projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: Add a skip_locking parameter to struct path, and make various funcs honor it
[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(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 b = read_node_slot(root, b, slot);
1356 }
1357 }
1358 if (!p->skip_locking)
1359 btrfs_tree_lock(b);
1360 unlock_up(p, level, lowest_unlock);
1361 } else {
1362 p->slots[level] = slot;
1363 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1364 sizeof(struct btrfs_item) + ins_len) {
1365 int sret = split_leaf(trans, root, key,
1366 p, ins_len, ret == 0);
1367 BUG_ON(sret > 0);
1368 if (sret)
1369 return sret;
1370 }
1371 unlock_up(p, level, lowest_unlock);
1372 return ret;
1373 }
1374 }
1375 return 1;
1376 }
1377
1378 /*
1379 * adjust the pointers going up the tree, starting at level
1380 * making sure the right key of each node is points to 'key'.
1381 * This is used after shifting pointers to the left, so it stops
1382 * fixing up pointers when a given leaf/node is not in slot 0 of the
1383 * higher levels
1384 *
1385 * If this fails to write a tree block, it returns -1, but continues
1386 * fixing up the blocks in ram so the tree is consistent.
1387 */
1388 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1389 struct btrfs_root *root, struct btrfs_path *path,
1390 struct btrfs_disk_key *key, int level)
1391 {
1392 int i;
1393 int ret = 0;
1394 struct extent_buffer *t;
1395
1396 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1397 int tslot = path->slots[i];
1398 if (!path->nodes[i])
1399 break;
1400 t = path->nodes[i];
1401 btrfs_set_node_key(t, key, tslot);
1402 btrfs_mark_buffer_dirty(path->nodes[i]);
1403 if (tslot != 0)
1404 break;
1405 }
1406 return ret;
1407 }
1408
1409 /*
1410 * try to push data from one node into the next node left in the
1411 * tree.
1412 *
1413 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1414 * error, and > 0 if there was no room in the left hand block.
1415 */
1416 static int push_node_left(struct btrfs_trans_handle *trans,
1417 struct btrfs_root *root, struct extent_buffer *dst,
1418 struct extent_buffer *src, int empty)
1419 {
1420 int push_items = 0;
1421 int src_nritems;
1422 int dst_nritems;
1423 int ret = 0;
1424
1425 src_nritems = btrfs_header_nritems(src);
1426 dst_nritems = btrfs_header_nritems(dst);
1427 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1428 WARN_ON(btrfs_header_generation(src) != trans->transid);
1429 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1430
1431 if (!empty && src_nritems <= 8)
1432 return 1;
1433
1434 if (push_items <= 0) {
1435 return 1;
1436 }
1437
1438 if (empty) {
1439 push_items = min(src_nritems, push_items);
1440 if (push_items < src_nritems) {
1441 /* leave at least 8 pointers in the node if
1442 * we aren't going to empty it
1443 */
1444 if (src_nritems - push_items < 8) {
1445 if (push_items <= 8)
1446 return 1;
1447 push_items -= 8;
1448 }
1449 }
1450 } else
1451 push_items = min(src_nritems - 8, push_items);
1452
1453 copy_extent_buffer(dst, src,
1454 btrfs_node_key_ptr_offset(dst_nritems),
1455 btrfs_node_key_ptr_offset(0),
1456 push_items * sizeof(struct btrfs_key_ptr));
1457
1458 if (push_items < src_nritems) {
1459 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1460 btrfs_node_key_ptr_offset(push_items),
1461 (src_nritems - push_items) *
1462 sizeof(struct btrfs_key_ptr));
1463 }
1464 btrfs_set_header_nritems(src, src_nritems - push_items);
1465 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1466 btrfs_mark_buffer_dirty(src);
1467 btrfs_mark_buffer_dirty(dst);
1468 return ret;
1469 }
1470
1471 /*
1472 * try to push data from one node into the next node right in the
1473 * tree.
1474 *
1475 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1476 * error, and > 0 if there was no room in the right hand block.
1477 *
1478 * this will only push up to 1/2 the contents of the left node over
1479 */
1480 static int balance_node_right(struct btrfs_trans_handle *trans,
1481 struct btrfs_root *root,
1482 struct extent_buffer *dst,
1483 struct extent_buffer *src)
1484 {
1485 int push_items = 0;
1486 int max_push;
1487 int src_nritems;
1488 int dst_nritems;
1489 int ret = 0;
1490
1491 WARN_ON(btrfs_header_generation(src) != trans->transid);
1492 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1493
1494 src_nritems = btrfs_header_nritems(src);
1495 dst_nritems = btrfs_header_nritems(dst);
1496 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1497 if (push_items <= 0) {
1498 return 1;
1499 }
1500
1501 if (src_nritems < 4) {
1502 return 1;
1503 }
1504
1505 max_push = src_nritems / 2 + 1;
1506 /* don't try to empty the node */
1507 if (max_push >= src_nritems) {
1508 return 1;
1509 }
1510
1511 if (max_push < push_items)
1512 push_items = max_push;
1513
1514 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1515 btrfs_node_key_ptr_offset(0),
1516 (dst_nritems) *
1517 sizeof(struct btrfs_key_ptr));
1518
1519 copy_extent_buffer(dst, src,
1520 btrfs_node_key_ptr_offset(0),
1521 btrfs_node_key_ptr_offset(src_nritems - push_items),
1522 push_items * sizeof(struct btrfs_key_ptr));
1523
1524 btrfs_set_header_nritems(src, src_nritems - push_items);
1525 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1526
1527 btrfs_mark_buffer_dirty(src);
1528 btrfs_mark_buffer_dirty(dst);
1529 return ret;
1530 }
1531
1532 /*
1533 * helper function to insert a new root level in the tree.
1534 * A new node is allocated, and a single item is inserted to
1535 * point to the existing root
1536 *
1537 * returns zero on success or < 0 on failure.
