ctree.c 40.9 KB
Newer Older
1
#include <linux/module.h>
2 3
#include "ctree.h"
#include "disk-io.h"
4
#include "transaction.h"
5

6 7 8 9 10
static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, struct btrfs_path *path, int level);
static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, struct btrfs_path *path, int data_size);
static int push_node_left(struct btrfs_trans_handle *trans, struct btrfs_root
C
Chris Mason 已提交
11
			  *root, struct buffer_head *dst, struct buffer_head
12 13
			  *src);
static int balance_node_right(struct btrfs_trans_handle *trans, struct
C
Chris Mason 已提交
14 15
			      btrfs_root *root, struct buffer_head *dst_buf,
			      struct buffer_head *src_buf);
16 17
static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
		   struct btrfs_path *path, int level, int slot);
18

C
Chris Mason 已提交
19
inline void btrfs_init_path(struct btrfs_path *p)
20 21 22 23
{
	memset(p, 0, sizeof(*p));
}

C
Chris Mason 已提交
24
void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
25 26
{
	int i;
C
Chris Mason 已提交
27
	for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
28 29
		if (!p->nodes[i])
			break;
C
Chris Mason 已提交
30
		btrfs_block_release(root, p->nodes[i]);
31
	}
C
Chris Mason 已提交
32
	memset(p, 0, sizeof(*p));
33 34
}

35
static int btrfs_cow_block(struct btrfs_trans_handle *trans, struct btrfs_root
C
Chris Mason 已提交
36 37
			   *root, struct buffer_head *buf, struct buffer_head
			   *parent, int parent_slot, struct buffer_head
38
			   **cow_ret)
C
Chris Mason 已提交
39
{
C
Chris Mason 已提交
40 41
	struct buffer_head *cow;
	struct btrfs_node *cow_node;
C
Chris Mason 已提交
42

43 44
	if (btrfs_header_generation(btrfs_buffer_header(buf)) ==
				    trans->transid) {
C
Chris Mason 已提交
45 46 47
		*cow_ret = buf;
		return 0;
	}
48
	cow = btrfs_alloc_free_block(trans, root);
C
Chris Mason 已提交
49 50 51
	cow_node = btrfs_buffer_node(cow);
	memcpy(cow_node, btrfs_buffer_node(buf), root->blocksize);
	btrfs_set_header_blocknr(&cow_node->header, cow->b_blocknr);
52
	btrfs_set_header_generation(&cow_node->header, trans->transid);
C
Chris Mason 已提交
53
	*cow_ret = cow;
54
	mark_buffer_dirty(cow);
55
	btrfs_inc_ref(trans, root, buf);
C
Chris Mason 已提交
56 57
	if (buf == root->node) {
		root->node = cow;
C
Chris Mason 已提交
58
		get_bh(cow);
59
		if (buf != root->commit_root)
C
Chris Mason 已提交
60
			btrfs_free_extent(trans, root, buf->b_blocknr, 1, 1);
C
Chris Mason 已提交
61
		btrfs_block_release(root, buf);
C
Chris Mason 已提交
62
	} else {
C
Chris Mason 已提交
63 64
		btrfs_set_node_blockptr(btrfs_buffer_node(parent), parent_slot,
					cow->b_blocknr);
65
		mark_buffer_dirty(parent);
C
Chris Mason 已提交
66
		btrfs_free_extent(trans, root, buf->b_blocknr, 1, 1);
C
Chris Mason 已提交
67
	}
C
Chris Mason 已提交
68
	btrfs_block_release(root, buf);
C
Chris Mason 已提交
69 70 71
	return 0;
}

C
Chris Mason 已提交
72 73 74 75 76
/*
 * 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
77 78
static inline unsigned int leaf_data_end(struct btrfs_root *root,
					 struct btrfs_leaf *leaf)
79
{
80
	u32 nr = btrfs_header_nritems(&leaf->header);
81
	if (nr == 0)
C
Chris Mason 已提交
82
		return BTRFS_LEAF_DATA_SIZE(root);
C
Chris Mason 已提交
83
	return btrfs_item_offset(leaf->items + nr - 1);
84 85
}

C
Chris Mason 已提交
86 87 88 89 90
/*
 * 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
91
int btrfs_leaf_free_space(struct btrfs_root *root, struct btrfs_leaf *leaf)
92
{
C
Chris Mason 已提交
93
	int data_end = leaf_data_end(root, leaf);
94
	int nritems = btrfs_header_nritems(&leaf->header);
95
	char *items_end = (char *)(leaf->items + nritems + 1);
C
Chris Mason 已提交
96
	return (char *)(btrfs_leaf_data(leaf) + data_end) - (char *)items_end;
97 98
}

C
Chris Mason 已提交
99 100 101
/*
 * compare two keys in a memcmp fashion
 */
C
Chris Mason 已提交
102
static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
103
{
C
Chris Mason 已提交
104 105 106 107 108
	struct btrfs_key k1;

	btrfs_disk_key_to_cpu(&k1, disk);

	if (k1.objectid > k2->objectid)
109
		return 1;
C
Chris Mason 已提交
110
	if (k1.objectid < k2->objectid)
111
		return -1;
112 113 114 115
	if (k1.offset > k2->offset)
		return 1;
	if (k1.offset < k2->offset)
		return -1;
C
Chris Mason 已提交
116 117 118 119
	if (k1.flags > k2->flags)
		return 1;
	if (k1.flags < k2->flags)
		return -1;
120 121
	return 0;
}
C
Chris Mason 已提交
122

C
Chris Mason 已提交
123 124
static int check_node(struct btrfs_root *root, struct btrfs_path *path,
		      int level)
C
Chris Mason 已提交
125 126
{
	int i;
C
Chris Mason 已提交
127
	struct btrfs_node *parent = NULL;
C
Chris Mason 已提交
128
	struct btrfs_node *node = btrfs_buffer_node(path->nodes[level]);
C
Chris Mason 已提交
129
	int parent_slot;
130
	u32 nritems = btrfs_header_nritems(&node->header);
C
Chris Mason 已提交
131 132

	if (path->nodes[level + 1])
C
Chris Mason 已提交
133
		parent = btrfs_buffer_node(path->nodes[level + 1]);
C
Chris Mason 已提交
134
	parent_slot = path->slots[level + 1];
135 136
	BUG_ON(nritems == 0);
	if (parent) {
C
Chris Mason 已提交
137
		struct btrfs_disk_key *parent_key;
C
Chris Mason 已提交
138 139
		parent_key = &parent->ptrs[parent_slot].key;
		BUG_ON(memcmp(parent_key, &node->ptrs[0].key,
C
Chris Mason 已提交
140
			      sizeof(struct btrfs_disk_key)));
141
		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
142
		       btrfs_header_blocknr(&node->header));
C
Chris Mason 已提交
143
	}
C
Chris Mason 已提交
144
	BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
145
	for (i = 0; nritems > 1 && i < nritems - 2; i++) {
C
Chris Mason 已提交
146
		struct btrfs_key cpukey;
C
Chris Mason 已提交
147 148
		btrfs_disk_key_to_cpu(&cpukey, &node->ptrs[i + 1].key);
		BUG_ON(comp_keys(&node->ptrs[i].key, &cpukey) >= 0);
C
Chris Mason 已提交
149 150 151 152
	}
	return 0;
}

