aboutsummaryrefslogtreecommitdiffstats
path: root/fs/btrfs/ctree.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs/ctree.c')
-rw-r--r--fs/btrfs/ctree.c3450
1 files changed, 3450 insertions, 0 deletions
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
new file mode 100644
index 000000000000..50aea8cb653a
--- /dev/null
+++ b/fs/btrfs/ctree.c
@@ -0,0 +1,3450 @@
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
26static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
27 *root, struct btrfs_path *path, int level);
28static 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);
31static 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);
34static 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);
38static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
39 struct btrfs_path *path, int level, int slot);
40
41inline void btrfs_init_path(struct btrfs_path *p)
42{
43 memset(p, 0, sizeof(*p));
44}
45
46struct 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
57void btrfs_free_path(struct btrfs_path *p)
58{
59 btrfs_release_path(NULL, p);
60 kmem_cache_free(btrfs_path_cachep, p);
61}
62
63void noinline btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
64{
65 int i;
66
67 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
68 p->slots[i] = 0;
69 if (!p->nodes[i])
70 continue;
71 if (p->locks[i]) {
72 btrfs_tree_unlock(p->nodes[i]);
73 p->locks[i] = 0;
74 }
75 free_extent_buffer(p->nodes[i]);
76 p->nodes[i] = NULL;
77 }
78}
79
80struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
81{
82 struct extent_buffer *eb;
83 spin_lock(&root->node_lock);
84 eb = root->node;
85 extent_buffer_get(eb);
86 spin_unlock(&root->node_lock);
87 return eb;
88}
89
90struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
91{
92 struct extent_buffer *eb;
93
94 while(1) {
95 eb = btrfs_root_node(root);
96 btrfs_tree_lock(eb);
97
98 spin_lock(&root->node_lock);
99 if (eb == root->node) {
100 spin_unlock(&root->node_lock);
101 break;
102 }
103 spin_unlock(&root->node_lock);
104
105 btrfs_tree_unlock(eb);
106 free_extent_buffer(eb);
107 }
108 return eb;
109}
110
111static void add_root_to_dirty_list(struct btrfs_root *root)
112{
113 if (root->track_dirty && list_empty(&root->dirty_list)) {
114 list_add(&root->dirty_list,
115 &root->fs_info->dirty_cowonly_roots);
116 }
117}
118
119int btrfs_copy_root(struct btrfs_trans_handle *trans,
120 struct btrfs_root *root,
121 struct extent_buffer *buf,
122 struct extent_buffer **cow_ret, u64 new_root_objectid)
123{
124 struct extent_buffer *cow;
125 u32 nritems;
126 int ret = 0;
127 int level;
128 struct btrfs_root *new_root;
129
130 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
131 if (!new_root)
132 return -ENOMEM;
133
134 memcpy(new_root, root, sizeof(*new_root));
135 new_root->root_key.objectid = new_root_objectid;
136
137 WARN_ON(root->ref_cows && trans->transid !=
138 root->fs_info->running_transaction->transid);
139 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
140
141 level = btrfs_header_level(buf);
142 nritems = btrfs_header_nritems(buf);
143
144 cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
145 new_root_objectid, trans->transid,
146 level, buf->start, 0);
147 if (IS_ERR(cow)) {
148 kfree(new_root);
149 return PTR_ERR(cow);
150 }
151
152 copy_extent_buffer(cow, buf, 0, 0, cow->len);
153 btrfs_set_header_bytenr(cow, cow->start);
154 btrfs_set_header_generation(cow, trans->transid);
155 btrfs_set_header_owner(cow, new_root_objectid);
156 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
157
158 WARN_ON(btrfs_header_generation(buf) > trans->transid);
159 ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
160 kfree(new_root);
161
162 if (ret)
163 return ret;
164
165 btrfs_mark_buffer_dirty(cow);
166 *cow_ret = cow;
167 return 0;
168}
169
170int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
171 struct btrfs_root *root,
172 struct extent_buffer *buf,
173 struct extent_buffer *parent, int parent_slot,
174 struct extent_buffer **cow_ret,
175 u64 search_start, u64 empty_size,
176 u64 prealloc_dest)
177{
178 u64 parent_start;
179 struct extent_buffer *cow;
180 u32 nritems;
181 int ret = 0;
182 int different_trans = 0;
183 int level;
184 int unlock_orig = 0;
185
186 if (*cow_ret == buf)
187 unlock_orig = 1;
188
189 WARN_ON(!btrfs_tree_locked(buf));
190
191 if (parent)
192 parent_start = parent->start;
193 else
194 parent_start = 0;
195
196 WARN_ON(root->ref_cows && trans->transid !=
197 root->fs_info->running_transaction->transid);
198 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
199
200 level = btrfs_header_level(buf);
201 nritems = btrfs_header_nritems(buf);
202
203 if (prealloc_dest) {
204 struct btrfs_key ins;
205
206 ins.objectid = prealloc_dest;
207 ins.offset = buf->len;
208 ins.type = BTRFS_EXTENT_ITEM_KEY;
209
210 ret = btrfs_alloc_reserved_extent(trans, root, parent_start,
211 root->root_key.objectid,
212 trans->transid, level, 0,
213 &ins);
214 BUG_ON(ret);
215 cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
216 buf->len);
217 } else {
218 cow = btrfs_alloc_free_block(trans, root, buf->len,
219 parent_start,
220 root->root_key.objectid,
221 trans->transid, level,
222 search_start, empty_size);
223 }
224 if (IS_ERR(cow))
225 return PTR_ERR(cow);
226
227 copy_extent_buffer(cow, buf, 0, 0, cow->len);
228 btrfs_set_header_bytenr(cow, cow->start);
229 btrfs_set_header_generation(cow, trans->transid);
230 btrfs_set_header_owner(cow, root->root_key.objectid);
231 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
232
233 WARN_ON(btrfs_header_generation(buf) > trans->transid);
234 if (btrfs_header_generation(buf) != trans->transid) {
235 u32 nr_extents;
236 different_trans = 1;
237 ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
238 if (ret)
239 return ret;
240
241 ret = btrfs_cache_ref(trans, root, buf, nr_extents);
242 WARN_ON(ret);
243 } else {
244 ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
245 if (ret)
246 return ret;
247 clean_tree_block(trans, root, buf);
248 }
249
250 if (buf == root->node) {
251 WARN_ON(parent && parent != buf);
252
253 spin_lock(&root->node_lock);
254 root->node = cow;
255 extent_buffer_get(cow);
256 spin_unlock(&root->node_lock);
257
258 if (buf != root->commit_root) {
259 btrfs_free_extent(trans, root, buf->start,
260 buf->len, buf->start,
261 root->root_key.objectid,
262 btrfs_header_generation(buf),
263 0, 0, 1);
264 }
265 free_extent_buffer(buf);
266 add_root_to_dirty_list(root);
267 } else {
268 btrfs_set_node_blockptr(parent, parent_slot,
269 cow->start);
270 WARN_ON(trans->transid == 0);
271 btrfs_set_node_ptr_generation(parent, parent_slot,
272 trans->transid);
273 btrfs_mark_buffer_dirty(parent);
274 WARN_ON(btrfs_header_generation(parent) != trans->transid);
275 btrfs_free_extent(trans, root, buf->start, buf->len,
276 parent_start, btrfs_header_owner(parent),
277 btrfs_header_generation(parent), 0, 0, 1);
278 }
279 if (unlock_orig)
280 btrfs_tree_unlock(buf);
281 free_extent_buffer(buf);
282 btrfs_mark_buffer_dirty(cow);
283 *cow_ret = cow;
284 return 0;
285}
286
287int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
288 struct btrfs_root *root, struct extent_buffer *buf,
289 struct extent_buffer *parent, int parent_slot,
290 struct extent_buffer **cow_ret, u64 prealloc_dest)
291{
292 u64 search_start;
293 u64 header_trans;
294 int ret;
295
296 if (trans->transaction != root->fs_info->running_transaction) {
297 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
298 root->fs_info->running_transaction->transid);
299 WARN_ON(1);
300 }
301 if (trans->transid != root->fs_info->generation) {
302 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
303 root->fs_info->generation);
304 WARN_ON(1);
305 }
306
307 header_trans = btrfs_header_generation(buf);
308 spin_lock(&root->fs_info->hash_lock);
309 if (header_trans == trans->transid &&
310 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
311 *cow_ret = buf;
312 spin_unlock(&root->fs_info->hash_lock);
313 WARN_ON(prealloc_dest);
314 return 0;
315 }
316 spin_unlock(&root->fs_info->hash_lock);
317 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
318 ret = __btrfs_cow_block(trans, root, buf, parent,
319 parent_slot, cow_ret, search_start, 0,
320 prealloc_dest);
321 return ret;
322}
323
324static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
325{
326 if (blocknr < other && other - (blocknr + blocksize) < 32768)
327 return 1;
328 if (blocknr > other && blocknr - (other + blocksize) < 32768)
329 return 1;
330 return 0;
331}
332
333/*
334 * compare two keys in a memcmp fashion
335 */
336static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
337{
338 struct btrfs_key k1;
339
340 btrfs_disk_key_to_cpu(&k1, disk);
341
342 if (k1.objectid > k2->objectid)
343 return 1;
344 if (k1.objectid < k2->objectid)
345 return -1;
346 if (k1.type > k2->type)
347 return 1;
348 if (k1.type < k2->type)
349 return -1;
350 if (k1.offset > k2->offset)
351 return 1;
352 if (k1.offset < k2->offset)
353 return -1;
354 return 0;
355}
356
357
358int btrfs_realloc_node(struct btrfs_trans_handle *trans,
359 struct btrfs_root *root, struct extent_buffer *parent,
360 int start_slot, int cache_only, u64 *last_ret,
361 struct btrfs_key *progress)
362{
363 struct extent_buffer *cur;
364 u64 blocknr;
365 u64 gen;
366 u64 search_start = *last_ret;
367 u64 last_block = 0;
368 u64 other;
369 u32 parent_nritems;
370 int end_slot;
371 int i;
372 int err = 0;
373 int parent_level;
374 int uptodate;
375 u32 blocksize;
376 int progress_passed = 0;
377 struct btrfs_disk_key disk_key;
378
379 parent_level = btrfs_header_level(parent);
380 if (cache_only && parent_level != 1)
381 return 0;
382
383 if (trans->transaction != root->fs_info->running_transaction) {
384 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
385 root->fs_info->running_transaction->transid);
386 WARN_ON(1);
387 }
388 if (trans->transid != root->fs_info->generation) {
389 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
390 root->fs_info->generation);
391 WARN_ON(1);
392 }
393
394 parent_nritems = btrfs_header_nritems(parent);
395 blocksize = btrfs_level_size(root, parent_level - 1);
396 end_slot = parent_nritems;
397
398 if (parent_nritems == 1)
399 return 0;
400
401 for (i = start_slot; i < end_slot; i++) {
402 int close = 1;
403
404 if (!parent->map_token) {
405 map_extent_buffer(parent,
406 btrfs_node_key_ptr_offset(i),
407 sizeof(struct btrfs_key_ptr),
408 &parent->map_token, &parent->kaddr,
409 &parent->map_start, &parent->map_len,
410 KM_USER1);
411 }
412 btrfs_node_key(parent, &disk_key, i);
413 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
414 continue;
415
416 progress_passed = 1;
417 blocknr = btrfs_node_blockptr(parent, i);
418 gen = btrfs_node_ptr_generation(parent, i);
419 if (last_block == 0)
420 last_block = blocknr;
421
422 if (i > 0) {
423 other = btrfs_node_blockptr(parent, i - 1);
424 close = close_blocks(blocknr, other, blocksize);
425 }
426 if (!close && i < end_slot - 2) {
427 other = btrfs_node_blockptr(parent, i + 1);
428 close = close_blocks(blocknr, other, blocksize);
429 }
430 if (close) {
431 last_block = blocknr;
432 continue;
433 }
434 if (parent->map_token) {
435 unmap_extent_buffer(parent, parent->map_token,
436 KM_USER1);
437 parent->map_token = NULL;
438 }
439
440 cur = btrfs_find_tree_block(root, blocknr, blocksize);
441 if (cur)
442 uptodate = btrfs_buffer_uptodate(cur, gen);
443 else
444 uptodate = 0;
445 if (!cur || !uptodate) {
446 if (cache_only) {
447 free_extent_buffer(cur);
448 continue;
449 }
450 if (!cur) {
451 cur = read_tree_block(root, blocknr,
452 blocksize, gen);
453 } else if (!uptodate) {
454 btrfs_read_buffer(cur, gen);
455 }
456 }
457 if (search_start == 0)
458 search_start = last_block;
459
460 btrfs_tree_lock(cur);
461 err = __btrfs_cow_block(trans, root, cur, parent, i,
462 &cur, search_start,
463 min(16 * blocksize,
464 (end_slot - i) * blocksize), 0);
465 if (err) {
466 btrfs_tree_unlock(cur);
467 free_extent_buffer(cur);
468 break;
469 }
470 search_start = cur->start;
471 last_block = cur->start;
472 *last_ret = search_start;
473 btrfs_tree_unlock(cur);
474 free_extent_buffer(cur);
475 }
476 if (parent->map_token) {
477 unmap_extent_buffer(parent, parent->map_token,
478 KM_USER1);
479 parent->map_token = NULL;
480 }
481 return err;
482}
483
484/*
485 * The leaf data grows from end-to-front in the node.
