projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: node balance optimizations
[deliverable/linux.git] / fs / btrfs / ctree.c
1 #include <linux/module.h>
2 #include "ctree.h"
3 #include "disk-io.h"
4 #include "transaction.h"
5
6 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
7 *root, struct btrfs_path *path, int level);
8 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
9 *root, struct btrfs_key *ins_key,
10 struct btrfs_path *path, int data_size);
11 static int push_node_left(struct btrfs_trans_handle *trans, struct btrfs_root
12 *root, struct buffer_head *dst, struct buffer_head
13 *src);
14 static int balance_node_right(struct btrfs_trans_handle *trans, struct
15 btrfs_root *root, struct buffer_head *dst_buf,
16 struct buffer_head *src_buf);
17 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
18 struct btrfs_path *path, int level, int slot);
19
20 inline void btrfs_init_path(struct btrfs_path *p)
21 {
22 memset(p, 0, sizeof(*p));
23 }
24
25 struct btrfs_path *btrfs_alloc_path(void)
26 {
27 struct btrfs_path *path;
28 path = kmem_cache_alloc(btrfs_path_cachep, GFP_NOFS);
29 if (path)
30 btrfs_init_path(path);
31 return path;
32 }
33
34 void btrfs_free_path(struct btrfs_path *p)
35 {
36 btrfs_release_path(NULL, p);
37 kmem_cache_free(btrfs_path_cachep, p);
38 }
39
40 void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
41 {
42 int i;
43 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
44 if (!p->nodes[i])
45 break;
46 btrfs_block_release(root, p->nodes[i]);
47 }
48 memset(p, 0, sizeof(*p));
49 }
50
51 static int btrfs_cow_block(struct btrfs_trans_handle *trans, struct btrfs_root
52 *root, struct buffer_head *buf, struct buffer_head
53 *parent, int parent_slot, struct buffer_head
54 **cow_ret)
55 {
56 struct buffer_head *cow;
57 struct btrfs_node *cow_node;
58
59 if (btrfs_header_generation(btrfs_buffer_header(buf)) ==
60 trans->transid) {
61 *cow_ret = buf;
62 return 0;
63 }
64 cow = btrfs_alloc_free_block(trans, root);
65 cow_node = btrfs_buffer_node(cow);
66 if (buf->b_size != root->blocksize || cow->b_size != root->blocksize)
67 WARN_ON(1);
68 memcpy(cow_node, btrfs_buffer_node(buf), root->blocksize);
69 btrfs_set_header_blocknr(&cow_node->header, bh_blocknr(cow));
70 btrfs_set_header_generation(&cow_node->header, trans->transid);
71 btrfs_inc_ref(trans, root, buf);
72 if (buf == root->node) {
73 root->node = cow;
74 get_bh(cow);
75 if (buf != root->commit_root) {
76 btrfs_free_extent(trans, root, bh_blocknr(buf), 1, 1);
77 }
78 btrfs_block_release(root, buf);
79 } else {
80 btrfs_set_node_blockptr(btrfs_buffer_node(parent), parent_slot,
81 bh_blocknr(cow));
82 btrfs_mark_buffer_dirty(parent);
83 btrfs_free_extent(trans, root, bh_blocknr(buf), 1, 1);
84 }
85 btrfs_block_release(root, buf);
86 mark_buffer_dirty(cow);
87 *cow_ret = cow;
88 return 0;
89 }
90
91 /*
92 * The leaf data grows from end-to-front in the node.
93 * this returns the address of the start of the last item,
94 * which is the stop of the leaf data stack
95 */
96 static inline unsigned int leaf_data_end(struct btrfs_root *root,
97 struct btrfs_leaf *leaf)
98 {
99 u32 nr = btrfs_header_nritems(&leaf->header);
100 if (nr == 0)
101 return BTRFS_LEAF_DATA_SIZE(root);
102 return btrfs_item_offset(leaf->items + nr - 1);
103 }
104
105 /*
106 * compare two keys in a memcmp fashion
107 */
108 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
109 {
110 struct btrfs_key k1;
111
112 btrfs_disk_key_to_cpu(&k1, disk);
113
114 if (k1.objectid > k2->objectid)
115 return 1;
116 if (k1.objectid < k2->objectid)
117 return -1;
118 if (k1.flags > k2->flags)
119 return 1;
120 if (k1.flags < k2->flags)
121 return -1;
122 if (k1.offset > k2->offset)
123 return 1;
124 if (k1.offset < k2->offset)
125 return -1;
126 return 0;
127 }
128
129 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
130 int level)
131 {
132 int i;
133 struct btrfs_node *parent = NULL;
134 struct btrfs_node *node = btrfs_buffer_node(path->nodes[level]);
135 int parent_slot;
136 u32 nritems = btrfs_header_nritems(&node->header);
137
138 if (path->nodes[level + 1])
139 parent = btrfs_buffer_node(path->nodes[level + 1]);
140 parent_slot = path->slots[level + 1];
141 BUG_ON(nritems == 0);
142 if (parent) {
143 struct btrfs_disk_key *parent_key;
144 parent_key = &parent->ptrs[parent_slot].key;
145 BUG_ON(memcmp(parent_key, &node->ptrs[0].key,
146 sizeof(struct btrfs_disk_key)));
147 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
148 btrfs_header_blocknr(&node->header));
149 }
150 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
151 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
152 struct btrfs_key cpukey;
153 btrfs_disk_key_to_cpu(&cpukey, &node->ptrs[i + 1].key);
154 BUG_ON(comp_keys(&node->ptrs[i].key, &cpukey) >= 0);
155 }
156 return 0;
157 }
158
159 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
160 int level)
161 {
162 int i;
163 struct btrfs_leaf *leaf = btrfs_buffer_leaf(path->nodes[level]);
164 struct btrfs_node *parent = NULL;
165 int parent_slot;
166 u32 nritems = btrfs_header_nritems(&leaf->header);
167
168 if (path->nodes[level + 1])
169 parent = btrfs_buffer_node(path->nodes[level + 1]);
170 parent_slot = path->slots[level + 1];
171 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
172
173 if (nritems == 0)
174 return 0;
175
176 if (parent) {
177 struct btrfs_disk_key *parent_key;
178 parent_key = &parent->ptrs[parent_slot].key;
179 BUG_ON(memcmp(parent_key, &leaf->items[0].key,
180 sizeof(struct btrfs_disk_key)));
181 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
182 btrfs_header_blocknr(&leaf->header));
183 }
184 for (i = 0; nritems > 1 && i < nritems - 2; i++) {
185 struct btrfs_key cpukey;
186 btrfs_disk_key_to_cpu(&cpukey, &leaf->items[i + 1].key);
187 BUG_ON(comp_keys(&leaf->items[i].key,
188 &cpukey) >= 0);
189 BUG_ON(btrfs_item_offset(leaf->items + i) !=
190 btrfs_item_end(leaf->items + i + 1));
191 if (i == 0) {
192 BUG_ON(btrfs_item_offset(leaf->items + i) +
193 btrfs_item_size(leaf->items + i) !=
194 BTRFS_LEAF_DATA_SIZE(root));
195 }
196 }
197 return 0;
198 }
199
200 static int check_block(struct btrfs_root *root, struct btrfs_path *path,
201 int level)
202 {
203 struct btrfs_node *node = btrfs_buffer_node(path->nodes[level]);
204 if (memcmp(node->header.fsid, root->fs_info->disk_super->fsid,
205 sizeof(node->header.fsid)))
206 BUG();
207 if (level == 0)
208 return check_leaf(root, path, level);
209 return check_node(root, path, level);
210 }
211
212 /*
213 * search for key in the array p. items p are item_size apart
214 * and there are 'max' items in p
215 * the slot in the array is returned via slot, and it points to
216 * the place where you would insert key if it is not found in
217 * the array.
