Btrfs: Make defrag check nodes against the progress key to prevent repeating work
[deliverable/linux.git] / fs / btrfs / ctree.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/sched.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "transaction.h"
23 #include "print-tree.h"
24
25 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
26 *root, struct btrfs_path *path, int level);
27 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
28 *root, struct btrfs_key *ins_key,
29 struct btrfs_path *path, int data_size, int extend);
30 static int push_node_left(struct btrfs_trans_handle *trans,
31 struct btrfs_root *root, struct extent_buffer *dst,
32 struct extent_buffer *src);
33 static int balance_node_right(struct btrfs_trans_handle *trans,
34 struct btrfs_root *root,
35 struct extent_buffer *dst_buf,
36 struct extent_buffer *src_buf);
37 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
38 struct btrfs_path *path, int level, int slot);
39
40 inline void btrfs_init_path(struct btrfs_path *p)
41 {
42 memset(p, 0, sizeof(*p));
43 }
44
45 struct btrfs_path *btrfs_alloc_path(void)
46 {
47 struct btrfs_path *path;
48 path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
49 if (path) {
50 btrfs_init_path(path);
51 path->reada = 1;
52 }
53 return path;
54 }
55
56 void btrfs_free_path(struct btrfs_path *p)
57 {
58 btrfs_release_path(NULL, p);
59 kmem_cache_free(btrfs_path_cachep, p);
60 }
61
62 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
63 {
64 int i;
65 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
66 if (!p->nodes[i])
67 break;
68 free_extent_buffer(p->nodes[i]);
69 }
70 memset(p, 0, sizeof(*p));
71 }
72
73 static int __btrfs_cow_block(struct btrfs_trans_handle *trans,
74 struct btrfs_root *root,
75 struct extent_buffer *buf,
76 struct extent_buffer *parent, int parent_slot,
77 struct extent_buffer **cow_ret,
78 u64 search_start, u64 empty_size)
79 {
80 struct extent_buffer *cow;
81 int ret = 0;
82 int different_trans = 0;
83
84 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
85
86 cow = btrfs_alloc_free_block(trans, root, buf->len,
87 search_start, empty_size);
88 if (IS_ERR(cow))
89 return PTR_ERR(cow);
90
91 copy_extent_buffer(cow, buf, 0, 0, cow->len);
92 btrfs_set_header_bytenr(cow, cow->start);
93 btrfs_set_header_generation(cow, trans->transid);
94 btrfs_set_header_owner(cow, root->root_key.objectid);
95
96 WARN_ON(btrfs_header_generation(buf) > trans->transid);
97 if (btrfs_header_generation(buf) != trans->transid) {
98 different_trans = 1;
99 ret = btrfs_inc_ref(trans, root, buf);
100 if (ret)
101 return ret;
102 } else {
103 clean_tree_block(trans, root, buf);
104 }
105
106 if (buf == root->node) {
107 root->node = cow;
108 extent_buffer_get(cow);
109 if (buf != root->commit_root) {
110 btrfs_free_extent(trans, root, buf->start,
111 buf->len, 1);
112 }
113 free_extent_buffer(buf);
114 } else {
115 btrfs_set_node_blockptr(parent, parent_slot,
116 cow->start);
117 btrfs_mark_buffer_dirty(parent);
118 WARN_ON(btrfs_header_generation(parent) != trans->transid);
119 btrfs_free_extent(trans, root, buf->start, buf->len, 1);
120 }
121 free_extent_buffer(buf);
122 btrfs_mark_buffer_dirty(cow);
123 *cow_ret = cow;
124 return 0;
125 }
126
127 int btrfs_cow_block(struct btrfs_trans_handle *trans,
128 struct btrfs_root *root, struct extent_buffer *buf,
129 struct extent_buffer *parent, int parent_slot,
130 struct extent_buffer **cow_ret)
131 {
132 u64 search_start;
133 int ret;
134 if (trans->transaction != root->fs_info->running_transaction) {
135 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
136 root->fs_info->running_transaction->transid);
137 WARN_ON(1);
138 }
139 if (trans->transid != root->fs_info->generation) {
140 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
141 root->fs_info->generation);
142 WARN_ON(1);
143 }
144 if (btrfs_header_generation(buf) == trans->transid) {
145 *cow_ret = buf;
146 return 0;
147 }
148
149 search_start = buf->start & ~((u64)BTRFS_BLOCK_GROUP_SIZE - 1);
150 ret = __btrfs_cow_block(trans, root, buf, parent,
151 parent_slot, cow_ret, search_start, 0);
152 return ret;
153 }
154
155 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
156 {
157 if (blocknr < other && other - (blocknr + blocksize) < 32768)
158 return 1;
159 if (blocknr > other && blocknr - (other + blocksize) < 32768)
160 return 1;
161 return 0;
162 }
163
164 /*
165 * compare two keys in a memcmp fashion
166 */
167 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
168 {
169 struct btrfs_key k1;
170
171 btrfs_disk_key_to_cpu(&k1, disk);
172
173 if (k1.objectid > k2->objectid)
174 return 1;
175 if (k1.objectid < k2->objectid)
176 return -1;
177 if (k1.type > k2->type)
178 return 1;
179 if (k1.type < k2->type)
180 return -1;
181 if (k1.offset > k2->offset)
182 return 1;
183 if (k1.offset < k2->offset)
184 return -1;
185 return 0;
186 }
187
188
189 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
190 struct btrfs_root *root, struct extent_buffer *parent,
191 int start_slot, int cache_only, u64 *last_ret,
192 struct btrfs_key *progress)
193 {
194 struct extent_buffer *cur;
195 struct extent_buffer *tmp;
196 u64 blocknr;
197 u64 search_start = *last_ret;
198 u64 last_block = 0;
199 u64 other;
200 u32 parent_nritems;
201 int end_slot;
202 int i;
203 int err = 0;
204 int parent_level;
205 int uptodate;
206 u32 blocksize;
207 int progress_passed = 0;
208 struct btrfs_disk_key disk_key;
209
210 parent_level = btrfs_header_level(parent);
211 if (cache_only && parent_level != 1)
212 return 0;
213
214 if (trans->transaction != root->fs_info->running_transaction) {
215 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
216 root->fs_info->running_transaction->transid);
217 WARN_ON(1);
218 }
219 if (trans->transid != root->fs_info->generation) {
220 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
221 root->fs_info->generation);
222 WARN_ON(1);
223 }
224
225 parent_nritems = btrfs_header_nritems(parent);
226 blocksize = btrfs_level_size(root, parent_level - 1);
227 end_slot = parent_nritems;
228
229 if (parent_nritems == 1)
230 return 0;
231
232 for (i = start_slot; i < end_slot; i++) {
233 int close = 1;
234
235 if (!parent->map_token) {
236 map_extent_buffer(parent,
237 btrfs_node_key_ptr_offset(i),
238 sizeof(struct btrfs_key_ptr),
239 &parent->map_token, &parent->kaddr,
240 &parent->map_start, &parent->map_len,
241 KM_USER1);
242 }
243 btrfs_node_key(parent, &disk_key, i);
244 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
245 continue;
246
247 progress_passed = 1;
248 blocknr = btrfs_node_blockptr(parent, i);
249 if (last_block == 0)
250 last_block = blocknr;
251
252 if (i > 0) {
253 other = btrfs_node_blockptr(parent, i - 1);
254 close = close_blocks(blocknr, other, blocksize);
255 }
256 if (close && i < end_slot - 2) {
257 other = btrfs_node_blockptr(parent, i + 1);
258 close = close_blocks(blocknr, other, blocksize);
259 }
260 if (close) {
261 last_block = blocknr;
262 continue;
263 }
264 if (parent->map_token) {
265 unmap_extent_buffer(parent, parent->map_token,
266 KM_USER1);
267 parent->map_token = NULL;
268 }
269
270 cur = btrfs_find_tree_block(root, blocknr, blocksize);
271 if (cur)
272 uptodate = btrfs_buffer_uptodate(cur);
273 else
274 uptodate = 0;
275 if (!cur || !uptodate) {
276 if (cache_only) {
277 free_extent_buffer(cur);
278 continue;
279 }
280 if (!cur) {
281 cur = read_tree_block(root, blocknr,
282 blocksize);
283 } else if (!uptodate) {
284 btrfs_read_buffer(cur);
285 }
286 }
287 if (search_start == 0)
288 search_start = last_block;
289
290 err = __btrfs_cow_block(trans, root, cur, parent, i,
291 &tmp, search_start,
292 min(16 * blocksize,
293 (end_slot - i) * blocksize));
294 if (err) {
295 free_extent_buffer(cur);
296 break;
297 }
298 search_start = tmp->start;
299 last_block = tmp->start;
300 *last_ret = search_start;
301 if (parent_level == 1)
302 btrfs_clear_buffer_defrag(tmp);
303 free_extent_buffer(tmp);
304 }
305 if (parent->map_token) {
306 unmap_extent_buffer(parent, parent->map_token,
307 KM_USER1);
308 parent->map_token = NULL;
309 }
310 return err;
311 }
312
313 /*
314 * The leaf data grows from end-to-front in the node.