C
Chris Mason 已提交
153 154
static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
		      int level)
C
Chris Mason 已提交
155 156
{
	int i;
C
Chris Mason 已提交
157
	struct btrfs_leaf *leaf = btrfs_buffer_leaf(path->nodes[level]);
C
Chris Mason 已提交
158
	struct btrfs_node *parent = NULL;
C
Chris Mason 已提交
159
	int parent_slot;
160
	u32 nritems = btrfs_header_nritems(&leaf->header);
C
Chris Mason 已提交
161 162

	if (path->nodes[level + 1])
C
Chris Mason 已提交
163
		parent = btrfs_buffer_node(path->nodes[level + 1]);
C
Chris Mason 已提交
164
	parent_slot = path->slots[level + 1];
C
Chris Mason 已提交
165
	BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
166 167 168 169 170

	if (nritems == 0)
		return 0;

	if (parent) {
C
Chris Mason 已提交
171
		struct btrfs_disk_key *parent_key;
C
Chris Mason 已提交
172
		parent_key = &parent->ptrs[parent_slot].key;
C
Chris Mason 已提交
173
		BUG_ON(memcmp(parent_key, &leaf->items[0].key,
C
Chris Mason 已提交
174
		       sizeof(struct btrfs_disk_key)));
175
		BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
176
		       btrfs_header_blocknr(&leaf->header));
C
Chris Mason 已提交
177
	}
178
	for (i = 0; nritems > 1 && i < nritems - 2; i++) {
C
Chris Mason 已提交
179 180
		struct btrfs_key cpukey;
		btrfs_disk_key_to_cpu(&cpukey, &leaf->items[i + 1].key);
C
Chris Mason 已提交
181
		BUG_ON(comp_keys(&leaf->items[i].key,
C
Chris Mason 已提交
182
		                 &cpukey) >= 0);
C
Chris Mason 已提交
183 184
		BUG_ON(btrfs_item_offset(leaf->items + i) !=
			btrfs_item_end(leaf->items + i + 1));
C
Chris Mason 已提交
185
		if (i == 0) {
C
Chris Mason 已提交
186 187
			BUG_ON(btrfs_item_offset(leaf->items + i) +
			       btrfs_item_size(leaf->items + i) !=
C
Chris Mason 已提交
188
			       BTRFS_LEAF_DATA_SIZE(root));
C
Chris Mason 已提交
189 190 191 192 193
		}
	}
	return 0;
}

C
Chris Mason 已提交
194 195
static int check_block(struct btrfs_root *root, struct btrfs_path *path,
			int level)
C
Chris Mason 已提交
196 197
{
	if (level == 0)
C
Chris Mason 已提交
198 199
		return check_leaf(root, path, level);
	return check_node(root, path, level);
C
Chris Mason 已提交
200 201
}

C
Chris Mason 已提交
202 203 204 205 206 207 208 209 210
/*
 * 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
211
static int generic_bin_search(char *p, int item_size, struct btrfs_key *key,
212 213 214 215 216 217
		       int max, int *slot)
{
	int low = 0;
	int high = max;
	int mid;
	int ret;
C
Chris Mason 已提交
218
	struct btrfs_disk_key *tmp;
219 220 221

	while(low < high) {
		mid = (low + high) / 2;
C
Chris Mason 已提交
222
		tmp = (struct btrfs_disk_key *)(p + mid * item_size);
223 224 225 226 227 228 229 230 231 232 233 234 235 236 237
		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
238 239 240 241
/*
 * simple bin_search frontend that does the right thing for
 * leaves vs nodes
 */
C
Chris Mason 已提交
242
static int bin_search(struct btrfs_node *c, struct btrfs_key *key, int *slot)
243
{
244
	if (btrfs_is_leaf(c)) {
C
Chris Mason 已提交
245
		struct btrfs_leaf *l = (struct btrfs_leaf *)c;
C
Chris Mason 已提交
246 247
		return generic_bin_search((void *)l->items,
					  sizeof(struct btrfs_item),
248 249
					  key, btrfs_header_nritems(&c->header),
					  slot);
250
	} else {
C
Chris Mason 已提交
251 252
		return generic_bin_search((void *)c->ptrs,
					  sizeof(struct btrfs_key_ptr),
253 254
					  key, btrfs_header_nritems(&c->header),
					  slot);
255 256 257 258
	}
	return -1;
}

C
Chris Mason 已提交
259 260
static struct buffer_head *read_node_slot(struct btrfs_root *root,
				   struct buffer_head *parent_buf,
261 262
				   int slot)
{
C
Chris Mason 已提交
263
	struct btrfs_node *node = btrfs_buffer_node(parent_buf);
264 265
	if (slot < 0)
		return NULL;
266
	if (slot >= btrfs_header_nritems(&node->header))
267
		return NULL;
268
	return read_tree_block(root, btrfs_node_blockptr(node, slot));
269 270
}

271 272
static int balance_level(struct btrfs_trans_handle *trans, struct btrfs_root
			 *root, struct btrfs_path *path, int level)
273
{
C
Chris Mason 已提交
274 275 276 277
	struct buffer_head *right_buf;
	struct buffer_head *mid_buf;
	struct buffer_head *left_buf;
	struct buffer_head *parent_buf = NULL;
C
Chris Mason 已提交
278 279 280 281
	struct btrfs_node *right = NULL;
	struct btrfs_node *mid;
	struct btrfs_node *left = NULL;
	struct btrfs_node *parent = NULL;
282 283 284 285
	int ret = 0;
	int wret;
	int pslot;
	int orig_slot = path->slots[level];
286
	u64 orig_ptr;
287 288 289 290 291

	if (level == 0)
		return 0;

	mid_buf = path->nodes[level];
C
Chris Mason 已提交
292
	mid = btrfs_buffer_node(mid_buf);
293
	orig_ptr = btrfs_node_blockptr(mid, orig_slot);
294

C
Chris Mason 已提交
295
	if (level < BTRFS_MAX_LEVEL - 1)
296 297 298
		parent_buf = path->nodes[level + 1];
	pslot = path->slots[level + 1];