218 *
219 * slot may point to max if the key is bigger than all of the keys
220 */
221 static int generic_bin_search(char *p, int item_size, struct btrfs_key *key,
222 int max, int *slot)
223 {
224 int low = 0;
225 int high = max;
226 int mid;
227 int ret;
228 struct btrfs_disk_key *tmp;
229
230 while(low < high) {
231 mid = (low + high) / 2;
232 tmp = (struct btrfs_disk_key *)(p + mid * item_size);
233 ret = comp_keys(tmp, key);
234
235 if (ret < 0)
236 low = mid + 1;
237 else if (ret > 0)
238 high = mid;
239 else {
240 *slot = mid;
241 return 0;
242 }
243 }
244 *slot = low;
245 return 1;
246 }
247
248 /*
249 * simple bin_search frontend that does the right thing for
250 * leaves vs nodes
251 */
252 static int bin_search(struct btrfs_node *c, struct btrfs_key *key, int *slot)
253 {
254 if (btrfs_is_leaf(c)) {
255 struct btrfs_leaf *l = (struct btrfs_leaf *)c;
256 return generic_bin_search((void *)l->items,
257 sizeof(struct btrfs_item),
258 key, btrfs_header_nritems(&c->header),
259 slot);
260 } else {
261 return generic_bin_search((void *)c->ptrs,
262 sizeof(struct btrfs_key_ptr),
263 key, btrfs_header_nritems(&c->header),
264 slot);
265 }
266 return -1;
267 }
268
269 static struct buffer_head *read_node_slot(struct btrfs_root *root,
270 struct buffer_head *parent_buf,
271 int slot)
272 {
273 struct btrfs_node *node = btrfs_buffer_node(parent_buf);
274 if (slot < 0)
275 return NULL;
276 if (slot >= btrfs_header_nritems(&node->header))
277 return NULL;
278 return read_tree_block(root, btrfs_node_blockptr(node, slot));
279 }
280
281 static int balance_level(struct btrfs_trans_handle *trans, struct btrfs_root
282 *root, struct btrfs_path *path, int level)
283 {
284 struct buffer_head *right_buf;
285 struct buffer_head *mid_buf;
286 struct buffer_head *left_buf;
287 struct buffer_head *parent_buf = NULL;
288 struct btrfs_node *right = NULL;
289 struct btrfs_node *mid;
290 struct btrfs_node *left = NULL;
291 struct btrfs_node *parent = NULL;
292 int ret = 0;
293 int wret;
294 int pslot;
295 int orig_slot = path->slots[level];
296 u64 orig_ptr;
297
298 if (level == 0)
299 return 0;
300
301 mid_buf = path->nodes[level];
302 mid = btrfs_buffer_node(mid_buf);
303 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
304
305 if (level < BTRFS_MAX_LEVEL - 1)
306 parent_buf = path->nodes[level + 1];
307 pslot = path->slots[level + 1];
308
309 /*
310 * deal with the case where there is only one pointer in the root
311 * by promoting the node below to a root
312 */
313 if (!parent_buf) {
314 struct buffer_head *child;
315 u64 blocknr = bh_blocknr(mid_buf);
316
317 if (btrfs_header_nritems(&mid->header) != 1)
318 return 0;
319
320 /* promote the child to a root */
321 child = read_node_slot(root, mid_buf, 0);
322 BUG_ON(!child);
323 root->node = child;
324 path->nodes[level] = NULL;
325 clean_tree_block(trans, root, mid_buf);
326 wait_on_buffer(mid_buf);
327 /* once for the path */
328 btrfs_block_release(root, mid_buf);
329 /* once for the root ptr */
330 btrfs_block_release(root, mid_buf);
331 return btrfs_free_extent(trans, root, blocknr, 1, 1);
332 }
333 parent = btrfs_buffer_node(parent_buf);
334
335 if (btrfs_header_nritems(&mid->header) >
336 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
337 return 0;
338
339 left_buf = read_node_slot(root, parent_buf, pslot - 1);
340 right_buf = read_node_slot(root, parent_buf, pslot + 1);
341
342 /* first, try to make some room in the middle buffer */
343 if (left_buf) {
344 btrfs_cow_block(trans, root, left_buf, parent_buf, pslot - 1,
345 &left_buf);
346 left = btrfs_buffer_node(left_buf);
347 orig_slot += btrfs_header_nritems(&left->header);
348 wret = push_node_left(trans, root, left_buf, mid_buf);
349 if (wret < 0)
350 ret = wret;
351 }
352
353 /*
354 * then try to empty the right most buffer into the middle
355 */
356 if (right_buf) {
357 btrfs_cow_block(trans, root, right_buf, parent_buf, pslot + 1,
358 &right_buf);
359 right = btrfs_buffer_node(right_buf);
360 wret = push_node_left(trans, root, mid_buf, right_buf);
361 if (wret < 0)
362 ret = wret;
363 if (btrfs_header_nritems(&right->header) == 0) {
364 u64 blocknr = bh_blocknr(right_buf);
365 clean_tree_block(trans, root, right_buf);
366 wait_on_buffer(right_buf);
367 btrfs_block_release(root, right_buf);
368 right_buf = NULL;
369 right = NULL;
370 wret = del_ptr(trans, root, path, level + 1, pslot +
371 1);
372 if (wret)
373 ret = wret;
374 wret = btrfs_free_extent(trans, root, blocknr, 1, 1);
375 if (wret)
376 ret = wret;
377 } else {
378 btrfs_memcpy(root, parent,
379 &parent->ptrs[pslot + 1].key,
380 &right->ptrs[0].key,
381 sizeof(struct btrfs_disk_key));
382 btrfs_mark_buffer_dirty(parent_buf);
383 }
384 }
385 if (btrfs_header_nritems(&mid->header) == 1) {
386 /*
387 * we're not allowed to leave a node with one item in the
388 * tree during a delete. A deletion from lower in the tree
389 * could try to delete the only pointer in this node.
390 * So, pull some keys from the left.
391 * There has to be a left pointer at this point because
392 * otherwise we would have pulled some pointers from the
393 * right
394 */
395 BUG_ON(!left_buf);
396 wret = balance_node_right(trans, root, mid_buf, left_buf);
397 if (wret < 0)
398 ret = wret;
399 BUG_ON(wret == 1);
400 }
401 if (btrfs_header_nritems(&mid->header) == 0) {
402 /* we've managed to empty the middle node, drop it */
403 u64 blocknr = bh_blocknr(mid_buf);
404 clean_tree_block(trans, root, mid_buf);
405 wait_on_buffer(mid_buf);
406 btrfs_block_release(root, mid_buf);
407 mid_buf = NULL;
408 mid = NULL;
409 wret = del_ptr(trans, root, path, level + 1, pslot);
410 if (wret)
411 ret = wret;
412 wret = btrfs_free_extent(trans, root, blocknr, 1, 1);
413 if (wret)
414 ret = wret;
415 } else {
416 /* update the parent key to reflect our changes */
417 btrfs_memcpy(root, parent,
418 &parent->ptrs[pslot].key, &mid->ptrs[0].key,
419 sizeof(struct btrfs_disk_key));
420 btrfs_mark_buffer_dirty(parent_buf);
421 }
422
423 /* update the path */
424 if (left_buf) {
425 if (btrfs_header_nritems(&left->header) > orig_slot) {
426 get_bh(left_buf);
427 path->nodes[level] = left_buf;
428 path->slots[level + 1] -= 1;
429 path->slots[level] = orig_slot;
430 if (mid_buf)
431 btrfs_block_release(root, mid_buf);
432 } else {
433 orig_slot -= btrfs_header_nritems(&left->header);
434 path->slots[level] = orig_slot;
435 }
436 }
437 /* double check we haven't messed things up */
438 check_block(root, path, level);
439 if (orig_ptr !=
440 btrfs_node_blockptr(btrfs_buffer_node(path->nodes[level]),
441 path->slots[level]))
442 BUG();
443
444 if (right_buf)
445 btrfs_block_release(root, right_buf);
446 if (left_buf)
447 btrfs_block_release(root, left_buf);
448 return ret;
449 }
450
451 /* returns zero if the push worked, non-zero otherwise */
452 static int push_nodes_for_insert(struct btrfs_trans_handle *trans,
453 struct btrfs_root *root,
454 struct btrfs_path *path, int level)
455 {
456 struct buffer_head *right_buf;
457 struct buffer_head *mid_buf;
458 struct buffer_head *left_buf;
459 struct buffer_head *parent_buf = NULL;
460 struct btrfs_node *right = NULL;
461 struct btrfs_node *mid;
462 struct btrfs_node *left = NULL;
463 struct btrfs_node *parent = NULL;
464 int ret = 0;
465 int wret;
466 int pslot;
467 int orig_slot = path->slots[level];
468 u64 orig_ptr;
469
470 if (level == 0)
471 return 1;
472
473 mid_buf = path->nodes[level];
474 mid = btrfs_buffer_node(mid_buf);
475 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
476
477 if (level < BTRFS_MAX_LEVEL - 1)
478 parent_buf = path->nodes[level + 1];
479 pslot = path->slots[level + 1];
480
481 if (!parent_buf)
