ctree.c 38.2 KB
Newer Older
1 2 3
#include <stdio.h>
#include <stdlib.h>
#include "kerncompat.h"
4 5 6
#include "radix-tree.h"
#include "ctree.h"
#include "disk-io.h"
C
Chris Mason 已提交
7
#include "print-tree.h"
8

C
Chris Mason 已提交
9
static int split_node(struct btrfs_root *root, struct btrfs_path *path,
C
Chris Mason 已提交
10
		      int level);
C
Chris Mason 已提交
11
static int split_leaf(struct btrfs_root *root, struct btrfs_path *path,
C
Chris Mason 已提交
12
		      int data_size);
C
Chris Mason 已提交
13 14 15 16 17 18
static int push_node_left(struct btrfs_root *root, struct btrfs_buffer *dst,
			  struct btrfs_buffer *src);
static int balance_node_right(struct btrfs_root *root,
			      struct btrfs_buffer *dst_buf,
			      struct btrfs_buffer *src_buf);
static int del_ptr(struct btrfs_root *root, struct btrfs_path *path, int level,
19
		   int slot);
20

C
Chris Mason 已提交
21
inline void btrfs_init_path(struct btrfs_path *p)
22 23 24 25
{
	memset(p, 0, sizeof(*p));
}

C
Chris Mason 已提交
26
void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
27 28
{
	int i;
C
Chris Mason 已提交
29
	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
30 31
		if (!p->nodes[i])
			break;
C
Chris Mason 已提交
32
		btrfs_block_release(root, p->nodes[i]);
33
	}
C
Chris Mason 已提交
34
	memset(p, 0, sizeof(*p));
35 36
}

C
Chris Mason 已提交
37
static int btrfs_cow_block(struct btrfs_root *root,
C
Chris Mason 已提交
38 39
		    struct btrfs_buffer *buf,
		    struct btrfs_buffer *parent,
C
Chris Mason 已提交
40
		    int parent_slot,
C
Chris Mason 已提交
41
		    struct btrfs_buffer **cow_ret)
C
Chris Mason 已提交
42
{
C
Chris Mason 已提交
43
	struct btrfs_buffer *cow;
C
Chris Mason 已提交
44 45 46 47 48

	if (!list_empty(&buf->dirty)) {
		*cow_ret = buf;
		return 0;
	}
C
Chris Mason 已提交
49
	cow = btrfs_alloc_free_block(root);
C
Chris Mason 已提交
50
	memcpy(&cow->node, &buf->node, sizeof(buf->node));
51
	btrfs_set_header_blocknr(&cow->node.header, cow->blocknr);
C
Chris Mason 已提交
52
	*cow_ret = cow;
53
	btrfs_inc_ref(root, buf);
C
Chris Mason 已提交
54 55 56
	if (buf == root->node) {
		root->node = cow;
		cow->count++;
57
		if (buf != root->commit_root)
C
Chris Mason 已提交
58 59
			btrfs_free_extent(root, buf->blocknr, 1);
		btrfs_block_release(root, buf);
C
Chris Mason 已提交
60
	} else {
61 62
		btrfs_set_node_blockptr(&parent->node, parent_slot,
					cow->blocknr);
C
Chris Mason 已提交
63
		BUG_ON(list_empty(&parent->dirty));
C
Chris Mason 已提交
64
		btrfs_free_extent(root, buf->blocknr, 1);
C
Chris Mason 已提交
65
	}
C
Chris Mason 已提交
66
	btrfs_block_release(root, buf);
C
Chris Mason 已提交
67 68 69
	return 0;
}

C
Chris Mason 已提交
70 71 72 73 74
/*
 * The leaf data grows from end-to-front in the node.
 * this returns the address of the start of the last item,
 * which is the stop of the leaf data stack
 */
C
Chris Mason 已提交
75
static inline unsigned int leaf_data_end(struct btrfs_leaf *leaf)
76
{
77
	u32 nr = btrfs_header_nritems(&leaf->header);
78
	if (nr == 0)
79
		return sizeof(leaf->data);
C
Chris Mason 已提交
80
	return btrfs_item_offset(leaf->items + nr - 1);
81 82
}

C
Chris Mason 已提交
83 84 85 86 87
/*
 * The space between the end of the leaf items and
 * the start of the leaf data.  IOW, how much room
 * the leaf has left for both items and data
 */
C
Chris Mason 已提交
88
int btrfs_leaf_free_space(struct btrfs_leaf *leaf)
89 90
{
	int data_end = leaf_data_end(leaf);
91
	int nritems = btrfs_header_nritems(&leaf->header);
92 93 94 95
	char *items_end = (char *)(leaf->items + nritems + 1);
	return (char *)(leaf->data + data_end) - (char *)items_end;
}

C
Chris Mason 已提交
96 97 98
/*
 * compare two keys in a memcmp fashion
 */
C
Chris Mason 已提交
99
static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
100
{
C
Chris Mason 已提交
101 102 103 104 105
	struct btrfs_key k1;

	btrfs_disk_key_to_cpu(&k1, disk);

	if (k1.objectid > k2->objectid)
106
		return 1;
C
Chris Mason 已提交
107
	if (k1.objectid < k2->objectid)
108
		return -1;
C
Chris Mason 已提交
109
	if (k1.flags > k2->flags)
110
		return 1;
C
Chris Mason 已提交
111
	if (k1.flags < k2->flags)
112
		return -1;
C
Chris Mason 已提交
113
	if (k1.offset > k2->offset)
114
		return 1;
C
Chris Mason 已提交
115
	if (k1.offset < k2->offset)
116 117 118
		return -1;
	return 0;
}
C
Chris Mason 已提交
119

C
Chris Mason 已提交
120
static int check_node(struct btrfs_path *path, int level)
C
Chris Mason 已提交
121 122
{
	int i;
C
Chris Mason 已提交
123 124
	struct btrfs_node *parent = NULL;
	struct btrfs_node *node = &path->nodes[level]->node;
C
Chris Mason 已提交
125
	int parent_slot;
126
	u32 nritems = btrfs_header_nritems(&node->header);
C
Chris Mason 已提交
127 128 129 130

	if (path->nodes[level + 1])
		parent = &path->nodes[level + 1]->node;
	parent_slot = path->slots[level + 1];
131 132
	BUG_ON(nritems == 0);
	if (parent) {
C
Chris Mason 已提交
133
		struct btrfs_disk_key *parent_key;
C
Chris Mason 已提交
134
		parent_key = &parent->keys[parent_slot];
C
Chris Mason 已提交
135 136
		BUG_ON(memcmp(parent_key, node->keys,
			      sizeof(struct btrfs_disk_key)));
137
		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
138
		       btrfs_header_blocknr(&node->header));
C
Chris Mason 已提交
139
	}
140 141
	BUG_ON(nritems > NODEPTRS_PER_BLOCK);
	for (i = 0; nritems > 1 && i < nritems - 2; i++) {
C
Chris Mason 已提交
142 143 144
		struct btrfs_key cpukey;
		btrfs_disk_key_to_cpu(&cpukey, &node->keys[i + 1]);
		BUG_ON(comp_keys(&node->keys[i], &cpukey) >= 0);
C
Chris Mason 已提交
145 146 147 148
	}
	return 0;
}