C
Chris Mason 已提交
299 300 301 302
	/*
	 * deal with the case where there is only one pointer in the root
	 * by promoting the node below to a root
	 */
303
	if (!parent_buf) {
C
Chris Mason 已提交
304 305
		struct buffer_head *child;
		u64 blocknr = mid_buf->b_blocknr;
306

307
		if (btrfs_header_nritems(&mid->header) != 1)
308 309 310 311 312 313 314 315
			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
316
		btrfs_block_release(root, mid_buf);
317
		/* once for the root ptr */
C
Chris Mason 已提交
318
		btrfs_block_release(root, mid_buf);
319 320
		clean_tree_block(trans, root, mid_buf);
		return btrfs_free_extent(trans, root, blocknr, 1, 1);
321
	}
C
Chris Mason 已提交
322
	parent = btrfs_buffer_node(parent_buf);
323

C
Chris Mason 已提交
324 325
	if (btrfs_header_nritems(&mid->header) >
	    BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
326 327 328 329
		return 0;

	left_buf = read_node_slot(root, parent_buf, pslot - 1);
	right_buf = read_node_slot(root, parent_buf, pslot + 1);
330 331

	/* first, try to make some room in the middle buffer */
332
	if (left_buf) {
333 334
		btrfs_cow_block(trans, root, left_buf, parent_buf, pslot - 1,
				&left_buf);
C
Chris Mason 已提交
335
		left = btrfs_buffer_node(left_buf);
336
		orig_slot += btrfs_header_nritems(&left->header);
337
		wret = push_node_left(trans, root, left_buf, mid_buf);
338 339
		if (wret < 0)
			ret = wret;
340
	}
341 342 343 344

	/*
	 * then try to empty the right most buffer into the middle
	 */
345
	if (right_buf) {
346 347
		btrfs_cow_block(trans, root, right_buf, parent_buf, pslot + 1,
				&right_buf);
C
Chris Mason 已提交
348
		right = btrfs_buffer_node(right_buf);
349
		wret = push_node_left(trans, root, mid_buf, right_buf);
350 351
		if (wret < 0)
			ret = wret;
352
		if (btrfs_header_nritems(&right->header) == 0) {
C
Chris Mason 已提交
353
			u64 blocknr = right_buf->b_blocknr;
C
Chris Mason 已提交
354
			btrfs_block_release(root, right_buf);
355
			clean_tree_block(trans, root, right_buf);
356 357
			right_buf = NULL;
			right = NULL;
358 359
			wret = del_ptr(trans, root, path, level + 1, pslot +
				       1);
360 361
			if (wret)
				ret = wret;
362
			wret = btrfs_free_extent(trans, root, blocknr, 1, 1);
363 364 365
			if (wret)
				ret = wret;
		} else {
C
Chris Mason 已提交
366 367
			memcpy(&parent->ptrs[pslot + 1].key,
				&right->ptrs[0].key,
C
Chris Mason 已提交
368
				sizeof(struct btrfs_disk_key));
369
			mark_buffer_dirty(parent_buf);
370 371
		}
	}
372
	if (btrfs_header_nritems(&mid->header) == 1) {
373 374 375 376 377 378 379 380 381 382
		/*
		 * 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);
383
		wret = balance_node_right(trans, root, mid_buf, left_buf);
384 385 386 387
		if (wret < 0)
			ret = wret;
		BUG_ON(wret == 1);
	}
388
	if (btrfs_header_nritems(&mid->header) == 0) {
389
		/* we've managed to empty the middle node, drop it */
C
Chris Mason 已提交
390
		u64 blocknr = mid_buf->b_blocknr;
C
Chris Mason 已提交
391
		btrfs_block_release(root, mid_buf);
392
		clean_tree_block(trans, root, mid_buf);
393 394
		mid_buf = NULL;
		mid = NULL;
395
		wret = del_ptr(trans, root, path, level + 1, pslot);
396 397
		if (wret)
			ret = wret;
398
		wret = btrfs_free_extent(trans, root, blocknr, 1, 1);
399 400
		if (wret)
			ret = wret;
401 402
	} else {
		/* update the parent key to reflect our changes */
C
Chris Mason 已提交
403
		memcpy(&parent->ptrs[pslot].key, &mid->ptrs[0].key,
C
Chris Mason 已提交
404
		       sizeof(struct btrfs_disk_key));
405
		mark_buffer_dirty(parent_buf);
406
	}
407

408
	/* update the path */
409
	if (left_buf) {
410
		if (btrfs_header_nritems(&left->header) > orig_slot) {
C
Chris Mason 已提交
411
			get_bh(left_buf);
412 413 414 415
			path->nodes[level] = left_buf;
			path->slots[level + 1] -= 1;
			path->slots[level] = orig_slot;
			if (mid_buf)
C
Chris Mason 已提交
416
				btrfs_block_release(root, mid_buf);
417
		} else {
418
			orig_slot -= btrfs_header_nritems(&left->header);
419 420 421
			path->slots[level] = orig_slot;
		}
	}
422
	/* double check we haven't messed things up */
C
Chris Mason 已提交
423
	check_block(root, path, level);
C
Chris Mason 已提交
424 425 426
	if (orig_ptr !=
	    btrfs_node_blockptr(btrfs_buffer_node(path->nodes[level]),
				path->slots[level]))
427
		BUG();
428 429

	if (right_buf)
C
Chris Mason 已提交
430
		btrfs_block_release(root, right_buf);
431
	if (left_buf)
C
Chris Mason 已提交
432
		btrfs_block_release(root, left_buf);
433 434 435
	return ret;
}

C
Chris Mason 已提交
436 437 438 439 440 441
/*
 * 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
442 443
 * 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
444 445 446 447
 *
 * 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
448
 */
449 450 451
int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, struct btrfs_key *key, struct btrfs_path *p, int
		      ins_len, int cow)
452
{
C
Chris Mason 已提交
453 454
	struct buffer_head *b;
	struct buffer_head *cow_buf;
C
Chris Mason 已提交
455
	struct btrfs_node *c;
456 457 458
	int slot;
	int ret;
	int level;
C
Chris Mason 已提交
459

