214 if (nritems) {
215 if (level == 0)
216 btrfs_item_key_to_cpu(buf, &first_key, 0);
217 else
218 btrfs_node_key_to_cpu(buf, &first_key, 0);
219 } else {
220 first_key.objectid = 0;
221 }
222 cow = btrfs_alloc_free_block(trans, root, buf->len,
223 root->root_key.objectid,
224 root_gen, first_key.objectid, level,
225 search_start, empty_size);
226 if (IS_ERR(cow))
227 return PTR_ERR(cow);
228
229 copy_extent_buffer(cow, buf, 0, 0, cow->len);
230 btrfs_set_header_bytenr(cow, cow->start);
231 btrfs_set_header_generation(cow, trans->transid);
232 btrfs_set_header_owner(cow, root->root_key.objectid);
233 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
234
235 WARN_ON(btrfs_header_generation(buf) > trans->transid);
236 if (btrfs_header_generation(buf) != trans->transid) {
237 different_trans = 1;
238 ret = btrfs_inc_ref(trans, root, buf);
239 if (ret)
240 return ret;
241 } else {
242 clean_tree_block(trans, root, buf);
243 }
244
245 if (buf == root->node) {
246 WARN_ON(parent && parent != buf);
247 root_gen = btrfs_header_generation(buf);
248
249 spin_lock(&root->node_lock);
250 root->node = cow;
251 extent_buffer_get(cow);
252 spin_unlock(&root->node_lock);
253
254 if (buf != root->commit_root) {
255 btrfs_free_extent(trans, root, buf->start,
256 buf->len, root->root_key.objectid,
257 root_gen, 0, 0, 1);
258 }
259 free_extent_buffer(buf);
260 add_root_to_dirty_list(root);
261 } else {
262 root_gen = btrfs_header_generation(parent);
263 btrfs_set_node_blockptr(parent, parent_slot,
264 cow->start);
265 WARN_ON(trans->transid == 0);
266 btrfs_set_node_ptr_generation(parent, parent_slot,
267 trans->transid);
268 btrfs_mark_buffer_dirty(parent);
269 WARN_ON(btrfs_header_generation(parent) != trans->transid);
270 btrfs_free_extent(trans, root, buf->start, buf->len,
271 btrfs_header_owner(parent), root_gen,
272 0, 0, 1);
273 }
274 if (unlock_orig)
275 btrfs_tree_unlock(buf);
276 free_extent_buffer(buf);
277 btrfs_mark_buffer_dirty(cow);
278 *cow_ret = cow;
279 return 0;
280 }
281
282 int btrfs_cow_block(struct btrfs_trans_handle *trans,
283 struct btrfs_root *root, struct extent_buffer *buf,
284 struct extent_buffer *parent, int parent_slot,
285 struct extent_buffer **cow_ret)
286 {
287 u64 search_start;
288 u64 header_trans;
289 int ret;
290
291 if (trans->transaction != root->fs_info->running_transaction) {
292 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
293 root->fs_info->running_transaction->transid);
294 WARN_ON(1);
295 }
296 if (trans->transid != root->fs_info->generation) {
297 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
298 root->fs_info->generation);
299 WARN_ON(1);
300 }
301
302 header_trans = btrfs_header_generation(buf);
303 spin_lock(&root->fs_info->hash_lock);
304 if (header_trans == trans->transid &&
305 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
306 *cow_ret = buf;
307 spin_unlock(&root->fs_info->hash_lock);
308 return 0;
309 }
310 spin_unlock(&root->fs_info->hash_lock);
311 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
312 ret = __btrfs_cow_block(trans, root, buf, parent,
313 parent_slot, cow_ret, search_start, 0);
314 return ret;
315 }
316
317 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
318 {
319 if (blocknr < other && other - (blocknr + blocksize) < 32768)
320 return 1;
321 if (blocknr > other && blocknr - (other + blocksize) < 32768)
322 return 1;
323 return 0;
324 }
325
326 /*
327 * compare two keys in a memcmp fashion
328 */
329 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
330 {
331 struct btrfs_key k1;
332
333 btrfs_disk_key_to_cpu(&k1, disk);
334
335 if (k1.objectid > k2->objectid)
336 return 1;
337 if (k1.objectid < k2->objectid)
338 return -1;
339 if (k1.type > k2->type)
340 return 1;
341 if (k1.type < k2->type)
342 return -1;
343 if (k1.offset > k2->offset)
344 return 1;
345 if (k1.offset < k2->offset)
346 return -1;
347 return 0;
348 }
349
350
351 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
352 struct btrfs_root *root, struct extent_buffer *parent,
353 int start_slot, int cache_only, u64 *last_ret,
354 struct btrfs_key *progress)
355 {
356 struct extent_buffer *cur;
357 struct extent_buffer *tmp;
358 u64 blocknr;
359 u64 gen;
360 u64 search_start = *last_ret;
361 u64 last_block = 0;
362 u64 other;
363 u32 parent_nritems;
364 int end_slot;
365 int i;
366 int err = 0;
367 int parent_level;
368 int uptodate;
369 u32 blocksize;
370 int progress_passed = 0;
371 struct btrfs_disk_key disk_key;
372
373 /* FIXME this code needs locking */
374 return 0;
375
376 parent_level = btrfs_header_level(parent);
377 if (cache_only && parent_level != 1)
378 return 0;
379
380 if (trans->transaction != root->fs_info->running_transaction) {
381 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
382 root->fs_info->running_transaction->transid);
383 WARN_ON(1);
384 }
385 if (trans->transid != root->fs_info->generation) {
386 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
387 root->fs_info->generation);
388 WARN_ON(1);
389 }
390
391 parent_nritems = btrfs_header_nritems(parent);
392 blocksize = btrfs_level_size(root, parent_level - 1);
393 end_slot = parent_nritems;
394
395 if (parent_nritems == 1)
396 return 0;
397
398 for (i = start_slot; i < end_slot; i++) {
399 int close = 1;
400
401 if (!parent->map_token) {
402 map_extent_buffer(parent,
403 btrfs_node_key_ptr_offset(i),
404 sizeof(struct btrfs_key_ptr),
405 &parent->map_token, &parent->kaddr,
406 &parent->map_start, &parent->map_len,
407 KM_USER1);
408 }
409 btrfs_node_key(parent, &disk_key, i);
410 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
411 continue;
412
413 progress_passed = 1;
414 blocknr = btrfs_node_blockptr(parent, i);
415 gen = btrfs_node_ptr_generation(parent, i);
416 if (last_block == 0)
417 last_block = blocknr;
418
419 if (i > 0) {
420 other = btrfs_node_blockptr(parent, i - 1);
421 close = close_blocks(blocknr, other, blocksize);
422 }
423 if (!close && i < end_slot - 2) {
424 other = btrfs_node_blockptr(parent, i + 1);
425 close = close_blocks(blocknr, other, blocksize);
426 }
427 if (close) {
428 last_block = blocknr;
429 continue;
430 }
431 if (parent->map_token) {
432 unmap_extent_buffer(parent, parent->map_token,
433 KM_USER1);
434 parent->map_token = NULL;
435 }
436
437 cur = btrfs_find_tree_block(root, blocknr, blocksize);
438 if (cur)
439 uptodate = btrfs_buffer_uptodate(cur, gen);
440 else
441 uptodate = 0;
442 if (!cur || !uptodate) {
443 if (cache_only) {
444 free_extent_buffer(cur);
445 continue;
446 }
447 if (!cur) {
448 cur = read_tree_block(root, blocknr,
449 blocksize, gen);
450 } else if (!uptodate) {
451 btrfs_read_buffer(cur, gen);
452 }
453 }
454 if (search_start == 0)
455 search_start = last_block;
456
457 err = __btrfs_cow_block(trans, root, cur, parent, i,
458 &tmp, search_start,
459 min(16 * blocksize,
460 (end_slot - i) * blocksize));
461 if (err) {
462 free_extent_buffer(cur);
463 break;
464 }
465 search_start = tmp->start;
466 last_block = tmp->start;
467 *last_ret = search_start;
468 if (parent_level == 1)
469 btrfs_clear_buffer_defrag(tmp);
470 free_extent_buffer(tmp);
471 }
472 if (parent->map_token) {
473 unmap_extent_buffer(parent, parent->map_token,
474 KM_USER1);
475 parent->map_token = NULL;
476 }
477 return err;
478 }
479
480 /*
481 * The leaf data grows from end-to-front in the node.
482 * this returns the address of the start of the last item,
483 * which is the stop of the leaf data stack
484 */
485 static inline unsigned int leaf_data_end(struct btrfs_root *root,
486 struct extent_buffer *leaf)
487 {
488 u32 nr = btrfs_header_nritems(leaf);
489 if (nr == 0)
490 return BTRFS_LEAF_DATA_SIZE(root);
491 return btrfs_item_offset_nr(leaf, nr - 1);
492 }
493
494 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
495 int level)
496 {
497 struct extent_buffer *parent = NULL;
498 struct extent_buffer *node = path->nodes[level];
499 struct btrfs_disk_key parent_key;
500 struct btrfs_disk_key node_key;
501 int parent_slot;
502 int slot;
503 struct btrfs_key cpukey;
504 u32 nritems = btrfs_header_nritems(node);
505
506 if (path->nodes[level + 1])
507 parent = path->nodes[level + 1];
508
509 slot = path->slots[level];
510 BUG_ON(nritems == 0);
511 if (parent) {
512 parent_slot = path->slots[level + 1];
513 btrfs_node_key(parent, &parent_key, parent_slot);
514 btrfs_node_key(node, &node_key, 0);
515 BUG_ON(memcmp(&parent_key, &node_key,
516 sizeof(struct btrfs_disk_key)));
517 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
518 btrfs_header_bytenr(node));
519 }
520 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
521 if (slot != 0) {
522 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
523 btrfs_node_key(node, &node_key, slot);
524 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
525 }
526 if (slot < nritems - 1) {
527 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
528 btrfs_node_key(node, &node_key, slot);
529 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
530 }
531 return 0;
532 }
533
534 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
535 int level)
536 {
537 struct extent_buffer *leaf = path->nodes[level];
538 struct extent_buffer *parent = NULL;
539 int parent_slot;
540 struct btrfs_key cpukey;
541 struct btrfs_disk_key parent_key;
542 struct btrfs_disk_key leaf_key;
543 int slot = path->slots[0];
544
545 u32 nritems = btrfs_header_nritems(leaf);
546
547 if (path->nodes[level + 1])
548 parent = path->nodes[level + 1];
549
550 if (nritems == 0)
551 return 0;
552
553 if (parent) {
554 parent_slot = path->slots[level + 1];
555 btrfs_node_key(parent, &parent_key, parent_slot);
556 btrfs_item_key(leaf, &leaf_key, 0);
557
558 BUG_ON(memcmp(&parent_key, &leaf_key,
559 sizeof(struct btrfs_disk_key)));
560 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
561 btrfs_header_bytenr(leaf));
562 }
563 #if 0
564 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
565 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
566 btrfs_item_key(leaf, &leaf_key, i);
567 if (comp_keys(&leaf_key, &cpukey) >= 0) {
568 btrfs_print_leaf(root, leaf);
569 printk("slot %d offset bad key\n", i);
570 BUG_ON(1);
571 }
572 if (btrfs_item_offset_nr(leaf, i) !=
573 btrfs_item_end_nr(leaf, i + 1)) {
574 btrfs_print_leaf(root, leaf);
575 printk("slot %d offset bad\n", i);
576 BUG_ON(1);
577 }
578 if (i == 0) {
579 if (btrfs_item_offset_nr(leaf, i) +
580 btrfs_item_size_nr(leaf, i) !=
581 BTRFS_LEAF_DATA_SIZE(root)) {
582 btrfs_print_leaf(root, leaf);
583 printk("slot %d first offset bad\n", i);
584 BUG_ON(1);
585 }
586 }
587 }
588 if (nritems > 0) {
589 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
590 btrfs_print_leaf(root, leaf);
591 printk("slot %d bad size \n", nritems - 1);
592 BUG_ON(1);
593 }
594 }
595 #endif
596 if (slot != 0 && slot < nritems - 1) {
597 btrfs_item_key(leaf, &leaf_key, slot);
598 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
599 if (comp_keys(&leaf_key, &cpukey) <= 0) {
600 btrfs_print_leaf(root, leaf);
601 printk("slot %d offset bad key\n", slot);
602 BUG_ON(1);
603 }
604 if (btrfs_item_offset_nr(leaf, slot - 1) !=
605 btrfs_item_end_nr(leaf, slot)) {
606 btrfs_print_leaf(root, leaf);
607 printk("slot %d offset bad\n", slot);
608 BUG_ON(1);
609 }
610 }
611 if (slot < nritems - 1) {
612 btrfs_item_key(leaf, &leaf_key, slot);
613 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
614 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
615 if (btrfs_item_offset_nr(leaf, slot) !=
616 btrfs_item_end_nr(leaf, slot + 1)) {
617 btrfs_print_leaf(root, leaf);
618 printk("slot %d offset bad\n", slot);
619 BUG_ON(1);
620 }
621 }
622 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
623 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
624 return 0;
625 }
626
627 static int noinline check_block(struct btrfs_root *root,
628 struct btrfs_path *path, int level)
629 {
630 u64 found_start;
631 return 0;
632 if (btrfs_header_level(path->nodes[level]) != level)
633 printk("warning: bad level %Lu wanted %d found %d\n",
634 path->nodes[level]->start, level,
635 btrfs_header_level(path->nodes[level]));
636 found_start = btrfs_header_bytenr(path->nodes[level]);
637 if (found_start != path->nodes[level]->start) {
638 printk("warning: bad bytentr %Lu found %Lu\n",
639 path->nodes[level]->start, found_start);
640 }
641 #if 0
642 struct extent_buffer *buf = path->nodes[level];
643
644 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
645 (unsigned long)btrfs_header_fsid(buf),
646 BTRFS_FSID_SIZE)) {
647 printk("warning bad block %Lu\n", buf->start);
648 return 1;
649 }
650 #endif
651 if (level == 0)
652 return check_leaf(root, path, level);
653 return check_node(root, path, level);
654 }
655
656 /*
657 * search for key in the extent_buffer. The items start at offset p,
658 * and they are item_size apart. There are 'max' items in p.