1538 */
1539 static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1540 struct btrfs_root *root,
1541 struct btrfs_path *path, int level)
1542 {
1543 u64 root_gen;
1544 u64 lower_gen;
1545 struct extent_buffer *lower;
1546 struct extent_buffer *c;
1547 struct extent_buffer *old;
1548 struct btrfs_disk_key lower_key;
1549
1550 BUG_ON(path->nodes[level]);
1551 BUG_ON(path->nodes[level-1] != root->node);
1552
1553 if (root->ref_cows)
1554 root_gen = trans->transid;
1555 else
1556 root_gen = 0;
1557
1558 lower = path->nodes[level-1];
1559 if (level == 1)
1560 btrfs_item_key(lower, &lower_key, 0);
1561 else
1562 btrfs_node_key(lower, &lower_key, 0);
1563
1564 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1565 root->root_key.objectid,
1566 root_gen, lower_key.objectid, level,
1567 root->node->start, 0);
1568 if (IS_ERR(c))
1569 return PTR_ERR(c);
1570
1571 memset_extent_buffer(c, 0, 0, root->nodesize);
1572 btrfs_set_header_nritems(c, 1);
1573 btrfs_set_header_level(c, level);
1574 btrfs_set_header_bytenr(c, c->start);
1575 btrfs_set_header_generation(c, trans->transid);
1576 btrfs_set_header_owner(c, root->root_key.objectid);
1577
1578 write_extent_buffer(c, root->fs_info->fsid,
1579 (unsigned long)btrfs_header_fsid(c),
1580 BTRFS_FSID_SIZE);
1581
1582 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1583 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1584 BTRFS_UUID_SIZE);
1585
1586 btrfs_set_node_key(c, &lower_key, 0);
1587 btrfs_set_node_blockptr(c, 0, lower->start);
1588 lower_gen = btrfs_header_generation(lower);
1589 WARN_ON(lower_gen == 0);
1590
1591 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1592
1593 btrfs_mark_buffer_dirty(c);
1594
1595 spin_lock(&root->node_lock);
1596 old = root->node;
1597 root->node = c;
1598 spin_unlock(&root->node_lock);
1599
1600 /* the super has an extra ref to root->node */
1601 free_extent_buffer(old);
1602
1603 add_root_to_dirty_list(root);
1604 extent_buffer_get(c);
1605 path->nodes[level] = c;
1606 path->locks[level] = 1;
1607 path->slots[level] = 0;
1608
1609 if (root->ref_cows && lower_gen != trans->transid) {
1610 struct btrfs_path *back_path = btrfs_alloc_path();
1611 int ret;
1612 mutex_lock(&root->fs_info->alloc_mutex);
1613 ret = btrfs_insert_extent_backref(trans,
1614 root->fs_info->extent_root,
1615 path, lower->start,
1616 root->root_key.objectid,
1617 trans->transid, 0, 0);
1618 BUG_ON(ret);
1619 mutex_unlock(&root->fs_info->alloc_mutex);
1620 btrfs_free_path(back_path);
1621 }
1622 return 0;
1623 }
1624
1625 /*
1626 * worker function to insert a single pointer in a node.
1627 * the node should have enough room for the pointer already
1628 *
1629 * slot and level indicate where you want the key to go, and
1630 * blocknr is the block the key points to.
1631 *
1632 * returns zero on success and < 0 on any error
1633 */
1634 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1635 *root, struct btrfs_path *path, struct btrfs_disk_key
1636 *key, u64 bytenr, int slot, int level)
1637 {
1638 struct extent_buffer *lower;
1639 int nritems;
1640
1641 BUG_ON(!path->nodes[level]);
1642 lower = path->nodes[level];
1643 nritems = btrfs_header_nritems(lower);
1644 if (slot > nritems)
1645 BUG();
1646 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1647 BUG();
1648 if (slot != nritems) {
1649 memmove_extent_buffer(lower,
1650 btrfs_node_key_ptr_offset(slot + 1),
1651 btrfs_node_key_ptr_offset(slot),
1652 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1653 }
1654 btrfs_set_node_key(lower, key, slot);
1655 btrfs_set_node_blockptr(lower, slot, bytenr);
1656 WARN_ON(trans->transid == 0);
1657 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1658 btrfs_set_header_nritems(lower, nritems + 1);
1659 btrfs_mark_buffer_dirty(lower);
1660 return 0;
1661 }
1662
1663 /*
1664 * split the node at the specified level in path in two.
1665 * The path is corrected to point to the appropriate node after the split
1666 *
1667 * Before splitting this tries to make some room in the node by pushing
1668 * left and right, if either one works, it returns right away.
1669 *
1670 * returns 0 on success and < 0 on failure
1671 */
1672 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1673 *root, struct btrfs_path *path, int level)
1674 {
1675 u64 root_gen;
1676 struct extent_buffer *c;
1677 struct extent_buffer *split;
1678 struct btrfs_disk_key disk_key;
1679 int mid;
1680 int ret;
1681 int wret;
1682 u32 c_nritems;
1683
1684 c = path->nodes[level];
1685 WARN_ON(btrfs_header_generation(c) != trans->transid);
1686 if (c == root->node) {
1687 /* trying to split the root, lets make a new one */
1688 ret = insert_new_root(trans, root, path, level + 1);
1689 if (ret)
1690 return ret;
1691 } else {
1692 ret = push_nodes_for_insert(trans, root, path, level);
1693 c = path->nodes[level];
1694 if (!ret && btrfs_header_nritems(c) <
1695 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1696 return 0;
1697 if (ret < 0)
1698 return ret;
1699 }
1700
1701 c_nritems = btrfs_header_nritems(c);
1702 if (root->ref_cows)
1703 root_gen = trans->transid;
1704 else
1705 root_gen = 0;
1706
1707 btrfs_node_key(c, &disk_key, 0);
1708 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1709 root->root_key.objectid,
1710 root_gen,
1711 btrfs_disk_key_objectid(&disk_key),
1712 level, c->start, 0);
1713 if (IS_ERR(split))
1714 return PTR_ERR(split);
1715
1716 btrfs_set_header_flags(split, btrfs_header_flags(c));
1717 btrfs_set_header_level(split, btrfs_header_level(c));
1718 btrfs_set_header_bytenr(split, split->start);
1719 btrfs_set_header_generation(split, trans->transid);
1720 btrfs_set_header_owner(split, root->root_key.objectid);
1721 btrfs_set_header_flags(split, 0);
1722 write_extent_buffer(split, root->fs_info->fsid,
1723 (unsigned long)btrfs_header_fsid(split),
1724 BTRFS_FSID_SIZE);
1725 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1726 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1727 BTRFS_UUID_SIZE);
1728
1729 mid = (c_nritems + 1) / 2;
1730
1731 copy_extent_buffer(split, c,
1732 btrfs_node_key_ptr_offset(0),
1733 btrfs_node_key_ptr_offset(mid),
1734 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1735 btrfs_set_header_nritems(split, c_nritems - mid);
1736 btrfs_set_header_nritems(c, mid);
1737 ret = 0;
1738
1739 btrfs_mark_buffer_dirty(c);
1740 btrfs_mark_buffer_dirty(split);
1741
1742 btrfs_node_key(split, &disk_key, 0);
1743 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1744 path->slots[level + 1] + 1,
1745 level + 1);
1746 if (wret)
1747 ret = wret;
1748
1749 if (path->slots[level] >= mid) {
1750 path->slots[level] -= mid;
1751 btrfs_tree_unlock(c);
1752 free_extent_buffer(c);
1753 path->nodes[level] = split;
1754 path->slots[level + 1] += 1;
1755 } else {
1756 btrfs_tree_unlock(split);
1757 free_extent_buffer(split);
1758 }
1759 return ret;
1760 }
1761
1762 /*
1763 * how many bytes are required to store the items in a leaf. start
1764 * and nr indicate which items in the leaf to check. This totals up the
1765 * space used both by the item structs and the item data
1766 */
1767 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1768 {
1769 int data_len;
1770 int nritems = btrfs_header_nritems(l);
1771 int end = min(nritems, start + nr) - 1;
1772
1773 if (!nr)
1774 return 0;
1775 data_len = btrfs_item_end_nr(l, start);
1776 data_len = data_len - btrfs_item_offset_nr(l, end);
1777 data_len += sizeof(struct btrfs_item) * nr;
1778 WARN_ON(data_len < 0);
1779 return data_len;
1780 }
1781
1782 /*
1783 * The space between the end of the leaf items and
1784 * the start of the leaf data. IOW, how much room
1785 * the leaf has left for both items and data
1786 */
1787 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1788 {
1789 int nritems = btrfs_header_nritems(leaf);
1790 int ret;
1791 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1792 if (ret < 0) {
1793 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1794 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1795 leaf_space_used(leaf, 0, nritems), nritems);
1796 }
1797 return ret;
1798 }
1799
1800 /*
1801 * push some data in the path leaf to the right, trying to free up at
1802 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1803 *
1804 * returns 1 if the push failed because the other node didn't have enough
1805 * room, 0 if everything worked out and < 0 if there were major errors.