C
Chris Mason 已提交
149
static int check_leaf(struct btrfs_path *path, int level)
C
Chris Mason 已提交
150 151
{
	int i;
C
Chris Mason 已提交
152 153
	struct btrfs_leaf *leaf = &path->nodes[level]->leaf;
	struct btrfs_node *parent = NULL;
C
Chris Mason 已提交
154
	int parent_slot;
155
	u32 nritems = btrfs_header_nritems(&leaf->header);
C
Chris Mason 已提交
156 157 158 159

	if (path->nodes[level + 1])
		parent = &path->nodes[level + 1]->node;
	parent_slot = path->slots[level + 1];
C
Chris Mason 已提交
160
	BUG_ON(btrfs_leaf_free_space(leaf) < 0);
161 162 163 164 165

	if (nritems == 0)
		return 0;

	if (parent) {
C
Chris Mason 已提交
166
		struct btrfs_disk_key *parent_key;
C
Chris Mason 已提交
167 168
		parent_key = &parent->keys[parent_slot];
		BUG_ON(memcmp(parent_key, &leaf->items[0].key,
C
Chris Mason 已提交
169
		       sizeof(struct btrfs_disk_key)));
170
		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
171
		       btrfs_header_blocknr(&leaf->header));
C
Chris Mason 已提交
172
	}
173
	for (i = 0; nritems > 1 && i < nritems - 2; i++) {
C
Chris Mason 已提交
174 175
		struct btrfs_key cpukey;
		btrfs_disk_key_to_cpu(&cpukey, &leaf->items[i + 1].key);
C
Chris Mason 已提交
176
		BUG_ON(comp_keys(&leaf->items[i].key,
C
Chris Mason 已提交
177
		                 &cpukey) >= 0);
C
Chris Mason 已提交
178 179
		BUG_ON(btrfs_item_offset(leaf->items + i) !=
			btrfs_item_end(leaf->items + i + 1));
C
Chris Mason 已提交
180
		if (i == 0) {
C
Chris Mason 已提交
181 182 183
			BUG_ON(btrfs_item_offset(leaf->items + i) +
			       btrfs_item_size(leaf->items + i) !=
			       LEAF_DATA_SIZE);
C
Chris Mason 已提交
184 185 186 187 188
		}
	}
	return 0;
}

C
Chris Mason 已提交
189
static int check_block(struct btrfs_path *path, int level)
C
Chris Mason 已提交
190 191 192 193 194 195
{
	if (level == 0)
		return check_leaf(path, level);
	return check_node(path, level);
}

C
Chris Mason 已提交
196 197 198 199 200 201 202 203 204
/*
 * search for key in the array p.  items p are item_size apart
 * and there are 'max' items in p
 * the slot in the array is returned via slot, and it points to
 * the place where you would insert key if it is not found in
 * the array.
 *
 * slot may point to max if the key is bigger than all of the keys
 */
C
Chris Mason 已提交
205
static int generic_bin_search(char *p, int item_size, struct btrfs_key *key,
206 207 208 209 210 211
		       int max, int *slot)
{
	int low = 0;
	int high = max;
	int mid;
	int ret;
C
Chris Mason 已提交
212
	struct btrfs_disk_key *tmp;
213 214 215

	while(low < high) {
		mid = (low + high) / 2;
C
Chris Mason 已提交
216
		tmp = (struct btrfs_disk_key *)(p + mid * item_size);
217 218 219 220 221 222 223 224 225 226 227 228 229 230 231
		ret = comp_keys(tmp, key);

		if (ret < 0)
			low = mid + 1;
		else if (ret > 0)
			high = mid;
		else {
			*slot = mid;
			return 0;
		}
	}
	*slot = low;
	return 1;
}

C
Chris Mason 已提交
232 233 234 235
/*
 * simple bin_search frontend that does the right thing for
 * leaves vs nodes
 */
C
Chris Mason 已提交
236
static int bin_search(struct btrfs_node *c, struct btrfs_key *key, int *slot)
237
{
238
	if (btrfs_is_leaf(c)) {
C
Chris Mason 已提交
239
		struct btrfs_leaf *l = (struct btrfs_leaf *)c;
C
Chris Mason 已提交
240 241
		return generic_bin_search((void *)l->items,
					  sizeof(struct btrfs_item),
242 243
					  key, btrfs_header_nritems(&c->header),
					  slot);
244
	} else {
C
Chris Mason 已提交
245 246
		return generic_bin_search((void *)c->keys,
					  sizeof(struct btrfs_disk_key),
247 248
					  key, btrfs_header_nritems(&c->header),
					  slot);
249 250 251 252
	}
	return -1;
}

C
Chris Mason 已提交
253
static struct btrfs_buffer *read_node_slot(struct btrfs_root *root,
C
Chris Mason 已提交
254
				   struct btrfs_buffer *parent_buf,
255 256
				   int slot)
{
C
Chris Mason 已提交
257
	struct btrfs_node *node = &parent_buf->node;
258 259
	if (slot < 0)
		return NULL;
260
	if (slot >= btrfs_header_nritems(&node->header))
261
		return NULL;
262
	return read_tree_block(root, btrfs_node_blockptr(node, slot));
263 264
}

C
Chris Mason 已提交
265
static int balance_level(struct btrfs_root *root, struct btrfs_path *path,
266 267
			int level)
{
C
Chris Mason 已提交
268 269 270 271 272 273 274 275
	struct btrfs_buffer *right_buf;
	struct btrfs_buffer *mid_buf;
	struct btrfs_buffer *left_buf;
	struct btrfs_buffer *parent_buf = NULL;
	struct btrfs_node *right = NULL;
	struct btrfs_node *mid;
	struct btrfs_node *left = NULL;
	struct btrfs_node *parent = NULL;
276 277 278 279
	int ret = 0;
	int wret;
	int pslot;
	int orig_slot = path->slots[level];
280
	u64 orig_ptr;
281 282 283 284 285 286

	if (level == 0)
		return 0;

	mid_buf = path->nodes[level];
	mid = &mid_buf->node;
287
	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
288

C
Chris Mason 已提交
289
	if (level < BTRFS_MAX_LEVEL - 1)
290 291 292 293
		parent_buf = path->nodes[level + 1];
	pslot = path->slots[level + 1];

	if (!parent_buf) {
C
Chris Mason 已提交
294
		struct btrfs_buffer *child;
295 296
		u64 blocknr = mid_buf->blocknr;

297
		if (btrfs_header_nritems(&mid->header) != 1)
298 299 300 301 302 303 304 305
			return 0;

		/* promote the child to a root */
		child = read_node_slot(root, mid_buf, 0);
		BUG_ON(!child);
		root->node = child;
		path->nodes[level] = NULL;
		/* once for the path */
C
Chris Mason 已提交
306
		btrfs_block_release(root, mid_buf);
307
		/* once for the root ptr */
C
Chris Mason 已提交
308
		btrfs_block_release(root, mid_buf);
309
		clean_tree_block(root, mid_buf);
C
Chris Mason 已提交
310
		return btrfs_free_extent(root, blocknr, 1);
311 312 313
	}
	parent = &parent_buf->node;

314
	if (btrfs_header_nritems(&mid->header) > NODEPTRS_PER_BLOCK / 4)
315 316 317 318
		return 0;

	left_buf = read_node_slot(root, parent_buf, pslot - 1);
	right_buf = read_node_slot(root, parent_buf, pslot + 1);
319 320

	/* first, try to make some room in the middle buffer */
321
	if (left_buf) {
C
Chris Mason 已提交
322 323
		btrfs_cow_block(root, left_buf, parent_buf,
				pslot - 1, &left_buf);
324
		left = &left_buf->node;
325
		orig_slot += btrfs_header_nritems(&left->header);
326 327 328
		wret = push_node_left(root, left_buf, mid_buf);
		if (wret < 0)
			ret = wret;
329
	}
330 331 332 333