486 * this returns the address of the start of the last item,
487 * which is the stop of the leaf data stack
488 */
489static inline unsigned int leaf_data_end(struct btrfs_root *root,
490 struct extent_buffer *leaf)
491{
492 u32 nr = btrfs_header_nritems(leaf);
493 if (nr == 0)
494 return BTRFS_LEAF_DATA_SIZE(root);
495 return btrfs_item_offset_nr(leaf, nr - 1);
496}
497
498static int check_node(struct btrfs_root *root, struct btrfs_path *path,
499 int level)
500{
501 struct extent_buffer *parent = NULL;
502 struct extent_buffer *node = path->nodes[level];
503 struct btrfs_disk_key parent_key;
504 struct btrfs_disk_key node_key;
505 int parent_slot;
506 int slot;
507 struct btrfs_key cpukey;
508 u32 nritems = btrfs_header_nritems(node);
509
510 if (path->nodes[level + 1])
511 parent = path->nodes[level + 1];
512
513 slot = path->slots[level];
514 BUG_ON(nritems == 0);
515 if (parent) {
516 parent_slot = path->slots[level + 1];
517 btrfs_node_key(parent, &parent_key, parent_slot);
518 btrfs_node_key(node, &node_key, 0);
519 BUG_ON(memcmp(&parent_key, &node_key,
520 sizeof(struct btrfs_disk_key)));
521 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
522 btrfs_header_bytenr(node));
523 }
524 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
525 if (slot != 0) {
526 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
527 btrfs_node_key(node, &node_key, slot);
528 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
529 }
530 if (slot < nritems - 1) {
531 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
532 btrfs_node_key(node, &node_key, slot);
533 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
534 }
535 return 0;
536}
537
538static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
539 int level)
540{
541 struct extent_buffer *leaf = path->nodes[level];
542 struct extent_buffer *parent = NULL;
543 int parent_slot;
544 struct btrfs_key cpukey;
545 struct btrfs_disk_key parent_key;
546 struct btrfs_disk_key leaf_key;
547 int slot = path->slots[0];
548
549 u32 nritems = btrfs_header_nritems(leaf);
550
551 if (path->nodes[level + 1])
552 parent = path->nodes[level + 1];
553
554 if (nritems == 0)
555 return 0;
556
557 if (parent) {
558 parent_slot = path->slots[level + 1];
559 btrfs_node_key(parent, &parent_key, parent_slot);
560 btrfs_item_key(leaf, &leaf_key, 0);
561
562 BUG_ON(memcmp(&parent_key, &leaf_key,
563 sizeof(struct btrfs_disk_key)));
564 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
565 btrfs_header_bytenr(leaf));
566 }
567#if 0
568 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
569 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
570 btrfs_item_key(leaf, &leaf_key, i);
571 if (comp_keys(&leaf_key, &cpukey) >= 0) {
572 btrfs_print_leaf(root, leaf);
573 printk("slot %d offset bad key\n", i);
574 BUG_ON(1);
575 }
576 if (btrfs_item_offset_nr(leaf, i) !=
577 btrfs_item_end_nr(leaf, i + 1)) {
578 btrfs_print_leaf(root, leaf);
579 printk("slot %d offset bad\n", i);
580 BUG_ON(1);
581 }
582 if (i == 0) {
583 if (btrfs_item_offset_nr(leaf, i) +
584 btrfs_item_size_nr(leaf, i) !=
585 BTRFS_LEAF_DATA_SIZE(root)) {
586 btrfs_print_leaf(root, leaf);
587 printk("slot %d first offset bad\n", i);
588 BUG_ON(1);
589 }
590 }
591 }
592 if (nritems > 0) {
593 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
594 btrfs_print_leaf(root, leaf);
595 printk("slot %d bad size \n", nritems - 1);
596 BUG_ON(1);
597 }
598 }
599#endif
600 if (slot != 0 && slot < nritems - 1) {
601 btrfs_item_key(leaf, &leaf_key, slot);
602 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
603 if (comp_keys(&leaf_key, &cpukey) <= 0) {
604 btrfs_print_leaf(root, leaf);
605 printk("slot %d offset bad key\n", slot);
606 BUG_ON(1);
607 }
608 if (btrfs_item_offset_nr(leaf, slot - 1) !=
609 btrfs_item_end_nr(leaf, slot)) {
610 btrfs_print_leaf(root, leaf);
611 printk("slot %d offset bad\n", slot);
612 BUG_ON(1);
613 }
614 }
615 if (slot < nritems - 1) {
616 btrfs_item_key(leaf, &leaf_key, slot);
617 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
618 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
619 if (btrfs_item_offset_nr(leaf, slot) !=
620 btrfs_item_end_nr(leaf, slot + 1)) {
621 btrfs_print_leaf(root, leaf);
622 printk("slot %d offset bad\n", slot);
623 BUG_ON(1);
624 }
625 }
626 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
627 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
628 return 0;
629}
630
631static int noinline check_block(struct btrfs_root *root,
632 struct btrfs_path *path, int level)
633{
634 u64 found_start;
635 return 0;
636 if (btrfs_header_level(path->nodes[level]) != level)
637 printk("warning: bad level %Lu wanted %d found %d\n",
638 path->nodes[level]->start, level,
639 btrfs_header_level(path->nodes[level]));
640 found_start = btrfs_header_bytenr(path->nodes[level]);
641 if (found_start != path->nodes[level]->start) {
642 printk("warning: bad bytentr %Lu found %Lu\n",
643 path->nodes[level]->start, found_start);
644 }
645#if 0
646 struct extent_buffer *buf = path->nodes[level];
647
648 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
649 (unsigned long)btrfs_header_fsid(buf),
650 BTRFS_FSID_SIZE)) {
651 printk("warning bad block %Lu\n", buf->start);
652 return 1;
653 }
654#endif
655 if (level == 0)
656 return check_leaf(root, path, level);
657 return check_node(root, path, level);
658}
659
660/*
661 * search for key in the extent_buffer. The items start at offset p,
662 * and they are item_size apart. There are 'max' items in p.
663 *
664 * the slot in the array is returned via slot, and it points to
665 * the place where you would insert key if it is not found in
666 * the array.
667 *
668 * slot may point to max if the key is bigger than all of the keys
669 */
670static noinline int generic_bin_search(struct extent_buffer *eb,
671 unsigned long p,
672 int item_size, struct btrfs_key *key,
673 int max, int *slot)
674{
675 int low = 0;
676 int high = max;
677 int mid;
678 int ret;
679 struct btrfs_disk_key *tmp = NULL;
680 struct btrfs_disk_key unaligned;
681 unsigned long offset;
682 char *map_token = NULL;
683 char *kaddr = NULL;
684 unsigned long map_start = 0;
685 unsigned long map_len = 0;
686 int err;
687
688 while(low < high) {
689 mid = (low + high) / 2;
690 offset = p + mid * item_size;
691
692 if (!map_token || offset < map_start ||
693 (offset + sizeof(struct btrfs_disk_key)) >
694 map_start + map_len) {
695 if (map_token) {
696 unmap_extent_buffer(eb, map_token, KM_USER0);
697 map_token = NULL;
698 }
699 err = map_extent_buffer(eb, offset,
700 sizeof(struct btrfs_disk_key),
701 &map_token, &kaddr,
702 &map_start, &map_len, KM_USER0);
703
704 if (!err) {
705 tmp = (struct btrfs_disk_key *)(kaddr + offset -
706 map_start);
707 } else {
708 read_extent_buffer(eb, &unaligned,
709 offset, sizeof(unaligned));
710 tmp = &unaligned;
711 }
712
713 } else {
714 tmp = (struct btrfs_disk_key *)(kaddr + offset -
715 map_start);
716 }
717 ret = comp_keys(tmp, key);
718
719 if (ret < 0)
720 low = mid + 1;
721 else if (ret > 0)
722 high = mid;
723 else {
724 *slot = mid;
725 if (map_token)
726 unmap_extent_buffer(eb, map_token, KM_USER0);
727 return 0;
728 }
729 }
730 *slot = low;
731 if (map_token)
732 unmap_extent_buffer(eb, map_token, KM_USER0);
733 return 1;
734}
735
736/*
737 * simple bin_search frontend that does the right thing for
738 * leaves vs nodes
739 */
740static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
741 int level, int *slot)
742{
743 if (level == 0) {
744 return generic_bin_search(eb,
745 offsetof(struct btrfs_leaf, items),
746 sizeof(struct btrfs_item),
747 key, btrfs_header_nritems(eb),
748 slot);
749 } else {
750 return generic_bin_search(eb,
751 offsetof(struct btrfs_node, ptrs),
752 sizeof(struct btrfs_key_ptr),
753 key, btrfs_header_nritems(eb),
754 slot);
755 }
756 return -1;
757}
758
759static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
760 struct extent_buffer *parent, int slot)
761{
762 int level = btrfs_header_level(parent);
763 if (slot < 0)
764 return NULL;
765 if (slot >= btrfs_header_nritems(parent))
766 return NULL;
767
768 BUG_ON(level == 0);
769
770 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
771 btrfs_level_size(root, level - 1),
772 btrfs_node_ptr_generation(parent, slot));
773}
774
775static noinline int balance_level(struct btrfs_trans_handle *trans,
776 struct btrfs_root *root,
777 struct btrfs_path *path, int level)
778{
779 struct extent_buffer *right = NULL;
780 struct extent_buffer *mid;
781 struct extent_buffer *left = NULL;
782 struct extent_buffer *parent = NULL;
783 int ret = 0;
784 int wret;
785 int pslot;
786 int orig_slot = path->slots[level];
787 int err_on_enospc = 0;
788 u64 orig_ptr;
789
790 if (level == 0)
791 return 0;
792
793 mid = path->nodes[level];
794 WARN_ON(!path->locks[level]);
795 WARN_ON(btrfs_header_generation(mid) != trans->transid);
796
797 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
798
799 if (level < BTRFS_MAX_LEVEL - 1)
800 parent = path->nodes[level + 1];
801 pslot = path->slots[level + 1];
802
803 /*
804 * deal with the case where there is only one pointer in the root
805 * by promoting the node below to a root
806 */
807 if (!parent) {
808 struct extent_buffer *child;
809
810 if (btrfs_header_nritems(mid) != 1)
811 return 0;
812
813 /* promote the child to a root */
814 child = read_node_slot(root, mid, 0);
815 btrfs_tree_lock(child);
816 BUG_ON(!child);
817 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
818 BUG_ON(ret);
819
820 spin_lock(&root->node_lock);
821 root->node = child;
822 spin_unlock(&root->node_lock);
823
824 ret = btrfs_update_extent_ref(trans, root, child->start,
825 mid->start, child->start,
826 root->root_key.objectid,
827 trans->transid, level - 1, 0);
828 BUG_ON(ret);
829
830 add_root_to_dirty_list(root);
831 btrfs_tree_unlock(child);
832 path->locks[level] = 0;
833 path->nodes[level] = NULL;
834 clean_tree_block(trans, root, mid);
835 btrfs_tree_unlock(mid);
836 /* once for the path */
837 free_extent_buffer(mid);
838 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
839 mid->start, root->root_key.objectid,
840 btrfs_header_generation(mid), 0, 0, 1);
841 /* once for the root ptr */
842 free_extent_buffer(mid);
843 return ret;
844 }
845 if (btrfs_header_nritems(mid) >
846 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
847 return 0;
848
849 if (btrfs_header_nritems(mid) < 2)
850 err_on_enospc = 1;
851
852 left = read_node_slot(root, parent, pslot - 1);
853 if (left) {
854 btrfs_tree_lock(left);
855 wret = btrfs_cow_block(trans, root, left,
856 parent, pslot - 1, &left, 0);
857 if (wret) {
858 ret = wret;
859 goto enospc;
860 }
861 }
862 right = read_node_slot(root, parent, pslot + 1);
863 if (right) {
864 btrfs_tree_lock(right);
865 wret = btrfs_cow_block(trans, root, right,
866 parent, pslot + 1, &right, 0);
867 if (wret) {
868 ret = wret;
869 goto enospc;
870 }
871 }
872
873 /* first, try to make some room in the middle buffer */
874 if (left) {
875 orig_slot += btrfs_header_nritems(left);
876 wret = push_node_left(trans, root, left, mid, 1);
877 if (wret < 0)
878 ret = wret;
879 if (btrfs_header_nritems(mid) < 2)
880 err_on_enospc = 1;
881 }
882
883 /*
884 * then try to empty the right most buffer into the middle
885 */
886 if (right) {
887 wret = push_node_left(trans, root, mid, right, 1);
888 if (wret < 0 && wret != -ENOSPC)
889 ret = wret;
890 if (btrfs_header_nritems(right) == 0) {
891 u64 bytenr = right->start;
892 u64 generation = btrfs_header_generation(parent);
893 u32 blocksize = right->len;
894
895 clean_tree_block(trans, root, right);
896 btrfs_tree_unlock(right);
897 free_extent_buffer(right);
898 right = NULL;
899 wret = del_ptr(trans, root, path, level + 1, pslot +
900 1);
901 if (wret)
902 ret = wret;
903 wret = btrfs_free_extent(trans, root, bytenr,
904 blocksize, parent->start,
905 btrfs_header_owner(parent),
906 generation, 0, 0, 1);
907 if (wret)
908 ret = wret;
909 } else {
910 struct btrfs_disk_key right_key;
911 btrfs_node_key(right, &right_key, 0);
912 btrfs_set_node_key(parent, &right_key, pslot + 1);
913 btrfs_mark_buffer_dirty(parent);
914 }
915 }
916 if (btrfs_header_nritems(mid) == 1) {
917 /*
918 * we're not allowed to leave a node with one item in the
919 * tree during a delete. A deletion from lower in the tree
920 * could try to delete the only pointer in this node.