315 * this returns the address of the start of the last item,
316 * which is the stop of the leaf data stack
317 */
318 static inline unsigned int leaf_data_end(struct btrfs_root *root,
319 struct extent_buffer *leaf)
320 {
321 u32 nr = btrfs_header_nritems(leaf);
322 if (nr == 0)
323 return BTRFS_LEAF_DATA_SIZE(root);
324 return btrfs_item_offset_nr(leaf, nr - 1);
325 }
326
327 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
328 int level)
329 {
330 struct extent_buffer *parent = NULL;
331 struct extent_buffer *node = path->nodes[level];
332 struct btrfs_disk_key parent_key;
333 struct btrfs_disk_key node_key;
334 int parent_slot;
335 int slot;
336 struct btrfs_key cpukey;
337 u32 nritems = btrfs_header_nritems(node);
338
339 if (path->nodes[level + 1])
340 parent = path->nodes[level + 1];
341
342 slot = path->slots[level];
343 BUG_ON(nritems == 0);
344 if (parent) {
345 parent_slot = path->slots[level + 1];
346 btrfs_node_key(parent, &parent_key, parent_slot);
347 btrfs_node_key(node, &node_key, 0);
348 BUG_ON(memcmp(&parent_key, &node_key,
349 sizeof(struct btrfs_disk_key)));
350 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
351 btrfs_header_bytenr(node));
352 }
353 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
354 if (slot != 0) {
355 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
356 btrfs_node_key(node, &node_key, slot);
357 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
358 }
359 if (slot < nritems - 1) {
360 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
361 btrfs_node_key(node, &node_key, slot);
362 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
363 }
364 return 0;
365 }
366
367 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
368 int level)
369 {
370 struct extent_buffer *leaf = path->nodes[level];
371 struct extent_buffer *parent = NULL;
372 int parent_slot;
373 struct btrfs_key cpukey;
374 struct btrfs_disk_key parent_key;
375 struct btrfs_disk_key leaf_key;
376 int slot = path->slots[0];
377
378 u32 nritems = btrfs_header_nritems(leaf);
379
380 if (path->nodes[level + 1])
381 parent = path->nodes[level + 1];
382
383 if (nritems == 0)
384 return 0;
385
386 if (parent) {
387 parent_slot = path->slots[level + 1];
388 btrfs_node_key(parent, &parent_key, parent_slot);
389 btrfs_item_key(leaf, &leaf_key, 0);
390
391 BUG_ON(memcmp(&parent_key, &leaf_key,
392 sizeof(struct btrfs_disk_key)));
393 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
394 btrfs_header_bytenr(leaf));
395 }
396 #if 0
397 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
398 btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
399 btrfs_item_key(leaf, &leaf_key, i);
400 if (comp_keys(&leaf_key, &cpukey) >= 0) {
401 btrfs_print_leaf(root, leaf);
402 printk("slot %d offset bad key\n", i);
403 BUG_ON(1);
404 }
405 if (btrfs_item_offset_nr(leaf, i) !=
406 btrfs_item_end_nr(leaf, i + 1)) {
407 btrfs_print_leaf(root, leaf);
408 printk("slot %d offset bad\n", i);
409 BUG_ON(1);
410 }
411 if (i == 0) {
412 if (btrfs_item_offset_nr(leaf, i) +
413 btrfs_item_size_nr(leaf, i) !=
414 BTRFS_LEAF_DATA_SIZE(root)) {
415 btrfs_print_leaf(root, leaf);
416 printk("slot %d first offset bad\n", i);
417 BUG_ON(1);
418 }
419 }
420 }
421 if (nritems > 0) {
422 if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
423 btrfs_print_leaf(root, leaf);
424 printk("slot %d bad size \n", nritems - 1);
425 BUG_ON(1);
426 }
427 }
428 #endif
429 if (slot != 0 && slot < nritems - 1) {
430 btrfs_item_key(leaf, &leaf_key, slot);
431 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
432 if (comp_keys(&leaf_key, &cpukey) <= 0) {
433 btrfs_print_leaf(root, leaf);
434 printk("slot %d offset bad key\n", slot);
435 BUG_ON(1);
436 }
437 if (btrfs_item_offset_nr(leaf, slot - 1) !=
438 btrfs_item_end_nr(leaf, slot)) {
439 btrfs_print_leaf(root, leaf);
440 printk("slot %d offset bad\n", slot);
441 BUG_ON(1);
442 }
443 }
444 if (slot < nritems - 1) {
445 btrfs_item_key(leaf, &leaf_key, slot);
446 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
447 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
448 if (btrfs_item_offset_nr(leaf, slot) !=
449 btrfs_item_end_nr(leaf, slot + 1)) {
450 btrfs_print_leaf(root, leaf);
451 printk("slot %d offset bad\n", slot);
452 BUG_ON(1);
453 }
454 }
455 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
456 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
457 return 0;
458 }
459
460 static int check_block(struct btrfs_root *root, struct btrfs_path *path,
461 int level)
462 {
463 return 0;
464 #if 0
465 struct extent_buffer *buf = path->nodes[level];
466
467 if (memcmp_extent_buffer(buf, root->fs_info->fsid,
468 (unsigned long)btrfs_header_fsid(buf),
469 BTRFS_FSID_SIZE)) {
470 printk("warning bad block %Lu\n", buf->start);
471 return 1;
472 }
473 #endif
474 if (level == 0)
475 return check_leaf(root, path, level);
476 return check_node(root, path, level);
477 }
478
479 /*
480 * search for key in the extent_buffer. The items start at offset p,
481 * and they are item_size apart. There are 'max' items in p.
482 *
483 * the slot in the array is returned via slot, and it points to
484 * the place where you would insert key if it is not found in
485 * the array.
486 *
487 * slot may point to max if the key is bigger than all of the keys
488 */
489 static int generic_bin_search(struct extent_buffer *eb, unsigned long p,
490 int item_size, struct btrfs_key *key,
491 int max, int *slot)
492 {
493 int low = 0;
494 int high = max;
495 int mid;
496 int ret;
497 struct btrfs_disk_key *tmp = NULL;
498 struct btrfs_disk_key unaligned;
499 unsigned long offset;
500 char *map_token = NULL;
501 char *kaddr = NULL;
502 unsigned long map_start = 0;
503 unsigned long map_len = 0;
504 int err;
505
506 while(low < high) {
507 mid = (low + high) / 2;
508 offset = p + mid * item_size;
509
510 if (!map_token || offset < map_start ||
511 (offset + sizeof(struct btrfs_disk_key)) >
512 map_start + map_len) {
513 if (map_token) {
514 unmap_extent_buffer(eb, map_token, KM_USER0);
515 map_token = NULL;
516 }
517 err = map_extent_buffer(eb, offset,
518 sizeof(struct btrfs_disk_key),
519 &map_token, &kaddr,
520 &map_start, &map_len, KM_USER0);
521
522 if (!err) {
523 tmp = (struct btrfs_disk_key *)(kaddr + offset -
524 map_start);
525 } else {
526 read_extent_buffer(eb, &unaligned,
527 offset, sizeof(unaligned));
528 tmp = &unaligned;
529 }
530
531 } else {
532 tmp = (struct btrfs_disk_key *)(kaddr + offset -
533 map_start);
534 }
535 ret = comp_keys(tmp, key);
536
537 if (ret < 0)
538 low = mid + 1;
539 else if (ret > 0)
540 high = mid;
541 else {
542 *slot = mid;
543 if (map_token)
544 unmap_extent_buffer(eb, map_token, KM_USER0);
545 return 0;
546 }
547 }
548 *slot = low;
549 if (map_token)
550 unmap_extent_buffer(eb, map_token, KM_USER0);
551 return 1;
552 }
553
554 /*
555 * simple bin_search frontend that does the right thing for
556 * leaves vs nodes
557 */
558 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
559 int level, int *slot)
560 {
561 if (level == 0) {
562 return generic_bin_search(eb,
563 offsetof(struct btrfs_leaf, items),
564 sizeof(struct btrfs_item),
565 key, btrfs_header_nritems(eb),
566 slot);
567 } else {
568 return generic_bin_search(eb,
569 offsetof(struct btrfs_node, ptrs),
570 sizeof(struct btrfs_key_ptr),
571 key, btrfs_header_nritems(eb),
572 slot);
573 }
574 return -1;
575 }
576
577 static struct extent_buffer *read_node_slot(struct btrfs_root *root,
578 struct extent_buffer *parent, int slot)
579 {
580 if (slot < 0)
581 return NULL;
582 if (slot >= btrfs_header_nritems(parent))
583 return NULL;
584 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
585 btrfs_level_size(root, btrfs_header_level(parent) - 1));
586 }
587
588 static int balance_level(struct btrfs_trans_handle *trans, struct btrfs_root
589 *root, struct btrfs_path *path, int level)
590 {
591 struct extent_buffer *right = NULL;
592 struct extent_buffer *mid;
593 struct extent_buffer *left = NULL;
594 struct extent_buffer *parent = NULL;
595 int ret = 0;
596 int wret;
597 int pslot;
598 int orig_slot = path->slots[level];
599 int err_on_enospc = 0;
600 u64 orig_ptr;
601
602 if (level == 0)
603 return 0;
604
605 mid = path->nodes[level];
606 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
607
608 if (level < BTRFS_MAX_LEVEL - 1)
609 parent = path->nodes[level + 1];
610 pslot = path->slots[level + 1];
611
612 /*
613 * deal with the case where there is only one pointer in the root
614 * by promoting the node below to a root
615 */
616 if (!parent) {
617 struct extent_buffer *child;
618
619 if (btrfs_header_nritems(mid) != 1)
620 return 0;
621
622 /* promote the child to a root */
623 child = read_node_slot(root, mid, 0);
624 BUG_ON(!child);
625 root->node = child;
626 path->nodes[level] = NULL;
627 clean_tree_block(trans, root, mid);
628 wait_on_tree_block_writeback(root, mid);
629 /* once for the path */
630 free_extent_buffer(mid);
631 ret = btrfs_free_extent(trans, root, mid->start, mid->len, 1);
632 /* once for the root ptr */
633 free_extent_buffer(mid);
634 return ret;
635 }
636 if (btrfs_header_nritems(mid) >
637 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
638 return 0;
639
640 if (btrfs_header_nritems(mid) < 2)
641 err_on_enospc = 1;
642
643 left = read_node_slot(root, parent, pslot - 1);
644 if (left) {
645 wret = btrfs_cow_block(trans, root, left,
646 parent, pslot - 1, &left);
647 if (wret) {
648 ret = wret;
649 goto enospc;
650 }
651 }
652 right = read_node_slot(root, parent, pslot + 1);
653 if (right) {
654 wret = btrfs_cow_block(trans, root, right,
655 parent, pslot + 1, &right);
656 if (wret) {
657 ret = wret;
658 goto enospc;
659 }
660 }
661
662 /* first, try to make some room in the middle buffer */
663 if (left) {
664 orig_slot += btrfs_header_nritems(left);
665 wret = push_node_left(trans, root, left, mid);
666 if (wret < 0)
667 ret = wret;
668 if (btrfs_header_nritems(mid) < 2)
669 err_on_enospc = 1;
670 }
671
672 /*
673 * then try to empty the right most buffer into the middle
674 */
675 if (right) {
676 wret = push_node_left(trans, root, mid, right);
677 if (wret < 0 && wret != -ENOSPC)
678 ret = wret;
679 if (btrfs_header_nritems(right) == 0) {
680 u64 bytenr = right->start;
681 u32 blocksize = right->len;
682
683 clean_tree_block(trans, root, right);
684 wait_on_tree_block_writeback(root, right);
685 free_extent_buffer(right);
686 right = NULL;
687 wret = del_ptr(trans, root, path, level + 1, pslot +
688 1);
689 if (wret)
690 ret = wret;
691 wret = btrfs_free_extent(trans, root, bytenr,
692 blocksize, 1);
693 if (wret)
694 ret = wret;
695 } else {
696 struct btrfs_disk_key right_key;
697 btrfs_node_key(right, &right_key, 0);
698 btrfs_set_node_key(parent, &right_key, pslot + 1);
699 btrfs_mark_buffer_dirty(parent);
700 }
701 }
702 if (btrfs_header_nritems(mid) == 1) {
703 /*
704 * we're not allowed to leave a node with one item in the
705 * tree during a delete. A deletion from lower in the tree
706 * could try to delete the only pointer in this node.