482 return 1;
483 parent = btrfs_buffer_node(parent_buf);
484
485 left_buf = read_node_slot(root, parent_buf, pslot - 1);
486
487 /* first, try to make some room in the middle buffer */
488 if (left_buf) {
489 u32 left_nr;
490 left = btrfs_buffer_node(left_buf);
491 left_nr = btrfs_header_nritems(&left->header);
492 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
493 wret = 1;
494 } else {
495 btrfs_cow_block(trans, root, left_buf, parent_buf,
496 pslot - 1, &left_buf);
497 left = btrfs_buffer_node(left_buf);
498 wret = push_node_left(trans, root, left_buf, mid_buf);
499 }
500 if (wret < 0)
501 ret = wret;
502 if (wret == 0) {
503 orig_slot += left_nr;
504 btrfs_memcpy(root, parent,
505 &parent->ptrs[pslot].key,
506 &mid->ptrs[0].key,
507 sizeof(struct btrfs_disk_key));
508 btrfs_mark_buffer_dirty(parent_buf);
509 if (btrfs_header_nritems(&left->header) > orig_slot) {
510 path->nodes[level] = left_buf;
511 path->slots[level + 1] -= 1;
512 path->slots[level] = orig_slot;
513 btrfs_block_release(root, mid_buf);
514 } else {
515 orig_slot -=
516 btrfs_header_nritems(&left->header);
517 path->slots[level] = orig_slot;
518 btrfs_block_release(root, left_buf);
519 }
520 check_node(root, path, level);
521 return 0;
522 }
523 btrfs_block_release(root, left_buf);
524 }
525 right_buf = read_node_slot(root, parent_buf, pslot + 1);
526
527 /*
528 * then try to empty the right most buffer into the middle
529 */
530 if (right_buf) {
531 u32 right_nr;
532 right = btrfs_buffer_node(right_buf);
533 right_nr = btrfs_header_nritems(&right->header);
534 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
535 wret = 1;
536 } else {
537 btrfs_cow_block(trans, root, right_buf,
538 parent_buf, pslot + 1, &right_buf);
539 right = btrfs_buffer_node(right_buf);
540 wret = balance_node_right(trans, root,
541 right_buf, mid_buf);
542 }
543 if (wret < 0)
544 ret = wret;
545 if (wret == 0) {
546 btrfs_memcpy(root, parent,
547 &parent->ptrs[pslot + 1].key,
548 &right->ptrs[0].key,
549 sizeof(struct btrfs_disk_key));
550 btrfs_mark_buffer_dirty(parent_buf);
551 if (btrfs_header_nritems(&mid->header) <= orig_slot) {
552 path->nodes[level] = right_buf;
553 path->slots[level + 1] += 1;
554 path->slots[level] = orig_slot -
555 btrfs_header_nritems(&mid->header);
556 btrfs_block_release(root, mid_buf);
557 } else {
558 btrfs_block_release(root, right_buf);
559 }
560 check_node(root, path, level);
561 return 0;
562 }
563 btrfs_block_release(root, right_buf);
564 }
565 check_node(root, path, level);
566 return 1;
567 }
568
569 /*
570 * look for key in the tree. path is filled in with nodes along the way
571 * if key is found, we return zero and you can find the item in the leaf
572 * level of the path (level 0)
573 *
574 * If the key isn't found, the path points to the slot where it should
575 * be inserted, and 1 is returned. If there are other errors during the
576 * search a negative error number is returned.
577 *
578 * if ins_len > 0, nodes and leaves will be split as we walk down the
579 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
580 * possible)
581 */
582 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
583 *root, struct btrfs_key *key, struct btrfs_path *p, int
584 ins_len, int cow)
585 {
586 struct buffer_head *b;
587 struct buffer_head *cow_buf;
588 struct btrfs_node *c;
589 int slot;
590 int ret;
591 int level;
592
593 WARN_ON(p->nodes[0] != NULL);
594 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
595 again:
596 b = root->node;
597 get_bh(b);
598 while (b) {
599 c = btrfs_buffer_node(b);
600 level = btrfs_header_level(&c->header);
601 if (cow) {
602 int wret;
603 wret = btrfs_cow_block(trans, root, b,
604 p->nodes[level + 1],
605 p->slots[level + 1],
606 &cow_buf);
607 b = cow_buf;
608 c = btrfs_buffer_node(b);
609 }
610 BUG_ON(!cow && ins_len);
611 if (level != btrfs_header_level(&c->header))
612 WARN_ON(1);
613 level = btrfs_header_level(&c->header);
614 p->nodes[level] = b;
615 ret = check_block(root, p, level);
616 if (ret)
617 return -1;
618 ret = bin_search(c, key, &slot);
619 if (!btrfs_is_leaf(c)) {
620 if (ret && slot > 0)
621 slot -= 1;
622 p->slots[level] = slot;
623 if (ins_len > 0 && btrfs_header_nritems(&c->header) >=
624 BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
625 int sret = split_node(trans, root, p, level);
626 BUG_ON(sret > 0);
627 if (sret)
628 return sret;
629 b = p->nodes[level];
630 c = btrfs_buffer_node(b);
631 slot = p->slots[level];
632 } else if (ins_len < 0) {
633 int sret = balance_level(trans, root, p,
634 level);
635 if (sret)
636 return sret;
637 b = p->nodes[level];
638 if (!b)
639 goto again;
640 c = btrfs_buffer_node(b);
641 slot = p->slots[level];
642 BUG_ON(btrfs_header_nritems(&c->header) == 1);
643 }
644 b = read_tree_block(root, btrfs_node_blockptr(c, slot));
645 } else {
646 struct btrfs_leaf *l = (struct btrfs_leaf *)c;
647 p->slots[level] = slot;
648 if (ins_len > 0 && btrfs_leaf_free_space(root, l) <
649 sizeof(struct btrfs_item) + ins_len) {
650 int sret = split_leaf(trans, root, key,
651 p, ins_len);
652 BUG_ON(sret > 0);
653 if (sret)
654 return sret;
655 }
656 return ret;
657 }
658 }
659 return 1;
660 }
661
662 /*
663 * adjust the pointers going up the tree, starting at level
664 * making sure the right key of each node is points to 'key'.
665 * This is used after shifting pointers to the left, so it stops
666 * fixing up pointers when a given leaf/node is not in slot 0 of the
667 * higher levels
668 *
669 * If this fails to write a tree block, it returns -1, but continues
670 * fixing up the blocks in ram so the tree is consistent.
671 */
672 static int fixup_low_keys(struct btrfs_trans_handle *trans, struct btrfs_root
673 *root, struct btrfs_path *path, struct btrfs_disk_key
674 *key, int level)
675 {
676 int i;
677 int ret = 0;
678 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
679 struct btrfs_node *t;
680 int tslot = path->slots[i];
681 if (!path->nodes[i])
682 break;
683 t = btrfs_buffer_node(path->nodes[i]);
684 btrfs_memcpy(root, t, &t->ptrs[tslot].key, key, sizeof(*key));
685 btrfs_mark_buffer_dirty(path->nodes[i]);
686 if (tslot != 0)
687 break;
688 }
689 return ret;
690 }
691
692 /*
693 * try to push data from one node into the next node left in the
694 * tree.
695 *
696 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
697 * error, and > 0 if there was no room in the left hand block.
698 */
699 static int push_node_left(struct btrfs_trans_handle *trans, struct btrfs_root
700 *root, struct buffer_head *dst_buf, struct
701 buffer_head *src_buf)
702 {
703 struct btrfs_node *src = btrfs_buffer_node(src_buf);
704 struct btrfs_node *dst = btrfs_buffer_node(dst_buf);
705 int push_items = 0;
706 int src_nritems;
707 int dst_nritems;
708 int ret = 0;
709
710 src_nritems = btrfs_header_nritems(&src->header);
711 dst_nritems = btrfs_header_nritems(&dst->header);
712 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
713 if (push_items <= 0) {
714 return 1;
715 }
716
717 if (src_nritems < push_items)
718 push_items = src_nritems;
719
720 btrfs_memcpy(root, dst, dst->ptrs + dst_nritems, src->ptrs,
721 push_items * sizeof(struct btrfs_key_ptr));
722 if (push_items < src_nritems) {
723 btrfs_memmove(root, src, src->ptrs, src->ptrs + push_items,
724 (src_nritems - push_items) *
725 sizeof(struct btrfs_key_ptr));
726 }
727 btrfs_set_header_nritems(&src->header, src_nritems - push_items);
728 btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
729 btrfs_mark_buffer_dirty(src_buf);
730 btrfs_mark_buffer_dirty(dst_buf);
731 return ret;
732 }
733
734 /*
735 * try to push data from one node into the next node right in the
736 * tree.