921 * So, pull some keys from the left.
922 * There has to be a left pointer at this point because
923 * otherwise we would have pulled some pointers from the
924 * right
925 */
926 BUG_ON(!left);
927 wret = balance_node_right(trans, root, mid, left);
928 if (wret < 0) {
929 ret = wret;
930 goto enospc;
931 }
932 if (wret == 1) {
933 wret = push_node_left(trans, root, left, mid, 1);
934 if (wret < 0)
935 ret = wret;
936 }
937 BUG_ON(wret == 1);
938 }
939 if (btrfs_header_nritems(mid) == 0) {
940 /* we've managed to empty the middle node, drop it */
941 u64 root_gen = btrfs_header_generation(parent);
942 u64 bytenr = mid->start;
943 u32 blocksize = mid->len;
944
945 clean_tree_block(trans, root, mid);
946 btrfs_tree_unlock(mid);
947 free_extent_buffer(mid);
948 mid = NULL;
949 wret = del_ptr(trans, root, path, level + 1, pslot);
950 if (wret)
951 ret = wret;
952 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
953 parent->start,
954 btrfs_header_owner(parent),
955 root_gen, 0, 0, 1);
956 if (wret)
957 ret = wret;
958 } else {
959 /* update the parent key to reflect our changes */
960 struct btrfs_disk_key mid_key;
961 btrfs_node_key(mid, &mid_key, 0);
962 btrfs_set_node_key(parent, &mid_key, pslot);
963 btrfs_mark_buffer_dirty(parent);
964 }
965
966 /* update the path */
967 if (left) {
968 if (btrfs_header_nritems(left) > orig_slot) {
969 extent_buffer_get(left);
970 /* left was locked after cow */
971 path->nodes[level] = left;
972 path->slots[level + 1] -= 1;
973 path->slots[level] = orig_slot;
974 if (mid) {
975 btrfs_tree_unlock(mid);
976 free_extent_buffer(mid);
977 }
978 } else {
979 orig_slot -= btrfs_header_nritems(left);
980 path->slots[level] = orig_slot;
981 }
982 }
983 /* double check we haven't messed things up */
984 check_block(root, path, level);
985 if (orig_ptr !=
986 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
987 BUG();
988enospc:
989 if (right) {
990 btrfs_tree_unlock(right);
991 free_extent_buffer(right);
992 }
993 if (left) {
994 if (path->nodes[level] != left)
995 btrfs_tree_unlock(left);
996 free_extent_buffer(left);
997 }
998 return ret;
999}
1000
1001/* returns zero if the push worked, non-zero otherwise */
1002static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
1003 struct btrfs_root *root,
1004 struct btrfs_path *path, int level)
1005{
1006 struct extent_buffer *right = NULL;
1007 struct extent_buffer *mid;
1008 struct extent_buffer *left = NULL;
1009 struct extent_buffer *parent = NULL;
1010 int ret = 0;
1011 int wret;
1012 int pslot;
1013 int orig_slot = path->slots[level];
1014 u64 orig_ptr;
1015
1016 if (level == 0)
1017 return 1;
1018
1019 mid = path->nodes[level];
1020 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1021 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1022
1023 if (level < BTRFS_MAX_LEVEL - 1)
1024 parent = path->nodes[level + 1];
1025 pslot = path->slots[level + 1];
1026
1027 if (!parent)
1028 return 1;
1029
1030 left = read_node_slot(root, parent, pslot - 1);
1031
1032 /* first, try to make some room in the middle buffer */
1033 if (left) {
1034 u32 left_nr;
1035
1036 btrfs_tree_lock(left);
1037 left_nr = btrfs_header_nritems(left);
1038 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1039 wret = 1;
1040 } else {
1041 ret = btrfs_cow_block(trans, root, left, parent,
1042 pslot - 1, &left, 0);
1043 if (ret)
1044 wret = 1;
1045 else {
1046 wret = push_node_left(trans, root,
1047 left, mid, 0);
1048 }
1049 }
1050 if (wret < 0)
1051 ret = wret;
1052 if (wret == 0) {
1053 struct btrfs_disk_key disk_key;
1054 orig_slot += left_nr;
1055 btrfs_node_key(mid, &disk_key, 0);
1056 btrfs_set_node_key(parent, &disk_key, pslot);
1057 btrfs_mark_buffer_dirty(parent);
1058 if (btrfs_header_nritems(left) > orig_slot) {
1059 path->nodes[level] = left;
1060 path->slots[level + 1] -= 1;
1061 path->slots[level] = orig_slot;
1062 btrfs_tree_unlock(mid);
1063 free_extent_buffer(mid);
1064 } else {
1065 orig_slot -=
1066 btrfs_header_nritems(left);
1067 path->slots[level] = orig_slot;
1068 btrfs_tree_unlock(left);
1069 free_extent_buffer(left);
1070 }
1071 return 0;
1072 }
1073 btrfs_tree_unlock(left);
1074 free_extent_buffer(left);
1075 }
1076 right = read_node_slot(root, parent, pslot + 1);
1077
1078 /*
1079 * then try to empty the right most buffer into the middle
1080 */
1081 if (right) {
1082 u32 right_nr;
1083 btrfs_tree_lock(right);
1084 right_nr = btrfs_header_nritems(right);
1085 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1086 wret = 1;
1087 } else {
1088 ret = btrfs_cow_block(trans, root, right,
1089 parent, pslot + 1,
1090 &right, 0);
1091 if (ret)
1092 wret = 1;
1093 else {
1094 wret = balance_node_right(trans, root,
1095 right, mid);
1096 }
1097 }
1098 if (wret < 0)
1099 ret = wret;
1100 if (wret == 0) {
1101 struct btrfs_disk_key disk_key;
1102
1103 btrfs_node_key(right, &disk_key, 0);
1104 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1105 btrfs_mark_buffer_dirty(parent);
1106
1107 if (btrfs_header_nritems(mid) <= orig_slot) {
1108 path->nodes[level] = right;
1109 path->slots[level + 1] += 1;
1110 path->slots[level] = orig_slot -
1111 btrfs_header_nritems(mid);
1112 btrfs_tree_unlock(mid);
1113 free_extent_buffer(mid);
1114 } else {
1115 btrfs_tree_unlock(right);
1116 free_extent_buffer(right);
1117 }
1118 return 0;
1119 }
1120 btrfs_tree_unlock(right);
1121 free_extent_buffer(right);
1122 }
1123 return 1;
1124}
1125
1126/*
1127 * readahead one full node of leaves
1128 */
1129static noinline void reada_for_search(struct btrfs_root *root,
1130 struct btrfs_path *path,
1131 int level, int slot, u64 objectid)
1132{
1133 struct extent_buffer *node;
1134 struct btrfs_disk_key disk_key;
1135 u32 nritems;
1136 u64 search;
1137 u64 lowest_read;
1138 u64 highest_read;
1139 u64 nread = 0;
1140 int direction = path->reada;
1141 struct extent_buffer *eb;
1142 u32 nr;
1143 u32 blocksize;
1144 u32 nscan = 0;
1145
1146 if (level != 1)
1147 return;
1148
1149 if (!path->nodes[level])
1150 return;
1151
1152 node = path->nodes[level];
1153
1154 search = btrfs_node_blockptr(node, slot);
1155 blocksize = btrfs_level_size(root, level - 1);
1156 eb = btrfs_find_tree_block(root, search, blocksize);
1157 if (eb) {
1158 free_extent_buffer(eb);
1159 return;
1160 }
1161
1162 highest_read = search;
1163 lowest_read = search;
1164
1165 nritems = btrfs_header_nritems(node);
1166 nr = slot;
1167 while(1) {
1168 if (direction < 0) {
1169 if (nr == 0)
1170 break;
1171 nr--;
1172 } else if (direction > 0) {
1173 nr++;
1174 if (nr >= nritems)
1175 break;
1176 }
1177 if (path->reada < 0 && objectid) {
1178 btrfs_node_key(node, &disk_key, nr);
1179 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1180 break;
1181 }
1182 search = btrfs_node_blockptr(node, nr);
1183 if ((search >= lowest_read && search <= highest_read) ||
1184 (search < lowest_read && lowest_read - search <= 32768) ||
1185 (search > highest_read && search - highest_read <= 32768)) {
1186 readahead_tree_block(root, search, blocksize,
1187 btrfs_node_ptr_generation(node, nr));
1188 nread += blocksize;
1189 }
1190 nscan++;
1191 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
1192 break;
1193 if(nread > (1024 * 1024) || nscan > 128)
1194 break;
1195
1196 if (search < lowest_read)
1197 lowest_read = search;
1198 if (search > highest_read)
1199 highest_read = search;
1200 }
1201}
1202
1203static noinline void unlock_up(struct btrfs_path *path, int level,
1204 int lowest_unlock)
1205{
1206 int i;
1207 int skip_level = level;
1208 int no_skips = 0;
1209 struct extent_buffer *t;
1210
1211 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1212 if (!path->nodes[i])
1213 break;
1214 if (!path->locks[i])
1215 break;
1216 if (!no_skips && path->slots[i] == 0) {
1217 skip_level = i + 1;
1218 continue;
1219 }
1220 if (!no_skips && path->keep_locks) {
1221 u32 nritems;
1222 t = path->nodes[i];
1223 nritems = btrfs_header_nritems(t);
1224 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1225 skip_level = i + 1;
1226 continue;
1227 }
1228 }
1229 if (skip_level < i && i >= lowest_unlock)
1230 no_skips = 1;
1231
1232 t = path->nodes[i];
1233 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1234 btrfs_tree_unlock(t);
1235 path->locks[i] = 0;
1236 }
1237 }
1238}
1239
1240/*
1241 * look for key in the tree. path is filled in with nodes along the way
1242 * if key is found, we return zero and you can find the item in the leaf
1243 * level of the path (level 0)
1244 *
1245 * If the key isn't found, the path points to the slot where it should
1246 * be inserted, and 1 is returned. If there are other errors during the
1247 * search a negative error number is returned.
1248 *
1249 * if ins_len > 0, nodes and leaves will be split as we walk down the
1250 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1251 * possible)
1252 */
1253int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1254 *root, struct btrfs_key *key, struct btrfs_path *p, int
1255 ins_len, int cow)
1256{
1257 struct extent_buffer *b;
1258 struct extent_buffer *tmp;
1259 int slot;
1260 int ret;
1261 int level;
1262 int should_reada = p->reada;
1263 int lowest_unlock = 1;
1264 int blocksize;
1265 u8 lowest_level = 0;
1266 u64 blocknr;
1267 u64 gen;
1268 struct btrfs_key prealloc_block;
1269
1270 lowest_level = p->lowest_level;
1271 WARN_ON(lowest_level && ins_len);
1272 WARN_ON(p->nodes[0] != NULL);
1273 WARN_ON(cow && root == root->fs_info->extent_root &&
1274 !mutex_is_locked(&root->fs_info->alloc_mutex));
1275 if (ins_len < 0)
1276 lowest_unlock = 2;
1277
1278 prealloc_block.objectid = 0;
1279
1280again:
1281 if (p->skip_locking)
1282 b = btrfs_root_node(root);
1283 else
1284 b = btrfs_lock_root_node(root);
1285
1286 while (b) {
1287 level = btrfs_header_level(b);
1288
1289 /*
1290 * setup the path here so we can release it under lock
1291 * contention with the cow code
1292 */
1293 p->nodes[level] = b;
1294 if (!p->skip_locking)
1295 p->locks[level] = 1;
1296
1297 if (cow) {
1298 int wret;
1299
1300 /* is a cow on this block not required */
1301 spin_lock(&root->fs_info->hash_lock);
1302 if (btrfs_header_generation(b) == trans->transid &&
1303 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1304 spin_unlock(&root->fs_info->hash_lock);
1305 goto cow_done;
1306 }
1307 spin_unlock(&root->fs_info->hash_lock);
1308
1309 /* ok, we have to cow, is our old prealloc the right
1310 * size?
1311 */
1312 if (prealloc_block.objectid &&
1313 prealloc_block.offset != b->len) {
1314 btrfs_free_reserved_extent(root,
1315 prealloc_block.objectid,
1316 prealloc_block.offset);
1317 prealloc_block.objectid = 0;
1318 }
1319
1320 /*
1321 * for higher level blocks, try not to allocate blocks
1322 * with the block and the parent locks held.