	/*
	 * then try to empty the right most buffer into the middle
	 */
334
	if (right_buf) {
C
Chris Mason 已提交
335 336
		btrfs_cow_block(root, right_buf, parent_buf,
				pslot + 1, &right_buf);
337 338 339 340
		right = &right_buf->node;
		wret = push_node_left(root, mid_buf, right_buf);
		if (wret < 0)
			ret = wret;
341
		if (btrfs_header_nritems(&right->header) == 0) {
342
			u64 blocknr = right_buf->blocknr;
C
Chris Mason 已提交
343
			btrfs_block_release(root, right_buf);
344
			clean_tree_block(root, right_buf);
345 346 347 348 349
			right_buf = NULL;
			right = NULL;
			wret = del_ptr(root, path, level + 1, pslot + 1);
			if (wret)
				ret = wret;
C
Chris Mason 已提交
350
			wret = btrfs_free_extent(root, blocknr, 1);
351 352 353 354
			if (wret)
				ret = wret;
		} else {
			memcpy(parent->keys + pslot + 1, right->keys,
C
Chris Mason 已提交
355
				sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
356
			BUG_ON(list_empty(&parent_buf->dirty));
357 358
		}
	}
359
	if (btrfs_header_nritems(&mid->header) == 1) {
360 361 362 363 364 365 366 367 368 369 370 371 372 373 374
		/*
		 * we're not allowed to leave a node with one item in the
		 * tree during a delete.  A deletion from lower in the tree
		 * could try to delete the only pointer in this node.
		 * So, pull some keys from the left.
		 * There has to be a left pointer at this point because
		 * otherwise we would have pulled some pointers from the
		 * right
		 */
		BUG_ON(!left_buf);
		wret = balance_node_right(root, mid_buf, left_buf);
		if (wret < 0)
			ret = wret;
		BUG_ON(wret == 1);
	}
375
	if (btrfs_header_nritems(&mid->header) == 0) {
376
		/* we've managed to empty the middle node, drop it */
377
		u64 blocknr = mid_buf->blocknr;
C
Chris Mason 已提交
378
		btrfs_block_release(root, mid_buf);
379
		clean_tree_block(root, mid_buf);
380 381 382 383 384
		mid_buf = NULL;
		mid = NULL;
		wret = del_ptr(root, path, level + 1, pslot);
		if (wret)
			ret = wret;
C
Chris Mason 已提交
385
		wret = btrfs_free_extent(root, blocknr, 1);
386 387
		if (wret)
			ret = wret;
388 389
	} else {
		/* update the parent key to reflect our changes */
C
Chris Mason 已提交
390 391
		memcpy(parent->keys + pslot, mid->keys,
		       sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
392
		BUG_ON(list_empty(&parent_buf->dirty));
393
	}
394

395
	/* update the path */
396
	if (left_buf) {
397
		if (btrfs_header_nritems(&left->header) > orig_slot) {
398 399 400 401 402
			left_buf->count++; // released below
			path->nodes[level] = left_buf;
			path->slots[level + 1] -= 1;
			path->slots[level] = orig_slot;
			if (mid_buf)
C
Chris Mason 已提交
403
				btrfs_block_release(root, mid_buf);
404
		} else {
405
			orig_slot -= btrfs_header_nritems(&left->header);
406 407 408
			path->slots[level] = orig_slot;
		}
	}
409 410
	/* double check we haven't messed things up */
	check_block(path, level);
411 412
	if (orig_ptr != btrfs_node_blockptr(&path->nodes[level]->node,
					    path->slots[level]))
413
		BUG();
414 415

	if (right_buf)
C
Chris Mason 已提交
416
		btrfs_block_release(root, right_buf);
417
	if (left_buf)
C
Chris Mason 已提交
418
		btrfs_block_release(root, left_buf);
419 420 421
	return ret;
}

C
Chris Mason 已提交
422 423 424 425 426 427
/*
 * look for key in the tree.  path is filled in with nodes along the way
 * if key is found, we return zero and you can find the item in the leaf
 * level of the path (level 0)
 *
 * If the key isn't found, the path points to the slot where it should
C
Chris Mason 已提交
428 429
 * be inserted, and 1 is returned.  If there are other errors during the
 * search a negative error number is returned.
C
Chris Mason 已提交
430 431 432 433
 *
 * if ins_len > 0, nodes and leaves will be split as we walk down the
 * tree.  if ins_len < 0, nodes will be merged as we walk down the tree (if
 * possible)
C
Chris Mason 已提交
434
 */
C
Chris Mason 已提交
435 436
int btrfs_search_slot(struct btrfs_root *root, struct btrfs_key *key,
		struct btrfs_path *p, int ins_len, int cow)
437
{
C
Chris Mason 已提交
438 439 440
	struct btrfs_buffer *b;
	struct btrfs_buffer *cow_buf;
	struct btrfs_node *c;
441 442 443
	int slot;
	int ret;
	int level;
C
Chris Mason 已提交
444

445 446
again:
	b = root->node;
447 448
	b->count++;
	while (b) {
449
		level = btrfs_header_level(&b->node.header);
C
Chris Mason 已提交
450 451 452 453 454 455 456
		if (cow) {
			int wret;
			wret = btrfs_cow_block(root, b, p->nodes[level + 1],
					       p->slots[level + 1], &cow_buf);
			b = cow_buf;
		}
		BUG_ON(!cow && ins_len);
457 458
		c = &b->node;
		p->nodes[level] = b;
C
Chris Mason 已提交
459 460 461
		ret = check_block(p, level);
		if (ret)
			return -1;
462
		ret = bin_search(c, key, &slot);
463
		if (!btrfs_is_leaf(c)) {
464 465 466
			if (ret && slot > 0)
				slot -= 1;
			p->slots[level] = slot;
467 468
			if (ins_len > 0 && btrfs_header_nritems(&c->header) ==
			    NODEPTRS_PER_BLOCK) {
C
Chris Mason 已提交
469 470 471 472 473 474 475
				int sret = split_node(root, p, level);
				BUG_ON(sret > 0);
				if (sret)
					return sret;
				b = p->nodes[level];
				c = &b->node;
				slot = p->slots[level];
476 477 478 479 480 481 482 483 484
			} else if (ins_len < 0) {
				int sret = balance_level(root, p, level);
				if (sret)
					return sret;
				b = p->nodes[level];
				if (!b)
					goto again;
				c = &b->node;
				slot = p->slots[level];
485
				BUG_ON(btrfs_header_nritems(&c->header) == 1);
C
Chris Mason 已提交
486
			}
487
			b = read_tree_block(root, btrfs_node_blockptr(c, slot));
488
		} else {
C
Chris Mason 已提交
489
			struct btrfs_leaf *l = (struct btrfs_leaf *)c;
490
			p->slots[level] = slot;
C
Chris Mason 已提交
491
			if (ins_len > 0 && btrfs_leaf_free_space(l) <
C
Chris Mason 已提交
492
			    sizeof(struct btrfs_item) + ins_len) {
C
Chris Mason 已提交
493 494 495 496 497
				int sret = split_leaf(root, p, ins_len);
				BUG_ON(sret > 0);
				if (sret)
					return sret;
			}
498
			BUG_ON(root->node->count == 1);
499 500 501
			return ret;
		}
	}
502
	BUG_ON(root->node->count == 1);
C
Chris Mason 已提交
503
	return 1;
504 505
}

C
Chris Mason 已提交
506 507 508 509 510 511
/*
 * adjust the pointers going up the tree, starting at level
 * making sure the right key of each node is points to 'key'.
 * This is used after shifting pointers to the left, so it stops
 * fixing up pointers when a given leaf/node is not in slot 0 of the
 * higher levels
C
Chris Mason 已提交
512 513 514
 *
 * If this fails to write a tree block, it returns -1, but continues
 * fixing up the blocks in ram so the tree is consistent.
C
Chris Mason 已提交
515
 */
C
Chris Mason 已提交
516 517
static int fixup_low_keys(struct btrfs_root *root,
			   struct btrfs_path *path, struct btrfs_disk_key *key,
518
			   int level)
519 520
{
	int i;
C
Chris Mason 已提交
521
	int ret = 0;
C
Chris Mason 已提交
522 523
	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
		struct btrfs_node *t;
524
		int tslot = path->slots[i];
525
		if (!path->nodes[i])
526
			break;
527
		t = &path->nodes[i]->node;
528
		memcpy(t->keys + tslot, key, sizeof(*key));
C
Chris Mason 已提交
529
		BUG_ON(list_empty(&path->nodes[i]->dirty));
530 531 532
		if (tslot != 0)
			break;
	}
C
Chris Mason 已提交
533
	return ret;
534 535
}

C
Chris Mason 已提交
536 537
/*
 * try to push data from one node into the next node left in the
538
 * tree.
C
Chris Mason 已提交
539 540 541
 *
 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
 * error, and > 0 if there was no room in the left hand block.
C
Chris Mason 已提交
542
 */
C
Chris Mason 已提交
543 544
static int push_node_left(struct btrfs_root *root, struct btrfs_buffer *dst_buf,
			  struct btrfs_buffer *src_buf)
545
{
C
Chris Mason 已提交
546 547
	struct btrfs_node *src = &src_buf->node;
	struct btrfs_node *dst = &dst_buf->node;
548
	int push_items = 0;
549 550
	int src_nritems;
	int dst_nritems;
C
Chris Mason 已提交
551
	int ret = 0;
552