659 *
660 * the slot in the array is returned via slot, and it points to
661 * the place where you would insert key if it is not found in
662 * the array.
663 *
664 * slot may point to max if the key is bigger than all of the keys
665 */
666 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
667 int item_size, struct btrfs_key *key,
668 int max, int *slot)
669 {
670 int low = 0;
671 int high = max;
672 int mid;
673 int ret;
674 struct btrfs_disk_key *tmp = NULL;
675 struct btrfs_disk_key unaligned;
676 unsigned long offset;
677 char *map_token = NULL;
678 char *kaddr = NULL;
679 unsigned long map_start = 0;
680 unsigned long map_len = 0;
681 int err;
682
683 while(low < high) {
684 mid = (low + high) / 2;
685 offset = p + mid * item_size;
686
687 if (!map_token || offset < map_start ||
688 (offset + sizeof(struct btrfs_disk_key)) >
689 map_start + map_len) {
690 if (map_token) {
691 unmap_extent_buffer(eb, map_token, KM_USER0);
692 map_token = NULL;
693 }
694 err = map_extent_buffer(eb, offset,
695 sizeof(struct btrfs_disk_key),
696 &map_token, &kaddr,
697 &map_start, &map_len, KM_USER0);
698
699 if (!err) {
700 tmp = (struct btrfs_disk_key *)(kaddr + offset -
701 map_start);
702 } else {
703 read_extent_buffer(eb, &unaligned,
704 offset, sizeof(unaligned));
705 tmp = &unaligned;
706 }
707
708 } else {
709 tmp = (struct btrfs_disk_key *)(kaddr + offset -
710 map_start);
711 }
712 ret = comp_keys(tmp, key);
713
714 if (ret < 0)
715 low = mid + 1;
716 else if (ret > 0)
717 high = mid;
718 else {
719 *slot = mid;
720 if (map_token)
721 unmap_extent_buffer(eb, map_token, KM_USER0);
722 return 0;
723 }
724 }
725 *slot = low;
726 if (map_token)
727 unmap_extent_buffer(eb, map_token, KM_USER0);
728 return 1;
729 }
730
731 /*
732 * simple bin_search frontend that does the right thing for
733 * leaves vs nodes
734 */
735 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
736 int level, int *slot)
737 {
738 if (level == 0) {
739 return generic_bin_search(eb,
740 offsetof(struct btrfs_leaf, items),
741 sizeof(struct btrfs_item),
742 key, btrfs_header_nritems(eb),
743 slot);
744 } else {
745 return generic_bin_search(eb,
746 offsetof(struct btrfs_node, ptrs),
747 sizeof(struct btrfs_key_ptr),
748 key, btrfs_header_nritems(eb),
749 slot);
750 }
751 return -1;
752 }
753
754 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
755 struct extent_buffer *parent, int slot)
756 {
757 int level = btrfs_header_level(parent);
758 if (slot < 0)
759 return NULL;
760 if (slot >= btrfs_header_nritems(parent))
761 return NULL;
762
763 BUG_ON(level == 0);
764
765 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
766 btrfs_level_size(root, level - 1),
767 btrfs_node_ptr_generation(parent, slot));
768 }
769
770 static int balance_level(struct btrfs_trans_handle *trans,
771 struct btrfs_root *root,
772 struct btrfs_path *path, int level)
773 {
774 struct extent_buffer *right = NULL;
775 struct extent_buffer *mid;
776 struct extent_buffer *left = NULL;
777 struct extent_buffer *parent = NULL;
778 int ret = 0;
779 int wret;
780 int pslot;
781 int orig_slot = path->slots[level];
782 int err_on_enospc = 0;
783 u64 orig_ptr;
784
785 if (level == 0)
786 return 0;
787
788 mid = path->nodes[level];
789 WARN_ON(!path->locks[level]);
790 WARN_ON(btrfs_header_generation(mid) != trans->transid);
791
792 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
793
794 if (level < BTRFS_MAX_LEVEL - 1)
795 parent = path->nodes[level + 1];
796 pslot = path->slots[level + 1];
797
798 /*
799 * deal with the case where there is only one pointer in the root
800 * by promoting the node below to a root
801 */
802 if (!parent) {
803 struct extent_buffer *child;
804
805 if (btrfs_header_nritems(mid) != 1)
806 return 0;
807
808 /* promote the child to a root */
809 child = read_node_slot(root, mid, 0);
810 btrfs_tree_lock(child);
811 BUG_ON(!child);
812 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
813 BUG_ON(ret);
814
815 spin_lock(&root->node_lock);
816 root->node = child;
817 spin_unlock(&root->node_lock);
818
819 add_root_to_dirty_list(root);
820 btrfs_tree_unlock(child);
821 path->locks[level] = 0;
822 path->nodes[level] = NULL;
823 clean_tree_block(trans, root, mid);
824 btrfs_tree_unlock(mid);
825 /* once for the path */
826 free_extent_buffer(mid);
827 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
828 root->root_key.objectid,
829 btrfs_header_generation(mid), 0, 0, 1);
830 /* once for the root ptr */
831 free_extent_buffer(mid);
832 return ret;
833 }
834 if (btrfs_header_nritems(mid) >
835 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
836 return 0;
837
838 if (btrfs_header_nritems(mid) < 2)
839 err_on_enospc = 1;
840
841 left = read_node_slot(root, parent, pslot - 1);
842 if (left) {
843 btrfs_tree_lock(left);
844 wret = btrfs_cow_block(trans, root, left,
845 parent, pslot - 1, &left);
846 if (wret) {
847 ret = wret;
848 goto enospc;
849 }
850 }
851 right = read_node_slot(root, parent, pslot + 1);
852 if (right) {
853 btrfs_tree_lock(right);
854 wret = btrfs_cow_block(trans, root, right,
855 parent, pslot + 1, &right);
856 if (wret) {
857 ret = wret;
858 goto enospc;
859 }
860 }
861
862 /* first, try to make some room in the middle buffer */
863 if (left) {
864 orig_slot += btrfs_header_nritems(left);
865 wret = push_node_left(trans, root, left, mid, 1);
866 if (wret < 0)
867 ret = wret;
868 if (btrfs_header_nritems(mid) < 2)
869 err_on_enospc = 1;
870 }
871
872 /*
873 * then try to empty the right most buffer into the middle
874 */
875 if (right) {
876 wret = push_node_left(trans, root, mid, right, 1);
877 if (wret < 0 && wret != -ENOSPC)
878 ret = wret;
879 if (btrfs_header_nritems(right) == 0) {
880 u64 bytenr = right->start;
881 u64 generation = btrfs_header_generation(parent);
882 u32 blocksize = right->len;
883
884 clean_tree_block(trans, root, right);
885 btrfs_tree_unlock(right);
886 free_extent_buffer(right);
887 right = NULL;
888 wret = del_ptr(trans, root, path, level + 1, pslot +
889 1);
890 if (wret)
891 ret = wret;
892 wret = btrfs_free_extent(trans, root, bytenr,
893 blocksize,
894 btrfs_header_owner(parent),
895 generation, 0, 0, 1);
896 if (wret)
897 ret = wret;
898 } else {
899 struct btrfs_disk_key right_key;
900 btrfs_node_key(right, &right_key, 0);
901 btrfs_set_node_key(parent, &right_key, pslot + 1);
902 btrfs_mark_buffer_dirty(parent);
903 }
904 }
905 if (btrfs_header_nritems(mid) == 1) {
906 /*
907 * we're not allowed to leave a node with one item in the
908 * tree during a delete. A deletion from lower in the tree
909 * could try to delete the only pointer in this node.
910 * So, pull some keys from the left.