1806 */
1807 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1808 *root, struct btrfs_path *path, int data_size,
1809 int empty)
1810 {
1811 struct extent_buffer *left = path->nodes[0];
1812 struct extent_buffer *right;
1813 struct extent_buffer *upper;
1814 struct btrfs_disk_key disk_key;
1815 int slot;
1816 u32 i;
1817 int free_space;
1818 int push_space = 0;
1819 int push_items = 0;
1820 struct btrfs_item *item;
1821 u32 left_nritems;
1822 u32 nr;
1823 u32 right_nritems;
1824 u32 data_end;
1825 u32 this_item_size;
1826 int ret;
1827
1828 slot = path->slots[1];
1829 if (!path->nodes[1]) {
1830 return 1;
1831 }
1832 upper = path->nodes[1];
1833 if (slot >= btrfs_header_nritems(upper) - 1)
1834 return 1;
1835
1836 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
1837
1838 right = read_node_slot(root, upper, slot + 1);
1839 btrfs_tree_lock(right);
1840 free_space = btrfs_leaf_free_space(root, right);
1841 if (free_space < data_size + sizeof(struct btrfs_item))
1842 goto out_unlock;
1843
1844 /* cow and double check */
1845 ret = btrfs_cow_block(trans, root, right, upper,
1846 slot + 1, &right);
1847 if (ret)
1848 goto out_unlock;
1849
1850 free_space = btrfs_leaf_free_space(root, right);
1851 if (free_space < data_size + sizeof(struct btrfs_item))
1852 goto out_unlock;
1853
1854 left_nritems = btrfs_header_nritems(left);
1855 if (left_nritems == 0)
1856 goto out_unlock;
1857
1858 if (empty)
1859 nr = 0;
1860 else
1861 nr = 1;
1862
1863 i = left_nritems - 1;
1864 while (i >= nr) {
1865 item = btrfs_item_nr(left, i);
1866
1867 if (path->slots[0] == i)
1868 push_space += data_size + sizeof(*item);
1869
1870 if (!left->map_token) {
1871 map_extent_buffer(left, (unsigned long)item,
1872 sizeof(struct btrfs_item),
1873 &left->map_token, &left->kaddr,
1874 &left->map_start, &left->map_len,
1875 KM_USER1);
1876 }
1877
1878 this_item_size = btrfs_item_size(left, item);
1879 if (this_item_size + sizeof(*item) + push_space > free_space)
1880 break;
1881 push_items++;
1882 push_space += this_item_size + sizeof(*item);
1883 if (i == 0)
1884 break;
1885 i--;
1886 }
1887 if (left->map_token) {
1888 unmap_extent_buffer(left, left->map_token, KM_USER1);
1889 left->map_token = NULL;
1890 }
1891
1892 if (push_items == 0)
1893 goto out_unlock;
1894
1895 if (!empty && push_items == left_nritems)
1896 WARN_ON(1);
1897
1898 /* push left to right */
1899 right_nritems = btrfs_header_nritems(right);
1900
1901 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1902 push_space -= leaf_data_end(root, left);
1903
1904 /* make room in the right data area */
1905 data_end = leaf_data_end(root, right);
1906 memmove_extent_buffer(right,
1907 btrfs_leaf_data(right) + data_end - push_space,
1908 btrfs_leaf_data(right) + data_end,
1909 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1910
1911 /* copy from the left data area */
1912 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1913 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1914 btrfs_leaf_data(left) + leaf_data_end(root, left),
1915 push_space);
1916
1917 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1918 btrfs_item_nr_offset(0),
1919 right_nritems * sizeof(struct btrfs_item));
1920
1921 /* copy the items from left to right */
1922 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1923 btrfs_item_nr_offset(left_nritems - push_items),
1924 push_items * sizeof(struct btrfs_item));
1925
1926 /* update the item pointers */
1927 right_nritems += push_items;
1928 btrfs_set_header_nritems(right, right_nritems);
1929 push_space = BTRFS_LEAF_DATA_SIZE(root);
1930 for (i = 0; i < right_nritems; i++) {
1931 item = btrfs_item_nr(right, i);
1932 if (!right->map_token) {
1933 map_extent_buffer(right, (unsigned long)item,
1934 sizeof(struct btrfs_item),
1935 &right->map_token, &right->kaddr,
1936 &right->map_start, &right->map_len,
1937 KM_USER1);
1938 }
1939 push_space -= btrfs_item_size(right, item);
1940 btrfs_set_item_offset(right, item, push_space);
1941 }
1942
1943 if (right->map_token) {
1944 unmap_extent_buffer(right, right->map_token, KM_USER1);
1945 right->map_token = NULL;
1946 }
1947 left_nritems -= push_items;
1948 btrfs_set_header_nritems(left, left_nritems);
1949
1950 if (left_nritems)
1951 btrfs_mark_buffer_dirty(left);
1952 btrfs_mark_buffer_dirty(right);
1953
1954 btrfs_item_key(right, &disk_key, 0);
1955 btrfs_set_node_key(upper, &disk_key, slot + 1);
1956 btrfs_mark_buffer_dirty(upper);
1957
1958 /* then fixup the leaf pointer in the path */
1959 if (path->slots[0] >= left_nritems) {
1960 path->slots[0] -= left_nritems;
1961 if (btrfs_header_nritems(path->nodes[0]) == 0)
1962 clean_tree_block(trans, root, path->nodes[0]);
1963 btrfs_tree_unlock(path->nodes[0]);
1964 free_extent_buffer(path->nodes[0]);
1965 path->nodes[0] = right;
1966 path->slots[1] += 1;
1967 } else {
1968 btrfs_tree_unlock(right);
1969 free_extent_buffer(right);
1970 }
1971 return 0;
1972
1973 out_unlock:
1974 btrfs_tree_unlock(right);
1975 free_extent_buffer(right);
1976 return 1;
1977 }
1978
1979 /*
1980 * push some data in the path leaf to the left, trying to free up at
1981 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1982 */
1983 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1984 *root, struct btrfs_path *path, int data_size,
1985 int empty)
1986 {
1987 struct btrfs_disk_key disk_key;
1988 struct extent_buffer *right = path->nodes[0];
1989 struct extent_buffer *left;
1990 int slot;
1991 int i;
1992 int free_space;
1993 int push_space = 0;
1994 int push_items = 0;
1995 struct btrfs_item *item;
1996 u32 old_left_nritems;
1997 u32 right_nritems;
1998 u32 nr;
1999 int ret = 0;
2000 int wret;
2001 u32 this_item_size;