553 554
	src_nritems = btrfs_header_nritems(&src->header);
	dst_nritems = btrfs_header_nritems(&dst->header);
555
	push_items = NODEPTRS_PER_BLOCK - dst_nritems;
556
	if (push_items <= 0) {
557
		return 1;
558
	}
559

560
	if (src_nritems < push_items)
561 562
		push_items = src_nritems;

563
	memcpy(dst->keys + dst_nritems, src->keys,
C
Chris Mason 已提交
564
		push_items * sizeof(struct btrfs_disk_key));
565
	memcpy(dst->blockptrs + dst_nritems, src->blockptrs,
566
		push_items * sizeof(u64));
567 568
	if (push_items < src_nritems) {
		memmove(src->keys, src->keys + push_items,
C
Chris Mason 已提交
569 570
			(src_nritems - push_items) *
			sizeof(struct btrfs_disk_key));
571 572 573
		memmove(src->blockptrs, src->blockptrs + push_items,
			(src_nritems - push_items) * sizeof(u64));
	}
574 575
	btrfs_set_header_nritems(&src->header, src_nritems - push_items);
	btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
C
Chris Mason 已提交
576 577
	BUG_ON(list_empty(&src_buf->dirty));
	BUG_ON(list_empty(&dst_buf->dirty));
578 579 580 581 582 583 584 585 586 587 588 589
	return ret;
}

/*
 * try to push data from one node into the next node right in the
 * tree.
 *
 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
 * error, and > 0 if there was no room in the right hand block.
 *
 * this will  only push up to 1/2 the contents of the left node over
 */
C
Chris Mason 已提交
590 591 592
static int balance_node_right(struct btrfs_root *root,
			      struct btrfs_buffer *dst_buf,
			      struct btrfs_buffer *src_buf)
593
{
C
Chris Mason 已提交
594 595
	struct btrfs_node *src = &src_buf->node;
	struct btrfs_node *dst = &dst_buf->node;
596 597 598 599 600 601
	int push_items = 0;
	int max_push;
	int src_nritems;
	int dst_nritems;
	int ret = 0;

602 603
	src_nritems = btrfs_header_nritems(&src->header);
	dst_nritems = btrfs_header_nritems(&dst->header);
604 605 606 607 608 609 610 611 612 613 614 615 616
	push_items = NODEPTRS_PER_BLOCK - dst_nritems;
	if (push_items <= 0) {
		return 1;
	}

	max_push = src_nritems / 2 + 1;
	/* don't try to empty the node */
	if (max_push > src_nritems)
		return 1;
	if (max_push < push_items)
		push_items = max_push;

	memmove(dst->keys + push_items, dst->keys,
C
Chris Mason 已提交
617
		dst_nritems * sizeof(struct btrfs_disk_key));
618 619 620
	memmove(dst->blockptrs + push_items, dst->blockptrs,
		dst_nritems * sizeof(u64));
	memcpy(dst->keys, src->keys + src_nritems - push_items,
C
Chris Mason 已提交
621
		push_items * sizeof(struct btrfs_disk_key));
622 623 624
	memcpy(dst->blockptrs, src->blockptrs + src_nritems - push_items,
		push_items * sizeof(u64));

625 626
	btrfs_set_header_nritems(&src->header, src_nritems - push_items);
	btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
627

C
Chris Mason 已提交
628 629
	BUG_ON(list_empty(&src_buf->dirty));
	BUG_ON(list_empty(&dst_buf->dirty));
C
Chris Mason 已提交
630
	return ret;
631 632
}

C
Chris Mason 已提交
633 634 635 636
/*
 * helper function to insert a new root level in the tree.
 * A new node is allocated, and a single item is inserted to
 * point to the existing root
C
Chris Mason 已提交
637 638
 *
 * returns zero on success or < 0 on failure.
C
Chris Mason 已提交
639
 */
C
Chris Mason 已提交
640 641
static int insert_new_root(struct btrfs_root *root,
			   struct btrfs_path *path, int level)
C
Chris Mason 已提交
642
{
C
Chris Mason 已提交
643 644 645
	struct btrfs_buffer *t;
	struct btrfs_node *lower;
	struct btrfs_node *c;
C
Chris Mason 已提交
646
	struct btrfs_disk_key *lower_key;
C
Chris Mason 已提交
647 648 649 650

	BUG_ON(path->nodes[level]);
	BUG_ON(path->nodes[level-1] != root->node);

C
Chris Mason 已提交
651
	t = btrfs_alloc_free_block(root);
C
Chris Mason 已提交
652 653
	c = &t->node;
	memset(c, 0, sizeof(c));
654 655 656 657 658
	btrfs_set_header_nritems(&c->header, 1);
	btrfs_set_header_level(&c->header, level);
	btrfs_set_header_blocknr(&c->header, t->blocknr);
	btrfs_set_header_parentid(&c->header,
	                       btrfs_header_parentid(&root->node->node.header));
C
Chris Mason 已提交
659
	lower = &path->nodes[level-1]->node;
660
	if (btrfs_is_leaf(lower))
C
Chris Mason 已提交
661
		lower_key = &((struct btrfs_leaf *)lower)->items[0].key;
C
Chris Mason 已提交
662 663
	else
		lower_key = lower->keys;
C
Chris Mason 已提交
664
	memcpy(c->keys, lower_key, sizeof(struct btrfs_disk_key));
665
	btrfs_set_node_blockptr(c, 0, path->nodes[level - 1]->blocknr);
C
Chris Mason 已提交
666
	/* the super has an extra ref to root->node */
C
Chris Mason 已提交
667
	btrfs_block_release(root, root->node);
C
Chris Mason 已提交
668 669 670 671 672 673 674
	root->node = t;
	t->count++;
	path->nodes[level] = t;
	path->slots[level] = 0;
	return 0;
}

C
Chris Mason 已提交
675 676 677
/*
 * worker function to insert a single pointer in a node.
 * the node should have enough room for the pointer already
C
Chris Mason 已提交
678
 *
C
Chris Mason 已提交
679 680
 * slot and level indicate where you want the key to go, and
 * blocknr is the block the key points to.
C
Chris Mason 已提交
681 682
 *
 * returns zero on success and < 0 on any error
C
Chris Mason 已提交
683
 */
C
Chris Mason 已提交
684 685
static int insert_ptr(struct btrfs_root *root,
		struct btrfs_path *path, struct btrfs_disk_key *key,
C
Chris Mason 已提交
686 687
		u64 blocknr, int slot, int level)
{
C
Chris Mason 已提交
688
	struct btrfs_node *lower;
C
Chris Mason 已提交
689
	int nritems;
C
Chris Mason 已提交
690 691

	BUG_ON(!path->nodes[level]);
C
Chris Mason 已提交
692
	lower = &path->nodes[level]->node;
693
	nritems = btrfs_header_nritems(&lower->header);
C
Chris Mason 已提交
694 695 696 697 698 699
	if (slot > nritems)
		BUG();
	if (nritems == NODEPTRS_PER_BLOCK)
		BUG();
	if (slot != nritems) {
		memmove(lower->keys + slot + 1, lower->keys + slot,
C
Chris Mason 已提交
700
			(nritems - slot) * sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
701 702 703
		memmove(lower->blockptrs + slot + 1, lower->blockptrs + slot,
			(nritems - slot) * sizeof(u64));
	}
C
Chris Mason 已提交
704
	memcpy(lower->keys + slot, key, sizeof(struct btrfs_disk_key));
705
	btrfs_set_node_blockptr(lower, slot, blocknr);
706
	btrfs_set_header_nritems(&lower->header, nritems + 1);
C
Chris Mason 已提交
707 708
	if (lower->keys[1].objectid == 0)
			BUG();
C
Chris Mason 已提交
709
	BUG_ON(list_empty(&path->nodes[level]->dirty));
C
Chris Mason 已提交
710 711 712
	return 0;
}