460 461
again:
	b = root->node;
C
Chris Mason 已提交
462
	get_bh(b);
463
	while (b) {
C
Chris Mason 已提交
464 465
		c = btrfs_buffer_node(b);
		level = btrfs_header_level(&c->header);
C
Chris Mason 已提交
466 467
		if (cow) {
			int wret;
C
Chris Mason 已提交
468 469 470
			wret = btrfs_cow_block(trans, root, b,
					       p->nodes[level + 1],
					       p->slots[level + 1],
471
					       &cow_buf);
C
Chris Mason 已提交
472 473 474
			b = cow_buf;
		}
		BUG_ON(!cow && ins_len);
C
Chris Mason 已提交
475
		c = btrfs_buffer_node(b);
476
		p->nodes[level] = b;
C
Chris Mason 已提交
477
		ret = check_block(root, p, level);
C
Chris Mason 已提交
478 479
		if (ret)
			return -1;
480
		ret = bin_search(c, key, &slot);
481
		if (!btrfs_is_leaf(c)) {
482 483 484
			if (ret && slot > 0)
				slot -= 1;
			p->slots[level] = slot;
485
			if (ins_len > 0 && btrfs_header_nritems(&c->header) ==
C
Chris Mason 已提交
486
			    BTRFS_NODEPTRS_PER_BLOCK(root)) {
487
				int sret = split_node(trans, root, p, level);
C
Chris Mason 已提交
488 489 490 491
				BUG_ON(sret > 0);
				if (sret)
					return sret;
				b = p->nodes[level];
C
Chris Mason 已提交
492
				c = btrfs_buffer_node(b);
C
Chris Mason 已提交
493
				slot = p->slots[level];
494
			} else if (ins_len < 0) {
495 496
				int sret = balance_level(trans, root, p,
							 level);
497 498 499 500 501
				if (sret)
					return sret;
				b = p->nodes[level];
				if (!b)
					goto again;
C
Chris Mason 已提交
502
				c = btrfs_buffer_node(b);
503
				slot = p->slots[level];
504
				BUG_ON(btrfs_header_nritems(&c->header) == 1);
C
Chris Mason 已提交
505
			}
506
			b = read_tree_block(root, btrfs_node_blockptr(c, slot));
507
		} else {
C
Chris Mason 已提交
508
			struct btrfs_leaf *l = (struct btrfs_leaf *)c;
509
			p->slots[level] = slot;
C
Chris Mason 已提交
510
			if (ins_len > 0 && btrfs_leaf_free_space(root, l) <
C
Chris Mason 已提交
511
			    sizeof(struct btrfs_item) + ins_len) {
512
				int sret = split_leaf(trans, root, p, ins_len);
C
Chris Mason 已提交
513 514 515 516
				BUG_ON(sret > 0);
				if (sret)
					return sret;
			}
517 518 519
			return ret;
		}
	}
C
Chris Mason 已提交
520
	return 1;
521 522
}

C
Chris Mason 已提交
523 524 525 526 527 528
/*
 * 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
529 530 531
 *
 * 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
532
 */
533 534 535
static int fixup_low_keys(struct btrfs_trans_handle *trans, struct btrfs_root
			  *root, struct btrfs_path *path, struct btrfs_disk_key
			  *key, int level)
536 537
{
	int i;
C
Chris Mason 已提交
538
	int ret = 0;
C
Chris Mason 已提交
539 540
	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
		struct btrfs_node *t;
541
		int tslot = path->slots[i];
542
		if (!path->nodes[i])
543
			break;
C
Chris Mason 已提交
544
		t = btrfs_buffer_node(path->nodes[i]);
C
Chris Mason 已提交
545
		memcpy(&t->ptrs[tslot].key, key, sizeof(*key));
546
		mark_buffer_dirty(path->nodes[i]);
547 548 549
		if (tslot != 0)
			break;
	}
C
Chris Mason 已提交
550
	return ret;
551 552
}

C
Chris Mason 已提交
553 554
/*
 * try to push data from one node into the next node left in the
555
 * tree.
C
Chris Mason 已提交
556 557 558
 *
 * 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
559
 */
560
static int push_node_left(struct btrfs_trans_handle *trans, struct btrfs_root
C
Chris Mason 已提交
561 562
			  *root, struct buffer_head *dst_buf, struct
			  buffer_head *src_buf)
563
{
C
Chris Mason 已提交
564 565
	struct btrfs_node *src = btrfs_buffer_node(src_buf);
	struct btrfs_node *dst = btrfs_buffer_node(dst_buf);
566
	int push_items = 0;
567 568
	int src_nritems;
	int dst_nritems;
C
Chris Mason 已提交
569
	int ret = 0;
570

571 572
	src_nritems = btrfs_header_nritems(&src->header);
	dst_nritems = btrfs_header_nritems(&dst->header);
C
Chris Mason 已提交
573
	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
574
	if (push_items <= 0) {
575
		return 1;
576
	}
577

578
	if (src_nritems < push_items)
579 580
		push_items = src_nritems;

C
Chris Mason 已提交
581 582
	memcpy(dst->ptrs + dst_nritems, src->ptrs,
		push_items * sizeof(struct btrfs_key_ptr));
583
	if (push_items < src_nritems) {
C
Chris Mason 已提交
584
		memmove(src->ptrs, src->ptrs + push_items,
C
Chris Mason 已提交
585
			(src_nritems - push_items) *
C
Chris Mason 已提交
586
			sizeof(struct btrfs_key_ptr));
587
	}
588 589
	btrfs_set_header_nritems(&src->header, src_nritems - push_items);
	btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
590 591
	mark_buffer_dirty(src_buf);
	mark_buffer_dirty(dst_buf);
592 593 594 595 596 597 598 599 600 601 602 603
	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
 */
604
static int balance_node_right(struct btrfs_trans_handle *trans, struct
C
Chris Mason 已提交
605 606
			      btrfs_root *root, struct buffer_head *dst_buf,
			      struct buffer_head *src_buf)
607
{
C
Chris Mason 已提交
608 609
	struct btrfs_node *src = btrfs_buffer_node(src_buf);
	struct btrfs_node *dst = btrfs_buffer_node(dst_buf);
610 611 612 613 614 615
	int push_items = 0;
	int max_push;
	int src_nritems;
	int dst_nritems;
	int ret = 0;

616 617
	src_nritems = btrfs_header_nritems(&src->header);
	dst_nritems = btrfs_header_nritems(&dst->header);
C
Chris Mason 已提交
618
	push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
619 620 621 622 623 624 625 626 627 628 629
	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;

C
Chris Mason 已提交
630 631 632 633
	memmove(dst->ptrs + push_items, dst->ptrs,
		dst_nritems * sizeof(struct btrfs_key_ptr));
	memcpy(dst->ptrs, src->ptrs + src_nritems - push_items,
		push_items * sizeof(struct btrfs_key_ptr));
634

635 636
	btrfs_set_header_nritems(&src->header, src_nritems - push_items);
	btrfs_set_header_nritems(&dst->header, dst_nritems + push_items);
637

638 639
	mark_buffer_dirty(src_buf);
	mark_buffer_dirty(dst_buf);
C
Chris Mason 已提交
640
	return ret;
641 642
}