2002 u32 old_left_item_size;
2003
2004 slot = path->slots[1];
2005 if (slot == 0)
2006 return 1;
2007 if (!path->nodes[1])
2008 return 1;
2009
2010 right_nritems = btrfs_header_nritems(right);
2011 if (right_nritems == 0) {
2012 return 1;
2013 }
2014
2015 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2016
2017 left = read_node_slot(root, path->nodes[1], slot - 1);
2018 btrfs_tree_lock(left);
2019 free_space = btrfs_leaf_free_space(root, left);
2020 if (free_space < data_size + sizeof(struct btrfs_item)) {
2021 ret = 1;
2022 goto out;
2023 }
2024
2025 /* cow and double check */
2026 ret = btrfs_cow_block(trans, root, left,
2027 path->nodes[1], slot - 1, &left);
2028 if (ret) {
2029 /* we hit -ENOSPC, but it isn't fatal here */
2030 ret = 1;
2031 goto out;
2032 }
2033
2034 free_space = btrfs_leaf_free_space(root, left);
2035 if (free_space < data_size + sizeof(struct btrfs_item)) {
2036 ret = 1;
2037 goto out;
2038 }
2039
2040 if (empty)
2041 nr = right_nritems;
2042 else
2043 nr = right_nritems - 1;
2044
2045 for (i = 0; i < nr; i++) {
2046 item = btrfs_item_nr(right, i);
2047 if (!right->map_token) {
2048 map_extent_buffer(right, (unsigned long)item,
2049 sizeof(struct btrfs_item),
2050 &right->map_token, &right->kaddr,
2051 &right->map_start, &right->map_len,
2052 KM_USER1);
2053 }
2054
2055 if (path->slots[0] == i)
2056 push_space += data_size + sizeof(*item);
2057
2058 this_item_size = btrfs_item_size(right, item);
2059 if (this_item_size + sizeof(*item) + push_space > free_space)
2060 break;
2061
2062 push_items++;
2063 push_space += this_item_size + sizeof(*item);
2064 }
2065
2066 if (right->map_token) {
2067 unmap_extent_buffer(right, right->map_token, KM_USER1);
2068 right->map_token = NULL;
2069 }
2070
2071 if (push_items == 0) {
2072 ret = 1;
2073 goto out;
2074 }
2075 if (!empty && push_items == btrfs_header_nritems(right))
2076 WARN_ON(1);
2077
2078 /* push data from right to left */
2079 copy_extent_buffer(left, right,
2080 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2081 btrfs_item_nr_offset(0),
2082 push_items * sizeof(struct btrfs_item));
2083
2084 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2085 btrfs_item_offset_nr(right, push_items -1);
2086
2087 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2088 leaf_data_end(root, left) - push_space,
2089 btrfs_leaf_data(right) +
2090 btrfs_item_offset_nr(right, push_items - 1),
2091 push_space);
2092 old_left_nritems = btrfs_header_nritems(left);
2093 BUG_ON(old_left_nritems < 0);
2094
2095 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2096 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2097 u32 ioff;
2098
2099 item = btrfs_item_nr(left, i);
2100 if (!left->map_token) {
2101 map_extent_buffer(left, (unsigned long)item,
2102 sizeof(struct btrfs_item),
2103 &left->map_token, &left->kaddr,
2104 &left->map_start, &left->map_len,
2105 KM_USER1);
2106 }
2107
2108 ioff = btrfs_item_offset(left, item);
2109 btrfs_set_item_offset(left, item,
2110 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2111 }
2112 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2113 if (left->map_token) {
2114 unmap_extent_buffer(left, left->map_token, KM_USER1);
2115 left->map_token = NULL;
2116 }
2117
2118 /* fixup right node */
2119 if (push_items > right_nritems) {
2120 printk("push items %d nr %u\n", push_items, right_nritems);
2121 WARN_ON(1);
2122 }
2123
2124 if (push_items < right_nritems) {
2125 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2126 leaf_data_end(root, right);
2127 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2128 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2129 btrfs_leaf_data(right) +
2130 leaf_data_end(root, right), push_space);
2131
2132 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2133 btrfs_item_nr_offset(push_items),
2134 (btrfs_header_nritems(right) - push_items) *
2135 sizeof(struct btrfs_item));
2136 }
2137 right_nritems -= push_items;
2138 btrfs_set_header_nritems(right, right_nritems);
2139 push_space = BTRFS_LEAF_DATA_SIZE(root);
2140 for (i = 0; i < right_nritems; i++) {
2141 item = btrfs_item_nr(right, i);
2142
2143 if (!right->map_token) {
2144 map_extent_buffer(right, (unsigned long)item,
2145 sizeof(struct btrfs_item),
2146 &right->map_token, &right->kaddr,
2147 &right->map_start, &right->map_len,
2148 KM_USER1);
2149 }
2150
2151 push_space = push_space - btrfs_item_size(right, item);
2152 btrfs_set_item_offset(right, item, push_space);
2153 }
2154 if (right->map_token) {
2155 unmap_extent_buffer(right, right->map_token, KM_USER1);
2156 right->map_token = NULL;
2157 }
2158
2159 btrfs_mark_buffer_dirty(left);
2160 if (right_nritems)
2161 btrfs_mark_buffer_dirty(right);
2162
2163 btrfs_item_key(right, &disk_key, 0);
2164 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2165 if (wret)
2166 ret = wret;
2167
2168 /* then fixup the leaf pointer in the path */
2169 if (path->slots[0] < push_items) {
2170 path->slots[0] += old_left_nritems;
2171 if (btrfs_header_nritems(path->nodes[0]) == 0)
2172 clean_tree_block(trans, root, path->nodes[0]);
2173 btrfs_tree_unlock(path->nodes[0]);
2174 free_extent_buffer(path->nodes[0]);
2175 path->nodes[0] = left;
2176 path->slots[1] -= 1;
2177 } else {
2178 btrfs_tree_unlock(left);
2179 free_extent_buffer(left);
2180 path->slots[0] -= push_items;
2181 }
2182 BUG_ON(path->slots[0] < 0);
2183 return ret;
2184 out:
2185 btrfs_tree_unlock(left);
2186 free_extent_buffer(left);
2187 return ret;
2188 }
2189
2190 /*
2191 * split the path's leaf in two, making sure there is at least data_size
2192 * available for the resulting leaf level of the path.
2193 *
2194 * returns 0 if all went well and < 0 on failure.