707 * So, pull some keys from the left.
708 * There has to be a left pointer at this point because
709 * otherwise we would have pulled some pointers from the
710 * right
711 */
712 BUG_ON(!left);
713 wret = balance_node_right(trans, root, mid, left);
714 if (wret < 0) {
715 ret = wret;
716 goto enospc;
717 }
718 BUG_ON(wret == 1);
719 }
720 if (btrfs_header_nritems(mid) == 0) {
721 /* we've managed to empty the middle node, drop it */
722 u64 bytenr = mid->start;
723 u32 blocksize = mid->len;
724 clean_tree_block(trans, root, mid);
725 wait_on_tree_block_writeback(root, mid);
726 free_extent_buffer(mid);
727 mid = NULL;
728 wret = del_ptr(trans, root, path, level + 1, pslot);
729 if (wret)
730 ret = wret;
731 wret = btrfs_free_extent(trans, root, bytenr, blocksize, 1);
732 if (wret)
733 ret = wret;
734 } else {
735 /* update the parent key to reflect our changes */
736 struct btrfs_disk_key mid_key;
737 btrfs_node_key(mid, &mid_key, 0);
738 btrfs_set_node_key(parent, &mid_key, pslot);
739 btrfs_mark_buffer_dirty(parent);
740 }
741
742 /* update the path */
743 if (left) {
744 if (btrfs_header_nritems(left) > orig_slot) {
745 extent_buffer_get(left);
746 path->nodes[level] = left;
747 path->slots[level + 1] -= 1;
748 path->slots[level] = orig_slot;
749 if (mid)
750 free_extent_buffer(mid);
751 } else {
752 orig_slot -= btrfs_header_nritems(left);
753 path->slots[level] = orig_slot;
754 }
755 }
756 /* double check we haven't messed things up */
757 check_block(root, path, level);
758 if (orig_ptr !=
759 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
760 BUG();
761 enospc:
762 if (right)
763 free_extent_buffer(right);
764 if (left)
765 free_extent_buffer(left);
766 return ret;
767 }
768
769 /* returns zero if the push worked, non-zero otherwise */
770 static int push_nodes_for_insert(struct btrfs_trans_handle *trans,
771 struct btrfs_root *root,
772 struct btrfs_path *path, int level)
773 {
774 struct extent_buffer *right = NULL;
775 struct extent_buffer *mid;
776 struct extent_buffer *left = NULL;
777 struct extent_buffer *parent = NULL;
778 int ret = 0;
779 int wret;
780 int pslot;
781 int orig_slot = path->slots[level];
782 u64 orig_ptr;
783
784 if (level == 0)
785 return 1;
786
787 mid = path->nodes[level];
788 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
789
790 if (level < BTRFS_MAX_LEVEL - 1)
791 parent = path->nodes[level + 1];
792 pslot = path->slots[level + 1];
793
794 if (!parent)
795 return 1;
796
797 left = read_node_slot(root, parent, pslot - 1);
798
799 /* first, try to make some room in the middle buffer */
800 if (left) {
801 u32 left_nr;
802 left_nr = btrfs_header_nritems(left);
803 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
804 wret = 1;
805 } else {
806 ret = btrfs_cow_block(trans, root, left, parent,
807 pslot - 1, &left);
808 if (ret)
809 wret = 1;
810 else {
811 wret = push_node_left(trans, root,
812 left, mid);
813 }
814 }
815 if (wret < 0)
816 ret = wret;
817 if (wret == 0) {
818 struct btrfs_disk_key disk_key;
819 orig_slot += left_nr;
820 btrfs_node_key(mid, &disk_key, 0);
821 btrfs_set_node_key(parent, &disk_key, pslot);
822 btrfs_mark_buffer_dirty(parent);
823 if (btrfs_header_nritems(left) > orig_slot) {
824 path->nodes[level] = left;
825 path->slots[level + 1] -= 1;
826 path->slots[level] = orig_slot;
827 free_extent_buffer(mid);
828 } else {
829 orig_slot -=
830 btrfs_header_nritems(left);
831 path->slots[level] = orig_slot;
832 free_extent_buffer(left);
833 }
834 return 0;
835 }
836 free_extent_buffer(left);
837 }
838 right= read_node_slot(root, parent, pslot + 1);
839
840 /*
841 * then try to empty the right most buffer into the middle
842 */
843 if (right) {
844 u32 right_nr;
845 right_nr = btrfs_header_nritems(right);
846 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
847 wret = 1;
848 } else {
849 ret = btrfs_cow_block(trans, root, right,
850 parent, pslot + 1,
851 &right);
852 if (ret)
853 wret = 1;
854 else {
855 wret = balance_node_right(trans, root,
856 right, mid);
857 }
858 }
859 if (wret < 0)
860 ret = wret;
861 if (wret == 0) {
862 struct btrfs_disk_key disk_key;
863
864 btrfs_node_key(right, &disk_key, 0);
865 btrfs_set_node_key(parent, &disk_key, pslot + 1);
866 btrfs_mark_buffer_dirty(parent);
867
868 if (btrfs_header_nritems(mid) <= orig_slot) {
869 path->nodes[level] = right;
870 path->slots[level + 1] += 1;
871 path->slots[level] = orig_slot -
872 btrfs_header_nritems(mid);
873 free_extent_buffer(mid);
874 } else {
875 free_extent_buffer(right);
876 }
877 return 0;
878 }
879 free_extent_buffer(right);
880 }
881 return 1;
882 }
883
884 /*
885 * readahead one full node of leaves
886 */
887 static void reada_for_search(struct btrfs_root *root, struct btrfs_path *path,
888 int level, int slot)
889 {
890 struct extent_buffer *node;
891 u32 nritems;
892 u64 search;
893 u64 lowest_read;
894 u64 highest_read;
895 u64 nread = 0;
896 int direction = path->reada;
897 struct extent_buffer *eb;
898 u32 nr;
899 u32 blocksize;
900 u32 nscan = 0;
901
902 if (level != 1)
903 return;
904
905 if (!path->nodes[level])
906 return;
907
908 node = path->nodes[level];
909 search = btrfs_node_blockptr(node, slot);
910 blocksize = btrfs_level_size(root, level - 1);
911 eb = btrfs_find_tree_block(root, search, blocksize);
912 if (eb) {
913 free_extent_buffer(eb);
914 return;
915 }
916
917 highest_read = search;
918 lowest_read = search;
919
920 nritems = btrfs_header_nritems(node);
921 nr = slot;
922 while(1) {
923 if (direction < 0) {
924 if (nr == 0)
925 break;
926 nr--;
927 } else if (direction > 0) {
928 nr++;
929 if (nr >= nritems)
930 break;
931 }
932 search = btrfs_node_blockptr(node, nr);
933 if ((search >= lowest_read && search <= highest_read) ||
934 (search < lowest_read && lowest_read - search <= 32768) ||
935 (search > highest_read && search - highest_read <= 32768)) {
936 readahead_tree_block(root, search, blocksize);
937 nread += blocksize;
938 }
939 nscan++;
940 if (path->reada < 2 && (nread > (256 * 1024) || nscan > 32))
941 break;
942 if(nread > (1024 * 1024) || nscan > 128)
943 break;
944
945 if (search < lowest_read)
946 lowest_read = search;
947 if (search > highest_read)
948 highest_read = search;
949 }
950 }
951 /*
952 * look for key in the tree. path is filled in with nodes along the way
953 * if key is found, we return zero and you can find the item in the leaf
954 * level of the path (level 0)
955 *
956 * If the key isn't found, the path points to the slot where it should
957 * be inserted, and 1 is returned. If there are other errors during the
958 * search a negative error number is returned.