1323 */
1324 if (level > 1 && !prealloc_block.objectid &&
1325 btrfs_path_lock_waiting(p, level)) {
1326 u32 size = b->len;
1327 u64 hint = b->start;
1328
1329 btrfs_release_path(root, p);
1330 ret = btrfs_reserve_extent(trans, root,
1331 size, size, 0,
1332 hint, (u64)-1,
1333 &prealloc_block, 0);
1334 BUG_ON(ret);
1335 goto again;
1336 }
1337
1338 wret = btrfs_cow_block(trans, root, b,
1339 p->nodes[level + 1],
1340 p->slots[level + 1],
1341 &b, prealloc_block.objectid);
1342 prealloc_block.objectid = 0;
1343 if (wret) {
1344 free_extent_buffer(b);
1345 ret = wret;
1346 goto done;
1347 }
1348 }
1349cow_done:
1350 BUG_ON(!cow && ins_len);
1351 if (level != btrfs_header_level(b))
1352 WARN_ON(1);
1353 level = btrfs_header_level(b);
1354
1355 p->nodes[level] = b;
1356 if (!p->skip_locking)
1357 p->locks[level] = 1;
1358
1359 ret = check_block(root, p, level);
1360 if (ret) {
1361 ret = -1;
1362 goto done;
1363 }
1364
1365 ret = bin_search(b, key, level, &slot);
1366 if (level != 0) {
1367 if (ret && slot > 0)
1368 slot -= 1;
1369 p->slots[level] = slot;
1370 if (ins_len > 0 && btrfs_header_nritems(b) >=
1371 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1372 int sret = split_node(trans, root, p, level);
1373 BUG_ON(sret > 0);
1374 if (sret) {
1375 ret = sret;
1376 goto done;
1377 }
1378 b = p->nodes[level];
1379 slot = p->slots[level];
1380 } else if (ins_len < 0) {
1381 int sret = balance_level(trans, root, p,
1382 level);
1383 if (sret) {
1384 ret = sret;
1385 goto done;
1386 }
1387 b = p->nodes[level];
1388 if (!b) {
1389 btrfs_release_path(NULL, p);
1390 goto again;
1391 }
1392 slot = p->slots[level];
1393 BUG_ON(btrfs_header_nritems(b) == 1);
1394 }
1395 unlock_up(p, level, lowest_unlock);
1396
1397 /* this is only true while dropping a snapshot */
1398 if (level == lowest_level) {
1399 break;
1400 }
1401
1402 blocknr = btrfs_node_blockptr(b, slot);
1403 gen = btrfs_node_ptr_generation(b, slot);
1404 blocksize = btrfs_level_size(root, level - 1);
1405
1406 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1407 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1408 b = tmp;
1409 } else {
1410 /*
1411 * reduce lock contention at high levels
1412 * of the btree by dropping locks before
1413 * we read.
1414 */
1415 if (level > 1) {
1416 btrfs_release_path(NULL, p);
1417 if (tmp)
1418 free_extent_buffer(tmp);
1419 if (should_reada)
1420 reada_for_search(root, p,
1421 level, slot,
1422 key->objectid);
1423
1424 tmp = read_tree_block(root, blocknr,
1425 blocksize, gen);
1426 if (tmp)
1427 free_extent_buffer(tmp);
1428 goto again;
1429 } else {
1430 if (tmp)
1431 free_extent_buffer(tmp);
1432 if (should_reada)
1433 reada_for_search(root, p,
1434 level, slot,
1435 key->objectid);
1436 b = read_node_slot(root, b, slot);
1437 }
1438 }
1439 if (!p->skip_locking)
1440 btrfs_tree_lock(b);
1441 } else {
1442 p->slots[level] = slot;
1443 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1444 sizeof(struct btrfs_item) + ins_len) {
1445 int sret = split_leaf(trans, root, key,
1446 p, ins_len, ret == 0);
1447 BUG_ON(sret > 0);
1448 if (sret) {
1449 ret = sret;
1450 goto done;
1451 }
1452 }
1453 unlock_up(p, level, lowest_unlock);
1454 goto done;
1455 }
1456 }
1457 ret = 1;
1458done:
1459 if (prealloc_block.objectid) {
1460 btrfs_free_reserved_extent(root,
1461 prealloc_block.objectid,
1462 prealloc_block.offset);
1463 }
1464
1465 return ret;
1466}
1467
1468/*
1469 * adjust the pointers going up the tree, starting at level
1470 * making sure the right key of each node is points to 'key'.
1471 * This is used after shifting pointers to the left, so it stops
1472 * fixing up pointers when a given leaf/node is not in slot 0 of the
1473 * higher levels
1474 *
1475 * If this fails to write a tree block, it returns -1, but continues
1476 * fixing up the blocks in ram so the tree is consistent.
1477 */
1478static int fixup_low_keys(struct btrfs_trans_handle *trans,
1479 struct btrfs_root *root, struct btrfs_path *path,
1480 struct btrfs_disk_key *key, int level)
1481{
1482 int i;
1483 int ret = 0;
1484 struct extent_buffer *t;
1485
1486 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1487 int tslot = path->slots[i];
1488 if (!path->nodes[i])
1489 break;
1490 t = path->nodes[i];
1491 btrfs_set_node_key(t, key, tslot);
1492 btrfs_mark_buffer_dirty(path->nodes[i]);
1493 if (tslot != 0)
1494 break;
1495 }
1496 return ret;
1497}
1498
1499/*
1500 * update item key.
1501 *
1502 * This function isn't completely safe. It's the caller's responsibility
1503 * that the new key won't break the order
1504 */
1505int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1506 struct btrfs_root *root, struct btrfs_path *path,
1507 struct btrfs_key *new_key)
1508{
1509 struct btrfs_disk_key disk_key;
1510 struct extent_buffer *eb;
1511 int slot;
1512
1513 eb = path->nodes[0];
1514 slot = path->slots[0];
1515 if (slot > 0) {
1516 btrfs_item_key(eb, &disk_key, slot - 1);
1517 if (comp_keys(&disk_key, new_key) >= 0)
1518 return -1;
1519 }
1520 if (slot < btrfs_header_nritems(eb) - 1) {
1521 btrfs_item_key(eb, &disk_key, slot + 1);
1522 if (comp_keys(&disk_key, new_key) <= 0)
1523 return -1;
1524 }
1525
1526 btrfs_cpu_key_to_disk(&disk_key, new_key);
1527 btrfs_set_item_key(eb, &disk_key, slot);
1528 btrfs_mark_buffer_dirty(eb);
1529 if (slot == 0)
1530 fixup_low_keys(trans, root, path, &disk_key, 1);
1531 return 0;
1532}
1533
1534/*
1535 * try to push data from one node into the next node left in the
1536 * tree.
1537 *
1538 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1539 * error, and > 0 if there was no room in the left hand block.
1540 */
1541static int push_node_left(struct btrfs_trans_handle *trans,
1542 struct btrfs_root *root, struct extent_buffer *dst,
1543 struct extent_buffer *src, int empty)
1544{
1545 int push_items = 0;
1546 int src_nritems;
1547 int dst_nritems;
1548 int ret = 0;
1549
1550 src_nritems = btrfs_header_nritems(src);
1551 dst_nritems = btrfs_header_nritems(dst);
1552 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1553 WARN_ON(btrfs_header_generation(src) != trans->transid);
1554 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1555
1556 if (!empty && src_nritems <= 8)
1557 return 1;
1558
1559 if (push_items <= 0) {
1560 return 1;
1561 }
1562
1563 if (empty) {
1564 push_items = min(src_nritems, push_items);
1565 if (push_items < src_nritems) {
1566 /* leave at least 8 pointers in the node if
1567 * we aren't going to empty it
1568 */
1569 if (src_nritems - push_items < 8) {
1570 if (push_items <= 8)
1571 return 1;
1572 push_items -= 8;
1573 }
1574 }
1575 } else
1576 push_items = min(src_nritems - 8, push_items);
1577
1578 copy_extent_buffer(dst, src,
1579 btrfs_node_key_ptr_offset(dst_nritems),
1580 btrfs_node_key_ptr_offset(0),
1581 push_items * sizeof(struct btrfs_key_ptr));
1582
1583 if (push_items < src_nritems) {
1584 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1585 btrfs_node_key_ptr_offset(push_items),
1586 (src_nritems - push_items) *
1587 sizeof(struct btrfs_key_ptr));
1588 }
1589 btrfs_set_header_nritems(src, src_nritems - push_items);
1590 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1591 btrfs_mark_buffer_dirty(src);
1592 btrfs_mark_buffer_dirty(dst);
1593
1594 ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1595 BUG_ON(ret);
1596
1597 return ret;
1598}
1599
1600/*
1601 * try to push data from one node into the next node right in the
1602 * tree.
1603 *
1604 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1605 * error, and > 0 if there was no room in the right hand block.
1606 *
1607 * this will only push up to 1/2 the contents of the left node over
1608 */
1609static int balance_node_right(struct btrfs_trans_handle *trans,
1610 struct btrfs_root *root,
1611 struct extent_buffer *dst,
1612 struct extent_buffer *src)
1613{
1614 int push_items = 0;
1615 int max_push;
1616 int src_nritems;
1617 int dst_nritems;
1618 int ret = 0;
1619
1620 WARN_ON(btrfs_header_generation(src) != trans->transid);
1621 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1622
1623 src_nritems = btrfs_header_nritems(src);
1624 dst_nritems = btrfs_header_nritems(dst);
1625 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1626 if (push_items <= 0) {
1627 return 1;
1628 }
1629
1630 if (src_nritems < 4) {
1631 return 1;
1632 }
1633
1634 max_push = src_nritems / 2 + 1;
1635 /* don't try to empty the node */
1636 if (max_push >= src_nritems) {
1637 return 1;
1638 }
1639
1640 if (max_push < push_items)
1641 push_items = max_push;
1642
1643 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1644 btrfs_node_key_ptr_offset(0),
1645 (dst_nritems) *
1646 sizeof(struct btrfs_key_ptr));
1647
1648 copy_extent_buffer(dst, src,
1649 btrfs_node_key_ptr_offset(0),
1650 btrfs_node_key_ptr_offset(src_nritems - push_items),
1651 push_items * sizeof(struct btrfs_key_ptr));
1652
1653 btrfs_set_header_nritems(src, src_nritems - push_items);
1654 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1655
1656 btrfs_mark_buffer_dirty(src);
1657 btrfs_mark_buffer_dirty(dst);
1658
1659 ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
1660 BUG_ON(ret);
1661
1662 return ret;
1663}
1664
1665/*
1666 * helper function to insert a new root level in the tree.
1667 * A new node is allocated, and a single item is inserted to
1668 * point to the existing root
1669 *
1670 * returns zero on success or < 0 on failure.
1671 */
1672static int noinline insert_new_root(struct btrfs_trans_handle *trans,
1673 struct btrfs_root *root,
1674 struct btrfs_path *path, int level)
1675{
1676 u64 lower_gen;
1677 struct extent_buffer *lower;
1678 struct extent_buffer *c;
1679 struct extent_buffer *old;
1680 struct btrfs_disk_key lower_key;
1681 int ret;
1682
1683 BUG_ON(path->nodes[level]);
1684 BUG_ON(path->nodes[level-1] != root->node);
1685
1686 lower = path->nodes[level-1];
1687 if (level == 1)
1688 btrfs_item_key(lower, &lower_key, 0);
1689 else
1690 btrfs_node_key(lower, &lower_key, 0);
1691
1692 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1693 root->root_key.objectid, trans->transid,
1694 level, root->node->start, 0);
1695 if (IS_ERR(c))
1696 return PTR_ERR(c);
1697
1698 memset_extent_buffer(c, 0, 0, root->nodesize);
1699 btrfs_set_header_nritems(c, 1);
1700 btrfs_set_header_level(c, level);
1701 btrfs_set_header_bytenr(c, c->start);
1702 btrfs_set_header_generation(c, trans->transid);
1703 btrfs_set_header_owner(c, root->root_key.objectid);
1704
1705 write_extent_buffer(c, root->fs_info->fsid,
1706 (unsigned long)btrfs_header_fsid(c),
1707 BTRFS_FSID_SIZE);
1708
1709 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
1710 (unsigned long)btrfs_header_chunk_tree_uuid(c),
1711 BTRFS_UUID_SIZE);
1712
1713 btrfs_set_node_key(c, &lower_key, 0);
1714 btrfs_set_node_blockptr(c, 0, lower->start);
1715 lower_gen = btrfs_header_generation(lower);
1716 WARN_ON(lower_gen != trans->transid);
1717
1718 btrfs_set_node_ptr_generation(c, 0, lower_gen);
1719
1720 btrfs_mark_buffer_dirty(c);
1721
1722 spin_lock(&root->node_lock);
1723 old = root->node;
1724 root->node = c;
1725 spin_unlock(&root->node_lock);
1726
1727 ret = btrfs_update_extent_ref(trans, root, lower->start,
1728 lower->start, c->start,
1729 root->root_key.objectid,
1730 trans->transid, level - 1, 0);
1731 BUG_ON(ret);
1732
1733 /* the super has an extra ref to root->node */
1734 free_extent_buffer(old);
1735
1736 add_root_to_dirty_list(root);
1737 extent_buffer_get(c);
1738 path->nodes[level] = c;
1739 path->locks[level] = 1;
1740 path->slots[level] = 0;
1741 return 0;
1742}
1743
1744/*
1745 * worker function to insert a single pointer in a node.
1746 * the node should have enough room for the pointer already
1747 *
1748 * slot and level indicate where you want the key to go, and
1749 * blocknr is the block the key points to.
1750 *
1751 * returns zero on success and < 0 on any error
1752 */
1753static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1754 *root, struct btrfs_path *path, struct btrfs_disk_key
1755 *key, u64 bytenr, int slot, int level)
1756{
1757 struct extent_buffer *lower;
1758 int nritems;
1759
1760 BUG_ON(!path->nodes[level]);
1761 lower = path->nodes[level];
1762 nritems = btrfs_header_nritems(lower);
1763 if (slot > nritems)
1764 BUG();
1765 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1766 BUG();
1767 if (slot != nritems) {
1768 memmove_extent_buffer(lower,
1769 btrfs_node_key_ptr_offset(slot + 1),
1770 btrfs_node_key_ptr_offset(slot),
1771 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1772 }
1773 btrfs_set_node_key(lower, key, slot);
1774 btrfs_set_node_blockptr(lower, slot, bytenr);
1775 WARN_ON(trans->transid == 0);
1776 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
1777 btrfs_set_header_nritems(lower, nritems + 1);
1778 btrfs_mark_buffer_dirty(lower);
1779 return 0;
1780}
1781
1782/*
1783 * split the node at the specified level in path in two.