737 *
738 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
739 * error, and > 0 if there was no room in the right hand block.
740 *
741 * this will only push up to 1/2 the contents of the left node over
742 */
743 static int balance_node_right(struct btrfs_trans_handle *trans, struct
744 btrfs_root *root, struct buffer_head *dst_buf,
745 struct buffer_head *src_buf)
746 {
747 struct btrfs_node *src = btrfs_buffer_node(src_buf);
748 struct btrfs_node *dst = btrfs_buffer_node(dst_buf);
749 int push_items = 0;
750 int max_push;
751 int src_nritems;
752 int dst_nritems;
753 int ret = 0;
754
755 src_nritems = btrfs_header_nritems(&src->header);
756 dst_nritems = btrfs_header_nritems(&dst->header);
757 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
758 if (push_items <= 0) {
759 return 1;
760 }
761
762 max_push = src_nritems / 2 + 1;
763 /* don't try to empty the node */
764 if (max_push > src_nritems)
765 return 1;
766 if (max_push < push_items)
767 push_items = max_push;
768
769 btrfs_memmove(root, dst, dst->ptrs + push_items, dst->ptrs,
770 dst_nritems * sizeof(struct btrfs_key_ptr));
771
772 btrfs_memcpy(root, dst, dst->ptrs,
773 src->ptrs + src_nritems - push_items,
774 push_items * sizeof(struct btrfs_key_ptr));
775
776 btrfs_set_header_nritems(&src->header, src_nritems - push_items);
777 btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
778
779 btrfs_mark_buffer_dirty(src_buf);
780 btrfs_mark_buffer_dirty(dst_buf);
781 return ret;
782 }
783
784 /*
785 * helper function to insert a new root level in the tree.
786 * A new node is allocated, and a single item is inserted to
787 * point to the existing root
788 *
789 * returns zero on success or < 0 on failure.
790 */
791 static int insert_new_root(struct btrfs_trans_handle *trans, struct btrfs_root
792 *root, struct btrfs_path *path, int level)
793 {
794 struct buffer_head *t;
795 struct btrfs_node *lower;
796 struct btrfs_node *c;
797 struct btrfs_disk_key *lower_key;
798
799 BUG_ON(path->nodes[level]);
800 BUG_ON(path->nodes[level-1] != root->node);
801
802 t = btrfs_alloc_free_block(trans, root);
803 c = btrfs_buffer_node(t);
804 memset(c, 0, root->blocksize);
805 btrfs_set_header_nritems(&c->header, 1);
806 btrfs_set_header_level(&c->header, level);
807 btrfs_set_header_blocknr(&c->header, bh_blocknr(t));
808 btrfs_set_header_generation(&c->header, trans->transid);
809 lower = btrfs_buffer_node(path->nodes[level-1]);
810 memcpy(c->header.fsid, root->fs_info->disk_super->fsid,
811 sizeof(c->header.fsid));
812 if (btrfs_is_leaf(lower))
813 lower_key = &((struct btrfs_leaf *)lower)->items[0].key;
814 else
815 lower_key = &lower->ptrs[0].key;
816 btrfs_memcpy(root, c, &c->ptrs[0].key, lower_key,
817 sizeof(struct btrfs_disk_key));
818 btrfs_set_node_blockptr(c, 0, bh_blocknr(path->nodes[level - 1]));
819
820 btrfs_mark_buffer_dirty(t);
821
822 /* the super has an extra ref to root->node */
823 btrfs_block_release(root, root->node);
824 root->node = t;
825 get_bh(t);
826 path->nodes[level] = t;
827 path->slots[level] = 0;
828 return 0;
829 }
830
831 /*
832 * worker function to insert a single pointer in a node.
833 * the node should have enough room for the pointer already
834 *
835 * slot and level indicate where you want the key to go, and
836 * blocknr is the block the key points to.
837 *
838 * returns zero on success and < 0 on any error
839 */
840 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
841 *root, struct btrfs_path *path, struct btrfs_disk_key
842 *key, u64 blocknr, int slot, int level)
843 {
844 struct btrfs_node *lower;
845 int nritems;
846
847 BUG_ON(!path->nodes[level]);
848 lower = btrfs_buffer_node(path->nodes[level]);
849 nritems = btrfs_header_nritems(&lower->header);
850 if (slot > nritems)
851 BUG();
852 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
853 BUG();
854 if (slot != nritems) {
855 btrfs_memmove(root, lower, lower->ptrs + slot + 1,
856 lower->ptrs + slot,
857 (nritems - slot) * sizeof(struct btrfs_key_ptr));
858 }
859 btrfs_memcpy(root, lower, &lower->ptrs[slot].key,
860 key, sizeof(struct btrfs_disk_key));
861 btrfs_set_node_blockptr(lower, slot, blocknr);
862 btrfs_set_header_nritems(&lower->header, nritems + 1);
863 btrfs_mark_buffer_dirty(path->nodes[level]);
864 return 0;
865 }
866
867 /*
868 * split the node at the specified level in path in two.
869 * The path is corrected to point to the appropriate node after the split
870 *
871 * Before splitting this tries to make some room in the node by pushing
872 * left and right, if either one works, it returns right away.
873 *
874 * returns 0 on success and < 0 on failure
875 */
876 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
877 *root, struct btrfs_path *path, int level)
878 {
879 struct buffer_head *t;
880 struct btrfs_node *c;
881 struct buffer_head *split_buffer;
882 struct btrfs_node *split;
883 int mid;
884 int ret;
885 int wret;
886 u32 c_nritems;
887
888 t = path->nodes[level];
889 c = btrfs_buffer_node(t);
890 if (t == root->node) {
891 /* trying to split the root, lets make a new one */
892 ret = insert_new_root(trans, root, path, level + 1);
893 if (ret)
894 return ret;
895 } else {
896 ret = push_nodes_for_insert(trans, root, path, level);
897 t = path->nodes[level];
898 c = btrfs_buffer_node(t);
899 if (!ret &&
900 btrfs_header_nritems(&c->header) <
901 BTRFS_NODEPTRS_PER_BLOCK(root) - 1)
902 return 0;
903 }
904
905 c_nritems = btrfs_header_nritems(&c->header);
906 split_buffer = btrfs_alloc_free_block(trans, root);
907 split = btrfs_buffer_node(split_buffer);
908 btrfs_set_header_flags(&split->header, btrfs_header_flags(&c->header));
909 btrfs_set_header_level(&split->header, btrfs_header_level(&c->header));
910 btrfs_set_header_blocknr(&split->header, bh_blocknr(split_buffer));
911 btrfs_set_header_generation(&split->header, trans->transid);
912 memcpy(split->header.fsid, root->fs_info->disk_super->fsid,
913 sizeof(split->header.fsid));
914 mid = (c_nritems + 1) / 2;
915 btrfs_memcpy(root, split, split->ptrs, c->ptrs + mid,
916 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
917 btrfs_set_header_nritems(&split->header, c_nritems - mid);
918 btrfs_set_header_nritems(&c->header, mid);
919 ret = 0;
920
921 btrfs_mark_buffer_dirty(t);
922 btrfs_mark_buffer_dirty(split_buffer);
923 wret = insert_ptr(trans, root, path, &split->ptrs[0].key,
924 bh_blocknr(split_buffer), path->slots[level + 1] + 1,
925 level + 1);
926 if (wret)
927 ret = wret;
928
929 if (path->slots[level] >= mid) {
930 path->slots[level] -= mid;
931 btrfs_block_release(root, t);
932 path->nodes[level] = split_buffer;
933 path->slots[level + 1] += 1;
934 } else {
935 btrfs_block_release(root, split_buffer);
936 }
937 return ret;
938 }
939
940 /*
941 * how many bytes are required to store the items in a leaf. start
942 * and nr indicate which items in the leaf to check. This totals up the
943 * space used both by the item structs and the item data
944 */
945 static int leaf_space_used(struct btrfs_leaf *l, int start, int nr)
946 {
947 int data_len;
948 int nritems = btrfs_header_nritems(&l->header);
949 int end = min(nritems, start + nr) - 1;
950
951 if (!nr)
952 return 0;
953 data_len = btrfs_item_end(l->items + start);
954 data_len = data_len - btrfs_item_offset(l->items + end);
955 data_len += sizeof(struct btrfs_item) * nr;
956 WARN_ON(data_len < 0);
957 return data_len;
958 }
959
960 /*
961 * The space between the end of the leaf items and
962 * the start of the leaf data. IOW, how much room
963 * the leaf has left for both items and data
964 */
965 int btrfs_leaf_free_space(struct btrfs_root *root, struct btrfs_leaf *leaf)
966 {
967 int nritems = btrfs_header_nritems(&leaf->header);
968 return BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
969 }
970
971 /*
972 * push some data in the path leaf to the right, trying to free up at
973 * least data_size bytes. returns zero if the push worked, nonzero otherwise
974 *
975 * returns 1 if the push failed because the other node didn't have enough
976 * room, 0 if everything worked out and < 0 if there were major errors.