911 * There has to be a left pointer at this point because
912 * otherwise we would have pulled some pointers from the
913 * right
914 */
915 BUG_ON(!left);
916 wret = balance_node_right(trans, root, mid, left);
917 if (wret < 0) {
918 ret = wret;
919 goto enospc;
920 }
921 if (wret == 1) {
922 wret = push_node_left(trans, root, left, mid, 1);
923 if (wret < 0)
924 ret = wret;
925 }
926 BUG_ON(wret == 1);
927 }
928 if (btrfs_header_nritems(mid) == 0) {
929 /* we've managed to empty the middle node, drop it */
930 u64 root_gen = btrfs_header_generation(parent);
931 u64 bytenr = mid->start;
932 u32 blocksize = mid->len;
933
934 clean_tree_block(trans, root, mid);
935 btrfs_tree_unlock(mid);
936 free_extent_buffer(mid);
937 mid = NULL;
938 wret = del_ptr(trans, root, path, level + 1, pslot);
939 if (wret)
940 ret = wret;
941 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
942 btrfs_header_owner(parent),
943 root_gen, 0, 0, 1);
944 if (wret)
945 ret = wret;
946 } else {
947 /* update the parent key to reflect our changes */
948 struct btrfs_disk_key mid_key;
949 btrfs_node_key(mid, &mid_key, 0);
950 btrfs_set_node_key(parent, &mid_key, pslot);
951 btrfs_mark_buffer_dirty(parent);
952 }
953
954 /* update the path */
955 if (left) {
956 if (btrfs_header_nritems(left) > orig_slot) {
957 extent_buffer_get(left);
958 /* left was locked after cow */
959 path->nodes[level] = left;
960 path->slots[level + 1] -= 1;
961 path->slots[level] = orig_slot;
962 if (mid) {
963 btrfs_tree_unlock(mid);
964 free_extent_buffer(mid);
965 }
966 } else {
967 orig_slot -= btrfs_header_nritems(left);
968 path->slots[level] = orig_slot;
969 }
970 }
971 /* double check we haven't messed things up */
972 check_block(root, path, level);
973 if (orig_ptr !=
974 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
975 BUG();
976 enospc:
977 if (right) {
978 btrfs_tree_unlock(right);
979 free_extent_buffer(right);
980 }
981 if (left) {
982 if (path->nodes[level] != left)
983 btrfs_tree_unlock(left);
984 free_extent_buffer(left);
985 }
986 return ret;
987 }
988
989 /* returns zero if the push worked, non-zero otherwise */
990 static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
991 struct btrfs_root *root,
992 struct btrfs_path *path, int level)
993 {
994 struct extent_buffer *right = NULL;
995 struct extent_buffer *mid;
996 struct extent_buffer *left = NULL;
997 struct extent_buffer *parent = NULL;
998 int ret = 0;
999 int wret;
1000 int pslot;
1001 int orig_slot = path->slots[level];
1002 u64 orig_ptr;
1003
1004 if (level == 0)
1005 return 1;
1006
1007 mid = path->nodes[level];
1008 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1009 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1010
1011 if (level < BTRFS_MAX_LEVEL - 1)
1012 parent = path->nodes[level + 1];
1013 pslot = path->slots[level + 1];
1014
1015 if (!parent)
1016 return 1;
1017
1018 left = read_node_slot(root, parent, pslot - 1);
1019
1020 /* first, try to make some room in the middle buffer */
1021 if (left) {
1022 u32 left_nr;
1023
1024 btrfs_tree_lock(left);
1025 left_nr = btrfs_header_nritems(left);
1026 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1027 wret = 1;
1028 } else {
1029 ret = btrfs_cow_block(trans, root, left, parent,
1030 pslot - 1, &left);
1031 if (ret)
1032 wret = 1;
1033 else {
1034 wret = push_node_left(trans, root,
1035 left, mid, 0);
1036 }
1037 }
1038 if (wret < 0)
1039 ret = wret;
1040 if (wret == 0) {
1041 struct btrfs_disk_key disk_key;
1042 orig_slot += left_nr;
1043 btrfs_node_key(mid, &disk_key, 0);
1044 btrfs_set_node_key(parent, &disk_key, pslot);
1045 btrfs_mark_buffer_dirty(parent);
1046 if (btrfs_header_nritems(left) > orig_slot) {
1047 path->nodes[level] = left;
1048 path->slots[level + 1] -= 1;
1049 path->slots[level] = orig_slot;
1050 btrfs_tree_unlock(mid);
1051 free_extent_buffer(mid);
1052 } else {
1053 orig_slot -=
1054 btrfs_header_nritems(left);
1055 path->slots[level] = orig_slot;
1056 btrfs_tree_unlock(left);
1057 free_extent_buffer(left);
1058 }
1059 return 0;
1060 }
1061 btrfs_tree_unlock(left);
1062 free_extent_buffer(left);
1063 }
1064 right = read_node_slot(root, parent, pslot + 1);
1065
1066 /*
1067 * then try to empty the right most buffer into the middle
1068 */
1069 if (right) {
1070 u32 right_nr;
1071 btrfs_tree_lock(right);
1072 right_nr = btrfs_header_nritems(right);
1073 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1074 wret = 1;
1075 } else {
1076 ret = btrfs_cow_block(trans, root, right,
1077 parent, pslot + 1,
1078 &right);
1079 if (ret)
1080 wret = 1;
1081 else {
1082 wret = balance_node_right(trans, root,
1083 right, mid);
1084 }
1085 }
1086 if (wret < 0)
1087 ret = wret;
1088 if (wret == 0) {
1089 struct btrfs_disk_key disk_key;
1090
1091 btrfs_node_key(right, &disk_key, 0);
1092 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1093 btrfs_mark_buffer_dirty(parent);
1094
1095 if (btrfs_header_nritems(mid) <= orig_slot) {
1096 path->nodes[level] = right;
1097 path->slots[level + 1] += 1;
1098 path->slots[level] = orig_slot -
1099 btrfs_header_nritems(mid);
1100 btrfs_tree_unlock(mid);
1101 free_extent_buffer(mid);
1102 } else {
1103 btrfs_tree_unlock(right);
1104 free_extent_buffer(right);
1105 }
1106 return 0;
1107 }
1108 btrfs_tree_unlock(right);
1109 free_extent_buffer(right);
1110 }
1111 return 1;
1112 }
1113
1114 /*
1115 * readahead one full node of leaves
1116 */
1117 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
1118 int level, int slot, u64 objectid)
1119 {
1120 struct extent_buffer *node;
1121 struct btrfs_disk_key disk_key;
1122 u32 nritems;
1123 u64 search;
1124 u64 lowest_read;
1125 u64 highest_read;
1126 u64 nread = 0;
1127 int direction = path->reada;
1128 struct extent_buffer *eb;
1129 u32 nr;
1130 u32 blocksize;
1131 u32 nscan = 0;
1132
1133 if (level != 1)
1134 return;
1135
1136 if (!path->nodes[level])
1137 return;
1138
1139 node = path->nodes[level];
1140
1141 search = btrfs_node_blockptr(node, slot);
1142 blocksize = btrfs_level_size(root, level - 1);
1143 eb = btrfs_find_tree_block(root, search, blocksize);
1144 if (eb) {
1145 free_extent_buffer(eb);
1146 return;
1147 }
1148
1149 highest_read = search;
1150 lowest_read = search;
1151
1152 nritems = btrfs_header_nritems(node);
1153 nr = slot;
1154 while(1) {
1155 if (direction < 0) {
1156 if (nr == 0)
1157 break;
1158 nr--;
1159 } else if (direction > 0) {
1160 nr++;
1161 if (nr >= nritems)
1162 break;
1163 }
1164 if (path->reada < 0 && objectid) {
1165 btrfs_node_key(node, &disk_key, nr);
1166 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1167 break;
1168 }
1169 search = btrfs_node_blockptr(node, nr);
1170 if ((search >= lowest_read && search <= highest_read) ||
1171 (search < lowest_read && lowest_read - search <= 32768) ||
1172 (search > highest_read && search - highest_read <= 32768)) {
1173 readahead_tree_block(root, search, blocksize,
1174 btrfs_node_ptr_generation(node, nr));
1175 nread += blocksize;
1176 }
1177 nscan++;
1178 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1179 break;
1180 if(nread > (1024 * 1024) || nscan > 128)
1181 break;
1182
1183 if (search < lowest_read)
1184 lowest_read = search;
1185 if (search > highest_read)
1186 highest_read = search;
1187 }
1188 }
1189
1190 static void unlock_up(struct btrfs_path *path, int level, int lowest_unlock)
1191 {
1192 int i;
1193 int skip_level = level;
1194 int no_skips = 0;
1195 struct extent_buffer *t;
1196
1197 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1198 if (!path->nodes[i])
1199 break;
1200 if (!path->locks[i])
1201 break;
1202 if (!no_skips && path->slots[i] == 0) {
1203 skip_level = i + 1;
1204 continue;
1205 }
1206 if (!no_skips && path->keep_locks) {
1207 u32 nritems;
1208 t = path->nodes[i];
1209 nritems = btrfs_header_nritems(t);
1210 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1211 skip_level = i + 1;
1212 continue;
1213 }
1214 }
1215 if (skip_level < i && i >= lowest_unlock)
1216 no_skips = 1;
1217
1218 t = path->nodes[i];
1219 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1220 btrfs_tree_unlock(t);
1221 path->locks[i] = 0;
1222 }
1223 }
1224 }
1225
1226 /*
1227 * look for key in the tree. path is filled in with nodes along the way
1228 * if key is found, we return zero and you can find the item in the leaf
1229 * level of the path (level 0)
1230 *
1231 * If the key isn't found, the path points to the slot where it should
1232 * be inserted, and 1 is returned. If there are other errors during the
1233 * search a negative error number is returned.
1234 *
1235 * if ins_len > 0, nodes and leaves will be split as we walk down the
1236 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1237 * possible)
1238 */
1239 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1240 *root, struct btrfs_key *key, struct btrfs_path *p, int
1241 ins_len, int cow)
1242 {
1243 struct extent_buffer *b;
1244 struct extent_buffer *tmp;
1245 int slot;
1246 int ret;
1247 int level;
1248 int should_reada = p->reada;
1249 int lowest_unlock = 1;
1250 int blocksize;
1251 u8 lowest_level = 0;
1252 u64 blocknr;
1253 u64 gen;
1254
1255 lowest_level = p->lowest_level;
1256 WARN_ON(lowest_level && ins_len);
1257 WARN_ON(p->nodes[0] != NULL);
1258 WARN_ON(cow && root == root->fs_info->extent_root &&
1259 !mutex_is_locked(&root->fs_info->alloc_mutex));
1260 WARN_ON(root == root->fs_info->chunk_root &&
1261 !mutex_is_locked(&root->fs_info->chunk_mutex));
1262 WARN_ON(root == root->fs_info->dev_root &&
1263 !mutex_is_locked(&root->fs_info->chunk_mutex));
1264 if (ins_len < 0)
1265 lowest_unlock = 2;
1266 again:
1267 if (p->skip_locking)
1268 b = btrfs_root_node(root);
1269 else
1270 b = btrfs_lock_root_node(root);
1271
1272 while (b) {
1273 level = btrfs_header_level(b);
1274 if (cow) {
1275 int wret;
1276 wret = btrfs_cow_block(trans, root, b,
1277 p->nodes[level + 1],
1278 p->slots[level + 1],
1279 &b);
1280 if (wret) {
1281 free_extent_buffer(b);
1282 return wret;
1283 }
1284 }
1285 BUG_ON(!cow && ins_len);
1286 if (level != btrfs_header_level(b))
1287 WARN_ON(1);
1288 level = btrfs_header_level(b);
1289 p->nodes[level] = b;
1290 if (!p->skip_locking)
1291 p->locks[level] = 1;
1292 ret = check_block(root, p, level);
1293 if (ret)
1294 return -1;
1295
1296 ret = bin_search(b, key, level, &slot);
1297 if (level != 0) {
1298 if (ret && slot > 0)
1299 slot -= 1;
1300 p->slots[level] = slot;
1301 if (ins_len > 0 && btrfs_header_nritems(b) >=
1302 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1303 int sret = split_node(trans, root, p, level);
1304 BUG_ON(sret > 0);
1305 if (sret)
1306 return sret;
1307 b = p->nodes[level];
1308 slot = p->slots[level];
1309 } else if (ins_len < 0) {
1310 int sret = balance_level(trans, root, p,
1311 level);
1312 if (sret)
1313 return sret;
1314 b = p->nodes[level];
1315 if (!b) {
1316 btrfs_release_path(NULL, p);
1317 goto again;
1318 }
1319 slot = p->slots[level];
1320 BUG_ON(btrfs_header_nritems(b) == 1);
1321 }
1322 /* this is only true while dropping a snapshot */
1323 if (level == lowest_level) {
1324 unlock_up(p, level, lowest_unlock);
1325 break;
1326 }
1327
1328 if (should_reada)
1329 reada_for_search(root, p, level, slot,
1330 key->objectid);
1331
1332 blocknr = btrfs_node_blockptr(b, slot);
1333 gen = btrfs_node_ptr_generation(b, slot);
1334 blocksize = btrfs_level_size(root, level - 1);
1335
1336 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1337 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1338 b = tmp;
1339 } else {
1340 /*
1341 * reduce lock contention at high levels
1342 * of the btree by dropping locks before
1343 * we read.