1784 * The path is corrected to point to the appropriate node after the split
1785 *
1786 * Before splitting this tries to make some room in the node by pushing
1787 * left and right, if either one works, it returns right away.
1788 *
1789 * returns 0 on success and < 0 on failure
1790 */
1791static noinline int split_node(struct btrfs_trans_handle *trans,
1792 struct btrfs_root *root,
1793 struct btrfs_path *path, int level)
1794{
1795 struct extent_buffer *c;
1796 struct extent_buffer *split;
1797 struct btrfs_disk_key disk_key;
1798 int mid;
1799 int ret;
1800 int wret;
1801 u32 c_nritems;
1802
1803 c = path->nodes[level];
1804 WARN_ON(btrfs_header_generation(c) != trans->transid);
1805 if (c == root->node) {
1806 /* trying to split the root, lets make a new one */
1807 ret = insert_new_root(trans, root, path, level + 1);
1808 if (ret)
1809 return ret;
1810 } else {
1811 ret = push_nodes_for_insert(trans, root, path, level);
1812 c = path->nodes[level];
1813 if (!ret && btrfs_header_nritems(c) <
1814 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
1815 return 0;
1816 if (ret < 0)
1817 return ret;
1818 }
1819
1820 c_nritems = btrfs_header_nritems(c);
1821
1822 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1823 path->nodes[level + 1]->start,
1824 root->root_key.objectid,
1825 trans->transid, level, c->start, 0);
1826 if (IS_ERR(split))
1827 return PTR_ERR(split);
1828
1829 btrfs_set_header_flags(split, btrfs_header_flags(c));
1830 btrfs_set_header_level(split, btrfs_header_level(c));
1831 btrfs_set_header_bytenr(split, split->start);
1832 btrfs_set_header_generation(split, trans->transid);
1833 btrfs_set_header_owner(split, root->root_key.objectid);
1834 btrfs_set_header_flags(split, 0);
1835 write_extent_buffer(split, root->fs_info->fsid,
1836 (unsigned long)btrfs_header_fsid(split),
1837 BTRFS_FSID_SIZE);
1838 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
1839 (unsigned long)btrfs_header_chunk_tree_uuid(split),
1840 BTRFS_UUID_SIZE);
1841
1842 mid = (c_nritems + 1) / 2;
1843
1844 copy_extent_buffer(split, c,
1845 btrfs_node_key_ptr_offset(0),
1846 btrfs_node_key_ptr_offset(mid),
1847 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1848 btrfs_set_header_nritems(split, c_nritems - mid);
1849 btrfs_set_header_nritems(c, mid);
1850 ret = 0;
1851
1852 btrfs_mark_buffer_dirty(c);
1853 btrfs_mark_buffer_dirty(split);
1854
1855 btrfs_node_key(split, &disk_key, 0);
1856 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1857 path->slots[level + 1] + 1,
1858 level + 1);
1859 if (wret)
1860 ret = wret;
1861
1862 ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
1863 BUG_ON(ret);
1864
1865 if (path->slots[level] >= mid) {
1866 path->slots[level] -= mid;
1867 btrfs_tree_unlock(c);
1868 free_extent_buffer(c);
1869 path->nodes[level] = split;
1870 path->slots[level + 1] += 1;
1871 } else {
1872 btrfs_tree_unlock(split);
1873 free_extent_buffer(split);
1874 }
1875 return ret;
1876}
1877
1878/*
1879 * how many bytes are required to store the items in a leaf. start
1880 * and nr indicate which items in the leaf to check. This totals up the
1881 * space used both by the item structs and the item data
1882 */
1883static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1884{
1885 int data_len;
1886 int nritems = btrfs_header_nritems(l);
1887 int end = min(nritems, start + nr) - 1;
1888
1889 if (!nr)
1890 return 0;
1891 data_len = btrfs_item_end_nr(l, start);
1892 data_len = data_len - btrfs_item_offset_nr(l, end);
1893 data_len += sizeof(struct btrfs_item) * nr;
1894 WARN_ON(data_len < 0);
1895 return data_len;
1896}
1897
1898/*
1899 * The space between the end of the leaf items and
1900 * the start of the leaf data. IOW, how much room
1901 * the leaf has left for both items and data
1902 */
1903int noinline btrfs_leaf_free_space(struct btrfs_root *root,
1904 struct extent_buffer *leaf)
1905{
1906 int nritems = btrfs_header_nritems(leaf);
1907 int ret;
1908 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1909 if (ret < 0) {
1910 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1911 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1912 leaf_space_used(leaf, 0, nritems), nritems);
1913 }
1914 return ret;
1915}
1916
1917/*
1918 * push some data in the path leaf to the right, trying to free up at
1919 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1920 *
1921 * returns 1 if the push failed because the other node didn't have enough
1922 * room, 0 if everything worked out and < 0 if there were major errors.
1923 */
1924static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1925 *root, struct btrfs_path *path, int data_size,
1926 int empty)
1927{
1928 struct extent_buffer *left = path->nodes[0];
1929 struct extent_buffer *right;
1930 struct extent_buffer *upper;
1931 struct btrfs_disk_key disk_key;
1932 int slot;
1933 u32 i;
1934 int free_space;
1935 int push_space = 0;
1936 int push_items = 0;
1937 struct btrfs_item *item;
1938 u32 left_nritems;
1939 u32 nr;
1940 u32 right_nritems;
1941 u32 data_end;
1942 u32 this_item_size;
1943 int ret;
1944
1945 slot = path->slots[1];
1946 if (!path->nodes[1]) {
1947 return 1;
1948 }
1949 upper = path->nodes[1];
1950 if (slot >= btrfs_header_nritems(upper) - 1)
1951 return 1;
1952
1953 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
1954
1955 right = read_node_slot(root, upper, slot + 1);
1956 btrfs_tree_lock(right);
1957 free_space = btrfs_leaf_free_space(root, right);
1958 if (free_space < data_size + sizeof(struct btrfs_item))
1959 goto out_unlock;
1960
1961 /* cow and double check */
1962 ret = btrfs_cow_block(trans, root, right, upper,
1963 slot + 1, &right, 0);
1964 if (ret)
1965 goto out_unlock;
1966
1967 free_space = btrfs_leaf_free_space(root, right);
1968 if (free_space < data_size + sizeof(struct btrfs_item))
1969 goto out_unlock;
1970
1971 left_nritems = btrfs_header_nritems(left);
1972 if (left_nritems == 0)
1973 goto out_unlock;
1974
1975 if (empty)
1976 nr = 0;
1977 else
1978 nr = 1;
1979
1980 if (path->slots[0] >= left_nritems)
1981 push_space += data_size + sizeof(*item);
1982
1983 i = left_nritems - 1;
1984 while (i >= nr) {
1985 item = btrfs_item_nr(left, i);
1986
1987 if (!empty && push_items > 0) {
1988 if (path->slots[0] > i)
1989 break;
1990 if (path->slots[0] == i) {
1991 int space = btrfs_leaf_free_space(root, left);
1992 if (space + push_space * 2 > free_space)
1993 break;
1994 }
1995 }
1996
1997 if (path->slots[0] == i)
1998 push_space += data_size + sizeof(*item);
1999
2000 if (!left->map_token) {
2001 map_extent_buffer(left, (unsigned long)item,
2002 sizeof(struct btrfs_item),
2003 &left->map_token, &left->kaddr,
2004 &left->map_start, &left->map_len,
2005 KM_USER1);
2006 }
2007
2008 this_item_size = btrfs_item_size(left, item);
2009 if (this_item_size + sizeof(*item) + push_space > free_space)
2010 break;
2011
2012 push_items++;
2013 push_space += this_item_size + sizeof(*item);
2014 if (i == 0)
2015 break;
2016 i--;
2017 }
2018 if (left->map_token) {
2019 unmap_extent_buffer(left, left->map_token, KM_USER1);
2020 left->map_token = NULL;
2021 }
2022
2023 if (push_items == 0)
2024 goto out_unlock;
2025
2026 if (!empty && push_items == left_nritems)
2027 WARN_ON(1);
2028
2029 /* push left to right */
2030 right_nritems = btrfs_header_nritems(right);
2031
2032 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2033 push_space -= leaf_data_end(root, left);
2034
2035 /* make room in the right data area */
2036 data_end = leaf_data_end(root, right);
2037 memmove_extent_buffer(right,
2038 btrfs_leaf_data(right) + data_end - push_space,
2039 btrfs_leaf_data(right) + data_end,
2040 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2041
2042 /* copy from the left data area */
2043 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2044 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2045 btrfs_leaf_data(left) + leaf_data_end(root, left),
2046 push_space);
2047
2048 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2049 btrfs_item_nr_offset(0),
2050 right_nritems * sizeof(struct btrfs_item));
2051
2052 /* copy the items from left to right */
2053 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2054 btrfs_item_nr_offset(left_nritems - push_items),
2055 push_items * sizeof(struct btrfs_item));
2056
2057 /* update the item pointers */
2058 right_nritems += push_items;
2059 btrfs_set_header_nritems(right, right_nritems);
2060 push_space = BTRFS_LEAF_DATA_SIZE(root);
2061 for (i = 0; i < right_nritems; i++) {
2062 item = btrfs_item_nr(right, i);
2063 if (!right->map_token) {
2064 map_extent_buffer(right, (unsigned long)item,
2065 sizeof(struct btrfs_item),
2066 &right->map_token, &right->kaddr,
2067 &right->map_start, &right->map_len,
2068 KM_USER1);
2069 }
2070 push_space -= btrfs_item_size(right, item);
2071 btrfs_set_item_offset(right, item, push_space);
2072 }
2073
2074 if (right->map_token) {
2075 unmap_extent_buffer(right, right->map_token, KM_USER1);
2076 right->map_token = NULL;
2077 }
2078 left_nritems -= push_items;
2079 btrfs_set_header_nritems(left, left_nritems);
2080
2081 if (left_nritems)
2082 btrfs_mark_buffer_dirty(left);
2083 btrfs_mark_buffer_dirty(right);
2084
2085 ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2086 BUG_ON(ret);
2087
2088 btrfs_item_key(right, &disk_key, 0);
2089 btrfs_set_node_key(upper, &disk_key, slot + 1);
2090 btrfs_mark_buffer_dirty(upper);
2091
2092 /* then fixup the leaf pointer in the path */
2093 if (path->slots[0] >= left_nritems) {
2094 path->slots[0] -= left_nritems;
2095 if (btrfs_header_nritems(path->nodes[0]) == 0)
2096 clean_tree_block(trans, root, path->nodes[0]);
2097 btrfs_tree_unlock(path->nodes[0]);
2098 free_extent_buffer(path->nodes[0]);
2099 path->nodes[0] = right;
2100 path->slots[1] += 1;
2101 } else {
2102 btrfs_tree_unlock(right);
2103 free_extent_buffer(right);
2104 }
2105 return 0;
2106
2107out_unlock:
2108 btrfs_tree_unlock(right);
2109 free_extent_buffer(right);
2110 return 1;
2111}
2112
2113/*
2114 * push some data in the path leaf to the left, trying to free up at
2115 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2116 */
2117static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2118 *root, struct btrfs_path *path, int data_size,
2119 int empty)
2120{
2121 struct btrfs_disk_key disk_key;
2122 struct extent_buffer *right = path->nodes[0];
2123 struct extent_buffer *left;
2124 int slot;
2125 int i;
2126 int free_space;
2127 int push_space = 0;
2128 int push_items = 0;
2129 struct btrfs_item *item;
2130 u32 old_left_nritems;
2131 u32 right_nritems;
2132 u32 nr;
2133 int ret = 0;
2134 int wret;
2135 u32 this_item_size;
2136 u32 old_left_item_size;
2137
2138 slot = path->slots[1];
2139 if (slot == 0)
2140 return 1;
2141 if (!path->nodes[1])
2142 return 1;
2143
2144 right_nritems = btrfs_header_nritems(right);
2145 if (right_nritems == 0) {
2146 return 1;
2147 }
2148
2149 WARN_ON(!btrfs_tree_locked(path->nodes[1]));
2150
2151 left = read_node_slot(root, path->nodes[1], slot - 1);
2152 btrfs_tree_lock(left);
2153 free_space = btrfs_leaf_free_space(root, left);
2154 if (free_space < data_size + sizeof(struct btrfs_item)) {
2155 ret = 1;
2156 goto out;
2157 }
2158
2159 /* cow and double check */
2160 ret = btrfs_cow_block(trans, root, left,
2161 path->nodes[1], slot - 1, &left, 0);
2162 if (ret) {
2163 /* we hit -ENOSPC, but it isn't fatal here */
2164 ret = 1;
2165 goto out;
2166 }
2167
2168 free_space = btrfs_leaf_free_space(root, left);
2169 if (free_space < data_size + sizeof(struct btrfs_item)) {
2170 ret = 1;
2171 goto out;
2172 }
2173
2174 if (empty)
2175 nr = right_nritems;
2176 else
2177 nr = right_nritems - 1;
2178
2179 for (i = 0; i < nr; i++) {
2180 item = btrfs_item_nr(right, i);
2181 if (!right->map_token) {
2182 map_extent_buffer(right, (unsigned long)item,
2183 sizeof(struct btrfs_item),
2184 &right->map_token, &right->kaddr,
2185 &right->map_start, &right->map_len,
2186 KM_USER1);
2187 }
2188
2189 if (!empty && push_items > 0) {