977 */
978 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
979 *root, struct btrfs_path *path, int data_size)
980 {
981 struct buffer_head *left_buf = path->nodes[0];
982 struct btrfs_leaf *left = btrfs_buffer_leaf(left_buf);
983 struct btrfs_leaf *right;
984 struct buffer_head *right_buf;
985 struct buffer_head *upper;
986 struct btrfs_node *upper_node;
987 int slot;
988 int i;
989 int free_space;
990 int push_space = 0;
991 int push_items = 0;
992 struct btrfs_item *item;
993 u32 left_nritems;
994 u32 right_nritems;
995
996 slot = path->slots[1];
997 if (!path->nodes[1]) {
998 return 1;
999 }
1000 upper = path->nodes[1];
1001 upper_node = btrfs_buffer_node(upper);
1002 if (slot >= btrfs_header_nritems(&upper_node->header) - 1) {
1003 return 1;
1004 }
1005 right_buf = read_tree_block(root,
1006 btrfs_node_blockptr(btrfs_buffer_node(upper), slot + 1));
1007 right = btrfs_buffer_leaf(right_buf);
1008 free_space = btrfs_leaf_free_space(root, right);
1009 if (free_space < data_size + sizeof(struct btrfs_item)) {
1010 btrfs_block_release(root, right_buf);
1011 return 1;
1012 }
1013 /* cow and double check */
1014 btrfs_cow_block(trans, root, right_buf, upper, slot + 1, &right_buf);
1015 right = btrfs_buffer_leaf(right_buf);
1016 free_space = btrfs_leaf_free_space(root, right);
1017 if (free_space < data_size + sizeof(struct btrfs_item)) {
1018 btrfs_block_release(root, right_buf);
1019 return 1;
1020 }
1021
1022 left_nritems = btrfs_header_nritems(&left->header);
1023 if (left_nritems == 0) {
1024 btrfs_block_release(root, right_buf);
1025 return 1;
1026 }
1027 for (i = left_nritems - 1; i >= 1; i--) {
1028 item = left->items + i;
1029 if (path->slots[0] == i)
1030 push_space += data_size + sizeof(*item);
1031 if (btrfs_item_size(item) + sizeof(*item) + push_space >
1032 free_space)
1033 break;
1034 push_items++;
1035 push_space += btrfs_item_size(item) + sizeof(*item);
1036 }
1037 if (push_items == 0) {
1038 btrfs_block_release(root, right_buf);
1039 return 1;
1040 }
1041 if (push_items == left_nritems)
1042 WARN_ON(1);
1043 right_nritems = btrfs_header_nritems(&right->header);
1044 /* push left to right */
1045 push_space = btrfs_item_end(left->items + left_nritems - push_items);
1046 push_space -= leaf_data_end(root, left);
1047 /* make room in the right data area */
1048 btrfs_memmove(root, right, btrfs_leaf_data(right) +
1049 leaf_data_end(root, right) - push_space,
1050 btrfs_leaf_data(right) +
1051 leaf_data_end(root, right), BTRFS_LEAF_DATA_SIZE(root) -
1052 leaf_data_end(root, right));
1053 /* copy from the left data area */
1054 btrfs_memcpy(root, right, btrfs_leaf_data(right) +
1055 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1056 btrfs_leaf_data(left) + leaf_data_end(root, left),
1057 push_space);
1058 btrfs_memmove(root, right, right->items + push_items, right->items,
1059 right_nritems * sizeof(struct btrfs_item));
1060 /* copy the items from left to right */
1061 btrfs_memcpy(root, right, right->items, left->items +
1062 left_nritems - push_items,
1063 push_items * sizeof(struct btrfs_item));
1064
1065 /* update the item pointers */
1066 right_nritems += push_items;
1067 btrfs_set_header_nritems(&right->header, right_nritems);
1068 push_space = BTRFS_LEAF_DATA_SIZE(root);
1069 for (i = 0; i < right_nritems; i++) {
1070 btrfs_set_item_offset(right->items + i, push_space -
1071 btrfs_item_size(right->items + i));
1072 push_space = btrfs_item_offset(right->items + i);
1073 }
1074 left_nritems -= push_items;
1075 btrfs_set_header_nritems(&left->header, left_nritems);
1076
1077 btrfs_mark_buffer_dirty(left_buf);
1078 btrfs_mark_buffer_dirty(right_buf);
1079
1080 btrfs_memcpy(root, upper_node, &upper_node->ptrs[slot + 1].key,
1081 &right->items[0].key, sizeof(struct btrfs_disk_key));
1082 btrfs_mark_buffer_dirty(upper);
1083
1084 /* then fixup the leaf pointer in the path */
1085 if (path->slots[0] >= left_nritems) {
1086 path->slots[0] -= left_nritems;
1087 btrfs_block_release(root, path->nodes[0]);
1088 path->nodes[0] = right_buf;
1089 path->slots[1] += 1;
1090 } else {
1091 btrfs_block_release(root, right_buf);
1092 }
1093 return 0;
1094 }
1095 /*
1096 * push some data in the path leaf to the left, trying to free up at
1097 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1098 */
1099 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
1100 *root, struct btrfs_path *path, int data_size)
1101 {
1102 struct buffer_head *right_buf = path->nodes[0];
1103 struct btrfs_leaf *right = btrfs_buffer_leaf(right_buf);
1104 struct buffer_head *t;
1105 struct btrfs_leaf *left;
1106 int slot;
1107 int i;
1108 int free_space;
1109 int push_space = 0;
1110 int push_items = 0;
1111 struct btrfs_item *item;
1112 u32 old_left_nritems;
1113 int ret = 0;
1114 int wret;
1115
1116 slot = path->slots[1];
1117 if (slot == 0) {
1118 return 1;
1119 }
1120 if (!path->nodes[1]) {
1121 return 1;
1122 }
1123 t = read_tree_block(root,
1124 btrfs_node_blockptr(btrfs_buffer_node(path->nodes[1]), slot - 1));
1125 left = btrfs_buffer_leaf(t);
1126 free_space = btrfs_leaf_free_space(root, left);
1127 if (free_space < data_size + sizeof(struct btrfs_item)) {
1128 btrfs_block_release(root, t);
1129 return 1;
1130 }
1131
1132 /* cow and double check */
1133 btrfs_cow_block(trans, root, t, path->nodes[1], slot - 1, &t);
1134 left = btrfs_buffer_leaf(t);
1135 free_space = btrfs_leaf_free_space(root, left);
1136 if (free_space < data_size + sizeof(struct btrfs_item)) {
1137 btrfs_block_release(root, t);
1138 return 1;
1139 }
1140
1141 if (btrfs_header_nritems(&right->header) == 0) {
1142 btrfs_block_release(root, t);
1143 return 1;
1144 }
1145
1146 for (i = 0; i < btrfs_header_nritems(&right->header) - 1; i++) {
1147 item = right->items + i;
1148 if (path->slots[0] == i)
1149 push_space += data_size + sizeof(*item);
1150 if (btrfs_item_size(item) + sizeof(*item) + push_space >
1151 free_space)
1152 break;
1153 push_items++;
1154 push_space += btrfs_item_size(item) + sizeof(*item);
1155 }
1156 if (push_items == 0) {
1157 btrfs_block_release(root, t);
1158 return 1;
1159 }
1160 if (push_items == btrfs_header_nritems(&right->header))
1161 WARN_ON(1);
1162 /* push data from right to left */
1163 btrfs_memcpy(root, left, left->items +
1164 btrfs_header_nritems(&left->header),
1165 right->items, push_items * sizeof(struct btrfs_item));
1166 push_space = BTRFS_LEAF_DATA_SIZE(root) -
1167 btrfs_item_offset(right->items + push_items -1);
1168 btrfs_memcpy(root, left, btrfs_leaf_data(left) +
1169 leaf_data_end(root, left) - push_space,
1170 btrfs_leaf_data(right) +
1171 btrfs_item_offset(right->items + push_items - 1),
1172 push_space);
1173 old_left_nritems = btrfs_header_nritems(&left->header);