1344 */
1345 if (level > 1) {
1346 btrfs_release_path(NULL, p);
1347 if (tmp)
1348 free_extent_buffer(tmp);
1349 tmp = read_tree_block(root, blocknr,
1350 blocksize, gen);
1351 if (tmp)
1352 free_extent_buffer(tmp);
1353 goto again;
1354 } else {
1355 if (tmp)
1356 free_extent_buffer(tmp);
1357 b = read_node_slot(root, b, slot);
1358 }
1359 }
1360 if (!p->skip_locking)
1361 btrfs_tree_lock(b);
1362 unlock_up(p, level, lowest_unlock);
1363 } else {
1364 p->slots[level] = slot;
1365 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1366 sizeof(struct btrfs_item) + ins_len) {
1367 int sret = split_leaf(trans, root, key,
1368 p, ins_len, ret == 0);
1369 BUG_ON(sret > 0);
1370 if (sret)
1371 return sret;
1372 }
1373 unlock_up(p, level, lowest_unlock);
1374 return ret;
1375 }
1376 }
1377 return 1;
1378 }
1379
1380 /*
1381 * adjust the pointers going up the tree, starting at level
1382 * making sure the right key of each node is points to 'key'.
1383 * This is used after shifting pointers to the left, so it stops
1384 * fixing up pointers when a given leaf/node is not in slot 0 of the
1385 * higher levels
1386 *
1387 * If this fails to write a tree block, it returns -1, but continues
1388 * fixing up the blocks in ram so the tree is consistent.
1389 */
1390 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1391 struct btrfs_root *root, struct btrfs_path *path,
1392 struct btrfs_disk_key *key, int level)
1393 {
1394 int i;
1395 int ret = 0;
1396 struct extent_buffer *t;
1397
1398 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1399 int tslot = path->slots[i];
1400 if (!path->nodes[i])
1401 break;
1402 t = path->nodes[i];
1403 btrfs_set_node_key(t, key, tslot);
1404 btrfs_mark_buffer_dirty(path->nodes[i]);
1405 if (tslot != 0)
1406 break;
1407 }
1408 return ret;
1409 }
1410
1411 /*
1412 * try to push data from one node into the next node left in the
1413 * tree.
1414 *
1415 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1416 * error, and > 0 if there was no room in the left hand block.
1417 */
1418 static int push_node_left(struct btrfs_trans_handle *trans,
1419 struct btrfs_root *root, struct extent_buffer *dst,
1420 struct extent_buffer *src, int empty)
1421 {
1422 int push_items = 0;
1423 int src_nritems;
1424 int dst_nritems;
1425 int ret = 0;
1426
1427 src_nritems = btrfs_header_nritems(src);
1428 dst_nritems = btrfs_header_nritems(dst);
1429 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1430 WARN_ON(btrfs_header_generation(src) != trans->transid);
1431 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1432
1433 if (!empty && src_nritems <= 8)
1434 return 1;
1435
1436 if (push_items <= 0) {
1437 return 1;
1438 }
1439
1440 if (empty) {
1441 push_items = min(src_nritems, push_items);
1442 if (push_items < src_nritems) {
1443 /* leave at least 8 pointers in the node if
1444 * we aren't going to empty it
1445 */
1446 if (src_nritems - push_items < 8) {
1447 if (push_items <= 8)
1448 return 1;
1449 push_items -= 8;
1450 }
1451 }
1452 } else
1453 push_items = min(src_nritems - 8, push_items);
1454
1455 copy_extent_buffer(dst, src,
1456 btrfs_node_key_ptr_offset(dst_nritems),
1457 btrfs_node_key_ptr_offset(0),
1458 push_items * sizeof(struct btrfs_key_ptr));
1459
1460 if (push_items < src_nritems) {
1461 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1462 btrfs_node_key_ptr_offset(push_items),
1463 (src_nritems - push_items) *
1464 sizeof(struct btrfs_key_ptr));
1465 }
1466 btrfs_set_header_nritems(src, src_nritems - push_items);
1467 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1468 btrfs_mark_buffer_dirty(src);
1469 btrfs_mark_buffer_dirty(dst);
1470 return ret;
1471 }
1472
1473 /*
1474 * try to push data from one node into the next node right in the
1475 * tree.
1476 *
1477 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1478 * error, and > 0 if there was no room in the right hand block.
1479 *
1480 * this will only push up to 1/2 the contents of the left node over
1481 */
1482 static int balance_node_right(struct btrfs_trans_handle *trans,
1483 struct btrfs_root *root,
1484 struct extent_buffer *dst,
1485 struct extent_buffer *src)
1486 {
1487 int push_items = 0;
1488 int max_push;
1489 int src_nritems;
1490 int dst_nritems;
1491 int ret = 0;
1492
1493 WARN_ON(btrfs_header_generation(src) != trans->transid);
1494 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1495
1496 src_nritems = btrfs_header_nritems(src);
1497 dst_nritems = btrfs_header_nritems(dst);
1498 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1499 if (push_items <= 0) {
1500 return 1;
1501 }
1502
1503 if (src_nritems < 4) {
1504 return 1;
1505 }
1506
1507 max_push = src_nritems / 2 + 1;
1508 /* don't try to empty the node */
1509 if (max_push >= src_nritems) {
1510 return 1;
1511 }
1512
1513 if (max_push < push_items)
1514 push_items = max_push;
1515
1516 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1517 btrfs_node_key_ptr_offset(0),
1518 (dst_nritems) *
1519 sizeof(struct btrfs_key_ptr));
1520
1521 copy_extent_buffer(dst, src,
1522 btrfs_node_key_ptr_offset(0),
1523 btrfs_node_key_ptr_offset(src_nritems - push_items),
1524 push_items * sizeof(struct btrfs_key_ptr));
1525
1526 btrfs_set_header_nritems(src, src_nritems - push_items);
1527 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1528
1529 btrfs_mark_buffer_dirty(src);
1530 btrfs_mark_buffer_dirty(dst);
1531 return ret;
1532 }
1533
1534 /*
1535 * helper function to insert a new root level in the tree.
1536 * A new node is allocated, and a single item is inserted to
1537 * point to the existing root
1538 *
1539 * returns zero on success or < 0 on failure.
1540 */
1541 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1542 struct btrfs_root *root,
1543 struct btrfs_path *path, int level)
1544 {
1545 u64 root_gen;
1546 u64 lower_gen;
1547 struct extent_buffer *lower;
1548 struct extent_buffer *c;
1549 struct extent_buffer *old;
1550 struct btrfs_disk_key lower_key;
1551
1552 BUG_ON(path->nodes[level]);
1553 BUG_ON(path->nodes[level-1] != root->node);
1554
1555 if (root->ref_cows)
1556 root_gen = trans->transid;
1557 else
1558 root_gen = 0;
1559
1560 lower = path->nodes[level-1];
1561 if (level == 1)
1562 btrfs_item_key(lower, &lower_key, 0);
1563 else
1564 btrfs_node_key(lower, &lower_key, 0);
1565
1566 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1567 root->root_key.objectid,
1568 root_gen, lower_key.objectid, level,
1569 root->node->start, 0);
1570 if (IS_ERR(c))
1571 return PTR_ERR(c);
1572
1573 memset_extent_buffer(c, 0, 0, root->nodesize);
1574 btrfs_set_header_nritems(c, 1);
1575 btrfs_set_header_level(c, level);
1576 btrfs_set_header_bytenr(c, c->start);
1577 btrfs_set_header_generation(c, trans->transid);
1578 btrfs_set_header_owner(c, root->root_key.objectid);
1579
1580 write_extent_buffer(c, root->fs_info->fsid,
1581 (unsigned long)btrfs_header_fsid(c),
1582 BTRFS_FSID_SIZE);
1583
1584 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1585 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1586 BTRFS_UUID_SIZE);
1587
1588 btrfs_set_node_key(c, &lower_key, 0);
1589 btrfs_set_node_blockptr(c, 0, lower->start);
1590 lower_gen = btrfs_header_generation(lower);
1591 WARN_ON(lower_gen == 0);
1592
1593 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1594
1595 btrfs_mark_buffer_dirty(c);
1596
1597 spin_lock(&root->node_lock);
1598 old = root->node;
1599 root->node = c;
1600 spin_unlock(&root->node_lock);
1601
1602 /* the super has an extra ref to root->node */
1603 free_extent_buffer(old);
1604
1605 add_root_to_dirty_list(root);
1606 extent_buffer_get(c);
1607 path->nodes[level] = c;
1608 path->locks[level] = 1;
1609 path->slots[level] = 0;
1610
1611 if (root->ref_cows && lower_gen != trans->transid) {
1612 struct btrfs_path *back_path = btrfs_alloc_path();
1613 int ret;
1614 mutex_lock(&root->fs_info->alloc_mutex);
1615 ret = btrfs_insert_extent_backref(trans,
1616 root->fs_info->extent_root,
1617 path, lower->start,
1618 root->root_key.objectid,
1619 trans->transid, 0, 0);
1620 BUG_ON(ret);
1621 mutex_unlock(&root->fs_info->alloc_mutex);
1622 btrfs_free_path(back_path);
1623 }
1624 return 0;
1625 }
1626
1627 /*
1628 * worker function to insert a single pointer in a node.
1629 * the node should have enough room for the pointer already
1630 *
1631 * slot and level indicate where you want the key to go, and
1632 * blocknr is the block the key points to.
1633 *
1634 * returns zero on success and < 0 on any error
1635 */
1636 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1637 *root, struct btrfs_path *path, struct btrfs_disk_key
1638 *key, u64 bytenr, int slot, int level)
1639 {
1640 struct extent_buffer *lower;
1641 int nritems;
1642
1643 BUG_ON(!path->nodes[level]);
1644 lower = path->nodes[level];
1645 nritems = btrfs_header_nritems(lower);
1646 if (slot > nritems)
1647 BUG();
1648 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1649 BUG();
1650 if (slot != nritems) {
1651 memmove_extent_buffer(lower,
1652 btrfs_node_key_ptr_offset(slot + 1),
1653 btrfs_node_key_ptr_offset(slot),
1654 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1655 }
1656 btrfs_set_node_key(lower, key, slot);
1657 btrfs_set_node_blockptr(lower, slot, bytenr);
1658 WARN_ON(trans->transid == 0);
1659 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1660 btrfs_set_header_nritems(lower, nritems + 1);
1661 btrfs_mark_buffer_dirty(lower);
1662 return 0;
1663 }
1664
1665 /*
1666 * split the node at the specified level in path in two.