C
Chris Mason 已提交
713 714 715 716 717 718
/*
 * split the node at the specified level in path in two.
 * The path is corrected to point to the appropriate node after the split
 *
 * Before splitting this tries to make some room in the node by pushing
 * left and right, if either one works, it returns right away.
C
Chris Mason 已提交
719 720
 *
 * returns 0 on success and < 0 on failure
C
Chris Mason 已提交
721
 */
C
Chris Mason 已提交
722
static int split_node(struct btrfs_root *root, struct btrfs_path *path,
C
Chris Mason 已提交
723
		      int level)
724
{
C
Chris Mason 已提交
725 726 727 728
	struct btrfs_buffer *t;
	struct btrfs_node *c;
	struct btrfs_buffer *split_buffer;
	struct btrfs_node *split;
729
	int mid;
C
Chris Mason 已提交
730
	int ret;
C
Chris Mason 已提交
731
	int wret;
732
	u32 c_nritems;
733

C
Chris Mason 已提交
734 735 736 737 738 739 740
	t = path->nodes[level];
	c = &t->node;
	if (t == root->node) {
		/* trying to split the root, lets make a new one */
		ret = insert_new_root(root, path, level + 1);
		if (ret)
			return ret;
741
	}
742
	c_nritems = btrfs_header_nritems(&c->header);
C
Chris Mason 已提交
743
	split_buffer = btrfs_alloc_free_block(root);
C
Chris Mason 已提交
744
	split = &split_buffer->node;
745 746 747 748 749
	btrfs_set_header_flags(&split->header, btrfs_header_flags(&c->header));
	btrfs_set_header_blocknr(&split->header, split_buffer->blocknr);
	btrfs_set_header_parentid(&split->header,
	                       btrfs_header_parentid(&root->node->node.header));
	mid = (c_nritems + 1) / 2;
C
Chris Mason 已提交
750
	memcpy(split->keys, c->keys + mid,
C
Chris Mason 已提交
751
		(c_nritems - mid) * sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
752
	memcpy(split->blockptrs, c->blockptrs + mid,
753 754 755
		(c_nritems - mid) * sizeof(u64));
	btrfs_set_header_nritems(&split->header, c_nritems - mid);
	btrfs_set_header_nritems(&c->header, mid);
C
Chris Mason 已提交
756 757
	ret = 0;

C
Chris Mason 已提交
758
	BUG_ON(list_empty(&t->dirty));
C
Chris Mason 已提交
759 760 761 762 763
	wret = insert_ptr(root, path, split->keys, split_buffer->blocknr,
			  path->slots[level + 1] + 1, level + 1);
	if (wret)
		ret = wret;

C
Chris Mason 已提交
764
	if (path->slots[level] >= mid) {
C
Chris Mason 已提交
765
		path->slots[level] -= mid;
C
Chris Mason 已提交
766
		btrfs_block_release(root, t);
C
Chris Mason 已提交
767 768 769
		path->nodes[level] = split_buffer;
		path->slots[level + 1] += 1;
	} else {
C
Chris Mason 已提交
770
		btrfs_block_release(root, split_buffer);
771
	}
C
Chris Mason 已提交
772
	return ret;
773 774
}

C
Chris Mason 已提交
775 776 777 778 779
/*
 * how many bytes are required to store the items in a leaf.  start
 * and nr indicate which items in the leaf to check.  This totals up the
 * space used both by the item structs and the item data
 */
C
Chris Mason 已提交
780
static int leaf_space_used(struct btrfs_leaf *l, int start, int nr)
781 782 783 784 785 786
{
	int data_len;
	int end = start + nr - 1;

	if (!nr)
		return 0;
C
Chris Mason 已提交
787 788 789
	data_len = btrfs_item_end(l->items + start);
	data_len = data_len - btrfs_item_offset(l->items + end);
	data_len += sizeof(struct btrfs_item) * nr;
790 791 792
	return data_len;
}

C
Chris Mason 已提交
793 794 795
/*
 * push some data in the path leaf to the right, trying to free up at
 * least data_size bytes.  returns zero if the push worked, nonzero otherwise
C
Chris Mason 已提交
796 797 798
 *
 * returns 1 if the push failed because the other node didn't have enough
 * room, 0 if everything worked out and < 0 if there were major errors.
C
Chris Mason 已提交
799
 */
C
Chris Mason 已提交
800
static int push_leaf_right(struct btrfs_root *root, struct btrfs_path *path,
C
Chris Mason 已提交
801
			   int data_size)
C
Chris Mason 已提交
802
{
C
Chris Mason 已提交
803 804 805 806 807
	struct btrfs_buffer *left_buf = path->nodes[0];
	struct btrfs_leaf *left = &left_buf->leaf;
	struct btrfs_leaf *right;
	struct btrfs_buffer *right_buf;
	struct btrfs_buffer *upper;
C
Chris Mason 已提交
808 809 810 811 812
	int slot;
	int i;
	int free_space;
	int push_space = 0;
	int push_items = 0;
C
Chris Mason 已提交
813
	struct btrfs_item *item;
814 815
	u32 left_nritems;
	u32 right_nritems;
C
Chris Mason 已提交
816 817 818 819 820 821

	slot = path->slots[1];
	if (!path->nodes[1]) {
		return 1;
	}
	upper = path->nodes[1];
822
	if (slot >= btrfs_header_nritems(&upper->node.header) - 1) {
C
Chris Mason 已提交
823 824
		return 1;
	}
825 826
	right_buf = read_tree_block(root, btrfs_node_blockptr(&upper->node,
							      slot + 1));
C
Chris Mason 已提交
827
	right = &right_buf->leaf;
C
Chris Mason 已提交
828
	free_space = btrfs_leaf_free_space(right);
C
Chris Mason 已提交
829
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
830
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
831 832
		return 1;
	}
C
Chris Mason 已提交
833 834 835
	/* cow and double check */
	btrfs_cow_block(root, right_buf, upper, slot + 1, &right_buf);
	right = &right_buf->leaf;
C
Chris Mason 已提交
836
	free_space = btrfs_leaf_free_space(right);
C
Chris Mason 已提交
837
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
838
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
839 840 841
		return 1;
	}

842 843
	left_nritems = btrfs_header_nritems(&left->header);
	for (i = left_nritems - 1; i >= 0; i--) {
C
Chris Mason 已提交
844 845 846
		item = left->items + i;
		if (path->slots[0] == i)
			push_space += data_size + sizeof(*item);
C
Chris Mason 已提交
847 848
		if (btrfs_item_size(item) + sizeof(*item) + push_space >
		    free_space)
C
Chris Mason 已提交
849 850
			break;
		push_items++;
C
Chris Mason 已提交
851
		push_space += btrfs_item_size(item) + sizeof(*item);
C
Chris Mason 已提交
852 853
	}
	if (push_items == 0) {
C
Chris Mason 已提交
854
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
855 856
		return 1;
	}
857
	right_nritems = btrfs_header_nritems(&right->header);
C
Chris Mason 已提交
858
	/* push left to right */
C
Chris Mason 已提交
859
	push_space = btrfs_item_end(left->items + left_nritems - push_items);
C
Chris Mason 已提交
860 861 862 863 864 865 866 867 868 869
	push_space -= leaf_data_end(left);
	/* make room in the right data area */
	memmove(right->data + leaf_data_end(right) - push_space,
		right->data + leaf_data_end(right),
		LEAF_DATA_SIZE - leaf_data_end(right));
	/* copy from the left data area */
	memcpy(right->data + LEAF_DATA_SIZE - push_space,
		left->data + leaf_data_end(left),
		push_space);
	memmove(right->items + push_items, right->items,
C
Chris Mason 已提交
870
		right_nritems * sizeof(struct btrfs_item));
C
Chris Mason 已提交
871
	/* copy the items from left to right */
872
	memcpy(right->items, left->items + left_nritems - push_items,
C
Chris Mason 已提交
873
		push_items * sizeof(struct btrfs_item));
C
Chris Mason 已提交
874 875

	/* update the item pointers */
876 877
	right_nritems += push_items;
	btrfs_set_header_nritems(&right->header, right_nritems);
C
Chris Mason 已提交
878
	push_space = LEAF_DATA_SIZE;
879
	for (i = 0; i < right_nritems; i++) {
C
Chris Mason 已提交
880 881 882
		btrfs_set_item_offset(right->items + i, push_space -
				      btrfs_item_size(right->items + i));
		push_space = btrfs_item_offset(right->items + i);
C
Chris Mason 已提交
883
	}
884 885
	left_nritems -= push_items;
	btrfs_set_header_nritems(&left->header, left_nritems);
C
Chris Mason 已提交
886