2195 */
2196 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
2197 *root, struct btrfs_key *ins_key,
2198 struct btrfs_path *path, int data_size, int extend)
2199 {
2200 u64 root_gen;
2201 struct extent_buffer *l;
2202 u32 nritems;
2203 int mid;
2204 int slot;
2205 struct extent_buffer *right;
2206 int space_needed = data_size + sizeof(struct btrfs_item);
2207 int data_copy_size;
2208 int rt_data_off;
2209 int i;
2210 int ret = 0;
2211 int wret;
2212 int double_split;
2213 int num_doubles = 0;
2214 struct btrfs_disk_key disk_key;
2215
2216 if (extend)
2217 space_needed = data_size;
2218
2219 if (root->ref_cows)
2220 root_gen = trans->transid;
2221 else
2222 root_gen = 0;
2223
2224 /* first try to make some room by pushing left and right */
2225 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2226 wret = push_leaf_right(trans, root, path, data_size, 0);
2227 if (wret < 0) {
2228 return wret;
2229 }
2230 if (wret) {
2231 wret = push_leaf_left(trans, root, path, data_size, 0);
2232 if (wret < 0)
2233 return wret;
2234 }
2235 l = path->nodes[0];
2236
2237 /* did the pushes work? */
2238 if (btrfs_leaf_free_space(root, l) >= space_needed)
2239 return 0;
2240 }
2241
2242 if (!path->nodes[1]) {
2243 ret = insert_new_root(trans, root, path, 1);
2244 if (ret)
2245 return ret;
2246 }
2247 again:
2248 double_split = 0;
2249 l = path->nodes[0];
2250 slot = path->slots[0];
2251 nritems = btrfs_header_nritems(l);
2252 mid = (nritems + 1)/ 2;
2253
2254 btrfs_item_key(l, &disk_key, 0);
2255
2256 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2257 root->root_key.objectid,
2258 root_gen, disk_key.objectid, 0,
2259 l->start, 0);
2260 if (IS_ERR(right)) {
2261 BUG_ON(1);
2262 return PTR_ERR(right);
2263 }
2264
2265 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2266 btrfs_set_header_bytenr(right, right->start);
2267 btrfs_set_header_generation(right, trans->transid);
2268 btrfs_set_header_owner(right, root->root_key.objectid);
2269 btrfs_set_header_level(right, 0);
2270 write_extent_buffer(right, root->fs_info->fsid,
2271 (unsigned long)btrfs_header_fsid(right),
2272 BTRFS_FSID_SIZE);
2273
2274 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2275 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2276 BTRFS_UUID_SIZE);
2277 if (mid <= slot) {
2278 if (nritems == 1 ||
2279 leaf_space_used(l, mid, nritems - mid) + space_needed >
2280 BTRFS_LEAF_DATA_SIZE(root)) {
2281 if (slot >= nritems) {
2282 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2283 btrfs_set_header_nritems(right, 0);
2284 wret = insert_ptr(trans, root, path,
2285 &disk_key, right->start,
2286 path->slots[1] + 1, 1);
2287 if (wret)
2288 ret = wret;
2289
2290 btrfs_tree_unlock(path->nodes[0]);
2291 free_extent_buffer(path->nodes[0]);
2292 path->nodes[0] = right;
2293 path->slots[0] = 0;
2294 path->slots[1] += 1;
2295 btrfs_mark_buffer_dirty(right);
2296 return ret;
2297 }
2298 mid = slot;
2299 if (mid != nritems &&
2300 leaf_space_used(l, mid, nritems - mid) +
2301 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2302 double_split = 1;
2303 }
2304 }
2305 } else {
2306 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2307 BTRFS_LEAF_DATA_SIZE(root)) {
2308 if (!extend && slot == 0) {
2309 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2310 btrfs_set_header_nritems(right, 0);
2311 wret = insert_ptr(trans, root, path,
2312 &disk_key,
2313 right->start,
2314 path->slots[1], 1);
2315 if (wret)
2316 ret = wret;
2317 btrfs_tree_unlock(path->nodes[0]);
2318 free_extent_buffer(path->nodes[0]);
2319 path->nodes[0] = right;
2320 path->slots[0] = 0;
2321 if (path->slots[1] == 0) {
2322 wret = fixup_low_keys(trans, root,
2323 path, &disk_key, 1);
2324 if (wret)
2325 ret = wret;
2326 }
2327 btrfs_mark_buffer_dirty(right);
2328 return ret;
2329 } else if (extend && slot == 0) {
2330 mid = 1;
2331 } else {
2332 mid = slot;
2333 if (mid != nritems &&
2334 leaf_space_used(l, mid, nritems - mid) +
2335 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2336 double_split = 1;
2337 }
2338 }
2339 }
2340 }
2341 nritems = nritems - mid;
2342 btrfs_set_header_nritems(right, nritems);
2343 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2344
2345 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2346 btrfs_item_nr_offset(mid),
2347 nritems * sizeof(struct btrfs_item));
2348
2349 copy_extent_buffer(right, l,
2350 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2351 data_copy_size, btrfs_leaf_data(l) +
2352 leaf_data_end(root, l), data_copy_size);
2353
2354 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2355 btrfs_item_end_nr(l, mid);
2356
2357 for (i = 0; i < nritems; i++) {
2358 struct btrfs_item *item = btrfs_item_nr(right, i);
2359 u32 ioff;
2360
2361 if (!right->map_token) {
2362 map_extent_buffer(right, (unsigned long)item,
2363 sizeof(struct btrfs_item),
2364 &right->map_token, &right->kaddr,
2365 &right->map_start, &right->map_len,
2366 KM_USER1);
2367 }
2368
2369 ioff = btrfs_item_offset(right, item);
2370 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2371 }
2372
2373 if (right->map_token) {
2374 unmap_extent_buffer(right, right->map_token, KM_USER1);
2375 right->map_token = NULL;
2376 }
2377
2378 btrfs_set_header_nritems(l, mid);
2379 ret = 0;
2380 btrfs_item_key(right, &disk_key, 0);
2381 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2382 path->slots[1] + 1, 1);
2383 if (wret)
2384 ret = wret;
2385
2386 btrfs_mark_buffer_dirty(right);
2387 btrfs_mark_buffer_dirty(l);
2388 BUG_ON(path->slots[0] != slot);
2389
2390 if (mid <= slot) {
2391 btrfs_tree_unlock(path->nodes[0]);
2392 free_extent_buffer(path->nodes[0]);
2393 path->nodes[0] = right;