959 *
960 * if ins_len > 0, nodes and leaves will be split as we walk down the
961 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
962 * possible)
963 */
964 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
965 *root, struct btrfs_key *key, struct btrfs_path *p, int
966 ins_len, int cow)
967 {
968 struct extent_buffer *b;
969 u64 bytenr;
970 int slot;
971 int ret;
972 int level;
973 int should_reada = p->reada;
974 u8 lowest_level = 0;
975
976 lowest_level = p->lowest_level;
977 WARN_ON(lowest_level && ins_len);
978 WARN_ON(p->nodes[0] != NULL);
979 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
980 again:
981 b = root->node;
982 extent_buffer_get(b);
983 while (b) {
984 level = btrfs_header_level(b);
985 if (cow) {
986 int wret;
987 wret = btrfs_cow_block(trans, root, b,
988 p->nodes[level + 1],
989 p->slots[level + 1],
990 &b);
991 if (wret) {
992 free_extent_buffer(b);
993 return wret;
994 }
995 }
996 BUG_ON(!cow && ins_len);
997 if (level != btrfs_header_level(b))
998 WARN_ON(1);
999 level = btrfs_header_level(b);
1000 p->nodes[level] = b;
1001 ret = check_block(root, p, level);
1002 if (ret)
1003 return -1;
1004 ret = bin_search(b, key, level, &slot);
1005 if (level != 0) {
1006 if (ret && slot > 0)
1007 slot -= 1;
1008 p->slots[level] = slot;
1009 if (ins_len > 0 && btrfs_header_nritems(b) >=
1010 BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1011 int sret = split_node(trans, root, p, level);
1012 BUG_ON(sret > 0);
1013 if (sret)
1014 return sret;
1015 b = p->nodes[level];
1016 slot = p->slots[level];
1017 } else if (ins_len < 0) {
1018 int sret = balance_level(trans, root, p,
1019 level);
1020 if (sret)
1021 return sret;
1022 b = p->nodes[level];
1023 if (!b) {
1024 btrfs_release_path(NULL, p);
1025 goto again;
1026 }
1027 slot = p->slots[level];
1028 BUG_ON(btrfs_header_nritems(b) == 1);
1029 }
1030 /* this is only true while dropping a snapshot */
1031 if (level == lowest_level)
1032 break;
1033 bytenr = btrfs_node_blockptr(b, slot);
1034 if (should_reada)
1035 reada_for_search(root, p, level, slot);
1036 b = read_tree_block(root, bytenr,
1037 btrfs_level_size(root, level - 1));
1038 } else {
1039 p->slots[level] = slot;
1040 if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
1041 sizeof(struct btrfs_item) + ins_len) {
1042 int sret = split_leaf(trans, root, key,
1043 p, ins_len, ret == 0);
1044 BUG_ON(sret > 0);
1045 if (sret)
1046 return sret;
1047 }
1048 return ret;
1049 }
1050 }
1051 return 1;
1052 }
1053
1054 /*
1055 * adjust the pointers going up the tree, starting at level
1056 * making sure the right key of each node is points to 'key'.
1057 * This is used after shifting pointers to the left, so it stops
1058 * fixing up pointers when a given leaf/node is not in slot 0 of the
1059 * higher levels
1060 *
1061 * If this fails to write a tree block, it returns -1, but continues
1062 * fixing up the blocks in ram so the tree is consistent.
1063 */
1064 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1065 struct btrfs_root *root, struct btrfs_path *path,
1066 struct btrfs_disk_key *key, int level)
1067 {
1068 int i;
1069 int ret = 0;
1070 struct extent_buffer *t;
1071
1072 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1073 int tslot = path->slots[i];
1074 if (!path->nodes[i])
1075 break;
1076 t = path->nodes[i];
1077 btrfs_set_node_key(t, key, tslot);
1078 btrfs_mark_buffer_dirty(path->nodes[i]);
1079 if (tslot != 0)
1080 break;
1081 }
1082 return ret;
1083 }
1084
1085 /*
1086 * try to push data from one node into the next node left in the
1087 * tree.
1088 *
1089 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1090 * error, and > 0 if there was no room in the left hand block.
1091 */
1092 static int push_node_left(struct btrfs_trans_handle *trans, struct btrfs_root
1093 *root, struct extent_buffer *dst,
1094 struct extent_buffer *src)
1095 {
1096 int push_items = 0;
1097 int src_nritems;
1098 int dst_nritems;
1099 int ret = 0;
1100
1101 src_nritems = btrfs_header_nritems(src);
1102 dst_nritems = btrfs_header_nritems(dst);
1103 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1104
1105 if (push_items <= 0) {
1106 return 1;
1107 }
1108
1109 if (src_nritems < push_items)
1110 push_items = src_nritems;
1111
1112 copy_extent_buffer(dst, src,
1113 btrfs_node_key_ptr_offset(dst_nritems),
1114 btrfs_node_key_ptr_offset(0),
1115 push_items * sizeof(struct btrfs_key_ptr));
1116
1117 if (push_items < src_nritems) {
1118 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1119 btrfs_node_key_ptr_offset(push_items),
1120 (src_nritems - push_items) *
1121 sizeof(struct btrfs_key_ptr));
1122 }
1123 btrfs_set_header_nritems(src, src_nritems - push_items);
1124 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1125 btrfs_mark_buffer_dirty(src);
1126 btrfs_mark_buffer_dirty(dst);
1127 return ret;
1128 }
1129
1130 /*
1131 * try to push data from one node into the next node right in the
1132 * tree.
1133 *
1134 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1135 * error, and > 0 if there was no room in the right hand block.
1136 *
1137 * this will only push up to 1/2 the contents of the left node over
1138 */
1139 static int balance_node_right(struct btrfs_trans_handle *trans,
1140 struct btrfs_root *root,
1141 struct extent_buffer *dst,
1142 struct extent_buffer *src)
1143 {
1144 int push_items = 0;
1145 int max_push;
1146 int src_nritems;
1147 int dst_nritems;
1148 int ret = 0;
1149
1150 src_nritems = btrfs_header_nritems(src);
1151 dst_nritems = btrfs_header_nritems(dst);
1152 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1153 if (push_items <= 0)
1154 return 1;
1155
1156 max_push = src_nritems / 2 + 1;
1157 /* don't try to empty the node */
1158 if (max_push >= src_nritems)
1159 return 1;
1160
1161 if (max_push < push_items)
1162 push_items = max_push;
1163
1164 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1165 btrfs_node_key_ptr_offset(0),
1166 (dst_nritems) *
1167 sizeof(struct btrfs_key_ptr));
1168
1169 copy_extent_buffer(dst, src,
1170 btrfs_node_key_ptr_offset(0),
1171 btrfs_node_key_ptr_offset(src_nritems - push_items),
1172 push_items * sizeof(struct btrfs_key_ptr));
1173
1174 btrfs_set_header_nritems(src, src_nritems - push_items);
1175 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1176
1177 btrfs_mark_buffer_dirty(src);
1178 btrfs_mark_buffer_dirty(dst);
1179 return ret;
1180 }
1181
1182 /*
1183 * helper function to insert a new root level in the tree.
1184 * A new node is allocated, and a single item is inserted to
1185 * point to the existing root
1186 *
1187 * returns zero on success or < 0 on failure.
1188 */
1189 static int insert_new_root(struct btrfs_trans_handle *trans,
1190 struct btrfs_root *root,
1191 struct btrfs_path *path, int level)
1192 {
1193 struct extent_buffer *lower;
1194 struct extent_buffer *c;
1195 struct btrfs_disk_key lower_key;
1196
1197 BUG_ON(path->nodes[level]);
1198 BUG_ON(path->nodes[level-1] != root->node);
1199
1200 c = btrfs_alloc_free_block(trans, root, root->nodesize,
1201 root->node->start, 0);
1202 if (IS_ERR(c))
1203 return PTR_ERR(c);
1204 memset_extent_buffer(c, 0, 0, root->nodesize);
1205 btrfs_set_header_nritems(c, 1);
1206 btrfs_set_header_level(c, level);
1207 btrfs_set_header_bytenr(c, c->start);
1208 btrfs_set_header_generation(c, trans->transid);
1209 btrfs_set_header_owner(c, root->root_key.objectid);
1210 lower = path->nodes[level-1];
1211
1212 write_extent_buffer(c, root->fs_info->fsid,
1213 (unsigned long)btrfs_header_fsid(c),
1214 BTRFS_FSID_SIZE);
1215 if (level == 1)
1216 btrfs_item_key(lower, &lower_key, 0);
1217 else
1218 btrfs_node_key(lower, &lower_key, 0);
1219 btrfs_set_node_key(c, &lower_key, 0);
1220 btrfs_set_node_blockptr(c, 0, lower->start);
1221
1222 btrfs_mark_buffer_dirty(c);
1223
1224 /* the super has an extra ref to root->node */
1225 free_extent_buffer(root->node);
1226 root->node = c;
1227 extent_buffer_get(c);
1228 path->nodes[level] = c;
1229 path->slots[level] = 0;
1230 return 0;
1231 }
1232
1233 /*
1234 * worker function to insert a single pointer in a node.
1235 * the node should have enough room for the pointer already
1236 *
1237 * slot and level indicate where you want the key to go, and
1238 * blocknr is the block the key points to.
1239 *
1240 * returns zero on success and < 0 on any error
1241 */
1242 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
1243 *root, struct btrfs_path *path, struct btrfs_disk_key
1244 *key, u64 bytenr, int slot, int level)
1245 {
1246 struct extent_buffer *lower;
1247 int nritems;
1248
1249 BUG_ON(!path->nodes[level]);
1250 lower = path->nodes[level];
1251 nritems = btrfs_header_nritems(lower);
1252 if (slot > nritems)
1253 BUG();
1254 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
1255 BUG();
1256 if (slot != nritems) {
1257 memmove_extent_buffer(lower,
1258 btrfs_node_key_ptr_offset(slot + 1),
1259 btrfs_node_key_ptr_offset(slot),
1260 (nritems - slot) * sizeof(struct btrfs_key_ptr));
1261 }
1262 btrfs_set_node_key(lower, key, slot);
1263 btrfs_set_node_blockptr(lower, slot, bytenr);
1264 btrfs_set_header_nritems(lower, nritems + 1);
1265 btrfs_mark_buffer_dirty(lower);
1266 return 0;
1267 }
1268
1269 /*
1270 * split the node at the specified level in path in two.
1271 * The path is corrected to point to the appropriate node after the split
1272 *
1273 * Before splitting this tries to make some room in the node by pushing
1274 * left and right, if either one works, it returns right away.