1667 * The path is corrected to point to the appropriate node after the split
1668 *
1669 * Before splitting this tries to make some room in the node by pushing
1670 * left and right, if either one works, it returns right away.
1671 *
1672 * returns 0 on success and < 0 on failure
1673 */
1674 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1675 *root, struct btrfs_path *path, int level)
1676 {
1677 u64 root_gen;
1678 struct extent_buffer *c;
1679 struct extent_buffer *split;
1680 struct btrfs_disk_key disk_key;
1681 int mid;
1682 int ret;
1683 int wret;
1684 u32 c_nritems;
1685
1686 c = path->nodes[level];
1687 WARN_ON(btrfs_header_generation(c) != trans->transid);
1688 if (c == root->node) {
1689 /* trying to split the root, lets make a new one */
1690 ret = insert_new_root(trans, root, path, level + 1);
1691 if (ret)
1692 return ret;
1693 } else {
1694 ret = push_nodes_for_insert(trans, root, path, level);
1695 c = path->nodes[level];
1696 if (!ret && btrfs_header_nritems(c) <
1697 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1698 return 0;
1699 if (ret < 0)
1700 return ret;
1701 }
1702
1703 c_nritems = btrfs_header_nritems(c);
1704 if (root->ref_cows)
1705 root_gen = trans->transid;
1706 else
1707 root_gen = 0;
1708
1709 btrfs_node_key(c, &disk_key, 0);
1710 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1711 root->root_key.objectid,
1712 root_gen,
1713 btrfs_disk_key_objectid(&disk_key),
1714 level, c->start, 0);
1715 if (IS_ERR(split))
1716 return PTR_ERR(split);
1717
1718 btrfs_set_header_flags(split, btrfs_header_flags(c));
1719 btrfs_set_header_level(split, btrfs_header_level(c));
1720 btrfs_set_header_bytenr(split, split->start);
1721 btrfs_set_header_generation(split, trans->transid);
1722 btrfs_set_header_owner(split, root->root_key.objectid);
1723 btrfs_set_header_flags(split, 0);
1724 write_extent_buffer(split, root->fs_info->fsid,
1725 (unsigned long)btrfs_header_fsid(split),
1726 BTRFS_FSID_SIZE);
1727 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1728 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1729 BTRFS_UUID_SIZE);
1730
1731 mid = (c_nritems + 1) / 2;
1732
1733 copy_extent_buffer(split, c,
1734 btrfs_node_key_ptr_offset(0),
1735 btrfs_node_key_ptr_offset(mid),
1736 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1737 btrfs_set_header_nritems(split, c_nritems - mid);
1738 btrfs_set_header_nritems(c, mid);
1739 ret = 0;
1740
1741 btrfs_mark_buffer_dirty(c);
1742 btrfs_mark_buffer_dirty(split);
1743
1744 btrfs_node_key(split, &disk_key, 0);
1745 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1746 path->slots[level + 1] + 1,
1747 level + 1);
1748 if (wret)
1749 ret = wret;
1750
1751 if (path->slots[level] >= mid) {
1752 path->slots[level] -= mid;
1753 btrfs_tree_unlock(c);
1754 free_extent_buffer(c);
1755 path->nodes[level] = split;
1756 path->slots[level + 1] += 1;
1757 } else {
1758 btrfs_tree_unlock(split);
1759 free_extent_buffer(split);
1760 }
1761 return ret;
1762 }
1763
1764 /*
1765 * how many bytes are required to store the items in a leaf. start
1766 * and nr indicate which items in the leaf to check. This totals up the
1767 * space used both by the item structs and the item data
1768 */
1769 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1770 {
1771 int data_len;
1772 int nritems = btrfs_header_nritems(l);
1773 int end = min(nritems, start + nr) - 1;
1774
1775 if (!nr)
1776 return 0;
1777 data_len = btrfs_item_end_nr(l, start);
1778 data_len = data_len - btrfs_item_offset_nr(l, end);
1779 data_len += sizeof(struct btrfs_item) * nr;
1780 WARN_ON(data_len < 0);
1781 return data_len;
1782 }
1783
1784 /*
1785 * The space between the end of the leaf items and
1786 * the start of the leaf data. IOW, how much room
1787 * the leaf has left for both items and data
1788 */
1789 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1790 {
1791 int nritems = btrfs_header_nritems(leaf);
1792 int ret;
1793 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1794 if (ret < 0) {
1795 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1796 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1797 leaf_space_used(leaf, 0, nritems), nritems);
1798 }
1799 return ret;
1800 }
1801
1802 /*
1803 * push some data in the path leaf to the right, trying to free up at
1804 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1805 *
1806 * returns 1 if the push failed because the other node didn't have enough
1807 * room, 0 if everything worked out and < 0 if there were major errors.
1808 */
1809 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1810 *root, struct btrfs_path *path, int data_size,
1811 int empty)
1812 {
1813 struct extent_buffer *left = path->nodes[0];
1814 struct extent_buffer *right;
1815 struct extent_buffer *upper;
1816 struct btrfs_disk_key disk_key;
1817 int slot;
1818 u32 i;
1819 int free_space;
1820 int push_space = 0;
1821 int push_items = 0;
1822 struct btrfs_item *item;
1823 u32 left_nritems;
1824 u32 nr;
1825 u32 right_nritems;
1826 u32 data_end;
1827 u32 this_item_size;
1828 int ret;
1829
1830 slot = path->slots[1];
1831 if (!path->nodes[1]) {
1832 return 1;
1833 }
1834 upper = path->nodes[1];
1835 if (slot >= btrfs_header_nritems(upper) - 1)
1836 return 1;
1837
1838 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
1839
1840 right = read_node_slot(root, upper, slot + 1);
1841 btrfs_tree_lock(right);
1842 free_space = btrfs_leaf_free_space(root, right);
1843 if (free_space < data_size + sizeof(struct btrfs_item))
1844 goto out_unlock;
1845
1846 /* cow and double check */
1847 ret = btrfs_cow_block(trans, root, right, upper,
1848 slot + 1, &right);
1849 if (ret)
1850 goto out_unlock;
1851
1852 free_space = btrfs_leaf_free_space(root, right);
1853 if (free_space < data_size + sizeof(struct btrfs_item))
1854 goto out_unlock;
1855
1856 left_nritems = btrfs_header_nritems(left);
1857 if (left_nritems == 0)
1858 goto out_unlock;
1859
1860 if (empty)
1861 nr = 0;
1862 else
1863 nr = 1;
1864
1865 i = left_nritems - 1;
1866 while (i >= nr) {
1867 item = btrfs_item_nr(left, i);
1868
1869 if (path->slots[0] == i)
1870 push_space += data_size + sizeof(*item);
1871
1872 if (!left->map_token) {
1873 map_extent_buffer(left, (unsigned long)item,
1874 sizeof(struct btrfs_item),
1875 &left->map_token, &left->kaddr,
1876 &left->map_start, &left->map_len,
1877 KM_USER1);
1878 }
1879
1880 this_item_size = btrfs_item_size(left, item);
1881 if (this_item_size + sizeof(*item) + push_space > free_space)
1882 break;
1883 push_items++;
1884 push_space += this_item_size + sizeof(*item);
1885 if (i == 0)
1886 break;
1887 i--;
1888 }
1889 if (left->map_token) {
1890 unmap_extent_buffer(left, left->map_token, KM_USER1);
1891 left->map_token = NULL;
1892 }
1893
1894 if (push_items == 0)
1895 goto out_unlock;
1896
1897 if (!empty && push_items == left_nritems)
1898 WARN_ON(1);
1899
1900 /* push left to right */
1901 right_nritems = btrfs_header_nritems(right);
1902
1903 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1904 push_space -= leaf_data_end(root, left);
1905
1906 /* make room in the right data area */
1907 data_end = leaf_data_end(root, right);
1908 memmove_extent_buffer(right,
1909 btrfs_leaf_data(right) + data_end - push_space,
1910 btrfs_leaf_data(right) + data_end,
1911 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1912
1913 /* copy from the left data area */
1914 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1915 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1916 btrfs_leaf_data(left) + leaf_data_end(root, left),
1917 push_space);
1918
1919 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1920 btrfs_item_nr_offset(0),
1921 right_nritems * sizeof(struct btrfs_item));
1922
1923 /* copy the items from left to right */
1924 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1925 btrfs_item_nr_offset(left_nritems - push_items),
1926 push_items * sizeof(struct btrfs_item));
1927
1928 /* update the item pointers */
1929 right_nritems += push_items;
1930 btrfs_set_header_nritems(right, right_nritems);
1931 push_space = BTRFS_LEAF_DATA_SIZE(root);
1932 for (i = 0; i < right_nritems; i++) {
1933 item = btrfs_item_nr(right, i);
1934 if (!right->map_token) {
1935 map_extent_buffer(right, (unsigned long)item,
1936 sizeof(struct btrfs_item),
1937 &right->map_token, &right->kaddr,
1938 &right->map_start, &right->map_len,
1939 KM_USER1);
1940 }
1941 push_space -= btrfs_item_size(right, item);
1942 btrfs_set_item_offset(right, item, push_space);
1943 }
1944
1945 if (right->map_token) {
1946 unmap_extent_buffer(right, right->map_token, KM_USER1);
1947 right->map_token = NULL;
1948 }
1949 left_nritems -= push_items;
1950 btrfs_set_header_nritems(left, left_nritems);
1951
1952 if (left_nritems)
1953 btrfs_mark_buffer_dirty(left);
1954 btrfs_mark_buffer_dirty(right);
1955
1956 btrfs_item_key(right, &disk_key, 0);
1957 btrfs_set_node_key(upper, &disk_key, slot + 1);
1958 btrfs_mark_buffer_dirty(upper);
1959
1960 /* then fixup the leaf pointer in the path */
1961 if (path->slots[0] >= left_nritems) {
1962 path->slots[0] -= left_nritems;
1963 if (btrfs_header_nritems(path->nodes[0]) == 0)
1964 clean_tree_block(trans, root, path->nodes[0]);
1965 btrfs_tree_unlock(path->nodes[0]);
1966 free_extent_buffer(path->nodes[0]);
1967 path->nodes[0] = right;
1968 path->slots[1] += 1;
1969 } else {
1970 btrfs_tree_unlock(right);
1971 free_extent_buffer(right);
1972 }
1973 return 0;
1974
1975 out_unlock:
1976 btrfs_tree_unlock(right);
1977 free_extent_buffer(right);
1978 return 1;
1979 }
1980
1981 /*
1982 * push some data in the path leaf to the left, trying to free up at
1983 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1984 */
1985 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1986 *root, struct btrfs_path *path, int data_size,
1987 int empty)
1988 {
1989 struct btrfs_disk_key disk_key;
1990 struct extent_buffer *right = path->nodes[0];
1991 struct extent_buffer *left;
1992 int slot;
1993 int i;
1994 int free_space;
1995 int push_space = 0;
1996 int push_items = 0;
1997 struct btrfs_item *item;
1998 u32 old_left_nritems;
1999 u32 right_nritems;
2000 u32 nr;
2001 int ret = 0;
2002 int wret;
2003 u32 this_item_size;
2004 u32 old_left_item_size;