2394 path->slots[0] -= mid;
2395 path->slots[1] += 1;
2396 } else {
2397 btrfs_tree_unlock(right);
2398 free_extent_buffer(right);
2399 }
2400
2401 BUG_ON(path->slots[0] < 0);
2402
2403 if (double_split) {
2404 BUG_ON(num_doubles != 0);
2405 num_doubles++;
2406 goto again;
2407 }
2408 return ret;
2409 }
2410
2411 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2412 struct btrfs_root *root,
2413 struct btrfs_path *path,
2414 u32 new_size, int from_end)
2415 {
2416 int ret = 0;
2417 int slot;
2418 int slot_orig;
2419 struct extent_buffer *leaf;
2420 struct btrfs_item *item;
2421 u32 nritems;
2422 unsigned int data_end;
2423 unsigned int old_data_start;
2424 unsigned int old_size;
2425 unsigned int size_diff;
2426 int i;
2427
2428 slot_orig = path->slots[0];
2429 leaf = path->nodes[0];
2430 slot = path->slots[0];
2431
2432 old_size = btrfs_item_size_nr(leaf, slot);
2433 if (old_size == new_size)
2434 return 0;
2435
2436 nritems = btrfs_header_nritems(leaf);
2437 data_end = leaf_data_end(root, leaf);
2438
2439 old_data_start = btrfs_item_offset_nr(leaf, slot);
2440
2441 size_diff = old_size - new_size;
2442
2443 BUG_ON(slot < 0);
2444 BUG_ON(slot >= nritems);
2445
2446 /*
2447 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2448 */
2449 /* first correct the data pointers */
2450 for (i = slot; i < nritems; i++) {
2451 u32 ioff;
2452 item = btrfs_item_nr(leaf, i);
2453
2454 if (!leaf->map_token) {
2455 map_extent_buffer(leaf, (unsigned long)item,
2456 sizeof(struct btrfs_item),
2457 &leaf->map_token, &leaf->kaddr,
2458 &leaf->map_start, &leaf->map_len,
2459 KM_USER1);
2460 }
2461
2462 ioff = btrfs_item_offset(leaf, item);
2463 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2464 }
2465
2466 if (leaf->map_token) {
2467 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2468 leaf->map_token = NULL;
2469 }
2470
2471 /* shift the data */
2472 if (from_end) {
2473 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2474 data_end + size_diff, btrfs_leaf_data(leaf) +
2475 data_end, old_data_start + new_size - data_end);
2476 } else {
2477 struct btrfs_disk_key disk_key;
2478 u64 offset;
2479
2480 btrfs_item_key(leaf, &disk_key, slot);
2481
2482 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2483 unsigned long ptr;
2484 struct btrfs_file_extent_item *fi;
2485
2486 fi = btrfs_item_ptr(leaf, slot,
2487 struct btrfs_file_extent_item);
2488 fi = (struct btrfs_file_extent_item *)(
2489 (unsigned long)fi - size_diff);
2490
2491 if (btrfs_file_extent_type(leaf, fi) ==
2492 BTRFS_FILE_EXTENT_INLINE) {
2493 ptr = btrfs_item_ptr_offset(leaf, slot);
2494 memmove_extent_buffer(leaf, ptr,
2495 (unsigned long)fi,
2496 offsetof(struct btrfs_file_extent_item,
2497 disk_bytenr));
2498 }
2499 }
2500
2501 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2502 data_end + size_diff, btrfs_leaf_data(leaf) +
2503 data_end, old_data_start - data_end);
2504
2505 offset = btrfs_disk_key_offset(&disk_key);
2506 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2507 btrfs_set_item_key(leaf, &disk_key, slot);
2508 if (slot == 0)
2509 fixup_low_keys(trans, root, path, &disk_key, 1);
2510 }
2511
2512 item = btrfs_item_nr(leaf, slot);
2513 btrfs_set_item_size(leaf, item, new_size);
2514 btrfs_mark_buffer_dirty(leaf);
2515
2516 ret = 0;
2517 if (btrfs_leaf_free_space(root, leaf) < 0) {
2518 btrfs_print_leaf(root, leaf);
2519 BUG();
2520 }
2521 return ret;
2522 }
2523
2524 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2525 struct btrfs_root *root, struct btrfs_path *path,
2526 u32 data_size)
2527 {
2528 int ret = 0;
2529 int slot;
2530 int slot_orig;
2531 struct extent_buffer *leaf;
2532 struct btrfs_item *item;
2533 u32 nritems;
2534 unsigned int data_end;
2535 unsigned int old_data;
2536 unsigned int old_size;
2537 int i;
2538
2539 slot_orig = path->slots[0];
2540 leaf = path->nodes[0];
2541
2542 nritems = btrfs_header_nritems(leaf);
2543 data_end = leaf_data_end(root, leaf);
2544
2545 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2546 btrfs_print_leaf(root, leaf);
2547 BUG();
2548 }
2549 slot = path->slots[0];
2550 old_data = btrfs_item_end_nr(leaf, slot);
2551
2552 BUG_ON(slot < 0);
2553 if (slot >= nritems) {
2554 btrfs_print_leaf(root, leaf);
2555 printk("slot %d too large, nritems %d\n", slot, nritems);
2556 BUG_ON(1);
2557 }
2558
2559 /*
2560 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2561 */
2562 /* first correct the data pointers */
2563 for (i = slot; i < nritems; i++) {
2564 u32 ioff;
2565 item = btrfs_item_nr(leaf, i);
2566
2567 if (!leaf->map_token) {
2568 map_extent_buffer(leaf, (unsigned long)item,
2569 sizeof(struct btrfs_item),
2570 &leaf->map_token, &leaf->kaddr,
2571 &leaf->map_start, &leaf->map_len,
2572 KM_USER1);
2573 }
2574 ioff = btrfs_item_offset(leaf, item);
2575 btrfs_set_item_offset(leaf, item, ioff - data_size);
2576 }
2577
2578 if (leaf->map_token) {
2579 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2580 leaf->map_token = NULL;
2581 }
2582
2583 /* shift the data */
2584 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2585 data_end - data_size, btrfs_leaf_data(leaf) +
2586 data_end, old_data - data_end);
2587
2588 data_end = old_data;
2589 old_size = btrfs_item_size_nr(leaf, slot);
2590 item = btrfs_item_nr(leaf, slot);
2591 btrfs_set_item_size(leaf, item, old_size + data_size);
2592 btrfs_mark_buffer_dirty(leaf);
2593
2594 ret = 0;
2595 if (btrfs_leaf_free_space(root, leaf) < 0) {
2596 btrfs_print_leaf(root, leaf);
2597 BUG();
2598 }
2599 return ret;
2600 }
2601
2602 /*
2603 * Given a key and some data, insert an item into the tree.
2604 * This does all the path init required, making room in the tree if needed.