2190 if (path->slots[0] < i)
2191 break;
2192 if (path->slots[0] == i) {
2193 int space = btrfs_leaf_free_space(root, right);
2194 if (space + push_space * 2 > free_space)
2195 break;
2196 }
2197 }
2198
2199 if (path->slots[0] == i)
2200 push_space += data_size + sizeof(*item);
2201
2202 this_item_size = btrfs_item_size(right, item);
2203 if (this_item_size + sizeof(*item) + push_space > free_space)
2204 break;
2205
2206 push_items++;
2207 push_space += this_item_size + sizeof(*item);
2208 }
2209
2210 if (right->map_token) {
2211 unmap_extent_buffer(right, right->map_token, KM_USER1);
2212 right->map_token = NULL;
2213 }
2214
2215 if (push_items == 0) {
2216 ret = 1;
2217 goto out;
2218 }
2219 if (!empty && push_items == btrfs_header_nritems(right))
2220 WARN_ON(1);
2221
2222 /* push data from right to left */
2223 copy_extent_buffer(left, right,
2224 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2225 btrfs_item_nr_offset(0),
2226 push_items * sizeof(struct btrfs_item));
2227
2228 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2229 btrfs_item_offset_nr(right, push_items -1);
2230
2231 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2232 leaf_data_end(root, left) - push_space,
2233 btrfs_leaf_data(right) +
2234 btrfs_item_offset_nr(right, push_items - 1),
2235 push_space);
2236 old_left_nritems = btrfs_header_nritems(left);
2237 BUG_ON(old_left_nritems < 0);
2238
2239 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2240 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2241 u32 ioff;
2242
2243 item = btrfs_item_nr(left, i);
2244 if (!left->map_token) {
2245 map_extent_buffer(left, (unsigned long)item,
2246 sizeof(struct btrfs_item),
2247 &left->map_token, &left->kaddr,
2248 &left->map_start, &left->map_len,
2249 KM_USER1);
2250 }
2251
2252 ioff = btrfs_item_offset(left, item);
2253 btrfs_set_item_offset(left, item,
2254 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2255 }
2256 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2257 if (left->map_token) {
2258 unmap_extent_buffer(left, left->map_token, KM_USER1);
2259 left->map_token = NULL;
2260 }
2261
2262 /* fixup right node */
2263 if (push_items > right_nritems) {
2264 printk("push items %d nr %u\n", push_items, right_nritems);
2265 WARN_ON(1);
2266 }
2267
2268 if (push_items < right_nritems) {
2269 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2270 leaf_data_end(root, right);
2271 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2272 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2273 btrfs_leaf_data(right) +
2274 leaf_data_end(root, right), push_space);
2275
2276 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2277 btrfs_item_nr_offset(push_items),
2278 (btrfs_header_nritems(right) - push_items) *
2279 sizeof(struct btrfs_item));
2280 }
2281 right_nritems -= push_items;
2282 btrfs_set_header_nritems(right, right_nritems);
2283 push_space = BTRFS_LEAF_DATA_SIZE(root);
2284 for (i = 0; i < right_nritems; i++) {
2285 item = btrfs_item_nr(right, i);
2286
2287 if (!right->map_token) {
2288 map_extent_buffer(right, (unsigned long)item,
2289 sizeof(struct btrfs_item),
2290 &right->map_token, &right->kaddr,
2291 &right->map_start, &right->map_len,
2292 KM_USER1);
2293 }
2294
2295 push_space = push_space - btrfs_item_size(right, item);
2296 btrfs_set_item_offset(right, item, push_space);
2297 }
2298 if (right->map_token) {
2299 unmap_extent_buffer(right, right->map_token, KM_USER1);
2300 right->map_token = NULL;
2301 }
2302
2303 btrfs_mark_buffer_dirty(left);
2304 if (right_nritems)
2305 btrfs_mark_buffer_dirty(right);
2306
2307 ret = btrfs_update_ref(trans, root, right, left,
2308 old_left_nritems, push_items);
2309 BUG_ON(ret);
2310
2311 btrfs_item_key(right, &disk_key, 0);
2312 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2313 if (wret)
2314 ret = wret;
2315
2316 /* then fixup the leaf pointer in the path */
2317 if (path->slots[0] < push_items) {
2318 path->slots[0] += old_left_nritems;
2319 if (btrfs_header_nritems(path->nodes[0]) == 0)
2320 clean_tree_block(trans, root, path->nodes[0]);
2321 btrfs_tree_unlock(path->nodes[0]);
2322 free_extent_buffer(path->nodes[0]);
2323 path->nodes[0] = left;
2324 path->slots[1] -= 1;
2325 } else {
2326 btrfs_tree_unlock(left);
2327 free_extent_buffer(left);
2328 path->slots[0] -= push_items;
2329 }
2330 BUG_ON(path->slots[0] < 0);
2331 return ret;
2332out:
2333 btrfs_tree_unlock(left);
2334 free_extent_buffer(left);
2335 return ret;
2336}
2337
2338/*
2339 * split the path's leaf in two, making sure there is at least data_size
2340 * available for the resulting leaf level of the path.
2341 *
2342 * returns 0 if all went well and < 0 on failure.
2343 */
2344static noinline int split_leaf(struct btrfs_trans_handle *trans,
2345 struct btrfs_root *root,
2346 struct btrfs_key *ins_key,
2347 struct btrfs_path *path, int data_size,
2348 int extend)
2349{
2350 struct extent_buffer *l;
2351 u32 nritems;
2352 int mid;
2353 int slot;
2354 struct extent_buffer *right;
2355 int space_needed = data_size + sizeof(struct btrfs_item);
2356 int data_copy_size;
2357 int rt_data_off;
2358 int i;
2359 int ret = 0;
2360 int wret;
2361 int double_split;
2362 int num_doubles = 0;
2363 struct btrfs_disk_key disk_key;
2364
2365 if (extend)
2366 space_needed = data_size;
2367
2368 /* first try to make some room by pushing left and right */
2369 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
2370 wret = push_leaf_right(trans, root, path, data_size, 0);
2371 if (wret < 0) {
2372 return wret;
2373 }
2374 if (wret) {
2375 wret = push_leaf_left(trans, root, path, data_size, 0);
2376 if (wret < 0)
2377 return wret;
2378 }
2379 l = path->nodes[0];
2380
2381 /* did the pushes work? */
2382 if (btrfs_leaf_free_space(root, l) >= space_needed)
2383 return 0;
2384 }
2385
2386 if (!path->nodes[1]) {
2387 ret = insert_new_root(trans, root, path, 1);
2388 if (ret)
2389 return ret;
2390 }
2391again:
2392 double_split = 0;
2393 l = path->nodes[0];
2394 slot = path->slots[0];
2395 nritems = btrfs_header_nritems(l);
2396 mid = (nritems + 1)/ 2;
2397
2398 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2399 path->nodes[1]->start,
2400 root->root_key.objectid,
2401 trans->transid, 0, l->start, 0);
2402 if (IS_ERR(right)) {
2403 BUG_ON(1);
2404 return PTR_ERR(right);
2405 }
2406
2407 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2408 btrfs_set_header_bytenr(right, right->start);
2409 btrfs_set_header_generation(right, trans->transid);
2410 btrfs_set_header_owner(right, root->root_key.objectid);
2411 btrfs_set_header_level(right, 0);
2412 write_extent_buffer(right, root->fs_info->fsid,
2413 (unsigned long)btrfs_header_fsid(right),
2414 BTRFS_FSID_SIZE);
2415
2416 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2417 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2418 BTRFS_UUID_SIZE);
2419 if (mid <= slot) {
2420 if (nritems == 1 ||
2421 leaf_space_used(l, mid, nritems - mid) + space_needed >
2422 BTRFS_LEAF_DATA_SIZE(root)) {
2423 if (slot >= nritems) {
2424 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2425 btrfs_set_header_nritems(right, 0);
2426 wret = insert_ptr(trans, root, path,
2427 &disk_key, right->start,
2428 path->slots[1] + 1, 1);
2429 if (wret)
2430 ret = wret;
2431
2432 btrfs_tree_unlock(path->nodes[0]);
2433 free_extent_buffer(path->nodes[0]);
2434 path->nodes[0] = right;
2435 path->slots[0] = 0;
2436 path->slots[1] += 1;
2437 btrfs_mark_buffer_dirty(right);
2438 return ret;
2439 }
2440 mid = slot;
2441 if (mid != nritems &&
2442 leaf_space_used(l, mid, nritems - mid) +
2443 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2444 double_split = 1;
2445 }
2446 }
2447 } else {
2448 if (leaf_space_used(l, 0, mid + 1) + space_needed >
2449 BTRFS_LEAF_DATA_SIZE(root)) {
2450 if (!extend && slot == 0) {
2451 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2452 btrfs_set_header_nritems(right, 0);
2453 wret = insert_ptr(trans, root, path,
2454 &disk_key,
2455 right->start,
2456 path->slots[1], 1);
2457 if (wret)
2458 ret = wret;
2459 btrfs_tree_unlock(path->nodes[0]);
2460 free_extent_buffer(path->nodes[0]);
2461 path->nodes[0] = right;
2462 path->slots[0] = 0;
2463 if (path->slots[1] == 0) {
2464 wret = fixup_low_keys(trans, root,
2465 path, &disk_key, 1);
2466 if (wret)
2467 ret = wret;
2468 }
2469 btrfs_mark_buffer_dirty(right);
2470 return ret;
2471 } else if (extend && slot == 0) {
2472 mid = 1;
2473 } else {
2474 mid = slot;
2475 if (mid != nritems &&
2476 leaf_space_used(l, mid, nritems - mid) +
2477 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
2478 double_split = 1;
2479 }
2480 }
2481 }
2482 }
2483 nritems = nritems - mid;
2484 btrfs_set_header_nritems(right, nritems);
2485 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2486
2487 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2488 btrfs_item_nr_offset(mid),
2489 nritems * sizeof(struct btrfs_item));
2490
2491 copy_extent_buffer(right, l,
2492 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2493 data_copy_size, btrfs_leaf_data(l) +
2494 leaf_data_end(root, l), data_copy_size);
2495
2496 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2497 btrfs_item_end_nr(l, mid);
2498
2499 for (i = 0; i < nritems; i++) {
2500 struct btrfs_item *item = btrfs_item_nr(right, i);
2501 u32 ioff;
2502
2503 if (!right->map_token) {
2504 map_extent_buffer(right, (unsigned long)item,
2505 sizeof(struct btrfs_item),
2506 &right->map_token, &right->kaddr,
2507 &right->map_start, &right->map_len,
2508 KM_USER1);
2509 }
2510
2511 ioff = btrfs_item_offset(right, item);
2512 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2513 }
2514
2515 if (right->map_token) {
2516 unmap_extent_buffer(right, right->map_token, KM_USER1);
2517 right->map_token = NULL;
2518 }
2519
2520 btrfs_set_header_nritems(l, mid);
2521 ret = 0;
2522 btrfs_item_key(right, &disk_key, 0);
2523 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2524 path->slots[1] + 1, 1);
2525 if (wret)
2526 ret = wret;
2527
2528 btrfs_mark_buffer_dirty(right);
2529 btrfs_mark_buffer_dirty(l);
2530 BUG_ON(path->slots[0] != slot);
2531
2532 ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2533 BUG_ON(ret);
2534
2535 if (mid <= slot) {
2536 btrfs_tree_unlock(path->nodes[0]);
2537 free_extent_buffer(path->nodes[0]);
2538 path->nodes[0] = right;
2539 path->slots[0] -= mid;
2540 path->slots[1] += 1;
2541 } else {
2542 btrfs_tree_unlock(right);
2543 free_extent_buffer(right);
2544 }
2545
2546 BUG_ON(path->slots[0] < 0);
2547
2548 if (double_split) {
2549 BUG_ON(num_doubles != 0);
2550 num_doubles++;
2551 goto again;
2552 }
2553 return ret;
2554}
2555
2556int btrfs_truncate_item(struct btrfs_trans_handle *trans,
2557 struct btrfs_root *root,
2558 struct btrfs_path *path,
2559 u32 new_size, int from_end)
2560{
2561 int ret = 0;
2562 int slot;
2563 int slot_orig;
2564 struct extent_buffer *leaf;
2565 struct btrfs_item *item;
2566 u32 nritems;
2567 unsigned int data_end;
2568 unsigned int old_data_start;
2569 unsigned int old_size;
2570 unsigned int size_diff;
2571 int i;
2572
2573 slot_orig = path->slots[0];
2574 leaf = path->nodes[0];
2575 slot = path->slots[0];
2576
2577 old_size = btrfs_item_size_nr(leaf, slot);
2578 if (old_size == new_size)
2579 return 0;
2580
2581 nritems = btrfs_header_nritems(leaf);
2582 data_end = leaf_data_end(root, leaf);
2583
2584 old_data_start = btrfs_item_offset_nr(leaf, slot);
2585
2586 size_diff = old_size - new_size;
2587
2588 BUG_ON(slot < 0);
2589 BUG_ON(slot >= nritems);
2590
2591 /*
2592 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2593 */
2594 /* first correct the data pointers */
2595 for (i = slot; i < nritems; i++) {
2596 u32 ioff;
2597 item = btrfs_item_nr(leaf, i);
2598
2599 if (!leaf->map_token) {
2600 map_extent_buffer(leaf, (unsigned long)item,
2601 sizeof(struct btrfs_item),
2602 &leaf->map_token, &leaf->kaddr,
2603 &leaf->map_start, &leaf->map_len,
2604 KM_USER1);
2605 }
2606
2607 ioff = btrfs_item_offset(leaf, item);