2005
2006 slot = path->slots[1];
2007 if (slot == 0)
2008 return 1;
2009 if (!path->nodes[1])
2010 return 1;
2011
2012 right_nritems = btrfs_header_nritems(right);
2013 if (right_nritems == 0) {
2014 return 1;
2015 }
2016
2017 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2018
2019 left = read_node_slot(root, path->nodes[1], slot - 1);
2020 btrfs_tree_lock(left);
2021 free_space = btrfs_leaf_free_space(root, left);
2022 if (free_space < data_size + sizeof(struct btrfs_item)) {
2023 ret = 1;
2024 goto out;
2025 }
2026
2027 /* cow and double check */
2028 ret = btrfs_cow_block(trans, root, left,
2029 path->nodes[1], slot - 1, &left);
2030 if (ret) {
2031 /* we hit -ENOSPC, but it isn't fatal here */
2032 ret = 1;
2033 goto out;
2034 }
2035
2036 free_space = btrfs_leaf_free_space(root, left);
2037 if (free_space < data_size + sizeof(struct btrfs_item)) {
2038 ret = 1;
2039 goto out;
2040 }
2041
2042 if (empty)
2043 nr = right_nritems;
2044 else
2045 nr = right_nritems - 1;
2046
2047 for (i = 0; i < nr; i++) {
2048 item = btrfs_item_nr(right, i);
2049 if (!right->map_token) {
2050 map_extent_buffer(right, (unsigned long)item,
2051 sizeof(struct btrfs_item),
2052 &right->map_token, &right->kaddr,
2053 &right->map_start, &right->map_len,
2054 KM_USER1);
2055 }
2056
2057 if (path->slots[0] == i)
2058 push_space += data_size + sizeof(*item);
2059
2060 this_item_size = btrfs_item_size(right, item);
2061 if (this_item_size + sizeof(*item) + push_space > free_space)
2062 break;
2063
2064 push_items++;
2065 push_space += this_item_size + sizeof(*item);
2066 }
2067
2068 if (right->map_token) {
2069 unmap_extent_buffer(right, right->map_token, KM_USER1);
2070 right->map_token = NULL;
2071 }
2072
2073 if (push_items == 0) {
2074 ret = 1;
2075 goto out;
2076 }
2077 if (!empty && push_items == btrfs_header_nritems(right))
2078 WARN_ON(1);
2079
2080 /* push data from right to left */
2081 copy_extent_buffer(left, right,
2082 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2083 btrfs_item_nr_offset(0),
2084 push_items * sizeof(struct btrfs_item));
2085
2086 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2087 btrfs_item_offset_nr(right, push_items -1);
2088
2089 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2090 leaf_data_end(root, left) - push_space,
2091 btrfs_leaf_data(right) +
2092 btrfs_item_offset_nr(right, push_items - 1),
2093 push_space);
2094 old_left_nritems = btrfs_header_nritems(left);
2095 BUG_ON(old_left_nritems < 0);
2096
2097 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2098 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2099 u32 ioff;
2100
2101 item = btrfs_item_nr(left, i);
2102 if (!left->map_token) {
2103 map_extent_buffer(left, (unsigned long)item,
2104 sizeof(struct btrfs_item),
2105 &left->map_token, &left->kaddr,
2106 &left->map_start, &left->map_len,
2107 KM_USER1);
2108 }
2109
2110 ioff = btrfs_item_offset(left, item);
2111 btrfs_set_item_offset(left, item,
2112 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2113 }
2114 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2115 if (left->map_token) {
2116 unmap_extent_buffer(left, left->map_token, KM_USER1);
2117 left->map_token = NULL;
2118 }
2119
2120 /* fixup right node */
2121 if (push_items > right_nritems) {
2122 printk("push items %d nr %u\n", push_items, right_nritems);
2123 WARN_ON(1);
2124 }
2125
2126 if (push_items < right_nritems) {
2127 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2128 leaf_data_end(root, right);
2129 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2130 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2131 btrfs_leaf_data(right) +
2132 leaf_data_end(root, right), push_space);
2133
2134 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2135 btrfs_item_nr_offset(push_items),
2136 (btrfs_header_nritems(right) - push_items) *
2137 sizeof(struct btrfs_item));
2138 }
2139 right_nritems -= push_items;
2140 btrfs_set_header_nritems(right, right_nritems);
2141 push_space = BTRFS_LEAF_DATA_SIZE(root);
2142 for (i = 0; i < right_nritems; i++) {
2143 item = btrfs_item_nr(right, i);
2144
2145 if (!right->map_token) {
2146 map_extent_buffer(right, (unsigned long)item,
2147 sizeof(struct btrfs_item),
2148 &right->map_token, &right->kaddr,
2149 &right->map_start, &right->map_len,
2150 KM_USER1);
2151 }
2152
2153 push_space = push_space - btrfs_item_size(right, item);
2154 btrfs_set_item_offset(right, item, push_space);
2155 }
2156 if (right->map_token) {
2157 unmap_extent_buffer(right, right->map_token, KM_USER1);
2158 right->map_token = NULL;
2159 }
2160
2161 btrfs_mark_buffer_dirty(left);
2162 if (right_nritems)
2163 btrfs_mark_buffer_dirty(right);
2164
2165 btrfs_item_key(right, &disk_key, 0);
2166 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2167 if (wret)
2168 ret = wret;
2169
2170 /* then fixup the leaf pointer in the path */
2171 if (path->slots[0] < push_items) {
2172 path->slots[0] += old_left_nritems;
2173 if (btrfs_header_nritems(path->nodes[0]) == 0)
2174 clean_tree_block(trans, root, path->nodes[0]);
2175 btrfs_tree_unlock(path->nodes[0]);
2176 free_extent_buffer(path->nodes[0]);
2177 path->nodes[0] = left;
2178 path->slots[1] -= 1;
2179 } else {
2180 btrfs_tree_unlock(left);
2181 free_extent_buffer(left);
2182 path->slots[0] -= push_items;
2183 }
2184 BUG_ON(path->slots[0] < 0);
2185 return ret;
2186 out:
2187 btrfs_tree_unlock(left);
2188 free_extent_buffer(left);
2189 return ret;
2190 }
2191
2192 /*
2193 * split the path's leaf in two, making sure there is at least data_size
2194 * available for the resulting leaf level of the path.
2195 *
2196 * returns 0 if all went well and < 0 on failure.
2197 */
2198 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
2199 *root, struct btrfs_key *ins_key,
2200 struct btrfs_path *path, int data_size, int extend)
2201 {
2202 u64 root_gen;
2203 struct extent_buffer *l;
2204 u32 nritems;
2205 int mid;
2206 int slot;
2207 struct extent_buffer *right;
2208 int space_needed = data_size + sizeof(struct btrfs_item);
2209 int data_copy_size;
2210 int rt_data_off;
2211 int i;
2212 int ret = 0;
2213 int wret;
2214 int double_split;
2215 int num_doubles = 0;
2216 struct btrfs_disk_key disk_key;
2217
2218 if (extend)
2219 space_needed = data_size;
2220
2221 if (root->ref_cows)
2222 root_gen = trans->transid;
2223 else
2224 root_gen = 0;
2225
2226 /* first try to make some room by pushing left and right */
2227 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2228 wret = push_leaf_right(trans, root, path, data_size, 0);
2229 if (wret < 0) {
2230 return wret;
2231 }
2232 if (wret) {
2233 wret = push_leaf_left(trans, root, path, data_size, 0);
2234 if (wret < 0)
2235 return wret;
2236 }
2237 l = path->nodes[0];
2238
2239 /* did the pushes work? */
2240 if (btrfs_leaf_free_space(root, l) >= space_needed)
2241 return 0;
2242 }
2243
2244 if (!path->nodes[1]) {
2245 ret = insert_new_root(trans, root, path, 1);
2246 if (ret)
2247 return ret;
2248 }
2249 again:
2250 double_split = 0;
2251 l = path->nodes[0];
2252 slot = path->slots[0];
2253 nritems = btrfs_header_nritems(l);
2254 mid = (nritems + 1)/ 2;
2255
2256 btrfs_item_key(l, &disk_key, 0);
2257
2258 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2259 root->root_key.objectid,
2260 root_gen, disk_key.objectid, 0,
2261 l->start, 0);
2262 if (IS_ERR(right)) {
2263 BUG_ON(1);
2264 return PTR_ERR(right);
2265 }
2266
2267 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2268 btrfs_set_header_bytenr(right, right->start);
2269 btrfs_set_header_generation(right, trans->transid);
2270 btrfs_set_header_owner(right, root->root_key.objectid);
2271 btrfs_set_header_level(right, 0);
2272 write_extent_buffer(right, root->fs_info->fsid,
2273 (unsigned long)btrfs_header_fsid(right),
2274 BTRFS_FSID_SIZE);
2275
2276 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2277 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2278 BTRFS_UUID_SIZE);
2279 if (mid <= slot) {
2280 if (nritems == 1 ||
2281 leaf_space_used(l, mid, nritems - mid) + space_needed >
2282 BTRFS_LEAF_DATA_SIZE(root)) {
2283 if (slot >= nritems) {
2284 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2285 btrfs_set_header_nritems(right, 0);
2286 wret = insert_ptr(trans, root, path,
2287 &disk_key, right->start,
2288 path->slots[1] + 1, 1);
2289 if (wret)
2290 ret = wret;
2291
2292 btrfs_tree_unlock(path->nodes[0]);
2293 free_extent_buffer(path->nodes[0]);
2294 path->nodes[0] = right;
2295 path->slots[0] = 0;
2296 path->slots[1] += 1;
2297 btrfs_mark_buffer_dirty(right);
2298 return ret;
2299 }
2300 mid = slot;
2301 if (mid != nritems &&
2302 leaf_space_used(l, mid, nritems - mid) +
2303 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2304 double_split = 1;
2305 }
2306 }
2307 } else {
2308 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2309 BTRFS_LEAF_DATA_SIZE(root)) {
2310 if (!extend && slot == 0) {
2311 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2312 btrfs_set_header_nritems(right, 0);
2313 wret = insert_ptr(trans, root, path,
2314 &disk_key,
2315 right->start,
2316 path->slots[1], 1);
2317 if (wret)
2318 ret = wret;
2319 btrfs_tree_unlock(path->nodes[0]);
2320 free_extent_buffer(path->nodes[0]);
2321 path->nodes[0] = right;
2322 path->slots[0] = 0;
2323 if (path->slots[1] == 0) {
2324 wret = fixup_low_keys(trans, root,
2325 path, &disk_key, 1);
2326 if (wret)
2327 ret = wret;
2328 }
2329 btrfs_mark_buffer_dirty(right);
2330 return ret;
2331 } else if (extend && slot == 0) {
2332 mid = 1;
2333 } else {
2334 mid = slot;
2335 if (mid != nritems &&
2336 leaf_space_used(l, mid, nritems - mid) +
2337 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2338 double_split = 1;
2339 }
2340 }
2341 }
2342 }
2343 nritems = nritems - mid;
2344 btrfs_set_header_nritems(right, nritems);
2345 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2346
2347 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2348 btrfs_item_nr_offset(mid),
2349 nritems * sizeof(struct btrfs_item));
2350
2351 copy_extent_buffer(right, l,
2352 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2353 data_copy_size, btrfs_leaf_data(l) +
2354 leaf_data_end(root, l), data_copy_size);
2355