C
Chris Mason 已提交
643 644 645 646
/*
 * 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
647 648
 *
 * returns zero on success or < 0 on failure.
C
Chris Mason 已提交
649
 */
650 651
static int insert_new_root(struct btrfs_trans_handle *trans, struct btrfs_root
			   *root, struct btrfs_path *path, int level)
C
Chris Mason 已提交
652
{
C
Chris Mason 已提交
653
	struct buffer_head *t;
C
Chris Mason 已提交
654 655
	struct btrfs_node *lower;
	struct btrfs_node *c;
C
Chris Mason 已提交
656
	struct btrfs_disk_key *lower_key;
C
Chris Mason 已提交
657 658 659 660

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

661
	t = btrfs_alloc_free_block(trans, root);
C
Chris Mason 已提交
662
	c = btrfs_buffer_node(t);
C
Chris Mason 已提交
663
	memset(c, 0, root->blocksize);
664 665
	btrfs_set_header_nritems(&c->header, 1);
	btrfs_set_header_level(&c->header, level);
C
Chris Mason 已提交
666
	btrfs_set_header_blocknr(&c->header, t->b_blocknr);
667
	btrfs_set_header_generation(&c->header, trans->transid);
668
	btrfs_set_header_parentid(&c->header,
C
Chris Mason 已提交
669 670
	      btrfs_header_parentid(btrfs_buffer_header(root->node)));
	lower = btrfs_buffer_node(path->nodes[level-1]);
671
	if (btrfs_is_leaf(lower))
C
Chris Mason 已提交
672
		lower_key = &((struct btrfs_leaf *)lower)->items[0].key;
C
Chris Mason 已提交
673
	else
C
Chris Mason 已提交
674 675
		lower_key = &lower->ptrs[0].key;
	memcpy(&c->ptrs[0].key, lower_key, sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
676
	btrfs_set_node_blockptr(c, 0, path->nodes[level - 1]->b_blocknr);
677 678 679

	mark_buffer_dirty(t);

C
Chris Mason 已提交
680
	/* the super has an extra ref to root->node */
C
Chris Mason 已提交
681
	btrfs_block_release(root, root->node);
C
Chris Mason 已提交
682
	root->node = t;
C
Chris Mason 已提交
683
	get_bh(t);
C
Chris Mason 已提交
684 685 686 687 688
	path->nodes[level] = t;
	path->slots[level] = 0;
	return 0;
}

C
Chris Mason 已提交
689 690 691
/*
 * 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
692
 *
C
Chris Mason 已提交
693 694
 * 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
695 696
 *
 * returns zero on success and < 0 on any error
C
Chris Mason 已提交
697
 */
698 699 700
static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, struct btrfs_path *path, struct btrfs_disk_key
		      *key, u64 blocknr, int slot, int level)
C
Chris Mason 已提交
701
{
C
Chris Mason 已提交
702
	struct btrfs_node *lower;
C
Chris Mason 已提交
703
	int nritems;
C
Chris Mason 已提交
704 705

	BUG_ON(!path->nodes[level]);
C
Chris Mason 已提交
706
	lower = btrfs_buffer_node(path->nodes[level]);
707
	nritems = btrfs_header_nritems(&lower->header);
C
Chris Mason 已提交
708 709
	if (slot > nritems)
		BUG();
C
Chris Mason 已提交
710
	if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
C
Chris Mason 已提交
711 712
		BUG();
	if (slot != nritems) {
C
Chris Mason 已提交
713 714
		memmove(lower->ptrs + slot + 1, lower->ptrs + slot,
			(nritems - slot) * sizeof(struct btrfs_key_ptr));
C
Chris Mason 已提交
715
	}
C
Chris Mason 已提交
716
	memcpy(&lower->ptrs[slot].key, key, sizeof(struct btrfs_disk_key));
717
	btrfs_set_node_blockptr(lower, slot, blocknr);
718
	btrfs_set_header_nritems(&lower->header, nritems + 1);
719
	mark_buffer_dirty(path->nodes[level]);
C
Chris Mason 已提交
720 721 722
	return 0;
}

C
Chris Mason 已提交
723 724 725 726 727 728
/*
 * 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
729 730
 *
 * returns 0 on success and < 0 on failure
C
Chris Mason 已提交
731
 */
732 733
static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, struct btrfs_path *path, int level)
734
{
C
Chris Mason 已提交
735
	struct buffer_head *t;
C
Chris Mason 已提交
736
	struct btrfs_node *c;
C
Chris Mason 已提交
737
	struct buffer_head *split_buffer;
C
Chris Mason 已提交
738
	struct btrfs_node *split;
739
	int mid;
C
Chris Mason 已提交
740
	int ret;
C
Chris Mason 已提交
741
	int wret;
742
	u32 c_nritems;
743

C
Chris Mason 已提交
744
	t = path->nodes[level];
C
Chris Mason 已提交
745
	c = btrfs_buffer_node(t);
C
Chris Mason 已提交
746 747
	if (t == root->node) {
		/* trying to split the root, lets make a new one */
748
		ret = insert_new_root(trans, root, path, level + 1);
C
Chris Mason 已提交
749 750
		if (ret)
			return ret;
751
	}
752
	c_nritems = btrfs_header_nritems(&c->header);
753
	split_buffer = btrfs_alloc_free_block(trans, root);
C
Chris Mason 已提交
754
	split = btrfs_buffer_node(split_buffer);
755
	btrfs_set_header_flags(&split->header, btrfs_header_flags(&c->header));
756
	btrfs_set_header_level(&split->header, btrfs_header_level(&c->header));
C
Chris Mason 已提交
757
	btrfs_set_header_blocknr(&split->header, split_buffer->b_blocknr);
758
	btrfs_set_header_generation(&split->header, trans->transid);
759
	btrfs_set_header_parentid(&split->header,
C
Chris Mason 已提交
760
	      btrfs_header_parentid(btrfs_buffer_header(root->node)));
761
	mid = (c_nritems + 1) / 2;
C
Chris Mason 已提交
762 763
	memcpy(split->ptrs, c->ptrs + mid,
		(c_nritems - mid) * sizeof(struct btrfs_key_ptr));
764 765
	btrfs_set_header_nritems(&split->header, c_nritems - mid);
	btrfs_set_header_nritems(&c->header, mid);
C
Chris Mason 已提交
766 767
	ret = 0;

768 769
	mark_buffer_dirty(t);
	mark_buffer_dirty(split_buffer);
770
	wret = insert_ptr(trans, root, path, &split->ptrs[0].key,
C
Chris Mason 已提交
771
			  split_buffer->b_blocknr, path->slots[level + 1] + 1,
C
Chris Mason 已提交
772
			  level + 1);
C
Chris Mason 已提交
773 774 775
	if (wret)
		ret = wret;

C
Chris Mason 已提交
776
	if (path->slots[level] >= mid) {
C
Chris Mason 已提交
777
		path->slots[level] -= mid;
C
Chris Mason 已提交
778
		btrfs_block_release(root, t);
C
Chris Mason 已提交
779 780 781
		path->nodes[level] = split_buffer;
		path->slots[level + 1] += 1;
	} else {
C
Chris Mason 已提交
782
		btrfs_block_release(root, split_buffer);
783
	}
C
Chris Mason 已提交
784
	return ret;
785 786
}