2608 btrfs_set_item_offset(leaf, item, ioff + size_diff);
2609 }
2610
2611 if (leaf->map_token) {
2612 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2613 leaf->map_token = NULL;
2614 }
2615
2616 /* shift the data */
2617 if (from_end) {
2618 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2619 data_end + size_diff, btrfs_leaf_data(leaf) +
2620 data_end, old_data_start + new_size - data_end);
2621 } else {
2622 struct btrfs_disk_key disk_key;
2623 u64 offset;
2624
2625 btrfs_item_key(leaf, &disk_key, slot);
2626
2627 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2628 unsigned long ptr;
2629 struct btrfs_file_extent_item *fi;
2630
2631 fi = btrfs_item_ptr(leaf, slot,
2632 struct btrfs_file_extent_item);
2633 fi = (struct btrfs_file_extent_item *)(
2634 (unsigned long)fi - size_diff);
2635
2636 if (btrfs_file_extent_type(leaf, fi) ==
2637 BTRFS_FILE_EXTENT_INLINE) {
2638 ptr = btrfs_item_ptr_offset(leaf, slot);
2639 memmove_extent_buffer(leaf, ptr,
2640 (unsigned long)fi,
2641 offsetof(struct btrfs_file_extent_item,
2642 disk_bytenr));
2643 }
2644 }
2645
2646 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2647 data_end + size_diff, btrfs_leaf_data(leaf) +
2648 data_end, old_data_start - data_end);
2649
2650 offset = btrfs_disk_key_offset(&disk_key);
2651 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2652 btrfs_set_item_key(leaf, &disk_key, slot);
2653 if (slot == 0)
2654 fixup_low_keys(trans, root, path, &disk_key, 1);
2655 }
2656
2657 item = btrfs_item_nr(leaf, slot);
2658 btrfs_set_item_size(leaf, item, new_size);
2659 btrfs_mark_buffer_dirty(leaf);
2660
2661 ret = 0;
2662 if (btrfs_leaf_free_space(root, leaf) < 0) {
2663 btrfs_print_leaf(root, leaf);
2664 BUG();
2665 }
2666 return ret;
2667}
2668
2669int btrfs_extend_item(struct btrfs_trans_handle *trans,
2670 struct btrfs_root *root, struct btrfs_path *path,
2671 u32 data_size)
2672{
2673 int ret = 0;
2674 int slot;
2675 int slot_orig;
2676 struct extent_buffer *leaf;
2677 struct btrfs_item *item;
2678 u32 nritems;
2679 unsigned int data_end;
2680 unsigned int old_data;
2681 unsigned int old_size;
2682 int i;
2683
2684 slot_orig = path->slots[0];
2685 leaf = path->nodes[0];
2686
2687 nritems = btrfs_header_nritems(leaf);
2688 data_end = leaf_data_end(root, leaf);
2689
2690 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2691 btrfs_print_leaf(root, leaf);
2692 BUG();
2693 }
2694 slot = path->slots[0];
2695 old_data = btrfs_item_end_nr(leaf, slot);
2696
2697 BUG_ON(slot < 0);
2698 if (slot >= nritems) {
2699 btrfs_print_leaf(root, leaf);
2700 printk("slot %d too large, nritems %d\n", slot, nritems);
2701 BUG_ON(1);
2702 }
2703
2704 /*
2705 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2706 */
2707 /* first correct the data pointers */
2708 for (i = slot; i < nritems; i++) {
2709 u32 ioff;
2710 item = btrfs_item_nr(leaf, i);
2711
2712 if (!leaf->map_token) {
2713 map_extent_buffer(leaf, (unsigned long)item,
2714 sizeof(struct btrfs_item),
2715 &leaf->map_token, &leaf->kaddr,
2716 &leaf->map_start, &leaf->map_len,
2717 KM_USER1);
2718 }
2719 ioff = btrfs_item_offset(leaf, item);
2720 btrfs_set_item_offset(leaf, item, ioff - data_size);
2721 }
2722
2723 if (leaf->map_token) {
2724 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2725 leaf->map_token = NULL;
2726 }
2727
2728 /* shift the data */
2729 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2730 data_end - data_size, btrfs_leaf_data(leaf) +
2731 data_end, old_data - data_end);
2732
2733 data_end = old_data;
2734 old_size = btrfs_item_size_nr(leaf, slot);
2735 item = btrfs_item_nr(leaf, slot);
2736 btrfs_set_item_size(leaf, item, old_size + data_size);
2737 btrfs_mark_buffer_dirty(leaf);
2738
2739 ret = 0;
2740 if (btrfs_leaf_free_space(root, leaf) < 0) {
2741 btrfs_print_leaf(root, leaf);
2742 BUG();
2743 }
2744 return ret;
2745}
2746
2747/*
2748 * Given a key and some data, insert an item into the tree.
2749 * This does all the path init required, making room in the tree if needed.
2750 */
2751int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
2752 struct btrfs_root *root,
2753 struct btrfs_path *path,
2754 struct btrfs_key *cpu_key, u32 *data_size,
2755 int nr)
2756{
2757 struct extent_buffer *leaf;
2758 struct btrfs_item *item;
2759 int ret = 0;
2760 int slot;
2761 int slot_orig;
2762 int i;
2763 u32 nritems;
2764 u32 total_size = 0;
2765 u32 total_data = 0;
2766 unsigned int data_end;
2767 struct btrfs_disk_key disk_key;
2768
2769 for (i = 0; i < nr; i++) {
2770 total_data += data_size[i];
2771 }
2772
2773 total_size = total_data + (nr * sizeof(struct btrfs_item));
2774 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
2775 if (ret == 0)
2776 return -EEXIST;
2777 if (ret < 0)
2778 goto out;
2779
2780 slot_orig = path->slots[0];
2781 leaf = path->nodes[0];
2782
2783 nritems = btrfs_header_nritems(leaf);
2784 data_end = leaf_data_end(root, leaf);
2785
2786 if (btrfs_leaf_free_space(root, leaf) < total_size) {
2787 btrfs_print_leaf(root, leaf);
2788 printk("not enough freespace need %u have %d\n",
2789 total_size, btrfs_leaf_free_space(root, leaf));
2790 BUG();
2791 }
2792
2793 slot = path->slots[0];
2794 BUG_ON(slot < 0);
2795
2796 if (slot != nritems) {
2797 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2798
2799 if (old_data < data_end) {
2800 btrfs_print_leaf(root, leaf);
2801 printk("slot %d old_data %d data_end %d\n",
2802 slot, old_data, data_end);
2803 BUG_ON(1);
2804 }
2805 /*
2806 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2807 */
2808 /* first correct the data pointers */
2809 WARN_ON(leaf->map_token);
2810 for (i = slot; i < nritems; i++) {
2811 u32 ioff;
2812
2813 item = btrfs_item_nr(leaf, i);
2814 if (!leaf->map_token) {
2815 map_extent_buffer(leaf, (unsigned long)item,
2816 sizeof(struct btrfs_item),
2817 &leaf->map_token, &leaf->kaddr,
2818 &leaf->map_start, &leaf->map_len,
2819 KM_USER1);
2820 }
2821
2822 ioff = btrfs_item_offset(leaf, item);
2823 btrfs_set_item_offset(leaf, item, ioff - total_data);
2824 }
2825 if (leaf->map_token) {
2826 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2827 leaf->map_token = NULL;
2828 }
2829
2830 /* shift the items */
2831 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
2832 btrfs_item_nr_offset(slot),
2833 (nritems - slot) * sizeof(struct btrfs_item));
2834
2835 /* shift the data */
2836 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2837 data_end - total_data, btrfs_leaf_data(leaf) +
2838 data_end, old_data - data_end);
2839 data_end = old_data;
2840 }
2841
2842 /* setup the item for the new data */
2843 for (i = 0; i < nr; i++) {
2844 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
2845 btrfs_set_item_key(leaf, &disk_key, slot + i);
2846 item = btrfs_item_nr(leaf, slot + i);
2847 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
2848 data_end -= data_size[i];
2849 btrfs_set_item_size(leaf, item, data_size[i]);
2850 }
2851 btrfs_set_header_nritems(leaf, nritems + nr);
2852 btrfs_mark_buffer_dirty(leaf);
2853
2854 ret = 0;
2855 if (slot == 0) {
2856 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2857 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2858 }
2859
2860 if (btrfs_leaf_free_space(root, leaf) < 0) {
2861 btrfs_print_leaf(root, leaf);
2862 BUG();
2863 }
2864out:
2865 return ret;
2866}
2867
2868/*
2869 * Given a key and some data, insert an item into the tree.
2870 * This does all the path init required, making room in the tree if needed.
2871 */
2872int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2873 *root, struct btrfs_key *cpu_key, void *data, u32
2874 data_size)
2875{
2876 int ret = 0;
2877 struct btrfs_path *path;
2878 struct extent_buffer *leaf;
2879 unsigned long ptr;
2880
2881 path = btrfs_alloc_path();
2882 BUG_ON(!path);
2883 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2884 if (!ret) {
2885 leaf = path->nodes[0];
2886 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2887 write_extent_buffer(leaf, data, ptr, data_size);
2888 btrfs_mark_buffer_dirty(leaf);
2889 }
2890 btrfs_free_path(path);
2891 return ret;
2892}
2893
2894/*
2895 * delete the pointer from a given node.
2896 *
2897 * If the delete empties a node, the node is removed from the tree,
2898 * continuing all the way the root if required. The root is converted into
2899 * a leaf if all the nodes are emptied.
2900 */
2901static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2902 struct btrfs_path *path, int level, int slot)
2903{
2904 struct extent_buffer *parent = path->nodes[level];
2905 u32 nritems;
2906 int ret = 0;
2907 int wret;
2908
2909 nritems = btrfs_header_nritems(parent);
2910 if (slot != nritems -1) {
2911 memmove_extent_buffer(parent,
2912 btrfs_node_key_ptr_offset(slot),
2913 btrfs_node_key_ptr_offset(slot + 1),
2914 sizeof(struct btrfs_key_ptr) *
2915 (nritems - slot - 1));
2916 }
2917 nritems--;
2918 btrfs_set_header_nritems(parent, nritems);
2919 if (nritems == 0 && parent == root->node) {
2920 BUG_ON(btrfs_header_level(root->node) != 1);
2921 /* just turn the root into a leaf and break */
2922 btrfs_set_header_level(root->node, 0);
2923 } else if (slot == 0) {
2924 struct btrfs_disk_key disk_key;
2925
2926 btrfs_node_key(parent, &disk_key, 0);
2927 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2928 if (wret)
2929 ret = wret;
2930 }
2931 btrfs_mark_buffer_dirty(parent);
2932 return ret;
2933}
2934
2935/*
2936 * delete the item at the leaf level in path. If that empties
2937 * the leaf, remove it from the tree
2938 */
2939int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2940 struct btrfs_path *path, int slot, int nr)
2941{
2942 struct extent_buffer *leaf;
2943 struct btrfs_item *item;
2944 int last_off;
2945 int dsize = 0;
2946 int ret = 0;
2947 int wret;
2948 int i;
2949 u32 nritems;
2950
2951 leaf = path->nodes[0];
2952 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
2953
2954 for (i = 0; i < nr; i++)
2955 dsize += btrfs_item_size_nr(leaf, slot + i);
2956
2957 nritems = btrfs_header_nritems(leaf);
2958
2959 if (slot + nr != nritems) {
2960 int data_end = leaf_data_end(root, leaf);
2961
2962 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2963 data_end + dsize,
2964 btrfs_leaf_data(leaf) + data_end,
2965 last_off - data_end);
2966
2967 for (i = slot + nr; i < nritems; i++) {
2968 u32 ioff;
2969
2970 item = btrfs_item_nr(leaf, i);
2971 if (!leaf->map_token) {
2972 map_extent_buffer(leaf, (unsigned long)item,
2973 sizeof(struct btrfs_item),
2974 &leaf->map_token, &leaf->kaddr,
2975 &leaf->map_start, &leaf->map_len,
2976 KM_USER1);
2977 }
2978 ioff = btrfs_item_offset(leaf, item);
2979 btrfs_set_item_offset(leaf, item, ioff + dsize);
2980 }
2981
2982 if (leaf->map_token) {
2983 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2984 leaf->map_token = NULL;
2985 }
2986
2987 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2988 btrfs_item_nr_offset(slot + nr),
2989 sizeof(struct btrfs_item) *
2990 (nritems - slot - nr));
2991 }
2992 btrfs_set_header_nritems(leaf, nritems - nr);
2993 nritems -= nr;
2994
2995 /* delete the leaf if we've emptied it */
2996 if (nritems == 0) {
2997 if (leaf == root->node) {
2998 btrfs_set_header_level(leaf, 0);
2999 } else {
3000 u64 root_gen = btrfs_header_generation(path->nodes[1]);
3001 wret = del_ptr(trans, root, path, 1, path->slots[1]);
3002 if (wret)
3003 ret = wret;
3004 wret = btrfs_free_extent(trans, root,
3005 leaf->start, leaf->len,
3006 path->nodes[1]->start,
3007 btrfs_header_owner(path->nodes[1]),
3008 root_gen, 0, 0, 1);
3009 if (wret)
3010 ret = wret;
3011 }
3012 } else {
3013 int used = leaf_space_used(leaf, 0, nritems);
3014 if (slot == 0) {
3015 struct btrfs_disk_key disk_key;
3016
3017 btrfs_item_key(leaf, &disk_key, 0);
3018 wret = fixup_low_keys(trans, root, path,
3019 &disk_key, 1);
3020 if (wret)
3021 ret = wret;
3022 }
3023
3024 /* delete the leaf if it is mostly empty */
3025 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
3026 /* push_leaf_left fixes the path.