2356 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2357 btrfs_item_end_nr(l, mid);
2358
2359 for (i = 0; i < nritems; i++) {
2360 struct btrfs_item *item = btrfs_item_nr(right, i);
2361 u32 ioff;
2362
2363 if (!right->map_token) {
2364 map_extent_buffer(right, (unsigned long)item,
2365 sizeof(struct btrfs_item),
2366 &right->map_token, &right->kaddr,
2367 &right->map_start, &right->map_len,
2368 KM_USER1);
2369 }
2370
2371 ioff = btrfs_item_offset(right, item);
2372 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2373 }
2374
2375 if (right->map_token) {
2376 unmap_extent_buffer(right, right->map_token, KM_USER1);
2377 right->map_token = NULL;
2378 }
2379
2380 btrfs_set_header_nritems(l, mid);
2381 ret = 0;
2382 btrfs_item_key(right, &disk_key, 0);
2383 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2384 path->slots[1] + 1, 1);
2385 if (wret)
2386 ret = wret;
2387
2388 btrfs_mark_buffer_dirty(right);
2389 btrfs_mark_buffer_dirty(l);
2390 BUG_ON(path->slots[0] != slot);
2391
2392 if (mid <= slot) {
2393 btrfs_tree_unlock(path->nodes[0]);
2394 free_extent_buffer(path->nodes[0]);
2395 path->nodes[0] = right;
2396 path->slots[0] -= mid;
2397 path->slots[1] += 1;
2398 } else {
2399 btrfs_tree_unlock(right);
2400 free_extent_buffer(right);
2401 }
2402
2403 BUG_ON(path->slots[0] < 0);
2404
2405 if (double_split) {
2406 BUG_ON(num_doubles != 0);
2407 num_doubles++;
2408 goto again;
2409 }
2410 return ret;
2411 }
2412
2413 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2414 struct btrfs_root *root,
2415 struct btrfs_path *path,
2416 u32 new_size, int from_end)
2417 {
2418 int ret = 0;
2419 int slot;
2420 int slot_orig;
2421 struct extent_buffer *leaf;
2422 struct btrfs_item *item;
2423 u32 nritems;
2424 unsigned int data_end;
2425 unsigned int old_data_start;
2426 unsigned int old_size;
2427 unsigned int size_diff;
2428 int i;
2429
2430 slot_orig = path->slots[0];
2431 leaf = path->nodes[0];
2432 slot = path->slots[0];
2433
2434 old_size = btrfs_item_size_nr(leaf, slot);
2435 if (old_size == new_size)
2436 return 0;
2437
2438 nritems = btrfs_header_nritems(leaf);
2439 data_end = leaf_data_end(root, leaf);
2440
2441 old_data_start = btrfs_item_offset_nr(leaf, slot);
2442
2443 size_diff = old_size - new_size;
2444
2445 BUG_ON(slot < 0);
2446 BUG_ON(slot >= nritems);
2447
2448 /*
2449 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2450 */
2451 /* first correct the data pointers */
2452 for (i = slot; i < nritems; i++) {
2453 u32 ioff;
2454 item = btrfs_item_nr(leaf, i);
2455
2456 if (!leaf->map_token) {
2457 map_extent_buffer(leaf, (unsigned long)item,
2458 sizeof(struct btrfs_item),
2459 &leaf->map_token, &leaf->kaddr,
2460 &leaf->map_start, &leaf->map_len,
2461 KM_USER1);
2462 }
2463
2464 ioff = btrfs_item_offset(leaf, item);
2465 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2466 }
2467
2468 if (leaf->map_token) {
2469 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2470 leaf->map_token = NULL;
2471 }
2472
2473 /* shift the data */
2474 if (from_end) {
2475 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2476 data_end + size_diff, btrfs_leaf_data(leaf) +
2477 data_end, old_data_start + new_size - data_end);
2478 } else {
2479 struct btrfs_disk_key disk_key;
2480 u64 offset;
2481
2482 btrfs_item_key(leaf, &disk_key, slot);
2483
2484 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2485 unsigned long ptr;
2486 struct btrfs_file_extent_item *fi;
2487
2488 fi = btrfs_item_ptr(leaf, slot,
2489 struct btrfs_file_extent_item);
2490 fi = (struct btrfs_file_extent_item *)(
2491 (unsigned long)fi - size_diff);
2492
2493 if (btrfs_file_extent_type(leaf, fi) ==
2494 BTRFS_FILE_EXTENT_INLINE) {
2495 ptr = btrfs_item_ptr_offset(leaf, slot);
2496 memmove_extent_buffer(leaf, ptr,
2497 (unsigned long)fi,
2498 offsetof(struct btrfs_file_extent_item,
2499 disk_bytenr));
2500 }
2501 }
2502
2503 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2504 data_end + size_diff, btrfs_leaf_data(leaf) +
2505 data_end, old_data_start - data_end);
2506
2507 offset = btrfs_disk_key_offset(&disk_key);
2508 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2509 btrfs_set_item_key(leaf, &disk_key, slot);
2510 if (slot == 0)
2511 fixup_low_keys(trans, root, path, &disk_key, 1);
2512 }
2513
2514 item = btrfs_item_nr(leaf, slot);
2515 btrfs_set_item_size(leaf, item, new_size);
2516 btrfs_mark_buffer_dirty(leaf);
2517
2518 ret = 0;
2519 if (btrfs_leaf_free_space(root, leaf) < 0) {
2520 btrfs_print_leaf(root, leaf);
2521 BUG();
2522 }
2523 return ret;
2524 }
2525
2526 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2527 struct btrfs_root *root, struct btrfs_path *path,
2528 u32 data_size)
2529 {
2530 int ret = 0;
2531 int slot;
2532 int slot_orig;
2533 struct extent_buffer *leaf;
2534 struct btrfs_item *item;
2535 u32 nritems;
2536 unsigned int data_end;
2537 unsigned int old_data;
2538 unsigned int old_size;
2539 int i;
2540
2541 slot_orig = path->slots[0];
2542 leaf = path->nodes[0];
2543
2544 nritems = btrfs_header_nritems(leaf);
2545 data_end = leaf_data_end(root, leaf);
2546
2547 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2548 btrfs_print_leaf(root, leaf);
2549 BUG();
2550 }
2551 slot = path->slots[0];
2552 old_data = btrfs_item_end_nr(leaf, slot);
2553
2554 BUG_ON(slot < 0);
2555 if (slot >= nritems) {
2556 btrfs_print_leaf(root, leaf);
2557 printk("slot %d too large, nritems %d\n", slot, nritems);
2558 BUG_ON(1);
2559 }
2560
2561 /*
2562 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2563 */
2564 /* first correct the data pointers */
2565 for (i = slot; i < nritems; i++) {
2566 u32 ioff;
2567 item = btrfs_item_nr(leaf, i);
2568
2569 if (!leaf->map_token) {
2570 map_extent_buffer(leaf, (unsigned long)item,
2571 sizeof(struct btrfs_item),
2572 &leaf->map_token, &leaf->kaddr,
2573 &leaf->map_start, &leaf->map_len,
2574 KM_USER1);
2575 }
2576 ioff = btrfs_item_offset(leaf, item);
2577 btrfs_set_item_offset(leaf, item, ioff - data_size);
2578 }
2579
2580 if (leaf->map_token) {
2581 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2582 leaf->map_token = NULL;
2583 }
2584
2585 /* shift the data */
2586 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2587 data_end - data_size, btrfs_leaf_data(leaf) +
2588 data_end, old_data - data_end);
2589
2590 data_end = old_data;
2591 old_size = btrfs_item_size_nr(leaf, slot);
2592 item = btrfs_item_nr(leaf, slot);
2593 btrfs_set_item_size(leaf, item, old_size + data_size);
2594 btrfs_mark_buffer_dirty(leaf);
2595
2596 ret = 0;
2597 if (btrfs_leaf_free_space(root, leaf) < 0) {
2598 btrfs_print_leaf(root, leaf);
2599 BUG();
2600 }
2601 return ret;
2602 }
2603
2604 /*
2605 * Given a key and some data, insert an item into the tree.
2606 * This does all the path init required, making room in the tree if needed.
2607 */
2608 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2609 struct btrfs_root *root,
2610 struct btrfs_path *path,
2611 struct btrfs_key *cpu_key, u32 *data_size,
2612 int nr)
2613 {
2614 struct extent_buffer *leaf;
2615 struct btrfs_item *item;
2616 int ret = 0;
2617 int slot;
2618 int slot_orig;
2619 int i;
2620 u32 nritems;
2621 u32 total_size = 0;
2622 u32 total_data = 0;
2623 unsigned int data_end;
2624 struct btrfs_disk_key disk_key;
2625
2626 for (i = 0; i < nr; i++) {
2627 total_data += data_size[i];
2628 }
2629
2630 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2631 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2632 if (ret == 0) {
2633 return -EEXIST;
2634 }
2635 if (ret < 0)
2636 goto out;
2637
2638 slot_orig = path->slots[0];
2639 leaf = path->nodes[0];
2640
2641 nritems = btrfs_header_nritems(leaf);
2642 data_end = leaf_data_end(root, leaf);
2643
2644 if (btrfs_leaf_free_space(root, leaf) <
2645 sizeof(struct btrfs_item) + total_size) {
2646 btrfs_print_leaf(root, leaf);
2647 printk("not enough freespace need %u have %d\n",
2648 total_size, btrfs_leaf_free_space(root, leaf));
2649 BUG();
2650 }
2651
2652 slot = path->slots[0];
2653 BUG_ON(slot < 0);
2654
2655 if (slot != nritems) {
2656 int i;
2657 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2658
2659 if (old_data < data_end) {
2660 btrfs_print_leaf(root, leaf);
2661 printk("slot %d old_data %d data_end %d\n",
2662 slot, old_data, data_end);
2663 BUG_ON(1);
2664 }
2665 /*
2666 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2667 */
2668 /* first correct the data pointers */
2669 WARN_ON(leaf->map_token);
2670 for (i = slot; i < nritems; i++) {
2671 u32 ioff;
2672
2673 item = btrfs_item_nr(leaf, i);
2674 if (!leaf->map_token) {
2675 map_extent_buffer(leaf, (unsigned long)item,
2676 sizeof(struct btrfs_item),
2677 &leaf->map_token, &leaf->kaddr,
2678 &leaf->map_start, &leaf->map_len,
2679 KM_USER1);
2680 }
2681
2682 ioff = btrfs_item_offset(leaf, item);
2683 btrfs_set_item_offset(leaf, item, ioff - total_data);
2684 }
2685 if (leaf->map_token) {
2686 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2687 leaf->map_token = NULL;
2688 }
2689
2690 /* shift the items */
2691 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2692 btrfs_item_nr_offset(slot),
2693 (nritems - slot) * sizeof(struct btrfs_item));
2694
2695 /* shift the data */
2696 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2697 data_end - total_data, btrfs_leaf_data(leaf) +
2698 data_end, old_data - data_end);
2699 data_end = old_data;
2700 }
2701
2702 /* setup the item for the new data */
2703 for (i = 0; i < nr; i++) {
2704 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2705 btrfs_set_item_key(leaf, &disk_key, slot + i);
2706 item = btrfs_item_nr(leaf, slot + i);
2707 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2708 data_end -= data_size[i];
2709 btrfs_set_item_size(leaf, item, data_size[i]);
2710 }
2711 btrfs_set_header_nritems(leaf, nritems + nr);
2712 btrfs_mark_buffer_dirty(leaf);
2713
2714 ret = 0;
2715 if (slot == 0) {
2716 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2717 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2718 }
2719
2720 if (btrfs_leaf_free_space(root, leaf) < 0) {
2721 btrfs_print_leaf(root, leaf);
2722 BUG();
2723 }
2724 out:
2725 return ret;
2726 }
2727
2728 /*
2729 * Given a key and some data, insert an item into the tree.