C
Chris Mason 已提交
787 788 789 790 791
/*
 * 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
792
static int leaf_space_used(struct btrfs_leaf *l, int start, int nr)
793 794 795 796 797 798
{
	int data_len;
	int end = start + nr - 1;

	if (!nr)
		return 0;
C
Chris Mason 已提交
799 800 801
	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;
802 803 804
	return data_len;
}

C
Chris Mason 已提交
805 806 807
/*
 * 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
808 809 810
 *
 * 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
811
 */
812 813
static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
			   *root, struct btrfs_path *path, int data_size)
C
Chris Mason 已提交
814
{
C
Chris Mason 已提交
815 816
	struct buffer_head *left_buf = path->nodes[0];
	struct btrfs_leaf *left = btrfs_buffer_leaf(left_buf);
C
Chris Mason 已提交
817
	struct btrfs_leaf *right;
C
Chris Mason 已提交
818 819 820
	struct buffer_head *right_buf;
	struct buffer_head *upper;
	struct btrfs_node *upper_node;
C
Chris Mason 已提交
821 822 823 824 825
	int slot;
	int i;
	int free_space;
	int push_space = 0;
	int push_items = 0;
C
Chris Mason 已提交
826
	struct btrfs_item *item;
827 828
	u32 left_nritems;
	u32 right_nritems;
C
Chris Mason 已提交
829 830 831 832 833 834

	slot = path->slots[1];
	if (!path->nodes[1]) {
		return 1;
	}
	upper = path->nodes[1];
C
Chris Mason 已提交
835 836
	upper_node = btrfs_buffer_node(upper);
	if (slot >= btrfs_header_nritems(&upper_node->header) - 1) {
C
Chris Mason 已提交
837 838
		return 1;
	}
C
Chris Mason 已提交
839 840 841
	right_buf = read_tree_block(root,
		    btrfs_node_blockptr(btrfs_buffer_node(upper), slot + 1));
	right = btrfs_buffer_leaf(right_buf);
C
Chris Mason 已提交
842
	free_space = btrfs_leaf_free_space(root, right);
C
Chris Mason 已提交
843
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
844
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
845 846
		return 1;
	}
C
Chris Mason 已提交
847
	/* cow and double check */
848
	btrfs_cow_block(trans, root, right_buf, upper, slot + 1, &right_buf);
C
Chris Mason 已提交
849
	right = btrfs_buffer_leaf(right_buf);
C
Chris Mason 已提交
850
	free_space = btrfs_leaf_free_space(root, right);
C
Chris Mason 已提交
851
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
852
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
853 854 855
		return 1;
	}

856 857
	left_nritems = btrfs_header_nritems(&left->header);
	for (i = left_nritems - 1; i >= 0; i--) {
C
Chris Mason 已提交
858 859 860
		item = left->items + i;
		if (path->slots[0] == i)
			push_space += data_size + sizeof(*item);
C
Chris Mason 已提交
861 862
		if (btrfs_item_size(item) + sizeof(*item) + push_space >
		    free_space)
C
Chris Mason 已提交
863 864
			break;
		push_items++;
C
Chris Mason 已提交
865
		push_space += btrfs_item_size(item) + sizeof(*item);
C
Chris Mason 已提交
866 867
	}
	if (push_items == 0) {
C
Chris Mason 已提交
868
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
869 870
		return 1;
	}
871
	right_nritems = btrfs_header_nritems(&right->header);
C
Chris Mason 已提交
872
	/* push left to right */
C
Chris Mason 已提交
873
	push_space = btrfs_item_end(left->items + left_nritems - push_items);
C
Chris Mason 已提交
874
	push_space -= leaf_data_end(root, left);
C
Chris Mason 已提交
875
	/* make room in the right data area */
C
Chris Mason 已提交
876 877 878
	memmove(btrfs_leaf_data(right) + leaf_data_end(root, right) -
		push_space, btrfs_leaf_data(right) + leaf_data_end(root, right),
		BTRFS_LEAF_DATA_SIZE(root) - leaf_data_end(root, right));
C
Chris Mason 已提交
879
	/* copy from the left data area */
C
Chris Mason 已提交
880 881
	memcpy(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) - push_space,
		btrfs_leaf_data(left) + leaf_data_end(root, left), push_space);
C
Chris Mason 已提交
882
	memmove(right->items + push_items, right->items,
C
Chris Mason 已提交
883
		right_nritems * sizeof(struct btrfs_item));
C
Chris Mason 已提交
884
	/* copy the items from left to right */
885
	memcpy(right->items, left->items + left_nritems - push_items,
C
Chris Mason 已提交
886
		push_items * sizeof(struct btrfs_item));
C
Chris Mason 已提交
887 888

	/* update the item pointers */
889 890
	right_nritems += push_items;
	btrfs_set_header_nritems(&right->header, right_nritems);
C
Chris Mason 已提交
891
	push_space = BTRFS_LEAF_DATA_SIZE(root);
892
	for (i = 0; i < right_nritems; i++) {
C
Chris Mason 已提交
893 894 895
		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
896
	}
897 898
	left_nritems -= push_items;
	btrfs_set_header_nritems(&left->header, left_nritems);
C
Chris Mason 已提交
899

900 901
	mark_buffer_dirty(left_buf);
	mark_buffer_dirty(right_buf);
C
Chris Mason 已提交
902
	memcpy(&upper_node->ptrs[slot + 1].key,
C
Chris Mason 已提交
903
		&right->items[0].key, sizeof(struct btrfs_disk_key));
904
	mark_buffer_dirty(upper);
C
Chris Mason 已提交
905