3027 * make sure the path still points to our leaf
3028 * for possible call to del_ptr below
3029 */
3030 slot = path->slots[1];
3031 extent_buffer_get(leaf);
3032
3033 wret = push_leaf_left(trans, root, path, 1, 1);
3034 if (wret < 0 && wret != -ENOSPC)
3035 ret = wret;
3036
3037 if (path->nodes[0] == leaf &&
3038 btrfs_header_nritems(leaf)) {
3039 wret = push_leaf_right(trans, root, path, 1, 1);
3040 if (wret < 0 && wret != -ENOSPC)
3041 ret = wret;
3042 }
3043
3044 if (btrfs_header_nritems(leaf) == 0) {
3045 u64 root_gen;
3046 u64 bytenr = leaf->start;
3047 u32 blocksize = leaf->len;
3048
3049 root_gen = btrfs_header_generation(
3050 path->nodes[1]);
3051
3052 wret = del_ptr(trans, root, path, 1, slot);
3053 if (wret)
3054 ret = wret;
3055
3056 free_extent_buffer(leaf);
3057 wret = btrfs_free_extent(trans, root, bytenr,
3058 blocksize, path->nodes[1]->start,
3059 btrfs_header_owner(path->nodes[1]),
3060 root_gen, 0, 0, 1);
3061 if (wret)
3062 ret = wret;
3063 } else {
3064 /* if we're still in the path, make sure
3065 * we're dirty. Otherwise, one of the
3066 * push_leaf functions must have already
3067 * dirtied this buffer
3068 */
3069 if (path->nodes[0] == leaf)
3070 btrfs_mark_buffer_dirty(leaf);
3071 free_extent_buffer(leaf);
3072 }
3073 } else {
3074 btrfs_mark_buffer_dirty(leaf);
3075 }
3076 }
3077 return ret;
3078}
3079
3080/*
3081 * search the tree again to find a leaf with lesser keys
3082 * returns 0 if it found something or 1 if there are no lesser leaves.
3083 * returns < 0 on io errors.
3084 */
3085int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3086{
3087 struct btrfs_key key;
3088 struct btrfs_disk_key found_key;
3089 int ret;
3090
3091 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3092
3093 if (key.offset > 0)
3094 key.offset--;
3095 else if (key.type > 0)
3096 key.type--;
3097 else if (key.objectid > 0)
3098 key.objectid--;
3099 else
3100 return 1;
3101
3102 btrfs_release_path(root, path);
3103 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3104 if (ret < 0)
3105 return ret;
3106 btrfs_item_key(path->nodes[0], &found_key, 0);
3107 ret = comp_keys(&found_key, &key);
3108 if (ret < 0)
3109 return 0;
3110 return 1;
3111}
3112
3113/*
3114 * A helper function to walk down the tree starting at min_key, and looking
3115 * for nodes or leaves that are either in cache or have a minimum
3116 * transaction id. This is used by the btree defrag code, but could
3117 * also be used to search for blocks that have changed since a given
3118 * transaction id.
3119 *
3120 * This does not cow, but it does stuff the starting key it finds back
3121 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3122 * key and get a writable path.
3123 *
3124 * This does lock as it descends, and path->keep_locks should be set
3125 * to 1 by the caller.
3126 *
3127 * This honors path->lowest_level to prevent descent past a given level
3128 * of the tree.
3129 *
3130 * returns zero if something useful was found, < 0 on error and 1 if there
3131 * was nothing in the tree that matched the search criteria.
3132 */
3133int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3134 struct btrfs_key *max_key,
3135 struct btrfs_path *path, int cache_only,
3136 u64 min_trans)
3137{
3138 struct extent_buffer *cur;
3139 struct btrfs_key found_key;
3140 int slot;
3141 int sret;
3142 u32 nritems;
3143 int level;
3144 int ret = 1;
3145
3146again:
3147 cur = btrfs_lock_root_node(root);
3148 level = btrfs_header_level(cur);
3149 WARN_ON(path->nodes[level]);
3150 path->nodes[level] = cur;
3151 path->locks[level] = 1;
3152
3153 if (btrfs_header_generation(cur) < min_trans) {
3154 ret = 1;
3155 goto out;
3156 }
3157 while(1) {
3158 nritems = btrfs_header_nritems(cur);
3159 level = btrfs_header_level(cur);
3160 sret = bin_search(cur, min_key, level, &slot);
3161
3162 /* at level = 0, we're done, setup the path and exit */
3163 if (level == 0) {
3164 if (slot >= nritems)
3165 goto find_next_key;
3166 ret = 0;
3167 path->slots[level] = slot;
3168 btrfs_item_key_to_cpu(cur, &found_key, slot);
3169 goto out;
3170 }
3171 if (sret && slot > 0)
3172 slot--;
3173 /*
3174 * check this node pointer against the cache_only and
3175 * min_trans parameters. If it isn't in cache or is too
3176 * old, skip to the next one.
3177 */
3178 while(slot < nritems) {
3179 u64 blockptr;
3180 u64 gen;
3181 struct extent_buffer *tmp;
3182 struct btrfs_disk_key disk_key;
3183
3184 blockptr = btrfs_node_blockptr(cur, slot);
3185 gen = btrfs_node_ptr_generation(cur, slot);
3186 if (gen < min_trans) {
3187 slot++;
3188 continue;
3189 }
3190 if (!cache_only)
3191 break;
3192
3193 if (max_key) {
3194 btrfs_node_key(cur, &disk_key, slot);
3195 if (comp_keys(&disk_key, max_key) >= 0) {
3196 ret = 1;
3197 goto out;
3198 }
3199 }
3200
3201 tmp = btrfs_find_tree_block(root, blockptr,
3202 btrfs_level_size(root, level - 1));
3203
3204 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3205 free_extent_buffer(tmp);
3206 break;
3207 }
3208 if (tmp)
3209 free_extent_buffer(tmp);
3210 slot++;
3211 }
3212find_next_key:
3213 /*
3214 * we didn't find a candidate key in this node, walk forward
3215 * and find another one
3216 */
3217 if (slot >= nritems) {
3218 path->slots[level] = slot;
3219 sret = btrfs_find_next_key(root, path, min_key, level,
3220 cache_only, min_trans);
3221 if (sret == 0) {
3222 btrfs_release_path(root, path);
3223 goto again;
3224 } else {
3225 goto out;
3226 }
3227 }
3228 /* save our key for returning back */
3229 btrfs_node_key_to_cpu(cur, &found_key, slot);
3230 path->slots[level] = slot;
3231 if (level == path->lowest_level) {
3232 ret = 0;
3233 unlock_up(path, level, 1);
3234 goto out;
3235 }
3236 cur = read_node_slot(root, cur, slot);
3237
3238 btrfs_tree_lock(cur);
3239 path->locks[level - 1] = 1;
3240 path->nodes[level - 1] = cur;
3241 unlock_up(path, level, 1);
3242 }
3243out:
3244 if (ret == 0)
3245 memcpy(min_key, &found_key, sizeof(found_key));
3246 return ret;
3247}
3248
3249/*
3250 * this is similar to btrfs_next_leaf, but does not try to preserve
3251 * and fixup the path. It looks for and returns the next key in the
3252 * tree based on the current path and the cache_only and min_trans
3253 * parameters.
3254 *
3255 * 0 is returned if another key is found, < 0 if there are any errors
3256 * and 1 is returned if there are no higher keys in the tree
3257 *
3258 * path->keep_locks should be set to 1 on the search made before
3259 * calling this function.
3260 */
3261int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3262 struct btrfs_key *key, int lowest_level,
3263 int cache_only, u64 min_trans)
3264{
3265 int level = lowest_level;
3266 int slot;
3267 struct extent_buffer *c;
3268
3269 while(level < BTRFS_MAX_LEVEL) {
3270 if (!path->nodes[level])
3271 return 1;
3272
3273 slot = path->slots[level] + 1;
3274 c = path->nodes[level];
3275next:
3276 if (slot >= btrfs_header_nritems(c)) {
3277 level++;
3278 if (level == BTRFS_MAX_LEVEL) {
3279 return 1;
3280 }
3281 continue;
3282 }
3283 if (level == 0)
3284 btrfs_item_key_to_cpu(c, key, slot);
3285 else {
3286 u64 blockptr = btrfs_node_blockptr(c, slot);
3287 u64 gen = btrfs_node_ptr_generation(c, slot);
3288
3289 if (cache_only) {
3290 struct extent_buffer *cur;
3291 cur = btrfs_find_tree_block(root, blockptr,
3292 btrfs_level_size(root, level - 1));
3293 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3294 slot++;
3295 if (cur)
3296 free_extent_buffer(cur);
3297 goto next;
3298 }
3299 free_extent_buffer(cur);
3300 }
3301 if (gen < min_trans) {
3302 slot++;
3303 goto next;
3304 }
3305 btrfs_node_key_to_cpu(c, key, slot);
3306 }
3307 return 0;
3308 }
3309 return 1;
3310}
3311
3312/*
3313 * search the tree again to find a leaf with greater keys
3314 * returns 0 if it found something or 1 if there are no greater leaves.
3315 * returns < 0 on io errors.
3316 */
3317int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3318{
3319 int slot;
3320 int level = 1;
3321 struct extent_buffer *c;
3322 struct extent_buffer *next = NULL;
3323 struct btrfs_key key;
3324 u32 nritems;
3325 int ret;
3326
3327 nritems = btrfs_header_nritems(path->nodes[0]);
3328 if (nritems == 0) {
3329 return 1;
3330 }
3331
3332 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3333
3334 btrfs_release_path(root, path);
3335 path->keep_locks = 1;
3336 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3337 path->keep_locks = 0;
3338
3339 if (ret < 0)
3340 return ret;
3341
3342 nritems = btrfs_header_nritems(path->nodes[0]);
3343 /*
3344 * by releasing the path above we dropped all our locks. A balance
3345 * could have added more items next to the key that used to be
3346 * at the very end of the block. So, check again here and
3347 * advance the path if there are now more items available.
3348 */
3349 if (nritems > 0 && path->slots[0] < nritems - 1) {
3350 path->slots[0]++;
3351 goto done;
3352 }
3353
3354 while(level < BTRFS_MAX_LEVEL) {
3355 if (!path->nodes[level])
3356 return 1;
3357
3358 slot = path->slots[level] + 1;
3359 c = path->nodes[level];
3360 if (slot >= btrfs_header_nritems(c)) {
3361 level++;
3362 if (level == BTRFS_MAX_LEVEL) {
3363 return 1;
3364 }
3365 continue;
3366 }
3367
3368 if (next) {
3369 btrfs_tree_unlock(next);
3370 free_extent_buffer(next);
3371 }
3372
3373 if (level == 1 && (path->locks[1] || path->skip_locking) &&
3374 path->reada)
3375 reada_for_search(root, path, level, slot, 0);
3376
3377 next = read_node_slot(root, c, slot);
3378 if (!path->skip_locking) {
3379 WARN_ON(!btrfs_tree_locked(c));
3380 btrfs_tree_lock(next);
3381 }
3382 break;
3383 }
3384 path->slots[level] = slot;
3385 while(1) {
3386 level--;
3387 c = path->nodes[level];
3388 if (path->locks[level])
3389 btrfs_tree_unlock(c);
3390 free_extent_buffer(c);
3391 path->nodes[level] = next;
3392 path->slots[level] = 0;
3393 if (!path->skip_locking)
3394 path->locks[level] = 1;
3395 if (!level)
3396 break;
3397 if (level == 1 && path->locks[1] && path->reada)
3398 reada_for_search(root, path, level, slot, 0);
3399 next = read_node_slot(root, next, 0);
3400 if (!path->skip_locking) {
3401 WARN_ON(!btrfs_tree_locked(path->nodes[level]));
3402 btrfs_tree_lock(next);
3403 }
3404 }
3405done:
3406 unlock_up(path, 0, 1);
3407 return 0;
3408}
3409
3410/*
3411 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
3412 * searching until it gets past min_objectid or finds an item of 'type'
3413 *
3414 * returns 0 if something is found, 1 if nothing was found and < 0 on error
3415 */
3416int btrfs_previous_item(struct btrfs_root *root,
3417 struct btrfs_path *path, u64 min_objectid,
3418 int type)
3419{
3420 struct btrfs_key found_key;
3421 struct extent_buffer *leaf;
3422 u32 nritems;
3423 int ret;
3424
3425 while(1) {
3426 if (path->slots[0] == 0) {
3427 ret = btrfs_prev_leaf(root, path);
3428 if (ret != 0)
3429 return ret;
3430 } else {
3431 path->slots[0]--;
3432 }
3433 leaf = path->nodes[0];
3434 nritems = btrfs_header_nritems(leaf);
3435 if (nritems == 0)
3436 return 1;
3437 if (path->slots[0] == nritems)
3438 path->slots[0]--;
3439
3440 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3441 if (found_key.type == type)
3442 return 0;
3443 if (found_key.objectid < min_objectid)
3444 break;
3445 if (found_key.objectid == min_objectid &&
3446 found_key.type < type)
3447 break;
3448 }
3449 return 1;
3450}
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-24 00:56:19 -0500