C
Chris Mason 已提交
887 888
	BUG_ON(list_empty(&left_buf->dirty));
	BUG_ON(list_empty(&right_buf->dirty));
C
Chris Mason 已提交
889
	memcpy(upper->node.keys + slot + 1,
C
Chris Mason 已提交
890
		&right->items[0].key, sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
891 892
	BUG_ON(list_empty(&upper->dirty));

C
Chris Mason 已提交
893
	/* then fixup the leaf pointer in the path */
894 895
	if (path->slots[0] >= left_nritems) {
		path->slots[0] -= left_nritems;
C
Chris Mason 已提交
896
		btrfs_block_release(root, path->nodes[0]);
C
Chris Mason 已提交
897 898 899
		path->nodes[0] = right_buf;
		path->slots[1] += 1;
	} else {
C
Chris Mason 已提交
900
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
901 902 903
	}
	return 0;
}
C
Chris Mason 已提交
904 905 906 907
/*
 * push some data in the path leaf to the left, trying to free up at
 * least data_size bytes.  returns zero if the push worked, nonzero otherwise
 */
C
Chris Mason 已提交
908
static int push_leaf_left(struct btrfs_root *root, struct btrfs_path *path,
C
Chris Mason 已提交
909
			  int data_size)
910
{
C
Chris Mason 已提交
911 912 913 914
	struct btrfs_buffer *right_buf = path->nodes[0];
	struct btrfs_leaf *right = &right_buf->leaf;
	struct btrfs_buffer *t;
	struct btrfs_leaf *left;
915 916 917 918 919
	int slot;
	int i;
	int free_space;
	int push_space = 0;
	int push_items = 0;
C
Chris Mason 已提交
920
	struct btrfs_item *item;
921
	u32 old_left_nritems;
C
Chris Mason 已提交
922 923
	int ret = 0;
	int wret;
924 925 926 927 928 929 930 931

	slot = path->slots[1];
	if (slot == 0) {
		return 1;
	}
	if (!path->nodes[1]) {
		return 1;
	}
932 933
	t = read_tree_block(root, btrfs_node_blockptr(&path->nodes[1]->node,
						      slot - 1));
934
	left = &t->leaf;
C
Chris Mason 已提交
935
	free_space = btrfs_leaf_free_space(left);
C
Chris Mason 已提交
936
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
937
		btrfs_block_release(root, t);
938 939
		return 1;
	}
C
Chris Mason 已提交
940 941 942 943

	/* cow and double check */
	btrfs_cow_block(root, t, path->nodes[1], slot - 1, &t);
	left = &t->leaf;
C
Chris Mason 已提交
944
	free_space = btrfs_leaf_free_space(left);
C
Chris Mason 已提交
945
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
946
		btrfs_block_release(root, t);
C
Chris Mason 已提交
947 948 949
		return 1;
	}

950
	for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
951 952 953
		item = right->items + i;
		if (path->slots[0] == i)
			push_space += data_size + sizeof(*item);
C
Chris Mason 已提交
954 955
		if (btrfs_item_size(item) + sizeof(*item) + push_space >
		    free_space)
956 957
			break;
		push_items++;
C
Chris Mason 已提交
958
		push_space += btrfs_item_size(item) + sizeof(*item);
959 960
	}
	if (push_items == 0) {
C
Chris Mason 已提交
961
		btrfs_block_release(root, t);
962 963 964
		return 1;
	}
	/* push data from right to left */
965
	memcpy(left->items + btrfs_header_nritems(&left->header),
C
Chris Mason 已提交
966 967 968
		right->items, push_items * sizeof(struct btrfs_item));
	push_space = LEAF_DATA_SIZE -
		     btrfs_item_offset(right->items + push_items -1);
969
	memcpy(left->data + leaf_data_end(left) - push_space,
C
Chris Mason 已提交
970
		right->data + btrfs_item_offset(right->items + push_items - 1),
971
		push_space);
972
	old_left_nritems = btrfs_header_nritems(&left->header);
973 974
	BUG_ON(old_left_nritems < 0);

C
Chris Mason 已提交
975 976 977 978 979
	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
		u16 ioff = btrfs_item_offset(left->items + i);
		btrfs_set_item_offset(left->items + i, ioff - (LEAF_DATA_SIZE -
				      btrfs_item_offset(left->items +
						        old_left_nritems - 1)));
980
	}
981
	btrfs_set_header_nritems(&left->header, old_left_nritems + push_items);
982 983

	/* fixup right node */
C
Chris Mason 已提交
984 985
	push_space = btrfs_item_offset(right->items + push_items - 1) -
		     leaf_data_end(right);
986 987 988
	memmove(right->data + LEAF_DATA_SIZE - push_space, right->data +
		leaf_data_end(right), push_space);
	memmove(right->items, right->items + push_items,
989
		(btrfs_header_nritems(&right->header) - push_items) *
C
Chris Mason 已提交
990
		sizeof(struct btrfs_item));
991 992 993
	btrfs_set_header_nritems(&right->header,
				 btrfs_header_nritems(&right->header) -
				 push_items);
994
	push_space = LEAF_DATA_SIZE;
995

996
	for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
C
Chris Mason 已提交
997 998 999
		btrfs_set_item_offset(right->items + i, push_space -
				      btrfs_item_size(right->items + i));
		push_space = btrfs_item_offset(right->items + i);
1000
	}
1001

C
Chris Mason 已提交
1002 1003
	BUG_ON(list_empty(&t->dirty));
	BUG_ON(list_empty(&right_buf->dirty));
1004

C
Chris Mason 已提交
1005 1006 1007
	wret = fixup_low_keys(root, path, &right->items[0].key, 1);
	if (wret)
		ret = wret;
1008 1009 1010 1011

	/* then fixup the leaf pointer in the path */
	if (path->slots[0] < push_items) {
		path->slots[0] += old_left_nritems;
C
Chris Mason 已提交
1012
		btrfs_block_release(root, path->nodes[0]);
1013
		path->nodes[0] = t;
1014 1015
		path->slots[1] -= 1;
	} else {
C
Chris Mason 已提交
1016
		btrfs_block_release(root, t);
1017 1018
		path->slots[0] -= push_items;
	}
1019
	BUG_ON(path->slots[0] < 0);
C
Chris Mason 已提交
1020
	return ret;
1021 1022
}

C
Chris Mason 已提交
1023 1024 1025
/*
 * split the path's leaf in two, making sure there is at least data_size
 * available for the resulting leaf level of the path.
C
Chris Mason 已提交
1026 1027
 *
 * returns 0 if all went well and < 0 on failure.
C
Chris Mason 已提交
1028
 */
C
Chris Mason 已提交
1029
static int split_leaf(struct btrfs_root *root, struct btrfs_path *path,
C
Chris Mason 已提交
1030
		      int data_size)
1031
{
C
Chris Mason 已提交
1032 1033
	struct btrfs_buffer *l_buf;
	struct btrfs_leaf *l;
1034
	u32 nritems;
1035 1036
	int mid;
	int slot;
C
Chris Mason 已提交
1037 1038
	struct btrfs_leaf *right;
	struct btrfs_buffer *right_buffer;
C
Chris Mason 已提交
1039
	int space_needed = data_size + sizeof(struct btrfs_item);
1040 1041 1042 1043
	int data_copy_size;
	int rt_data_off;
	int i;
	int ret;
C
Chris Mason 已提交
1044 1045
	int wret;

C
Chris Mason 已提交
1046 1047 1048 1049 1050 1051 1052 1053 1054
	wret = push_leaf_left(root, path, data_size);
	if (wret < 0)
		return wret;
	if (wret) {
		wret = push_leaf_right(root, path, data_size);
		if (wret < 0)
			return wret;
	}

C
Chris Mason 已提交
1055 1056 1057 1058
	l_buf = path->nodes[0];
	l = &l_buf->leaf;

	/* did the pushes work? */
C
Chris Mason 已提交
1059
	if (btrfs_leaf_free_space(l) >= sizeof(struct btrfs_item) + data_size)
C
Chris Mason 已提交
1060 1061
		return 0;