C
Chris Mason 已提交
906
	/* then fixup the leaf pointer in the path */
907 908
	if (path->slots[0] >= left_nritems) {
		path->slots[0] -= left_nritems;
C
Chris Mason 已提交
909
		btrfs_block_release(root, path->nodes[0]);
C
Chris Mason 已提交
910 911 912
		path->nodes[0] = right_buf;
		path->slots[1] += 1;
	} else {
C
Chris Mason 已提交
913
		btrfs_block_release(root, right_buf);
C
Chris Mason 已提交
914 915 916
	}
	return 0;
}
C
Chris Mason 已提交
917 918 919 920
/*
 * 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
 */
921 922
static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
			  *root, struct btrfs_path *path, int data_size)
923
{
C
Chris Mason 已提交
924 925 926
	struct buffer_head *right_buf = path->nodes[0];
	struct btrfs_leaf *right = btrfs_buffer_leaf(right_buf);
	struct buffer_head *t;
C
Chris Mason 已提交
927
	struct btrfs_leaf *left;
928 929 930 931 932
	int slot;
	int i;
	int free_space;
	int push_space = 0;
	int push_items = 0;
C
Chris Mason 已提交
933
	struct btrfs_item *item;
934
	u32 old_left_nritems;
C
Chris Mason 已提交
935 936
	int ret = 0;
	int wret;
937 938 939 940 941 942 943 944

	slot = path->slots[1];
	if (slot == 0) {
		return 1;
	}
	if (!path->nodes[1]) {
		return 1;
	}
C
Chris Mason 已提交
945 946 947
	t = read_tree_block(root,
	    btrfs_node_blockptr(btrfs_buffer_node(path->nodes[1]), slot - 1));
	left = btrfs_buffer_leaf(t);
C
Chris Mason 已提交
948
	free_space = btrfs_leaf_free_space(root, left);
C
Chris Mason 已提交
949
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
950
		btrfs_block_release(root, t);
951 952
		return 1;
	}
C
Chris Mason 已提交
953 954

	/* cow and double check */
955
	btrfs_cow_block(trans, root, t, path->nodes[1], slot - 1, &t);
C
Chris Mason 已提交
956
	left = btrfs_buffer_leaf(t);
C
Chris Mason 已提交
957
	free_space = btrfs_leaf_free_space(root, left);
C
Chris Mason 已提交
958
	if (free_space < data_size + sizeof(struct btrfs_item)) {
C
Chris Mason 已提交
959
		btrfs_block_release(root, t);
C
Chris Mason 已提交
960 961 962
		return 1;
	}

963
	for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
964 965 966
		item = right->items + i;
		if (path->slots[0] == i)
			push_space += data_size + sizeof(*item);
C
Chris Mason 已提交
967 968
		if (btrfs_item_size(item) + sizeof(*item) + push_space >
		    free_space)
969 970
			break;
		push_items++;
C
Chris Mason 已提交
971
		push_space += btrfs_item_size(item) + sizeof(*item);
972 973
	}
	if (push_items == 0) {
C
Chris Mason 已提交
974
		btrfs_block_release(root, t);
975 976 977
		return 1;
	}
	/* push data from right to left */
978
	memcpy(left->items + btrfs_header_nritems(&left->header),
C
Chris Mason 已提交
979
		right->items, push_items * sizeof(struct btrfs_item));
C
Chris Mason 已提交
980
	push_space = BTRFS_LEAF_DATA_SIZE(root) -
C
Chris Mason 已提交
981
		     btrfs_item_offset(right->items + push_items -1);
C
Chris Mason 已提交
982 983 984
	memcpy(btrfs_leaf_data(left) + leaf_data_end(root, left) - push_space,
		btrfs_leaf_data(right) +
		btrfs_item_offset(right->items + push_items - 1),
985
		push_space);
986
	old_left_nritems = btrfs_header_nritems(&left->header);
987 988
	BUG_ON(old_left_nritems < 0);

C
Chris Mason 已提交
989
	for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
C
Chris Mason 已提交
990 991 992
		u32 ioff = btrfs_item_offset(left->items + i);
		btrfs_set_item_offset(left->items + i, ioff -
				     (BTRFS_LEAF_DATA_SIZE(root) -
C
Chris Mason 已提交
993 994
				      btrfs_item_offset(left->items +
						        old_left_nritems - 1)));
995
	}
996
	btrfs_set_header_nritems(&left->header, old_left_nritems + push_items);
997 998

	/* fixup right node */
C
Chris Mason 已提交
999
	push_space = btrfs_item_offset(right->items + push_items - 1) -
C
Chris Mason 已提交
1000 1001 1002 1003
		     leaf_data_end(root, right);
	memmove(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
		push_space, btrfs_leaf_data(right) +
		leaf_data_end(root, right), push_space);
1004
	memmove(right->items, right->items + push_items,
1005
		(btrfs_header_nritems(&right->header) - push_items) *
C
Chris Mason 已提交
1006
		sizeof(struct btrfs_item));
1007 1008 1009
	btrfs_set_header_nritems(&right->header,
				 btrfs_header_nritems(&right->header) -
				 push_items);
C
Chris Mason 已提交
1010
	push_space = BTRFS_LEAF_DATA_SIZE(root);
1011

1012
	for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
C
Chris Mason 已提交
1013 1014 1015
		btrfs_set_item_offset(right->items + i, push_space -
				      btrfs_item_size(right->items + i));
		push_space = btrfs_item_offset(right->items + i);
1016
	}
1017

1018 1019
	mark_buffer_dirty(t);
	mark_buffer_dirty(right_buf);
1020

1021
	wret = fixup_low_keys(trans, root, path, &right->items[0].key, 1);
C
Chris Mason 已提交
1022 1023
	if (wret)
		ret = wret;
1024 1025 1026 1027

	/* 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
1028
		btrfs_block_release(root, path->nodes[0]);
1029
		path->nodes[0] = t;
1030 1031
		path->slots[1] -= 1;
	} else {
C
Chris Mason 已提交
1032
		btrfs_block_release(root, t);
1033 1034
		path->slots[0] -= push_items;
	}
1035
	BUG_ON(path->slots[0] < 0);
C
Chris Mason 已提交
1036
	return ret;
1037 1038
}

C
Chris Mason 已提交
1039 1040 1041
/*
 * 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
1042 1043
 *
 * returns 0 if all went well and < 0 on failure.
C
Chris Mason 已提交
1044
 */
1045 1046
static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, struct btrfs_path *path, int data_size)
1047
{
C
Chris Mason 已提交
1048
	struct buffer_head *l_buf;
C
Chris Mason 已提交
1049
	struct btrfs_leaf *l;
1050
	u32 nritems;
1051 1052
	int mid;
	int slot;
C
Chris Mason 已提交
1053
	struct btrfs_leaf *right;
C
Chris Mason 已提交
1054
	struct buffer_head *right_buffer;
C
Chris Mason 已提交
1055
	int space_needed = data_size + sizeof(struct btrfs_item);
1056 1057 1058 1059
	int data_copy_size;
	int rt_data_off;
	int i;
	int ret;
C
Chris Mason 已提交
1060 1061
	int wret;

C
Chris Mason 已提交
1062
	/* first try to make some room by pushing left and right */
1063
	wret = push_leaf_left(trans, root, path, data_size);
C
Chris Mason 已提交
1064 1065 1066
	if (wret < 0)
		return wret;
	if (wret) {
1067
		wret = push_leaf_right(trans, root, path, data_size);
C
Chris Mason 已提交
1068 1069 1070
		if (wret < 0)
			return wret;
	}
C
Chris Mason 已提交
1071
	l_buf = path->nodes[0];
C
Chris Mason 已提交
1072
	l = btrfs_buffer_leaf(l_buf);
C
Chris Mason 已提交
1073 1074