1275 *
1276 * returns 0 on success and < 0 on failure
1277 */
1278 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
1279 *root, struct btrfs_path *path, int level)
1280 {
1281 struct extent_buffer *c;
1282 struct extent_buffer *split;
1283 struct btrfs_disk_key disk_key;
1284 int mid;
1285 int ret;
1286 int wret;
1287 u32 c_nritems;
1288
1289 c = path->nodes[level];
1290 if (c == root->node) {
1291 /* trying to split the root, lets make a new one */
1292 ret = insert_new_root(trans, root, path, level + 1);
1293 if (ret)
1294 return ret;
1295 } else {
1296 ret = push_nodes_for_insert(trans, root, path, level);
1297 c = path->nodes[level];
1298 if (!ret && btrfs_header_nritems(c) <
1299 BTRFS_NODEPTRS_PER_BLOCK(root) - 1)
1300 return 0;
1301 if (ret < 0)
1302 return ret;
1303 }
1304
1305 c_nritems = btrfs_header_nritems(c);
1306 split = btrfs_alloc_free_block(trans, root, root->nodesize,
1307 c->start, 0);
1308 if (IS_ERR(split))
1309 return PTR_ERR(split);
1310
1311 btrfs_set_header_flags(split, btrfs_header_flags(c));
1312 btrfs_set_header_level(split, btrfs_header_level(c));
1313 btrfs_set_header_bytenr(split, split->start);
1314 btrfs_set_header_generation(split, trans->transid);
1315 btrfs_set_header_owner(split, root->root_key.objectid);
1316 write_extent_buffer(split, root->fs_info->fsid,
1317 (unsigned long)btrfs_header_fsid(split),
1318 BTRFS_FSID_SIZE);
1319
1320 mid = (c_nritems + 1) / 2;
1321
1322 copy_extent_buffer(split, c,
1323 btrfs_node_key_ptr_offset(0),
1324 btrfs_node_key_ptr_offset(mid),
1325 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
1326 btrfs_set_header_nritems(split, c_nritems - mid);
1327 btrfs_set_header_nritems(c, mid);
1328 ret = 0;
1329
1330 btrfs_mark_buffer_dirty(c);
1331 btrfs_mark_buffer_dirty(split);
1332
1333 btrfs_node_key(split, &disk_key, 0);
1334 wret = insert_ptr(trans, root, path, &disk_key, split->start,
1335 path->slots[level + 1] + 1,
1336 level + 1);
1337 if (wret)
1338 ret = wret;
1339
1340 if (path->slots[level] >= mid) {
1341 path->slots[level] -= mid;
1342 free_extent_buffer(c);
1343 path->nodes[level] = split;
1344 path->slots[level + 1] += 1;
1345 } else {
1346 free_extent_buffer(split);
1347 }
1348 return ret;
1349 }
1350
1351 /*
1352 * how many bytes are required to store the items in a leaf. start
1353 * and nr indicate which items in the leaf to check. This totals up the
1354 * space used both by the item structs and the item data
1355 */
1356 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
1357 {
1358 int data_len;
1359 int nritems = btrfs_header_nritems(l);
1360 int end = min(nritems, start + nr) - 1;
1361
1362 if (!nr)
1363 return 0;
1364 data_len = btrfs_item_end_nr(l, start);
1365 data_len = data_len - btrfs_item_offset_nr(l, end);
1366 data_len += sizeof(struct btrfs_item) * nr;
1367 WARN_ON(data_len < 0);
1368 return data_len;
1369 }
1370
1371 /*
1372 * The space between the end of the leaf items and
1373 * the start of the leaf data. IOW, how much room
1374 * the leaf has left for both items and data
1375 */
1376 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf)
1377 {
1378 int nritems = btrfs_header_nritems(leaf);
1379 int ret;
1380 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
1381 if (ret < 0) {
1382 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1383 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
1384 leaf_space_used(leaf, 0, nritems), nritems);
1385 }
1386 return ret;
1387 }
1388
1389 /*
1390 * push some data in the path leaf to the right, trying to free up at
1391 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1392 *
1393 * returns 1 if the push failed because the other node didn't have enough
1394 * room, 0 if everything worked out and < 0 if there were major errors.
1395 */
1396 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
1397 *root, struct btrfs_path *path, int data_size)
1398 {
1399 struct extent_buffer *left = path->nodes[0];
1400 struct extent_buffer *right;
1401 struct extent_buffer *upper;
1402 struct btrfs_disk_key disk_key;
1403 int slot;
1404 int i;
1405 int free_space;
1406 int push_space = 0;
1407 int push_items = 0;
1408 struct btrfs_item *item;
1409 u32 left_nritems;
1410 u32 right_nritems;
1411 u32 data_end;
1412 u32 this_item_size;
1413 int ret;
1414
1415 slot = path->slots[1];
1416 if (!path->nodes[1]) {
1417 return 1;
1418 }
1419 upper = path->nodes[1];
1420 if (slot >= btrfs_header_nritems(upper) - 1)
1421 return 1;
1422
1423 right = read_tree_block(root, btrfs_node_blockptr(upper, slot + 1),
1424 root->leafsize);
1425 free_space = btrfs_leaf_free_space(root, right);
1426 if (free_space < data_size + sizeof(struct btrfs_item)) {
1427 free_extent_buffer(right);
1428 return 1;
1429 }
1430
1431 /* cow and double check */
1432 ret = btrfs_cow_block(trans, root, right, upper,
1433 slot + 1, &right);
1434 if (ret) {
1435 free_extent_buffer(right);
1436 return 1;
1437 }
1438 free_space = btrfs_leaf_free_space(root, right);
1439 if (free_space < data_size + sizeof(struct btrfs_item)) {
1440 free_extent_buffer(right);
1441 return 1;
1442 }
1443
1444 left_nritems = btrfs_header_nritems(left);
1445 if (left_nritems == 0) {
1446 free_extent_buffer(right);
1447 return 1;
1448 }
1449
1450 for (i = left_nritems - 1; i >= 1; i--) {
1451 item = btrfs_item_nr(left, i);
1452
1453 if (path->slots[0] == i)
1454 push_space += data_size + sizeof(*item);
1455
1456 if (!left->map_token) {
1457 map_extent_buffer(left, (unsigned long)item,
1458 sizeof(struct btrfs_item),
1459 &left->map_token, &left->kaddr,
1460 &left->map_start, &left->map_len,
1461 KM_USER1);
1462 }
1463
1464 this_item_size = btrfs_item_size(left, item);
1465 if (this_item_size + sizeof(*item) + push_space > free_space)
1466 break;
1467 push_items++;
1468 push_space += this_item_size + sizeof(*item);
1469 }
1470 if (left->map_token) {
1471 unmap_extent_buffer(left, left->map_token, KM_USER1);
1472 left->map_token = NULL;
1473 }
1474
1475 if (push_items == 0) {
1476 free_extent_buffer(right);
1477 return 1;
1478 }
1479
1480 if (push_items == left_nritems)
1481 WARN_ON(1);
1482
1483 /* push left to right */
1484 right_nritems = btrfs_header_nritems(right);
1485 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
1486 push_space -= leaf_data_end(root, left);
1487
1488 /* make room in the right data area */
1489 data_end = leaf_data_end(root, right);
1490 memmove_extent_buffer(right,
1491 btrfs_leaf_data(right) + data_end - push_space,
1492 btrfs_leaf_data(right) + data_end,
1493 BTRFS_LEAF_DATA_SIZE(root) - data_end);
1494
1495 /* copy from the left data area */
1496 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
1497 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1498 btrfs_leaf_data(left) + leaf_data_end(root, left),
1499 push_space);
1500
1501 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
1502 btrfs_item_nr_offset(0),
1503 right_nritems * sizeof(struct btrfs_item));
1504
1505 /* copy the items from left to right */
1506 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
1507 btrfs_item_nr_offset(left_nritems - push_items),
1508 push_items * sizeof(struct btrfs_item));
1509
1510 /* update the item pointers */
1511 right_nritems += push_items;
1512 btrfs_set_header_nritems(right, right_nritems);
1513 push_space = BTRFS_LEAF_DATA_SIZE(root);
1514
1515 for (i = 0; i < right_nritems; i++) {
1516 item = btrfs_item_nr(right, i);
1517 if (!right->map_token) {
1518 map_extent_buffer(right, (unsigned long)item,
1519 sizeof(struct btrfs_item),
1520 &right->map_token, &right->kaddr,
1521 &right->map_start, &right->map_len,
1522 KM_USER1);
1523 }
1524 push_space -= btrfs_item_size(right, item);
1525 btrfs_set_item_offset(right, item, push_space);
1526 }
1527
1528 if (right->map_token) {
1529 unmap_extent_buffer(right, right->map_token, KM_USER1);
1530 right->map_token = NULL;
1531 }
1532 left_nritems -= push_items;
1533 btrfs_set_header_nritems(left, left_nritems);
1534
1535 btrfs_mark_buffer_dirty(left);
1536 btrfs_mark_buffer_dirty(right);
1537
1538 btrfs_item_key(right, &disk_key, 0);
1539 btrfs_set_node_key(upper, &disk_key, slot + 1);
1540 btrfs_mark_buffer_dirty(upper);
1541
1542 /* then fixup the leaf pointer in the path */
1543 if (path->slots[0] >= left_nritems) {
1544 path->slots[0] -= left_nritems;
1545 free_extent_buffer(path->nodes[0]);
1546 path->nodes[0] = right;