C
Chris Mason 已提交
1062 1063 1064 1065 1066
	if (!path->nodes[1]) {
		ret = insert_new_root(root, path, 1);
		if (ret)
			return ret;
	}
1067
	slot = path->slots[0];
1068
	nritems = btrfs_header_nritems(&l->header);
1069
	mid = (nritems + 1)/ 2;
C
Chris Mason 已提交
1070
	right_buffer = btrfs_alloc_free_block(root);
1071 1072 1073
	BUG_ON(!right_buffer);
	BUG_ON(mid == nritems);
	right = &right_buffer->leaf;
1074 1075
	memset(right, 0, sizeof(*right));
	if (mid <= slot) {
C
Chris Mason 已提交
1076
		/* FIXME, just alloc a new leaf here */
1077 1078 1079 1080
		if (leaf_space_used(l, mid, nritems - mid) + space_needed >
			LEAF_DATA_SIZE)
			BUG();
	} else {
C
Chris Mason 已提交
1081
		/* FIXME, just alloc a new leaf here */
1082 1083 1084 1085
		if (leaf_space_used(l, 0, mid + 1) + space_needed >
			LEAF_DATA_SIZE)
			BUG();
	}
1086 1087 1088 1089 1090
	btrfs_set_header_nritems(&right->header, nritems - mid);
	btrfs_set_header_blocknr(&right->header, right_buffer->blocknr);
	btrfs_set_header_level(&right->header, 0);
	btrfs_set_header_parentid(&right->header,
	                       btrfs_header_parentid(&root->node->node.header));
C
Chris Mason 已提交
1091
	data_copy_size = btrfs_item_end(l->items + mid) - leaf_data_end(l);
1092
	memcpy(right->items, l->items + mid,
C
Chris Mason 已提交
1093
	       (nritems - mid) * sizeof(struct btrfs_item));
1094 1095
	memcpy(right->data + LEAF_DATA_SIZE - data_copy_size,
	       l->data + leaf_data_end(l), data_copy_size);
C
Chris Mason 已提交
1096
	rt_data_off = LEAF_DATA_SIZE - btrfs_item_end(l->items + mid);
C
Chris Mason 已提交
1097

C
Chris Mason 已提交
1098 1099 1100 1101
	for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
		u16 ioff = btrfs_item_offset(right->items + i);
		btrfs_set_item_offset(right->items + i, ioff + rt_data_off);
	}
C
Chris Mason 已提交
1102

1103
	btrfs_set_header_nritems(&l->header, mid);
C
Chris Mason 已提交
1104 1105
	ret = 0;
	wret = insert_ptr(root, path, &right->items[0].key,
C
Chris Mason 已提交
1106
			  right_buffer->blocknr, path->slots[1] + 1, 1);
C
Chris Mason 已提交
1107 1108
	if (wret)
		ret = wret;
C
Chris Mason 已提交
1109 1110
	BUG_ON(list_empty(&right_buffer->dirty));
	BUG_ON(list_empty(&l_buf->dirty));
1111
	BUG_ON(path->slots[0] != slot);
1112
	if (mid <= slot) {
C
Chris Mason 已提交
1113
		btrfs_block_release(root, path->nodes[0]);
1114
		path->nodes[0] = right_buffer;
1115 1116
		path->slots[0] -= mid;
		path->slots[1] += 1;
1117
	} else
C
Chris Mason 已提交
1118
		btrfs_block_release(root, right_buffer);
1119
	BUG_ON(path->slots[0] < 0);
1120 1121 1122
	return ret;
}

C
Chris Mason 已提交
1123 1124 1125 1126
/*
 * Given a key and some data, insert an item into the tree.
 * This does all the path init required, making room in the tree if needed.
 */
C
Chris Mason 已提交
1127
int btrfs_insert_item(struct btrfs_root *root, struct btrfs_key *cpu_key,
1128 1129
			  void *data, int data_size)
{
C
Chris Mason 已提交
1130
	int ret = 0;
1131
	int slot;
1132
	int slot_orig;
C
Chris Mason 已提交
1133 1134
	struct btrfs_leaf *leaf;
	struct btrfs_buffer *leaf_buf;
1135
	u32 nritems;
1136
	unsigned int data_end;
C
Chris Mason 已提交
1137
	struct btrfs_path path;
C
Chris Mason 已提交
1138 1139 1140
	struct btrfs_disk_key disk_key;

	btrfs_cpu_key_to_disk(&disk_key, cpu_key);
1141

C
Chris Mason 已提交
1142
	/* create a root if there isn't one */
C
Chris Mason 已提交
1143
	if (!root->node)
C
Chris Mason 已提交
1144
		BUG();
C
Chris Mason 已提交
1145 1146
	btrfs_init_path(&path);
	ret = btrfs_search_slot(root, cpu_key, &path, data_size, 1);
1147
	if (ret == 0) {
C
Chris Mason 已提交
1148
		btrfs_release_path(root, &path);
1149
		return -EEXIST;
C
Chris Mason 已提交
1150
	}
1151 1152
	if (ret < 0)
		goto out;
1153

1154 1155 1156
	slot_orig = path.slots[0];
	leaf_buf = path.nodes[0];
	leaf = &leaf_buf->leaf;
C
Chris Mason 已提交
1157

1158
	nritems = btrfs_header_nritems(&leaf->header);
1159
	data_end = leaf_data_end(leaf);
1160

C
Chris Mason 已提交
1161 1162
	if (btrfs_leaf_free_space(leaf) <
	    sizeof(struct btrfs_item) + data_size)
1163 1164 1165
		BUG();

	slot = path.slots[0];
1166
	BUG_ON(slot < 0);
1167 1168
	if (slot != nritems) {
		int i;
C
Chris Mason 已提交
1169
		unsigned int old_data = btrfs_item_end(leaf->items + slot);
1170 1171 1172 1173 1174

		/*
		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
		 */
		/* first correct the data pointers */
C
Chris Mason 已提交
1175 1176 1177 1178 1179
		for (i = slot; i < nritems; i++) {
			u16 ioff = btrfs_item_offset(leaf->items + i);
			btrfs_set_item_offset(leaf->items + i,
					      ioff - data_size);
		}
1180 1181 1182

		/* shift the items */
		memmove(leaf->items + slot + 1, leaf->items + slot,
C
Chris Mason 已提交
1183
		        (nritems - slot) * sizeof(struct btrfs_item));
1184 1185 1186 1187 1188 1189

		/* shift the data */
		memmove(leaf->data + data_end - data_size, leaf->data +
		        data_end, old_data - data_end);
		data_end = old_data;
	}
C
Chris Mason 已提交
1190
	/* copy the new data in */
C
Chris Mason 已提交
1191 1192
	memcpy(&leaf->items[slot].key, &disk_key,
		sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
1193 1194
	btrfs_set_item_offset(leaf->items + slot, data_end - data_size);
	btrfs_set_item_size(leaf->items + slot, data_size);
1195
	memcpy(leaf->data + data_end - data_size, data, data_size);
1196
	btrfs_set_header_nritems(&leaf->header, nritems + 1);
C
Chris Mason 已提交
1197 1198

	ret = 0;
1199
	if (slot == 0)
C
Chris Mason 已提交
1200
		ret = fixup_low_keys(root, &path, &disk_key, 1);
C
Chris Mason 已提交
1201

C
Chris Mason 已提交
1202
	BUG_ON(list_empty(&leaf_buf->dirty));
C
Chris Mason 已提交
1203
	if (btrfs_leaf_free_space(leaf) < 0)
1204
		BUG();
1205
	check_leaf(&path, 0);
1206
out:
C
Chris Mason 已提交
1207
	btrfs_release_path(root, &path);
C
Chris Mason 已提交
1208
	return ret;
1209 1210
}

C
Chris Mason 已提交
1211
/*
C
Chris Mason 已提交
1212
 * delete the pointer from a given node.
C
Chris Mason 已提交
1213 1214 1215 1216 1217
 *
 * If the delete empties a node, the node is removed from the tree,
 * continuing all the way the root if required.  The root is converted into
 * a leaf if all the nodes are emptied.
 */
C
Chris Mason 已提交
1218
static int del_ptr(struct btrfs_root *root, struct btrfs_path *path, int level,
1219
		   int slot)
1220
{
C
Chris Mason 已提交
1221 1222
	struct btrfs_node *node;
	struct btrfs_buffer *parent = path->nodes[level];
1223
	u32 nritems;
C
Chris Mason 已提交
1224
	int ret = 0;
1225
	int wret;
1226