2605 */
2606 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2607 struct btrfs_root *root,
2608 struct btrfs_path *path,
2609 struct btrfs_key *cpu_key, u32 *data_size,
2610 int nr)
2611 {
2612 struct extent_buffer *leaf;
2613 struct btrfs_item *item;
2614 int ret = 0;
2615 int slot;
2616 int slot_orig;
2617 int i;
2618 u32 nritems;
2619 u32 total_size = 0;
2620 u32 total_data = 0;
2621 unsigned int data_end;
2622 struct btrfs_disk_key disk_key;
2623
2624 for (i = 0; i < nr; i++) {
2625 total_data += data_size[i];
2626 }
2627
2628 total_size = total_data + (nr - 1) * sizeof(struct btrfs_item);
2629 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2630 if (ret == 0) {
2631 return -EEXIST;
2632 }
2633 if (ret < 0)
2634 goto out;
2635
2636 slot_orig = path->slots[0];
2637 leaf = path->nodes[0];
2638
2639 nritems = btrfs_header_nritems(leaf);
2640 data_end = leaf_data_end(root, leaf);
2641
2642 if (btrfs_leaf_free_space(root, leaf) <
2643 sizeof(struct btrfs_item) + total_size) {
2644 btrfs_print_leaf(root, leaf);
2645 printk("not enough freespace need %u have %d\n",
2646 total_size, btrfs_leaf_free_space(root, leaf));
2647 BUG();
2648 }
2649
2650 slot = path->slots[0];
2651 BUG_ON(slot < 0);
2652
2653 if (slot != nritems) {
2654 int i;
2655 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2656
2657 if (old_data < data_end) {
2658 btrfs_print_leaf(root, leaf);
2659 printk("slot %d old_data %d data_end %d\n",
2660 slot, old_data, data_end);
2661 BUG_ON(1);
2662 }
2663 /*
2664 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2665 */
2666 /* first correct the data pointers */
2667 WARN_ON(leaf->map_token);
2668 for (i = slot; i < nritems; i++) {
2669 u32 ioff;
2670
2671 item = btrfs_item_nr(leaf, i);
2672 if (!leaf->map_token) {
2673 map_extent_buffer(leaf, (unsigned long)item,
2674 sizeof(struct btrfs_item),
2675 &leaf->map_token, &leaf->kaddr,
2676 &leaf->map_start, &leaf->map_len,
2677 KM_USER1);
2678 }
2679
2680 ioff = btrfs_item_offset(leaf, item);
2681 btrfs_set_item_offset(leaf, item, ioff - total_data);
2682 }
2683 if (leaf->map_token) {
2684 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2685 leaf->map_token = NULL;
2686 }
2687
2688 /* shift the items */
2689 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2690 btrfs_item_nr_offset(slot),
2691 (nritems - slot) * sizeof(struct btrfs_item));
2692
2693 /* shift the data */
2694 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2695 data_end - total_data, btrfs_leaf_data(leaf) +
2696 data_end, old_data - data_end);
2697 data_end = old_data;
2698 }
2699
2700 /* setup the item for the new data */
2701 for (i = 0; i < nr; i++) {
2702 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2703 btrfs_set_item_key(leaf, &disk_key, slot + i);
2704 item = btrfs_item_nr(leaf, slot + i);
2705 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2706 data_end -= data_size[i];
2707 btrfs_set_item_size(leaf, item, data_size[i]);
2708 }
2709 btrfs_set_header_nritems(leaf, nritems + nr);
2710 btrfs_mark_buffer_dirty(leaf);
2711
2712 ret = 0;
2713 if (slot == 0) {
2714 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2715 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2716 }
2717
2718 if (btrfs_leaf_free_space(root, leaf) < 0) {
2719 btrfs_print_leaf(root, leaf);
2720 BUG();
2721 }
2722 out:
2723 return ret;
2724 }
2725
2726 /*
2727 * Given a key and some data, insert an item into the tree.
2728 * This does all the path init required, making room in the tree if needed.
2729 */
2730 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2731 *root, struct btrfs_key *cpu_key, void *data, u32
2732 data_size)
2733 {
2734 int ret = 0;
2735 struct btrfs_path *path;
2736 struct extent_buffer *leaf;
2737 unsigned long ptr;
2738
2739 path = btrfs_alloc_path();
2740 BUG_ON(!path);
2741 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2742 if (!ret) {
2743 leaf = path->nodes[0];
2744 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2745 write_extent_buffer(leaf, data, ptr, data_size);
2746 btrfs_mark_buffer_dirty(leaf);
2747 }
2748 btrfs_free_path(path);
2749 return ret;
2750 }
2751
2752 /*
2753 * delete the pointer from a given node.
2754 *
2755 * If the delete empties a node, the node is removed from the tree,
2756 * continuing all the way the root if required. The root is converted into
2757 * a leaf if all the nodes are emptied.
2758 */
2759 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2760 struct btrfs_path *path, int level, int slot)
2761 {
2762 struct extent_buffer *parent = path->nodes[level];
2763 u32 nritems;
2764 int ret = 0;
2765 int wret;
2766
2767 nritems = btrfs_header_nritems(parent);
2768 if (slot != nritems -1) {
2769 memmove_extent_buffer(parent,
2770 btrfs_node_key_ptr_offset(slot),
2771 btrfs_node_key_ptr_offset(slot + 1),
2772 sizeof(struct btrfs_key_ptr) *
2773 (nritems - slot - 1));
2774 }
2775 nritems--;
2776 btrfs_set_header_nritems(parent, nritems);
2777 if (nritems == 0 && parent == root->node) {
2778 BUG_ON(btrfs_header_level(root->node) != 1);
2779 /* just turn the root into a leaf and break */
2780 btrfs_set_header_level(root->node, 0);
2781 } else if (slot == 0) {
2782 struct btrfs_disk_key disk_key;
2783
2784 btrfs_node_key(parent, &disk_key, 0);
2785 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2786 if (wret)
2787 ret = wret;
2788 }
2789 btrfs_mark_buffer_dirty(parent);
2790 return ret;
2791 }
2792
2793 /*
2794 * delete the item at the leaf level in path. If that empties
2795 * the leaf, remove it from the tree
2796 */
2797 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2798 struct btrfs_path *path, int slot, int nr)
2799 {
2800 struct extent_buffer *leaf;
2801 struct btrfs_item *item;
2802 int last_off;
2803 int dsize = 0;
2804 int ret = 0;
2805 int wret;
2806 int i;
2807 u32 nritems;
2808
2809 leaf = path->nodes[0];
2810 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2811
2812 for (i = 0; i < nr; i++)
2813 dsize += btrfs_item_size_nr(leaf, slot + i);
2814
2815 nritems = btrfs_header_nritems(leaf);
2816
2817 if (slot + nr != nritems) {
2818 int i;
2819 int data_end = leaf_data_end(root, leaf);
2820
2821 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2822 data_end + dsize,
2823 btrfs_leaf_data(leaf) + data_end,
2824 last_off - data_end);
2825
2826 for (i = slot + nr; i < nritems; i++) {
2827 u32 ioff;
2828
2829 item = btrfs_item_nr(leaf, i);
2830 if (!leaf->map_token) {
2831 map_extent_buffer(leaf, (unsigned long)item,
2832 sizeof(struct btrfs_item),
2833 &leaf->map_token, &leaf->kaddr,
2834 &leaf->map_start, &leaf->map_len,
2835 KM_USER1);
2836 }
2837 ioff = btrfs_item_offset(leaf, item);
2838 btrfs_set_item_offset(leaf, item, ioff + dsize);
2839 }
2840
2841 if (leaf->map_token) {
2842 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2843 leaf->map_token = NULL;
2844 }
2845
2846 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2847 btrfs_item_nr_offset(slot + nr),
2848 sizeof(struct btrfs_item) *
2849 (nritems - slot - nr));
2850 }
2851 btrfs_set_header_nritems(leaf, nritems - nr);
2852 nritems -= nr;
2853
2854 /* delete the leaf if we've emptied it */
2855 if (nritems == 0) {
2856 if (leaf == root->node) {
2857 btrfs_set_header_level(leaf, 0);
2858 } else {
2859 u64 root_gen = btrfs_header_generation(path->nodes[1]);
2860 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2861 if (wret)
2862 ret = wret;
2863 wret = btrfs_free_extent(trans, root,
2864 leaf->start, leaf->len,
2865 btrfs_header_owner(path->nodes[1]),
2866 root_gen, 0, 0, 1);
2867 if (wret)
2868 ret = wret;
2869 }
2870 } else {
2871 int used = leaf_space_used(leaf, 0, nritems);
2872 if (slot == 0) {
2873 struct btrfs_disk_key disk_key;
2874
2875 btrfs_item_key(leaf, &disk_key, 0);
2876 wret = fixup_low_keys(trans, root, path,
2877 &disk_key, 1);
2878 if (wret)
2879 ret = wret;
2880 }
2881
2882 /* delete the leaf if it is mostly empty */
2883 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
2884 /* push_leaf_left fixes the path.