1174 BUG_ON(old_left_nritems < 0);
1175
1176 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
1177 u32 ioff = btrfs_item_offset(left->items + i);
1178 btrfs_set_item_offset(left->items + i, ioff -
1179 (BTRFS_LEAF_DATA_SIZE(root) -
1180 btrfs_item_offset(left->items +
1181 old_left_nritems - 1)));
1182 }
1183 btrfs_set_header_nritems(&left->header, old_left_nritems + push_items);
1184
1185 /* fixup right node */
1186 push_space = btrfs_item_offset(right->items + push_items - 1) -
1187 leaf_data_end(root, right);
1188 btrfs_memmove(root, right, btrfs_leaf_data(right) +
1189 BTRFS_LEAF_DATA_SIZE(root) - push_space,
1190 btrfs_leaf_data(right) +
1191 leaf_data_end(root, right), push_space);
1192 btrfs_memmove(root, right, right->items, right->items + push_items,
1193 (btrfs_header_nritems(&right->header) - push_items) *
1194 sizeof(struct btrfs_item));
1195 btrfs_set_header_nritems(&right->header,
1196 btrfs_header_nritems(&right->header) -
1197 push_items);
1198 push_space = BTRFS_LEAF_DATA_SIZE(root);
1199
1200 for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
1201 btrfs_set_item_offset(right->items + i, push_space -
1202 btrfs_item_size(right->items + i));
1203 push_space = btrfs_item_offset(right->items + i);
1204 }
1205
1206 btrfs_mark_buffer_dirty(t);
1207 btrfs_mark_buffer_dirty(right_buf);
1208 wret = fixup_low_keys(trans, root, path, &right->items[0].key, 1);
1209 if (wret)
1210 ret = wret;
1211
1212 /* then fixup the leaf pointer in the path */
1213 if (path->slots[0] < push_items) {
1214 path->slots[0] += old_left_nritems;
1215 btrfs_block_release(root, path->nodes[0]);
1216 path->nodes[0] = t;
1217 path->slots[1] -= 1;
1218 } else {
1219 btrfs_block_release(root, t);
1220 path->slots[0] -= push_items;
1221 }
1222 BUG_ON(path->slots[0] < 0);
1223 return ret;
1224 }
1225
1226 /*
1227 * split the path's leaf in two, making sure there is at least data_size
1228 * available for the resulting leaf level of the path.
1229 *
1230 * returns 0 if all went well and < 0 on failure.
1231 */
1232 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
1233 *root, struct btrfs_key *ins_key,
1234 struct btrfs_path *path, int data_size)
1235 {
1236 struct buffer_head *l_buf;
1237 struct btrfs_leaf *l;
1238 u32 nritems;
1239 int mid;
1240 int slot;
1241 struct btrfs_leaf *right;
1242 struct buffer_head *right_buffer;
1243 int space_needed = data_size + sizeof(struct btrfs_item);
1244 int data_copy_size;
1245 int rt_data_off;
1246 int i;
1247 int ret = 0;
1248 int wret;
1249 int double_split = 0;
1250 struct btrfs_disk_key disk_key;
1251
1252 /* first try to make some room by pushing left and right */
1253 wret = push_leaf_left(trans, root, path, data_size);
1254 if (wret < 0)
1255 return wret;
1256 if (wret) {
1257 wret = push_leaf_right(trans, root, path, data_size);
1258 if (wret < 0)
1259 return wret;
1260 }
1261 l_buf = path->nodes[0];
1262 l = btrfs_buffer_leaf(l_buf);
1263
1264 /* did the pushes work? */
1265 if (btrfs_leaf_free_space(root, l) >=
1266 sizeof(struct btrfs_item) + data_size)
1267 return 0;
1268
1269 if (!path->nodes[1]) {
1270 ret = insert_new_root(trans, root, path, 1);
1271 if (ret)
1272 return ret;
1273 }
1274 slot = path->slots[0];
1275 nritems = btrfs_header_nritems(&l->header);
1276 mid = (nritems + 1)/ 2;
1277 right_buffer = btrfs_alloc_free_block(trans, root);
1278 BUG_ON(!right_buffer);
1279 right = btrfs_buffer_leaf(right_buffer);
1280 memset(&right->header, 0, sizeof(right->header));
1281 btrfs_set_header_blocknr(&right->header, bh_blocknr(right_buffer));
1282 btrfs_set_header_generation(&right->header, trans->transid);
1283 btrfs_set_header_level(&right->header, 0);
1284 memcpy(right->header.fsid, root->fs_info->disk_super->fsid,
1285 sizeof(right->header.fsid));
1286 if (mid <= slot) {
1287 if (nritems == 1 ||
1288 leaf_space_used(l, mid, nritems - mid) + space_needed >
1289 BTRFS_LEAF_DATA_SIZE(root)) {
1290 if (slot >= nritems) {
1291 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1292 btrfs_set_header_nritems(&right->header, 0);
1293 wret = insert_ptr(trans, root, path,
1294 &disk_key,
1295 bh_blocknr(right_buffer),
1296 path->slots[1] + 1, 1);
1297 if (wret)
1298 ret = wret;
1299 btrfs_block_release(root, path->nodes[0]);
1300 path->nodes[0] = right_buffer;
1301 path->slots[0] = 0;
1302 path->slots[1] += 1;
1303 return ret;
1304 }
1305 mid = slot;
1306 double_split = 1;
1307 }
1308 } else {
1309 if (leaf_space_used(l, 0, mid + 1) + space_needed >
1310 BTRFS_LEAF_DATA_SIZE(root)) {
1311 if (slot == 0) {
1312 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1313 btrfs_set_header_nritems(&right->header, 0);
1314 wret = insert_ptr(trans, root, path,
1315 &disk_key,
1316 bh_blocknr(right_buffer),
1317 path->slots[1] - 1, 1);
1318 if (wret)
1319 ret = wret;
1320 btrfs_block_release(root, path->nodes[0]);
1321 path->nodes[0] = right_buffer;
1322 path->slots[0] = 0;
1323 path->slots[1] -= 1;
1324 if (path->slots[1] == 0) {
1325 wret = fixup_low_keys(trans, root,
1326 path, &disk_key, 1);
1327 if (wret)
1328 ret = wret;
1329 }
1330 return ret;
1331 }
1332 mid = slot;
1333 double_split = 1;
1334 }
1335 }
1336 btrfs_set_header_nritems(&right->header, nritems - mid);
1337 data_copy_size = btrfs_item_end(l->items + mid) -
1338 leaf_data_end(root, l);
1339 btrfs_memcpy(root, right, right->items, l->items + mid,
1340 (nritems - mid) * sizeof(struct btrfs_item));
1341 btrfs_memcpy(root, right,
1342 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
1343 data_copy_size, btrfs_leaf_data(l) +
1344 leaf_data_end(root, l), data_copy_size);
1345 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
1346 btrfs_item_end(l->items + mid);
1347
1348 for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
1349 u32 ioff = btrfs_item_offset(right->items + i);
1350 btrfs_set_item_offset(right->items + i, ioff + rt_data_off);
1351 }
1352
1353 btrfs_set_header_nritems(&l->header, mid);
1354 ret = 0;
1355 wret = insert_ptr(trans, root, path, &right->items[0].key,
1356 bh_blocknr(right_buffer), path->slots[1] + 1, 1);
1357 if (wret)
1358 ret = wret;
1359 btrfs_mark_buffer_dirty(right_buffer);
1360 btrfs_mark_buffer_dirty(l_buf);
1361 BUG_ON(path->slots[0] != slot);
1362 if (mid <= slot) {
1363 btrfs_block_release(root, path->nodes[0]);
1364 path->nodes[0] = right_buffer;
1365 path->slots[0] -= mid;
1366 path->slots[1] += 1;
1367 } else
1368 btrfs_block_release(root, right_buffer);
1369 BUG_ON(path->slots[0] < 0);
1370
1371 if (!double_split)
1372 return ret;
1373 right_buffer = btrfs_alloc_free_block(trans, root);
1374 BUG_ON(!right_buffer);
1375 right = btrfs_buffer_leaf(right_buffer);
1376 memset(&right->header, 0, sizeof(right->header));