1227
	node = &parent->node;
1228
	nritems = btrfs_header_nritems(&node->header);
1229 1230
	if (slot != nritems -1) {
		memmove(node->keys + slot, node->keys + slot + 1,
C
Chris Mason 已提交
1231
			sizeof(struct btrfs_disk_key) * (nritems - slot - 1));
1232 1233 1234 1235
		memmove(node->blockptrs + slot,
			node->blockptrs + slot + 1,
			sizeof(u64) * (nritems - slot - 1));
	}
1236 1237 1238 1239
	nritems--;
	btrfs_set_header_nritems(&node->header, nritems);
	if (nritems == 0 && parent == root->node) {
		BUG_ON(btrfs_header_level(&root->node->node.header) != 1);
1240
		/* just turn the root into a leaf and break */
1241
		btrfs_set_header_level(&root->node->node.header, 0);
1242 1243
	} else if (slot == 0) {
		wret = fixup_low_keys(root, path, node->keys, level + 1);
C
Chris Mason 已提交
1244 1245
		if (wret)
			ret = wret;
1246
	}
C
Chris Mason 已提交
1247
	BUG_ON(list_empty(&parent->dirty));
C
Chris Mason 已提交
1248
	return ret;
1249 1250
}

C
Chris Mason 已提交
1251 1252 1253 1254
/*
 * delete the item at the leaf level in path.  If that empties
 * the leaf, remove it from the tree
 */
C
Chris Mason 已提交
1255
int btrfs_del_item(struct btrfs_root *root, struct btrfs_path *path)
1256 1257
{
	int slot;
C
Chris Mason 已提交
1258 1259
	struct btrfs_leaf *leaf;
	struct btrfs_buffer *leaf_buf;
1260 1261
	int doff;
	int dsize;
C
Chris Mason 已提交
1262 1263
	int ret = 0;
	int wret;
1264
	u32 nritems;
1265

1266 1267
	leaf_buf = path->nodes[0];
	leaf = &leaf_buf->leaf;
1268
	slot = path->slots[0];
C
Chris Mason 已提交
1269 1270
	doff = btrfs_item_offset(leaf->items + slot);
	dsize = btrfs_item_size(leaf->items + slot);
1271
	nritems = btrfs_header_nritems(&leaf->header);
1272

1273
	if (slot != nritems - 1) {
1274 1275 1276 1277 1278
		int i;
		int data_end = leaf_data_end(leaf);
		memmove(leaf->data + data_end + dsize,
			leaf->data + data_end,
			doff - data_end);
C
Chris Mason 已提交
1279 1280 1281 1282
		for (i = slot + 1; i < nritems; i++) {
			u16 ioff = btrfs_item_offset(leaf->items + i);
			btrfs_set_item_offset(leaf->items + i, ioff + dsize);
		}
1283
		memmove(leaf->items + slot, leaf->items + slot + 1,
C
Chris Mason 已提交
1284
			sizeof(struct btrfs_item) *
1285
			(nritems - slot - 1));
1286
	}
1287 1288
	btrfs_set_header_nritems(&leaf->header, nritems - 1);
	nritems--;
C
Chris Mason 已提交
1289
	/* delete the leaf if we've emptied it */
1290
	if (nritems == 0) {
1291
		if (leaf_buf == root->node) {
1292
			btrfs_set_header_level(&leaf->header, 0);
C
Chris Mason 已提交
1293
			BUG_ON(list_empty(&leaf_buf->dirty));
1294
		} else {
1295
			clean_tree_block(root, leaf_buf);
1296
			wret = del_ptr(root, path, 1, path->slots[1]);
C
Chris Mason 已提交
1297 1298
			if (wret)
				ret = wret;
C
Chris Mason 已提交
1299
			wret = btrfs_free_extent(root, leaf_buf->blocknr, 1);
C
Chris Mason 已提交
1300 1301
			if (wret)
				ret = wret;
1302
		}
1303
	} else {
1304
		int used = leaf_space_used(leaf, 0, nritems);
C
Chris Mason 已提交
1305 1306 1307 1308 1309 1310
		if (slot == 0) {
			wret = fixup_low_keys(root, path,
						   &leaf->items[0].key, 1);
			if (wret)
				ret = wret;
		}
C
Chris Mason 已提交
1311
		BUG_ON(list_empty(&leaf_buf->dirty));
C
Chris Mason 已提交
1312

C
Chris Mason 已提交
1313
		/* delete the leaf if it is mostly empty */
C
Chris Mason 已提交
1314
		if (used < LEAF_DATA_SIZE / 3) {
1315 1316 1317 1318
			/* push_leaf_left fixes the path.
			 * make sure the path still points to our leaf
			 * for possible call to del_ptr below
			 */
1319
			slot = path->slots[1];
1320
			leaf_buf->count++;
C
Chris Mason 已提交
1321 1322 1323
			wret = push_leaf_left(root, path, 1);
			if (wret < 0)
				ret = wret;
1324
			if (path->nodes[0] == leaf_buf &&
1325
			    btrfs_header_nritems(&leaf->header)) {
C
Chris Mason 已提交
1326 1327 1328 1329
				wret = push_leaf_right(root, path, 1);
				if (wret < 0)
					ret = wret;
			}
1330
			if (btrfs_header_nritems(&leaf->header) == 0) {
C
Chris Mason 已提交
1331
				u64 blocknr = leaf_buf->blocknr;
1332
				clean_tree_block(root, leaf_buf);
1333
				wret = del_ptr(root, path, 1, slot);
C
Chris Mason 已提交
1334 1335
				if (wret)
					ret = wret;
C
Chris Mason 已提交
1336 1337
				btrfs_block_release(root, leaf_buf);
				wret = btrfs_free_extent(root, blocknr, 1);
C
Chris Mason 已提交
1338 1339
				if (wret)
					ret = wret;
C
Chris Mason 已提交
1340
			} else {
C
Chris Mason 已提交
1341
				btrfs_block_release(root, leaf_buf);
1342 1343 1344
			}
		}
	}
C
Chris Mason 已提交
1345
	return ret;
1346 1347
}

C
Chris Mason 已提交
1348 1349
/*
 * walk up the tree as far as required to find the next leaf.
C
Chris Mason 已提交
1350 1351
 * returns 0 if it found something or 1 if there are no greater leaves.
 * returns < 0 on io errors.
C
Chris Mason 已提交
1352
 */
C
Chris Mason 已提交
1353
int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
1354 1355 1356 1357
{
	int slot;
	int level = 1;
	u64 blocknr;
C
Chris Mason 已提交
1358 1359
	struct btrfs_buffer *c;
	struct btrfs_buffer *next = NULL;
1360

C
Chris Mason 已提交
1361
	while(level < BTRFS_MAX_LEVEL) {
1362
		if (!path->nodes[level])
C
Chris Mason 已提交
1363
			return 1;
1364 1365
		slot = path->slots[level] + 1;
		c = path->nodes[level];
1366
		if (slot >= btrfs_header_nritems(&c->node.header)) {
1367 1368 1369
			level++;
			continue;
		}
1370
		blocknr = btrfs_node_blockptr(&c->node, slot);
C
Chris Mason 已提交
1371
		if (next)
C
Chris Mason 已提交
1372
			btrfs_block_release(root, next);
1373 1374 1375 1376 1377 1378 1379
		next = read_tree_block(root, blocknr);
		break;
	}
	path->slots[level] = slot;
	while(1) {
		level--;
		c = path->nodes[level];
C
Chris Mason 已提交
1380
		btrfs_block_release(root, c);
1381 1382 1383 1384
		path->nodes[level] = next;
		path->slots[level] = 0;
		if (!level)
			break;
1385 1386
		next = read_tree_block(root,
				       btrfs_node_blockptr(&next->node, 0));
1387 1388 1389 1390
	}
	return 0;
}

C
Chris Mason 已提交
1391