2730 * This does all the path init required, making room in the tree if needed.
2731 */
2732 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2733 *root, struct btrfs_key *cpu_key, void *data, u32
2734 data_size)
2735 {
2736 int ret = 0;
2737 struct btrfs_path *path;
2738 struct extent_buffer *leaf;
2739 unsigned long ptr;
2740
2741 path = btrfs_alloc_path();
2742 BUG_ON(!path);
2743 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2744 if (!ret) {
2745 leaf = path->nodes[0];
2746 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2747 write_extent_buffer(leaf, data, ptr, data_size);
2748 btrfs_mark_buffer_dirty(leaf);
2749 }
2750 btrfs_free_path(path);
2751 return ret;
2752 }
2753
2754 /*
2755 * delete the pointer from a given node.
2756 *
2757 * If the delete empties a node, the node is removed from the tree,
2758 * continuing all the way the root if required. The root is converted into
2759 * a leaf if all the nodes are emptied.
2760 */
2761 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2762 struct btrfs_path *path, int level, int slot)
2763 {
2764 struct extent_buffer *parent = path->nodes[level];
2765 u32 nritems;
2766 int ret = 0;
2767 int wret;
2768
2769 nritems = btrfs_header_nritems(parent);
2770 if (slot != nritems -1) {
2771 memmove_extent_buffer(parent,
2772 btrfs_node_key_ptr_offset(slot),
2773 btrfs_node_key_ptr_offset(slot + 1),
2774 sizeof(struct btrfs_key_ptr) *
2775 (nritems - slot - 1));
2776 }
2777 nritems--;
2778 btrfs_set_header_nritems(parent, nritems);
2779 if (nritems == 0 && parent == root->node) {
2780 BUG_ON(btrfs_header_level(root->node) != 1);
2781 /* just turn the root into a leaf and break */
2782 btrfs_set_header_level(root->node, 0);
2783 } else if (slot == 0) {
2784 struct btrfs_disk_key disk_key;
2785
2786 btrfs_node_key(parent, &disk_key, 0);
2787 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2788 if (wret)
2789 ret = wret;
2790 }
2791 btrfs_mark_buffer_dirty(parent);
2792 return ret;
2793 }
2794
2795 /*
2796 * delete the item at the leaf level in path. If that empties
2797 * the leaf, remove it from the tree
2798 */
2799 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2800 struct btrfs_path *path, int slot, int nr)
2801 {
2802 struct extent_buffer *leaf;
2803 struct btrfs_item *item;
2804 int last_off;
2805 int dsize = 0;
2806 int ret = 0;
2807 int wret;
2808 int i;
2809 u32 nritems;
2810
2811 leaf = path->nodes[0];
2812 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2813
2814 for (i = 0; i < nr; i++)
2815 dsize += btrfs_item_size_nr(leaf, slot + i);
2816
2817 nritems = btrfs_header_nritems(leaf);
2818
2819 if (slot + nr != nritems) {
2820 int i;
2821 int data_end = leaf_data_end(root, leaf);
2822
2823 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2824 data_end + dsize,
2825 btrfs_leaf_data(leaf) + data_end,
2826 last_off - data_end);
2827
2828 for (i = slot + nr; i < nritems; i++) {
2829 u32 ioff;
2830
2831 item = btrfs_item_nr(leaf, i);
2832 if (!leaf->map_token) {
2833 map_extent_buffer(leaf, (unsigned long)item,
2834 sizeof(struct btrfs_item),
2835 &leaf->map_token, &leaf->kaddr,
2836 &leaf->map_start, &leaf->map_len,
2837 KM_USER1);
2838 }
2839 ioff = btrfs_item_offset(leaf, item);
2840 btrfs_set_item_offset(leaf, item, ioff + dsize);
2841 }
2842
2843 if (leaf->map_token) {
2844 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2845 leaf->map_token = NULL;
2846 }
2847
2848 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2849 btrfs_item_nr_offset(slot + nr),
2850 sizeof(struct btrfs_item) *
2851 (nritems - slot - nr));
2852 }
2853 btrfs_set_header_nritems(leaf, nritems - nr);
2854 nritems -= nr;
2855
2856 /* delete the leaf if we've emptied it */
2857 if (nritems == 0) {
2858 if (leaf == root->node) {
2859 btrfs_set_header_level(leaf, 0);
2860 } else {
2861 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2862 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2863 if (wret)
2864 ret = wret;
2865 wret = btrfs_free_extent(trans, root,
2866 leaf->start, leaf->len,
2867 btrfs_header_owner(path->nodes[1]),
2868 root_gen, 0, 0, 1);
2869 if (wret)
2870 ret = wret;
2871 }
2872 } else {
2873 int used = leaf_space_used(leaf, 0, nritems);
2874 if (slot == 0) {
2875 struct btrfs_disk_key disk_key;
2876
2877 btrfs_item_key(leaf, &disk_key, 0);
2878 wret = fixup_low_keys(trans, root, path,
2879 &disk_key, 1);
2880 if (wret)
2881 ret = wret;
2882 }
2883
2884 /* delete the leaf if it is mostly empty */
2885 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2886 /* push_leaf_left fixes the path.
2887 * make sure the path still points to our leaf
2888 * for possible call to del_ptr below
2889 */
2890 slot = path->slots[1];
2891 extent_buffer_get(leaf);
2892
2893 wret = push_leaf_left(trans, root, path, 1, 1);
2894 if (wret < 0 && wret != -ENOSPC)
2895 ret = wret;
2896
2897 if (path->nodes[0] == leaf &&
2898 btrfs_header_nritems(leaf)) {
2899 wret = push_leaf_right(trans, root, path, 1, 1);
2900 if (wret < 0 && wret != -ENOSPC)
2901 ret = wret;
2902 }
2903
2904 if (btrfs_header_nritems(leaf) == 0) {
2905 u64 root_gen;
2906 u64 bytenr = leaf->start;
2907 u32 blocksize = leaf->len;
2908
2909 root_gen = btrfs_header_generation(
2910 path->nodes[1]);
2911
2912 wret = del_ptr(trans, root, path, 1, slot);
2913 if (wret)
2914 ret = wret;
2915
2916 free_extent_buffer(leaf);
2917 wret = btrfs_free_extent(trans, root, bytenr,
2918 blocksize,
2919 btrfs_header_owner(path->nodes[1]),
2920 root_gen, 0, 0, 1);
2921 if (wret)
2922 ret = wret;
2923 } else {
2924 /* if we're still in the path, make sure
2925 * we're dirty. Otherwise, one of the
2926 * push_leaf functions must have already
2927 * dirtied this buffer
2928 */
2929 if (path->nodes[0] == leaf)
2930 btrfs_mark_buffer_dirty(leaf);
2931 free_extent_buffer(leaf);
2932 }
2933 } else {
2934 btrfs_mark_buffer_dirty(leaf);
2935 }
2936 }
2937 return ret;
2938 }
2939
2940 /*
2941 * search the tree again to find a leaf with lesser keys
2942 * returns 0 if it found something or 1 if there are no lesser leaves.
2943 * returns < 0 on io errors.
2944 */
2945 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2946 {
2947 struct btrfs_key key;
2948 struct btrfs_disk_key found_key;
2949 int ret;
2950
2951 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
2952
2953 if (key.offset > 0)
2954 key.offset--;
2955 else if (key.type > 0)
2956 key.type--;
2957 else if (key.objectid > 0)
2958 key.objectid--;
2959 else
2960 return 1;
2961
2962 btrfs_release_path(root, path);
2963 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2964 if (ret < 0)
2965 return ret;
2966 btrfs_item_key(path->nodes[0], &found_key, 0);
2967 ret = comp_keys(&found_key, &key);
2968 if (ret < 0)
2969 return 0;
2970 return 1;
2971 }
2972
2973 /*
2974 * search the tree again to find a leaf with greater keys
2975 * returns 0 if it found something or 1 if there are no greater leaves.
2976 * returns < 0 on io errors.
2977 */
2978 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2979 {
2980 int slot;
2981 int level = 1;
2982 struct extent_buffer *c;
2983 struct extent_buffer *next = NULL;
2984 struct btrfs_key key;
2985 u32 nritems;
2986 int ret;
2987
2988 nritems = btrfs_header_nritems(path->nodes[0]);
2989 if (nritems == 0) {
2990 return 1;
2991 }
2992
2993 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
2994
2995 btrfs_release_path(root, path);
2996 path->keep_locks = 1;
2997 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2998 path->keep_locks = 0;
2999
3000 if (ret < 0)
3001 return ret;
3002
3003 nritems = btrfs_header_nritems(path->nodes[0]);
3004 /*
3005 * by releasing the path above we dropped all our locks. A balance
3006 * could have added more items next to the key that used to be
3007 * at the very end of the block. So, check again here and
3008 * advance the path if there are now more items available.
3009 */
3010 if (nritems > 0 && path->slots[0] < nritems - 1) {
3011 path->slots[0]++;
3012 goto done;
3013 }
3014
3015 while(level < BTRFS_MAX_LEVEL) {
3016 if (!path->nodes[level])
3017 return 1;
3018
3019 slot = path->slots[level] + 1;
3020 c = path->nodes[level];
3021 if (slot >= btrfs_header_nritems(c)) {
3022 level++;
3023 if (level == BTRFS_MAX_LEVEL) {
3024 return 1;
3025 }
3026 continue;
3027 }
3028
3029 if (next) {
3030 btrfs_tree_unlock(next);
3031 free_extent_buffer(next);
3032 }
3033
3034 if (level == 1 && path->locks[1] && path->reada)
3035 reada_for_search(root, path, level, slot, 0);
3036
3037 next = read_node_slot(root, c, slot);
3038 if (!path->skip_locking) {
3039 WARN_ON(!btrfs_tree_locked(c));
3040 btrfs_tree_lock(next);
3041 }
3042 break;
3043 }
3044 path->slots[level] = slot;
3045 while(1) {
3046 level--;
3047 c = path->nodes[level];
3048 if (path->locks[level])
3049 btrfs_tree_unlock(c);
3050 free_extent_buffer(c);
3051 path->nodes[level] = next;
3052 path->slots[level] = 0;
3053 if (!path->skip_locking)
3054 path->locks[level] = 1;
3055 if (!level)
3056 break;
3057 if (level == 1 && path->locks[1] && path->reada)
3058 reada_for_search(root, path, level, slot, 0);
3059 next = read_node_slot(root, next, 0);
3060 if (!path->skip_locking) {
3061 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3062 btrfs_tree_lock(next);
3063 }
3064 }
3065 done:
3066 unlock_up(path, 0, 1);
3067 return 0;
3068 }
3069
3070 int btrfs_previous_item(struct btrfs_root *root,
3071 struct btrfs_path *path, u64 min_objectid,
3072 int type)
3073 {
3074 struct btrfs_key found_key;
3075 struct extent_buffer *leaf;
3076 int ret;
3077
3078 while(1) {
3079 if (path->slots[0] == 0) {
3080 ret = btrfs_prev_leaf(root, path);
3081 if (ret != 0)
3082 return ret;
3083 } else {
3084 path->slots[0]--;
3085 }
3086 leaf = path->nodes[0];
3087 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3088 if (found_key.type == type)
3089 return 0;
3090 }
3091 return 1;
3092 }
3093
Linux kernel - Deliverables
This page took 0.098519 seconds and 5 git commands to generate.