2885 * make sure the path still points to our leaf
2886 * for possible call to del_ptr below
2887 */
2888 slot = path->slots[1];
2889 extent_buffer_get(leaf);
2890
2891 wret = push_leaf_left(trans, root, path, 1, 1);
2892 if (wret < 0 && wret != -ENOSPC)
2893 ret = wret;
2894
2895 if (path->nodes[0] == leaf &&
2896 btrfs_header_nritems(leaf)) {
2897 wret = push_leaf_right(trans, root, path, 1, 1);
2898 if (wret < 0 && wret != -ENOSPC)
2899 ret = wret;
2900 }
2901
2902 if (btrfs_header_nritems(leaf) == 0) {
2903 u64 root_gen;
2904 u64 bytenr = leaf->start;
2905 u32 blocksize = leaf->len;
2906
2907 root_gen = btrfs_header_generation(
2908 path->nodes[1]);
2909
2910 wret = del_ptr(trans, root, path, 1, slot);
2911 if (wret)
2912 ret = wret;
2913
2914 free_extent_buffer(leaf);
2915 wret = btrfs_free_extent(trans, root, bytenr,
2916 blocksize,
2917 btrfs_header_owner(path->nodes[1]),
2918 root_gen, 0, 0, 1);
2919 if (wret)
2920 ret = wret;
2921 } else {
2922 /* if we're still in the path, make sure
2923 * we're dirty. Otherwise, one of the
2924 * push_leaf functions must have already
2925 * dirtied this buffer
2926 */
2927 if (path->nodes[0] == leaf)
2928 btrfs_mark_buffer_dirty(leaf);
2929 free_extent_buffer(leaf);
2930 }
2931 } else {
2932 btrfs_mark_buffer_dirty(leaf);
2933 }
2934 }
2935 return ret;
2936 }
2937
2938 /*
2939 * search the tree again to find a leaf with lesser keys
2940 * returns 0 if it found something or 1 if there are no lesser leaves.
2941 * returns < 0 on io errors.
2942 */
2943 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
2944 {
2945 struct btrfs_key key;
2946 struct btrfs_disk_key found_key;
2947 int ret;
2948
2949 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
2950
2951 if (key.offset > 0)
2952 key.offset--;
2953 else if (key.type > 0)
2954 key.type--;
2955 else if (key.objectid > 0)
2956 key.objectid--;
2957 else
2958 return 1;
2959
2960 btrfs_release_path(root, path);
2961 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2962 if (ret < 0)
2963 return ret;
2964 btrfs_item_key(path->nodes[0], &found_key, 0);
2965 ret = comp_keys(&found_key, &key);
2966 if (ret < 0)
2967 return 0;
2968 return 1;
2969 }
2970
2971 /*
2972 * search the tree again to find a leaf with greater keys
2973 * returns 0 if it found something or 1 if there are no greater leaves.
2974 * returns < 0 on io errors.
2975 */
2976 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2977 {
2978 int slot;
2979 int level = 1;
2980 struct extent_buffer *c;
2981 struct extent_buffer *next = NULL;
2982 struct btrfs_key key;
2983 u32 nritems;
2984 int ret;
2985
2986 nritems = btrfs_header_nritems(path->nodes[0]);
2987 if (nritems == 0) {
2988 return 1;
2989 }
2990
2991 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
2992
2993 btrfs_release_path(root, path);
2994 path->keep_locks = 1;
2995 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2996 path->keep_locks = 0;
2997
2998 if (ret < 0)
2999 return ret;
3000
3001 nritems = btrfs_header_nritems(path->nodes[0]);
3002 /*
3003 * by releasing the path above we dropped all our locks. A balance
3004 * could have added more items next to the key that used to be
3005 * at the very end of the block. So, check again here and
3006 * advance the path if there are now more items available.
3007 */
3008 if (nritems > 0 && path->slots[0] < nritems - 1) {
3009 path->slots[0]++;
3010 goto done;
3011 }
3012
3013 while(level < BTRFS_MAX_LEVEL) {
3014 if (!path->nodes[level])
3015 return 1;
3016
3017 slot = path->slots[level] + 1;
3018 c = path->nodes[level];
3019 if (slot >= btrfs_header_nritems(c)) {
3020 level++;
3021 if (level == BTRFS_MAX_LEVEL) {
3022 return 1;
3023 }
3024 continue;
3025 }
3026
3027 if (next) {
3028 btrfs_tree_unlock(next);
3029 free_extent_buffer(next);
3030 }
3031
3032 if (level == 1 && path->locks[1] && path->reada)
3033 reada_for_search(root, path, level, slot, 0);
3034
3035 next = read_node_slot(root, c, slot);
3036 if (!path->skip_locking) {
3037 WARN_ON(!btrfs_tree_locked(c));
3038 btrfs_tree_lock(next);
3039 }
3040 break;
3041 }
3042 path->slots[level] = slot;
3043 while(1) {
3044 level--;
3045 c = path->nodes[level];
3046 if (path->locks[level])
3047 btrfs_tree_unlock(c);
3048 free_extent_buffer(c);
3049 path->nodes[level] = next;
3050 path->slots[level] = 0;
3051 path->locks[level] = 1;
3052 if (!level)
3053 break;
3054 if (level == 1 && path->locks[1] && path->reada)
3055 reada_for_search(root, path, level, slot, 0);
3056 next = read_node_slot(root, next, 0);
3057 if (!path->skip_locking) {
3058 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3059 btrfs_tree_lock(next);
3060 }
3061 }
3062 done:
3063 unlock_up(path, 0, 1);
3064 return 0;
3065 }
3066
3067 int btrfs_previous_item(struct btrfs_root *root,
3068 struct btrfs_path *path, u64 min_objectid,
3069 int type)
3070 {
3071 struct btrfs_key found_key;
3072 struct extent_buffer *leaf;
3073 int ret;
3074
3075 while(1) {
3076 if (path->slots[0] == 0) {
3077 ret = btrfs_prev_leaf(root, path);
3078 if (ret != 0)
3079 return ret;
3080 } else {
3081 path->slots[0]--;
3082 }
3083 leaf = path->nodes[0];
3084 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3085 if (found_key.type == type)
3086 return 0;
3087 }
3088 return 1;
3089 }
3090
Linux kernel - Deliverables
This page took 0.106525 seconds and 6 git commands to generate.