1547 path->slots[1] += 1;
1548 } else {
1549 free_extent_buffer(right);
1550 }
1551 return 0;
1552 }
1553 /*
1554 * push some data in the path leaf to the left, trying to free up at
1555 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1556 */
1557 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1558 *root, struct btrfs_path *path, int data_size)
1559 {
1560 struct btrfs_disk_key disk_key;
1561 struct extent_buffer *right = path->nodes[0];
1562 struct extent_buffer *left;
1563 int slot;
1564 int i;
1565 int free_space;
1566 int push_space = 0;
1567 int push_items = 0;
1568 struct btrfs_item *item;
1569 u32 old_left_nritems;
1570 u32 right_nritems;
1571 int ret = 0;
1572 int wret;
1573 u32 this_item_size;
1574 u32 old_left_item_size;
1575
1576 slot = path->slots[1];
1577 if (slot == 0)
1578 return 1;
1579 if (!path->nodes[1])
1580 return 1;
1581
1582 right_nritems = btrfs_header_nritems(right);
1583 if (right_nritems == 0) {
1584 return 1;
1585 }
1586
1587 left = read_tree_block(root, btrfs_node_blockptr(path->nodes[1],
1588 slot - 1), root->leafsize);
1589 free_space = btrfs_leaf_free_space(root, left);
1590 if (free_space < data_size + sizeof(struct btrfs_item)) {
1591 free_extent_buffer(left);
1592 return 1;
1593 }
1594
1595 /* cow and double check */
1596 ret = btrfs_cow_block(trans, root, left,
1597 path->nodes[1], slot - 1, &left);
1598 if (ret) {
1599 /* we hit -ENOSPC, but it isn't fatal here */
1600 free_extent_buffer(left);
1601 return 1;
1602 }
1603
1604 free_space = btrfs_leaf_free_space(root, left);
1605 if (free_space < data_size + sizeof(struct btrfs_item)) {
1606 free_extent_buffer(left);
1607 return 1;
1608 }
1609
1610 for (i = 0; i < right_nritems - 1; i++) {
1611 item = btrfs_item_nr(right, i);
1612 if (!right->map_token) {
1613 map_extent_buffer(right, (unsigned long)item,
1614 sizeof(struct btrfs_item),
1615 &right->map_token, &right->kaddr,
1616 &right->map_start, &right->map_len,
1617 KM_USER1);
1618 }
1619
1620 if (path->slots[0] == i)
1621 push_space += data_size + sizeof(*item);
1622
1623 this_item_size = btrfs_item_size(right, item);
1624 if (this_item_size + sizeof(*item) + push_space > free_space)
1625 break;
1626
1627 push_items++;
1628 push_space += this_item_size + sizeof(*item);
1629 }
1630
1631 if (right->map_token) {
1632 unmap_extent_buffer(right, right->map_token, KM_USER1);
1633 right->map_token = NULL;
1634 }
1635
1636 if (push_items == 0) {
1637 free_extent_buffer(left);
1638 return 1;
1639 }
1640 if (push_items == btrfs_header_nritems(right))
1641 WARN_ON(1);
1642
1643 /* push data from right to left */
1644 copy_extent_buffer(left, right,
1645 btrfs_item_nr_offset(btrfs_header_nritems(left)),
1646 btrfs_item_nr_offset(0),
1647 push_items * sizeof(struct btrfs_item));
1648
1649 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1650 btrfs_item_offset_nr(right, push_items -1);
1651
1652 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
1653 leaf_data_end(root, left) - push_space,
1654 btrfs_leaf_data(right) +
1655 btrfs_item_offset_nr(right, push_items - 1),
1656 push_space);
1657 old_left_nritems = btrfs_header_nritems(left);
1658 BUG_ON(old_left_nritems < 0);
1659
1660 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
1661 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1662 u32 ioff;
1663
1664 item = btrfs_item_nr(left, i);
1665 if (!left->map_token) {
1666 map_extent_buffer(left, (unsigned long)item,
1667 sizeof(struct btrfs_item),
1668 &left->map_token, &left->kaddr,
1669 &left->map_start, &left->map_len,
1670 KM_USER1);
1671 }
1672
1673 ioff = btrfs_item_offset(left, item);
1674 btrfs_set_item_offset(left, item,
1675 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
1676 }
1677 btrfs_set_header_nritems(left, old_left_nritems + push_items);
1678 if (left->map_token) {
1679 unmap_extent_buffer(left, left->map_token, KM_USER1);
1680 left->map_token = NULL;
1681 }
1682
1683 /* fixup right node */
1684 push_space = btrfs_item_offset_nr(right, push_items - 1) -
1685 leaf_data_end(root, right);
1686 memmove_extent_buffer(right, btrfs_leaf_data(right) +
1687 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1688 btrfs_leaf_data(right) +
1689 leaf_data_end(root, right), push_space);
1690
1691 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
1692 btrfs_item_nr_offset(push_items),
1693 (btrfs_header_nritems(right) - push_items) *
1694 sizeof(struct btrfs_item));
1695
1696 right_nritems = btrfs_header_nritems(right) - push_items;
1697 btrfs_set_header_nritems(right, right_nritems);
1698 push_space = BTRFS_LEAF_DATA_SIZE(root);
1699
1700 for (i = 0; i < right_nritems; i++) {
1701 item = btrfs_item_nr(right, i);
1702
1703 if (!right->map_token) {
1704 map_extent_buffer(right, (unsigned long)item,
1705 sizeof(struct btrfs_item),
1706 &right->map_token, &right->kaddr,
1707 &right->map_start, &right->map_len,
1708 KM_USER1);
1709 }
1710
1711 push_space = push_space - btrfs_item_size(right, item);
1712 btrfs_set_item_offset(right, item, push_space);
1713 }
1714 if (right->map_token) {
1715 unmap_extent_buffer(right, right->map_token, KM_USER1);
1716 right->map_token = NULL;
1717 }
1718
1719 btrfs_mark_buffer_dirty(left);
1720 btrfs_mark_buffer_dirty(right);
1721
1722 btrfs_item_key(right, &disk_key, 0);
1723 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
1724 if (wret)
1725 ret = wret;
1726
1727 /* then fixup the leaf pointer in the path */
1728 if (path->slots[0] < push_items) {
1729 path->slots[0] += old_left_nritems;
1730 free_extent_buffer(path->nodes[0]);
1731 path->nodes[0] = left;
1732 path->slots[1] -= 1;
1733 } else {
1734 free_extent_buffer(left);
1735 path->slots[0] -= push_items;
1736 }
1737 BUG_ON(path->slots[0] < 0);
1738 return ret;
1739 }
1740
1741 /*
1742 * split the path's leaf in two, making sure there is at least data_size
1743 * available for the resulting leaf level of the path.
1744 *
1745 * returns 0 if all went well and < 0 on failure.
1746 */
1747 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
1748 *root, struct btrfs_key *ins_key,
1749 struct btrfs_path *path, int data_size, int extend)
1750 {
1751 struct extent_buffer *l;
1752 u32 nritems;
1753 int mid;
1754 int slot;
1755 struct extent_buffer *right;
1756 int space_needed = data_size + sizeof(struct btrfs_item);
1757 int data_copy_size;
1758 int rt_data_off;
1759 int i;
1760 int ret = 0;
1761 int wret;
1762 int double_split;
1763 int num_doubles = 0;
1764 struct btrfs_disk_key disk_key;
1765
1766 if (extend)
1767 space_needed = data_size;
1768
1769 /* first try to make some room by pushing left and right */
1770 if (ins_key->type != BTRFS_DIR_ITEM_KEY) {
1771 wret = push_leaf_right(trans, root, path, data_size);
1772 if (wret < 0) {
1773 return wret;
1774 }
1775 if (wret) {
1776 wret = push_leaf_left(trans, root, path, data_size);
1777 if (wret < 0)
1778 return wret;
1779 }
1780 l = path->nodes[0];
1781
1782 /* did the pushes work? */
1783 if (btrfs_leaf_free_space(root, l) >= space_needed)
1784 return 0;
1785 }
1786
1787 if (!path->nodes[1]) {
1788 ret = insert_new_root(trans, root, path, 1);
1789 if (ret)
1790 return ret;
1791 }
1792 again:
1793 double_split = 0;
1794 l = path->nodes[0];
1795 slot = path->slots[0];
1796 nritems = btrfs_header_nritems(l);
1797 mid = (nritems + 1)/ 2;
1798
1799 right = btrfs_alloc_free_block(trans, root, root->leafsize,
1800 l->start, 0);
1801 if (IS_ERR(right))
1802 return PTR_ERR(right);
1803
1804 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
1805 btrfs_set_header_bytenr(right, right->start);
1806 btrfs_set_header_generation(right, trans->transid);
1807 btrfs_set_header_owner(right, root->root_key.objectid);
1808 btrfs_set_header_level(right, 0);
1809 write_extent_buffer(right, root->fs_info->fsid,
1810 (unsigned long)btrfs_header_fsid(right),
1811 BTRFS_FSID_SIZE);
1812 if (mid <= slot) {
1813 if (nritems == 1 ||
1814 leaf_space_used(l, mid, nritems - mid) + space_needed >
1815 BTRFS_LEAF_DATA_SIZE(root)) {
1816 if (slot >= nritems) {
1817 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1818 btrfs_set_header_nritems(right, 0);
1819 wret = insert_ptr(trans, root, path,
1820 &disk_key, right->start,
1821 path->slots[1] + 1, 1);
1822 if (wret)
1823 ret = wret;
1824 free_extent_buffer(path->nodes[0]);
1825 path->nodes[0] = right;
1826 path->slots[0] = 0;
1827 path->slots[1] += 1;