1377 btrfs_set_header_blocknr(&right->header, bh_blocknr(right_buffer));
1378 btrfs_set_header_generation(&right->header, trans->transid);
1379 btrfs_set_header_level(&right->header, 0);
1380 memcpy(right->header.fsid, root->fs_info->disk_super->fsid,
1381 sizeof(right->header.fsid));
1382 btrfs_cpu_key_to_disk(&disk_key, ins_key);
1383 btrfs_set_header_nritems(&right->header, 0);
1384 wret = insert_ptr(trans, root, path,
1385 &disk_key,
1386 bh_blocknr(right_buffer),
1387 path->slots[1], 1);
1388 if (wret)
1389 ret = wret;
1390 if (path->slots[1] == 0) {
1391 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
1392 if (wret)
1393 ret = wret;
1394 }
1395 btrfs_block_release(root, path->nodes[0]);
1396 path->nodes[0] = right_buffer;
1397 path->slots[0] = 0;
1398 check_node(root, path, 1);
1399 check_leaf(root, path, 0);
1400 return ret;
1401 }
1402
1403 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root,
1405 struct btrfs_path *path,
1406 u32 new_size)
1407 {
1408 int ret = 0;
1409 int slot;
1410 int slot_orig;
1411 struct btrfs_leaf *leaf;
1412 struct buffer_head *leaf_buf;
1413 u32 nritems;
1414 unsigned int data_end;
1415 unsigned int old_data_start;
1416 unsigned int old_size;
1417 unsigned int size_diff;
1418 int i;
1419
1420 slot_orig = path->slots[0];
1421 leaf_buf = path->nodes[0];
1422 leaf = btrfs_buffer_leaf(leaf_buf);
1423
1424 nritems = btrfs_header_nritems(&leaf->header);
1425 data_end = leaf_data_end(root, leaf);
1426
1427 slot = path->slots[0];
1428 old_data_start = btrfs_item_offset(leaf->items + slot);
1429 old_size = btrfs_item_size(leaf->items + slot);
1430 BUG_ON(old_size <= new_size);
1431 size_diff = old_size - new_size;
1432
1433 BUG_ON(slot < 0);
1434 BUG_ON(slot >= nritems);
1435
1436 /*
1437 * item0..itemN ... dataN.offset..dataN.size .. data0.size
1438 */
1439 /* first correct the data pointers */
1440 for (i = slot; i < nritems; i++) {
1441 u32 ioff = btrfs_item_offset(leaf->items + i);
1442 btrfs_set_item_offset(leaf->items + i,
1443 ioff + size_diff);
1444 }
1445 /* shift the data */
1446 btrfs_memmove(root, leaf, btrfs_leaf_data(leaf) +
1447 data_end + size_diff, btrfs_leaf_data(leaf) +
1448 data_end, old_data_start + new_size - data_end);
1449 btrfs_set_item_size(leaf->items + slot, new_size);
1450 btrfs_mark_buffer_dirty(leaf_buf);
1451
1452 ret = 0;
1453 if (btrfs_leaf_free_space(root, leaf) < 0)
1454 BUG();
1455 check_leaf(root, path, 0);
1456 return ret;
1457 }
1458
1459 int btrfs_extend_item(struct btrfs_trans_handle *trans, struct btrfs_root
1460 *root, struct btrfs_path *path, u32 data_size)
1461 {
1462 int ret = 0;
1463 int slot;
1464 int slot_orig;
1465 struct btrfs_leaf *leaf;
1466 struct buffer_head *leaf_buf;
1467 u32 nritems;
1468 unsigned int data_end;
1469 unsigned int old_data;
1470 unsigned int old_size;
1471 int i;
1472
1473 slot_orig = path->slots[0];
1474 leaf_buf = path->nodes[0];
1475 leaf = btrfs_buffer_leaf(leaf_buf);
1476
1477 nritems = btrfs_header_nritems(&leaf->header);
1478 data_end = leaf_data_end(root, leaf);
1479
1480 if (btrfs_leaf_free_space(root, leaf) < data_size)
1481 BUG();
1482 slot = path->slots[0];
1483 old_data = btrfs_item_end(leaf->items + slot);
1484
1485 BUG_ON(slot < 0);
1486 BUG_ON(slot >= nritems);
1487
1488 /*
1489 * item0..itemN ... dataN.offset..dataN.size .. data0.size
1490 */
1491 /* first correct the data pointers */
1492 for (i = slot; i < nritems; i++) {
1493 u32 ioff = btrfs_item_offset(leaf->items + i);
1494 btrfs_set_item_offset(leaf->items + i,
1495 ioff - data_size);
1496 }
1497 /* shift the data */
1498 btrfs_memmove(root, leaf, btrfs_leaf_data(leaf) +
1499 data_end - data_size, btrfs_leaf_data(leaf) +
1500 data_end, old_data - data_end);
1501 data_end = old_data;
1502 old_size = btrfs_item_size(leaf->items + slot);
1503 btrfs_set_item_size(leaf->items + slot, old_size + data_size);
1504 btrfs_mark_buffer_dirty(leaf_buf);
1505
1506 ret = 0;
1507 if (btrfs_leaf_free_space(root, leaf) < 0)
1508 BUG();
1509 check_leaf(root, path, 0);
1510 return ret;
1511 }
1512
1513 /*
1514 * Given a key and some data, insert an item into the tree.
1515 * This does all the path init required, making room in the tree if needed.
1516 */
1517 int btrfs_insert_empty_item(struct btrfs_trans_handle *trans, struct btrfs_root
1518 *root, struct btrfs_path *path, struct btrfs_key
1519 *cpu_key, u32 data_size)
1520 {
1521 int ret = 0;
1522 int slot;
1523 int slot_orig;
1524 struct btrfs_leaf *leaf;
1525 struct buffer_head *leaf_buf;
1526 u32 nritems;
1527 unsigned int data_end;
1528 struct btrfs_disk_key disk_key;
1529
1530 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
1531
1532 /* create a root if there isn't one */
1533 if (!root->node)
1534 BUG();
1535 ret = btrfs_search_slot(trans, root, cpu_key, path, data_size, 1);
1536 if (ret == 0) {
1537 return -EEXIST;
1538 }
1539 if (ret < 0)
1540 goto out;
1541
1542 slot_orig = path->slots[0];
1543 leaf_buf = path->nodes[0];
1544 leaf = btrfs_buffer_leaf(leaf_buf);
1545
1546 nritems = btrfs_header_nritems(&leaf->header);
1547 data_end = leaf_data_end(root, leaf);
1548
1549 if (btrfs_leaf_free_space(root, leaf) <
1550 sizeof(struct btrfs_item) + data_size) {
1551 BUG();
1552 }
1553 slot = path->slots[0];
1554 BUG_ON(slot < 0);
1555 if (slot != nritems) {
1556 int i;
1557 unsigned int old_data = btrfs_item_end(leaf->items + slot);
1558
1559 /*
1560 * item0..itemN ... dataN.offset..dataN.size .. data0.size
1561 */
1562 /* first correct the data pointers */
1563 for (i = slot; i < nritems; i++) {
1564 u32 ioff = btrfs_item_offset(leaf->items + i);
1565 btrfs_set_item_offset(leaf->items + i,
1566 ioff - data_size);
1567 }
1568
1569 /* shift the items */
1570 btrfs_memmove(root, leaf, leaf->items + slot + 1,
1571 leaf->items + slot,
1572 (nritems - slot) * sizeof(struct btrfs_item));
1573
1574 /* shift the data */
1575 btrfs_memmove(root, leaf, btrfs_leaf_data(leaf) +
1576 data_end - data_size, btrfs_leaf_data(leaf) +
1577 data_end, old_data - data_end);
1578 data_end = old_data;
1579 }
1580 /* setup the item for the new data */
1581 btrfs_memcpy(root, leaf, &leaf->items[slot].key, &disk_key,
1582 sizeof(struct btrfs_disk_key));
1583 btrfs_set_item_offset(leaf->items + slot, data_end - data_size);
1584 btrfs_set_item_size(leaf->items + slot, data_size);
1585 btrfs_set_header_nritems(&leaf->header, nritems + 1);
1586 btrfs_mark_buffer_dirty(leaf_buf);
1587
1588 ret = 0;
1589 if (slot == 0)
1590 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
1591
1592 if (btrfs_leaf_free_space(root, leaf) < 0)
1593 BUG();
1594 check_leaf(root, path, 0);
1595 out:
1596 return ret;
1597 }
1598
1599 /*
1600 * Given a key and some data, insert an item into the tree.