1828 return ret;
1829 }
1830 mid = slot;
1831 if (mid != nritems &&
1832 leaf_space_used(l, mid, nritems - mid) +
1833 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
1834 double_split = 1;
1835 }
1836 }
1837 } else {
1838 if (leaf_space_used(l, 0, mid + 1) + space_needed >
1839 BTRFS_LEAF_DATA_SIZE(root)) {
1840 if (!extend && slot == 0) {
1841 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1842 btrfs_set_header_nritems(right, 0);
1843 wret = insert_ptr(trans, root, path,
1844 &disk_key,
1845 right->start,
1846 path->slots[1], 1);
1847 if (wret)
1848 ret = wret;
1849 free_extent_buffer(path->nodes[0]);
1850 path->nodes[0] = right;
1851 path->slots[0] = 0;
1852 if (path->slots[1] == 0) {
1853 wret = fixup_low_keys(trans, root,
1854 path, &disk_key, 1);
1855 if (wret)
1856 ret = wret;
1857 }
1858 return ret;
1859 } else if (extend && slot == 0) {
1860 mid = 1;
1861 } else {
1862 mid = slot;
1863 if (mid != nritems &&
1864 leaf_space_used(l, mid, nritems - mid) +
1865 space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
1866 double_split = 1;
1867 }
1868 }
1869 }
1870 }
1871 nritems = nritems - mid;
1872 btrfs_set_header_nritems(right, nritems);
1873 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
1874
1875 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
1876 btrfs_item_nr_offset(mid),
1877 nritems * sizeof(struct btrfs_item));
1878
1879 copy_extent_buffer(right, l,
1880 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
1881 data_copy_size, btrfs_leaf_data(l) +
1882 leaf_data_end(root, l), data_copy_size);
1883
1884 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
1885 btrfs_item_end_nr(l, mid);
1886
1887 for (i = 0; i < nritems; i++) {
1888 struct btrfs_item *item = btrfs_item_nr(right, i);
1889 u32 ioff;
1890
1891 if (!right->map_token) {
1892 map_extent_buffer(right, (unsigned long)item,
1893 sizeof(struct btrfs_item),
1894 &right->map_token, &right->kaddr,
1895 &right->map_start, &right->map_len,
1896 KM_USER1);
1897 }
1898
1899 ioff = btrfs_item_offset(right, item);
1900 btrfs_set_item_offset(right, item, ioff + rt_data_off);
1901 }
1902
1903 if (right->map_token) {
1904 unmap_extent_buffer(right, right->map_token, KM_USER1);
1905 right->map_token = NULL;
1906 }
1907
1908 btrfs_set_header_nritems(l, mid);
1909 ret = 0;
1910 btrfs_item_key(right, &disk_key, 0);
1911 wret = insert_ptr(trans, root, path, &disk_key, right->start,
1912 path->slots[1] + 1, 1);
1913 if (wret)
1914 ret = wret;
1915
1916 btrfs_mark_buffer_dirty(right);
1917 btrfs_mark_buffer_dirty(l);
1918 BUG_ON(path->slots[0] != slot);
1919
1920 if (mid <= slot) {
1921 free_extent_buffer(path->nodes[0]);
1922 path->nodes[0] = right;
1923 path->slots[0] -= mid;
1924 path->slots[1] += 1;
1925 } else
1926 free_extent_buffer(right);
1927
1928 BUG_ON(path->slots[0] < 0);
1929
1930 if (double_split) {
1931 BUG_ON(num_doubles != 0);
1932 num_doubles++;
1933 goto again;
1934 }
1935 return ret;
1936 }
1937
1938 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
1939 struct btrfs_root *root,
1940 struct btrfs_path *path,
1941 u32 new_size, int from_end)
1942 {
1943 int ret = 0;
1944 int slot;
1945 int slot_orig;
1946 struct extent_buffer *leaf;
1947 struct btrfs_item *item;
1948 u32 nritems;
1949 unsigned int data_end;
1950 unsigned int old_data_start;
1951 unsigned int old_size;
1952 unsigned int size_diff;
1953 int i;
1954
1955 slot_orig = path->slots[0];
1956 leaf = path->nodes[0];
1957 slot = path->slots[0];
1958
1959 old_size = btrfs_item_size_nr(leaf, slot);
1960 if (old_size == new_size)
1961 return 0;
1962
1963 nritems = btrfs_header_nritems(leaf);
1964 data_end = leaf_data_end(root, leaf);
1965
1966 old_data_start = btrfs_item_offset_nr(leaf, slot);
1967
1968 size_diff = old_size - new_size;
1969
1970 BUG_ON(slot < 0);
1971 BUG_ON(slot >= nritems);
1972
1973 /*
1974 * item0..itemN ... dataN.offset..dataN.size .. data0.size
1975 */
1976 /* first correct the data pointers */
1977 for (i = slot; i < nritems; i++) {
1978 u32 ioff;
1979 item = btrfs_item_nr(leaf, i);
1980
1981 if (!leaf->map_token) {
1982 map_extent_buffer(leaf, (unsigned long)item,
1983 sizeof(struct btrfs_item),
1984 &leaf->map_token, &leaf->kaddr,
1985 &leaf->map_start, &leaf->map_len,
1986 KM_USER1);
1987 }
1988
1989 ioff = btrfs_item_offset(leaf, item);
1990 btrfs_set_item_offset(leaf, item, ioff + size_diff);
1991 }
1992
1993 if (leaf->map_token) {
1994 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
1995 leaf->map_token = NULL;
1996 }
1997
1998 /* shift the data */
1999 if (from_end) {
2000 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2001 data_end + size_diff, btrfs_leaf_data(leaf) +
2002 data_end, old_data_start + new_size - data_end);
2003 } else {
2004 struct btrfs_disk_key disk_key;
2005 u64 offset;
2006
2007 btrfs_item_key(leaf, &disk_key, slot);
2008
2009 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
2010 unsigned long ptr;
2011 struct btrfs_file_extent_item *fi;
2012
2013 fi = btrfs_item_ptr(leaf, slot,
2014 struct btrfs_file_extent_item);
2015 fi = (struct btrfs_file_extent_item *)(
2016 (unsigned long)fi - size_diff);
2017
2018 if (btrfs_file_extent_type(leaf, fi) ==
2019 BTRFS_FILE_EXTENT_INLINE) {
2020 ptr = btrfs_item_ptr_offset(leaf, slot);
2021 memmove_extent_buffer(leaf, ptr,
2022 (unsigned long)fi,
2023 offsetof(struct btrfs_file_extent_item,
2024 disk_bytenr));
2025 }
2026 }
2027
2028 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2029 data_end + size_diff, btrfs_leaf_data(leaf) +
2030 data_end, old_data_start - data_end);
2031
2032 offset = btrfs_disk_key_offset(&disk_key);
2033 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
2034 btrfs_set_item_key(leaf, &disk_key, slot);
2035 if (slot == 0)
2036 fixup_low_keys(trans, root, path, &disk_key, 1);
2037 }
2038
2039 item = btrfs_item_nr(leaf, slot);
2040 btrfs_set_item_size(leaf, item, new_size);
2041 btrfs_mark_buffer_dirty(leaf);
2042
2043 ret = 0;
2044 if (btrfs_leaf_free_space(root, leaf) < 0) {
2045 btrfs_print_leaf(root, leaf);
2046 BUG();
2047 }
2048 return ret;
2049 }
2050
2051 int btrfs_extend_item(struct btrfs_trans_handle *trans,
2052 struct btrfs_root *root, struct btrfs_path *path,
2053 u32 data_size)
2054 {
2055 int ret = 0;
2056 int slot;
2057 int slot_orig;
2058 struct extent_buffer *leaf;
2059 struct btrfs_item *item;
2060 u32 nritems;
2061 unsigned int data_end;
2062 unsigned int old_data;
2063 unsigned int old_size;
2064 int i;
2065
2066 slot_orig = path->slots[0];
2067 leaf = path->nodes[0];
2068
2069 nritems = btrfs_header_nritems(leaf);
2070 data_end = leaf_data_end(root, leaf);
2071
2072 if (btrfs_leaf_free_space(root, leaf) < data_size) {
2073 btrfs_print_leaf(root, leaf);
2074 BUG();
2075 }
2076 slot = path->slots[0];
2077 old_data = btrfs_item_end_nr(leaf, slot);
2078
2079 BUG_ON(slot < 0);
2080 if (slot >= nritems) {
2081 btrfs_print_leaf(root, leaf);
2082 printk("slot %d too large, nritems %d\n", slot, nritems);
2083 BUG_ON(1);
2084 }
2085
2086 /*
2087 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2088 */
2089 /* first correct the data pointers */
2090 for (i = slot; i < nritems; i++) {
2091 u32 ioff;
2092 item = btrfs_item_nr(leaf, i);
2093
2094 if (!leaf->map_token) {
2095 map_extent_buffer(leaf, (unsigned long)item,
2096 sizeof(struct btrfs_item),
2097 &leaf->map_token, &leaf->kaddr,
2098 &leaf->map_start, &leaf->map_len,
2099 KM_USER1);
2100 }
2101 ioff = btrfs_item_offset(leaf, item);
2102 btrfs_set_item_offset(leaf, item, ioff - data_size);
2103 }
2104
2105 if (leaf->map_token) {
2106 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2107 leaf->map_token = NULL;
2108 }
2109
2110 /* shift the data */
2111 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2112 data_end - data_size, btrfs_leaf_data(leaf) +
2113 data_end, old_data - data_end);
2114
2115 data_end = old_data;
2116 old_size = btrfs_item_size_nr(leaf, slot);
2117 item = btrfs_item_nr(leaf, slot);
2118 btrfs_set_item_size(leaf, item, old_size + data_size);
2119 btrfs_mark_buffer_dirty(leaf);
2120
2121 ret = 0;
2122 if (btrfs_leaf_free_space(root, leaf) < 0) {
2123 btrfs_print_leaf(root, leaf);
2124 BUG();
2125 }
2126 return ret;
2127 }
2128
2129 /*
2130 * Given a key and some data, insert an item into the tree.
2131 * This does all the path init required, making room in the tree if needed.
2132 */
2133 int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
2134 struct btrfs_root *root,
2135 struct btrfs_path *path,
2136 struct btrfs_key *cpu_key, u32 data_size)
2137 {
2138 struct extent_buffer *leaf;
2139 struct btrfs_item *item;
2140 int ret = 0;
2141 int slot;
2142 int slot_orig;
2143 u32 nritems;
2144 unsigned int data_end;
2145 struct btrfs_disk_key disk_key;
2146
2147 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
2148
2149 /* create a root if there isn't one */
2150 if (!root->node)
2151 BUG();
2152
2153 ret = btrfs_search_slot(trans, root, cpu_key, path, data_size, 1);
2154 if (ret == 0) {
2155 return -EEXIST;
2156 }
2157 if (ret < 0)
2158 goto out;
2159
2160 slot_orig = path->slots[0];
2161 leaf = path->nodes[0];
2162
2163 nritems = btrfs_header_nritems(leaf);
2164 data_end = leaf_data_end(root, leaf);
2165
2166 if (btrfs_leaf_free_space(root, leaf) <
2167 sizeof(struct btrfs_item) + data_size) {
2168 btrfs_print_leaf(root, leaf);
2169 printk("not enough freespace need %u have %d\n",
2170 data_size, btrfs_leaf_free_space(root, leaf));
2171 BUG();
2172 }
2173
2174 slot = path->slots[0];
2175 BUG_ON(slot < 0);
2176
2177 if (slot != nritems) {
2178 int i;
2179 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
2180
2181 if (old_data < data_end) {
2182 btrfs_print_leaf(root, leaf);
2183 printk("slot %d old_data %d data_end %d\n",
2184 slot, old_data, data_end);
2185 BUG_ON(1);
2186 }
2187 /*
2188 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2189 */
2190 /* first correct the data pointers */
2191 WARN_ON(leaf->map_token);
2192 for (i = slot; i < nritems; i++) {
2193 u32 ioff;
2194
2195 item = btrfs_item_nr(leaf, i);
2196 if (!leaf->map_token) {
2197 map_extent_buffer(leaf, (unsigned long)item,
2198 sizeof(struct btrfs_item),
2199 &leaf->map_token, &leaf->kaddr,
2200 &leaf->map_start, &leaf->map_len,
2201 KM_USER1);
2202 }
2203
2204 ioff = btrfs_item_offset(leaf, item);
2205 btrfs_set_item_offset(leaf, item, ioff - data_size);
2206 }
2207 if (leaf->map_token) {
2208 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2209 leaf->map_token = NULL;
2210 }
2211
2212 /* shift the items */
2213 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2214 btrfs_item_nr_offset(slot),
2215 (nritems - slot) * sizeof(struct btrfs_item));
2216
2217 /* shift the data */
2218 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2219 data_end - data_size, btrfs_leaf_data(leaf) +
2220 data_end, old_data - data_end);
2221 data_end = old_data;
2222 }
2223
2224 /* setup the item for the new data */
2225 btrfs_set_item_key(leaf, &disk_key, slot);
2226 item = btrfs_item_nr(leaf, slot);
2227 btrfs_set_item_offset(leaf, item, data_end - data_size);
2228 btrfs_set_item_size(leaf, item, data_size);
2229 btrfs_set_header_nritems(leaf, nritems + 1);
2230 btrfs_mark_buffer_dirty(leaf);
2231
2232 ret = 0;
2233 if (slot == 0)
2234 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
2235
2236 if (btrfs_leaf_free_space(root, leaf) < 0) {
2237 btrfs_print_leaf(root, leaf);
2238 BUG();
2239 }
2240 out:
2241 return ret;
2242 }
2243
2244 /*
2245 * Given a key and some data, insert an item into the tree.
2246 * This does all the path init required, making room in the tree if needed.
2247 */
2248 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
2249 *root, struct btrfs_key *cpu_key, void *data, u32
2250 data_size)
2251 {
2252 int ret = 0;
2253 struct btrfs_path *path;
2254 struct extent_buffer *leaf;
2255 unsigned long ptr;
2256
2257 path = btrfs_alloc_path();
2258 BUG_ON(!path);
2259 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
2260 if (!ret) {
2261 leaf = path->nodes[0];
2262 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2263 write_extent_buffer(leaf, data, ptr, data_size);
2264 btrfs_mark_buffer_dirty(leaf);
2265 }
2266 btrfs_free_path(path);
2267 return ret;
2268 }
2269
2270 /*
2271 * delete the pointer from a given node.
2272 *
2273 * If the delete empties a node, the node is removed from the tree,
2274 * continuing all the way the root if required. The root is converted into
2275 * a leaf if all the nodes are emptied.
2276 */
2277 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2278 struct btrfs_path *path, int level, int slot)
2279 {
2280 struct extent_buffer *parent = path->nodes[level];
2281 u32 nritems;
2282 int ret = 0;
2283 int wret;
2284
2285 nritems = btrfs_header_nritems(parent);
2286 if (slot != nritems -1) {
2287 memmove_extent_buffer(parent,
2288 btrfs_node_key_ptr_offset(slot),
2289 btrfs_node_key_ptr_offset(slot + 1),
2290 sizeof(struct btrfs_key_ptr) *
2291 (nritems - slot - 1));
2292 }
2293 nritems--;
2294 btrfs_set_header_nritems(parent, nritems);
2295 if (nritems == 0 && parent == root->node) {
2296 BUG_ON(btrfs_header_level(root->node) != 1);
2297 /* just turn the root into a leaf and break */
2298 btrfs_set_header_level(root->node, 0);
2299 } else if (slot == 0) {
2300 struct btrfs_disk_key disk_key;
2301
2302 btrfs_node_key(parent, &disk_key, 0);
2303 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
2304 if (wret)
2305 ret = wret;
2306 }
2307 btrfs_mark_buffer_dirty(parent);
2308 return ret;
2309 }
2310
2311 /*
2312 * delete the item at the leaf level in path. If that empties
2313 * the leaf, remove it from the tree
2314 */
2315 int btrfs_del_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
2316 struct btrfs_path *path)
2317 {
2318 int slot;
2319 struct extent_buffer *leaf;
2320 struct btrfs_item *item;
2321 int doff;
2322 int dsize;
2323 int ret = 0;
2324 int wret;
2325 u32 nritems;
2326
2327 leaf = path->nodes[0];
2328 slot = path->slots[0];
2329 doff = btrfs_item_offset_nr(leaf, slot);
2330 dsize = btrfs_item_size_nr(leaf, slot);
2331 nritems = btrfs_header_nritems(leaf);
2332
2333 if (slot != nritems - 1) {
2334 int i;
2335 int data_end = leaf_data_end(root, leaf);
2336
2337 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
2338 data_end + dsize,
2339 btrfs_leaf_data(leaf) + data_end,
2340 doff - data_end);
2341
2342 for (i = slot + 1; i < nritems; i++) {
2343 u32 ioff;
2344
2345 item = btrfs_item_nr(leaf, i);
2346 if (!leaf->map_token) {
2347 map_extent_buffer(leaf, (unsigned long)item,
2348 sizeof(struct btrfs_item),
2349 &leaf->map_token, &leaf->kaddr,
2350 &leaf->map_start, &leaf->map_len,
2351 KM_USER1);
2352 }
2353 ioff = btrfs_item_offset(leaf, item);
2354 btrfs_set_item_offset(leaf, item, ioff + dsize);
2355 }
2356
2357 if (leaf->map_token) {
2358 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2359 leaf->map_token = NULL;
2360 }
2361
2362 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
2363 btrfs_item_nr_offset(slot + 1),
2364 sizeof(struct btrfs_item) *
2365 (nritems - slot - 1));
2366 }
2367 btrfs_set_header_nritems(leaf, nritems - 1);
2368 nritems--;
2369
2370 /* delete the leaf if we've emptied it */
2371 if (nritems == 0) {
2372 if (leaf == root->node) {
2373 btrfs_set_header_level(leaf, 0);
2374 } else {
2375 clean_tree_block(trans, root, leaf);
2376 wait_on_tree_block_writeback(root, leaf);
2377 wret = del_ptr(trans, root, path, 1, path->slots[1]);
2378 if (wret)
2379 ret = wret;
2380 wret = btrfs_free_extent(trans, root,
2381 leaf->start, leaf->len, 1);
2382 if (wret)
2383 ret = wret;
2384 }
2385 } else {
2386 int used = leaf_space_used(leaf, 0, nritems);
2387 if (slot == 0) {
2388 struct btrfs_disk_key disk_key;
2389
2390 btrfs_item_key(leaf, &disk_key, 0);
2391 wret = fixup_low_keys(trans, root, path,
2392 &disk_key, 1);
2393 if (wret)
2394 ret = wret;
2395 }
2396
2397 /* delete the leaf if it is mostly empty */
2398 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
2399 /* push_leaf_left fixes the path.
2400 * make sure the path still points to our leaf
2401 * for possible call to del_ptr below
2402 */
2403 slot = path->slots[1];
2404 extent_buffer_get(leaf);
2405
2406 wret = push_leaf_right(trans, root, path, 1);
2407 if (wret < 0 && wret != -ENOSPC)
2408 ret = wret;
2409
2410 if (path->nodes[0] == leaf &&
2411 btrfs_header_nritems(leaf)) {
2412 wret = push_leaf_left(trans, root, path, 1);
2413 if (wret < 0 && wret != -ENOSPC)
2414 ret = wret;
2415 }
2416
2417 if (btrfs_header_nritems(leaf) == 0) {
2418 u64 bytenr = leaf->start;
2419 u32 blocksize = leaf->len;
2420
2421 clean_tree_block(trans, root, leaf);
2422 wait_on_tree_block_writeback(root, leaf);
2423
2424 wret = del_ptr(trans, root, path, 1, slot);
2425 if (wret)
2426 ret = wret;
2427
2428 free_extent_buffer(leaf);
2429 wret = btrfs_free_extent(trans, root, bytenr,
2430 blocksize, 1);
2431 if (wret)
2432 ret = wret;
2433 } else {
2434 btrfs_mark_buffer_dirty(leaf);
2435 free_extent_buffer(leaf);
2436 }
2437 } else {
2438 btrfs_mark_buffer_dirty(leaf);
2439 }
2440 }
2441 return ret;
2442 }
2443
2444 /*
2445 * walk up the tree as far as required to find the next leaf.
2446 * returns 0 if it found something or 1 if there are no greater leaves.
2447 * returns < 0 on io errors.
2448 */
2449 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
2450 {
2451 int slot;
2452 int level = 1;
2453 u64 bytenr;
2454 struct extent_buffer *c;
2455 struct extent_buffer *next = NULL;
2456
2457 while(level < BTRFS_MAX_LEVEL) {
2458 if (!path->nodes[level])
2459 return 1;
2460
2461 slot = path->slots[level] + 1;
2462 c = path->nodes[level];
2463 if (slot >= btrfs_header_nritems(c)) {
2464 level++;
2465 continue;
2466 }
2467
2468 bytenr = btrfs_node_blockptr(c, slot);
2469 if (next)
2470 free_extent_buffer(next);
2471
2472 if (path->reada)
2473 reada_for_search(root, path, level, slot);
2474
2475 next = read_tree_block(root, bytenr,
2476 btrfs_level_size(root, level -1));
2477 break;
2478 }
2479 path->slots[level] = slot;
2480 while(1) {
2481 level--;
2482 c = path->nodes[level];
2483 free_extent_buffer(c);
2484 path->nodes[level] = next;
2485 path->slots[level] = 0;
2486 if (!level)
2487 break;
2488 if (path->reada)
2489 reada_for_search(root, path, level, 0);
2490 next = read_tree_block(root, btrfs_node_blockptr(next, 0),
2491 btrfs_level_size(root, level - 1));
2492 }
2493 return 0;
2494 }
This page took 0.082373 seconds and 6 git commands to generate.