1601 * This does all the path init required, making room in the tree if needed.
1602 */
1603 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
1604 *root, struct btrfs_key *cpu_key, void *data, u32
1605 data_size)
1606 {
1607 int ret = 0;
1608 struct btrfs_path *path;
1609 u8 *ptr;
1610
1611 path = btrfs_alloc_path();
1612 BUG_ON(!path);
1613 btrfs_init_path(path);
1614 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
1615 if (!ret) {
1616 ptr = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1617 path->slots[0], u8);
1618 btrfs_memcpy(root, path->nodes[0]->b_data,
1619 ptr, data, data_size);
1620 btrfs_mark_buffer_dirty(path->nodes[0]);
1621 }
1622 btrfs_release_path(root, path);
1623 btrfs_free_path(path);
1624 return ret;
1625 }
1626
1627 /*
1628 * delete the pointer from a given node.
1629 *
1630 * If the delete empties a node, the node is removed from the tree,
1631 * continuing all the way the root if required. The root is converted into
1632 * a leaf if all the nodes are emptied.
1633 */
1634 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1635 struct btrfs_path *path, int level, int slot)
1636 {
1637 struct btrfs_node *node;
1638 struct buffer_head *parent = path->nodes[level];
1639 u32 nritems;
1640 int ret = 0;
1641 int wret;
1642
1643 node = btrfs_buffer_node(parent);
1644 nritems = btrfs_header_nritems(&node->header);
1645 if (slot != nritems -1) {
1646 btrfs_memmove(root, node, node->ptrs + slot,
1647 node->ptrs + slot + 1,
1648 sizeof(struct btrfs_key_ptr) *
1649 (nritems - slot - 1));
1650 }
1651 nritems--;
1652 btrfs_set_header_nritems(&node->header, nritems);
1653 if (nritems == 0 && parent == root->node) {
1654 struct btrfs_header *header = btrfs_buffer_header(root->node);
1655 BUG_ON(btrfs_header_level(header) != 1);
1656 /* just turn the root into a leaf and break */
1657 btrfs_set_header_level(header, 0);
1658 } else if (slot == 0) {
1659 wret = fixup_low_keys(trans, root, path, &node->ptrs[0].key,
1660 level + 1);
1661 if (wret)
1662 ret = wret;
1663 }
1664 btrfs_mark_buffer_dirty(parent);
1665 return ret;
1666 }
1667
1668 /*
1669 * delete the item at the leaf level in path. If that empties
1670 * the leaf, remove it from the tree
1671 */
1672 int btrfs_del_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1673 struct btrfs_path *path)
1674 {
1675 int slot;
1676 struct btrfs_leaf *leaf;
1677 struct buffer_head *leaf_buf;
1678 int doff;
1679 int dsize;
1680 int ret = 0;
1681 int wret;
1682 u32 nritems;
1683
1684 leaf_buf = path->nodes[0];
1685 leaf = btrfs_buffer_leaf(leaf_buf);
1686 slot = path->slots[0];
1687 doff = btrfs_item_offset(leaf->items + slot);
1688 dsize = btrfs_item_size(leaf->items + slot);
1689 nritems = btrfs_header_nritems(&leaf->header);
1690
1691 if (slot != nritems - 1) {
1692 int i;
1693 int data_end = leaf_data_end(root, leaf);
1694 btrfs_memmove(root, leaf, btrfs_leaf_data(leaf) +
1695 data_end + dsize,
1696 btrfs_leaf_data(leaf) + data_end,
1697 doff - data_end);
1698 for (i = slot + 1; i < nritems; i++) {
1699 u32 ioff = btrfs_item_offset(leaf->items + i);
1700 btrfs_set_item_offset(leaf->items + i, ioff + dsize);
1701 }
1702 btrfs_memmove(root, leaf, leaf->items + slot,
1703 leaf->items + slot + 1,
1704 sizeof(struct btrfs_item) *
1705 (nritems - slot - 1));
1706 }
1707 btrfs_set_header_nritems(&leaf->header, nritems - 1);
1708 nritems--;
1709 /* delete the leaf if we've emptied it */
1710 if (nritems == 0) {
1711 if (leaf_buf == root->node) {
1712 btrfs_set_header_level(&leaf->header, 0);
1713 } else {
1714 clean_tree_block(trans, root, leaf_buf);
1715 wait_on_buffer(leaf_buf);
1716 wret = del_ptr(trans, root, path, 1, path->slots[1]);
1717 if (wret)
1718 ret = wret;
1719 wret = btrfs_free_extent(trans, root,
1720 bh_blocknr(leaf_buf), 1, 1);
1721 if (wret)
1722 ret = wret;
1723 }
1724 } else {
1725 int used = leaf_space_used(leaf, 0, nritems);
1726 if (slot == 0) {
1727 wret = fixup_low_keys(trans, root, path,
1728 &leaf->items[0].key, 1);
1729 if (wret)
1730 ret = wret;
1731 }
1732
1733 /* delete the leaf if it is mostly empty */
1734 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
1735 /* push_leaf_left fixes the path.
1736 * make sure the path still points to our leaf
1737 * for possible call to del_ptr below
1738 */
1739 slot = path->slots[1];
1740 get_bh(leaf_buf);
1741 wret = push_leaf_left(trans, root, path, 1);
1742 if (wret < 0)
1743 ret = wret;
1744 if (path->nodes[0] == leaf_buf &&
1745 btrfs_header_nritems(&leaf->header)) {
1746 wret = push_leaf_right(trans, root, path, 1);
1747 if (wret < 0)
1748 ret = wret;
1749 }
1750 if (btrfs_header_nritems(&leaf->header) == 0) {
1751 u64 blocknr = bh_blocknr(leaf_buf);
1752 clean_tree_block(trans, root, leaf_buf);
1753 wait_on_buffer(leaf_buf);
1754 wret = del_ptr(trans, root, path, 1, slot);
1755 if (wret)
1756 ret = wret;
1757 btrfs_block_release(root, leaf_buf);
1758 wret = btrfs_free_extent(trans, root, blocknr,
1759 1, 1);
1760 if (wret)
1761 ret = wret;
1762 } else {
1763 btrfs_mark_buffer_dirty(leaf_buf);
1764 btrfs_block_release(root, leaf_buf);
1765 }
1766 } else {
1767 btrfs_mark_buffer_dirty(leaf_buf);
1768 }
1769 }
1770 return ret;
1771 }
1772
1773 /*
1774 * walk up the tree as far as required to find the next leaf.
1775 * returns 0 if it found something or 1 if there are no greater leaves.
1776 * returns < 0 on io errors.
1777 */
1778 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
1779 {
1780 int slot;
1781 int level = 1;
1782 u64 blocknr;
1783 struct buffer_head *c;
1784 struct btrfs_node *c_node;
1785 struct buffer_head *next = NULL;
1786
1787 while(level < BTRFS_MAX_LEVEL) {
1788 if (!path->nodes[level])
1789 return 1;
1790 slot = path->slots[level] + 1;
1791 c = path->nodes[level];
1792 c_node = btrfs_buffer_node(c);
1793 if (slot >= btrfs_header_nritems(&c_node->header)) {
1794 level++;
1795 continue;
1796 }
1797 blocknr = btrfs_node_blockptr(c_node, slot);
1798 if (next)
1799 btrfs_block_release(root, next);
1800 next = read_tree_block(root, blocknr);
1801 break;
1802 }
1803 path->slots[level] = slot;
1804 while(1) {
1805 level--;
1806 c = path->nodes[level];
1807 btrfs_block_release(root, c);
1808 path->nodes[level] = next;
1809 path->slots[level] = 0;
1810 if (!level)
1811 break;
1812 next = read_tree_block(root,
1813 btrfs_node_blockptr(btrfs_buffer_node(next), 0));
1814 }
1815 return 0;
1816 }
Linux kernel - Deliverables
This page took 0.077798 seconds and 6 git commands to generate.