	/* did the pushes work? */
C
Chris Mason 已提交
1075 1076
	if (btrfs_leaf_free_space(root, l) >=
	    sizeof(struct btrfs_item) + data_size)
C
Chris Mason 已提交
1077 1078
		return 0;

C
Chris Mason 已提交
1079
	if (!path->nodes[1]) {
1080
		ret = insert_new_root(trans, root, path, 1);
C
Chris Mason 已提交
1081 1082 1083
		if (ret)
			return ret;
	}
1084
	slot = path->slots[0];
1085
	nritems = btrfs_header_nritems(&l->header);
1086
	mid = (nritems + 1)/ 2;
1087
	right_buffer = btrfs_alloc_free_block(trans, root);
1088 1089
	BUG_ON(!right_buffer);
	BUG_ON(mid == nritems);
C
Chris Mason 已提交
1090
	right = btrfs_buffer_leaf(right_buffer);
C
Chris Mason 已提交
1091
	memset(&right->header, 0, sizeof(right->header));
1092
	if (mid <= slot) {
C
Chris Mason 已提交
1093
		/* FIXME, just alloc a new leaf here */
1094
		if (leaf_space_used(l, mid, nritems - mid) + space_needed >
C
Chris Mason 已提交
1095
			BTRFS_LEAF_DATA_SIZE(root))
1096 1097
			BUG();
	} else {
C
Chris Mason 已提交
1098
		/* FIXME, just alloc a new leaf here */
1099
		if (leaf_space_used(l, 0, mid + 1) + space_needed >
C
Chris Mason 已提交
1100
			BTRFS_LEAF_DATA_SIZE(root))
1101 1102
			BUG();
	}
1103
	btrfs_set_header_nritems(&right->header, nritems - mid);
C
Chris Mason 已提交
1104
	btrfs_set_header_blocknr(&right->header, right_buffer->b_blocknr);
1105
	btrfs_set_header_generation(&right->header, trans->transid);
1106 1107
	btrfs_set_header_level(&right->header, 0);
	btrfs_set_header_parentid(&right->header,
C
Chris Mason 已提交
1108
	      btrfs_header_parentid(btrfs_buffer_header(root->node)));
C
Chris Mason 已提交
1109 1110
	data_copy_size = btrfs_item_end(l->items + mid) -
			 leaf_data_end(root, l);
1111
	memcpy(right->items, l->items + mid,
C
Chris Mason 已提交
1112
	       (nritems - mid) * sizeof(struct btrfs_item));
C
Chris Mason 已提交
1113 1114 1115 1116 1117
	memcpy(btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
		data_copy_size, btrfs_leaf_data(l) +
		leaf_data_end(root, l), data_copy_size);
	rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
		      btrfs_item_end(l->items + mid);
C
Chris Mason 已提交
1118

C
Chris Mason 已提交
1119
	for (i = 0; i < btrfs_header_nritems(&right->header); i++) {
C
Chris Mason 已提交
1120
		u32 ioff = btrfs_item_offset(right->items + i);
C
Chris Mason 已提交
1121 1122
		btrfs_set_item_offset(right->items + i, ioff + rt_data_off);
	}
C
Chris Mason 已提交
1123

1124
	btrfs_set_header_nritems(&l->header, mid);
C
Chris Mason 已提交
1125
	ret = 0;
1126
	wret = insert_ptr(trans, root, path, &right->items[0].key,
C
Chris Mason 已提交
1127
			  right_buffer->b_blocknr, path->slots[1] + 1, 1);
C
Chris Mason 已提交
1128 1129
	if (wret)
		ret = wret;
1130 1131
	mark_buffer_dirty(right_buffer);
	mark_buffer_dirty(l_buf);
1132
	BUG_ON(path->slots[0] != slot);
1133
	if (mid <= slot) {
C
Chris Mason 已提交
1134
		btrfs_block_release(root, path->nodes[0]);
1135
		path->nodes[0] = right_buffer;
1136 1137
		path->slots[0] -= mid;
		path->slots[1] += 1;
1138
	} else
C
Chris Mason 已提交
1139
		btrfs_block_release(root, right_buffer);
1140
	BUG_ON(path->slots[0] < 0);
1141 1142 1143
	return ret;
}

C
Chris Mason 已提交
1144 1145 1146 1147
/*
 * 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.
 */
1148 1149 1150
int btrfs_insert_empty_item(struct btrfs_trans_handle *trans, struct btrfs_root
			    *root, struct btrfs_path *path, struct btrfs_key
			    *cpu_key, u32 data_size)
1151
{
C
Chris Mason 已提交
1152
	int ret = 0;
1153
	int slot;
1154
	int slot_orig;
C
Chris Mason 已提交
1155
	struct btrfs_leaf *leaf;
C
Chris Mason 已提交
1156
	struct buffer_head *leaf_buf;
1157
	u32 nritems;
1158
	unsigned int data_end;
C
Chris Mason 已提交
1159 1160 1161
	struct btrfs_disk_key disk_key;

	btrfs_cpu_key_to_disk(&disk_key, cpu_key);
1162

C
Chris Mason 已提交
1163
	/* create a root if there isn't one */
C
Chris Mason 已提交
1164
	if (!root->node)
C
Chris Mason 已提交
1165
		BUG();
1166
	ret = btrfs_search_slot(trans, root, cpu_key, path, data_size, 1);
1167
	if (ret == 0) {
1168
		return -EEXIST;
C
Chris Mason 已提交
1169
	}
1170 1171
	if (ret < 0)
		goto out;
1172

1173 1174
	slot_orig = path->slots[0];
	leaf_buf = path->nodes[0];
C
Chris Mason 已提交
1175
	leaf = btrfs_buffer_leaf(leaf_buf);
C
Chris Mason 已提交
1176

1177
	nritems = btrfs_header_nritems(&leaf->header);
C
Chris Mason 已提交
1178
	data_end = leaf_data_end(root, leaf);
1179

C
Chris Mason 已提交
1180
	if (btrfs_leaf_free_space(root, leaf) <
C
Chris Mason 已提交
1181
	    sizeof(struct btrfs_item) + data_size)
1182 1183
		BUG();

1184
	slot = path->slots[0];
1185
	BUG_ON(slot < 0);
1186 1187
	if (slot != nritems) {
		int i;
C
Chris Mason 已提交
1188
		unsigned int old_data = btrfs_item_end(leaf->items + slot);
1189 1190 1191 1192 1193

		/*
		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
		 */
		/* first correct the data pointers */
C
Chris Mason 已提交
1194
		for (i = slot; i < nritems; i++) {
C
Chris Mason 已提交
1195
			u32 ioff = btrfs_item_offset(leaf->items + i);
C
Chris Mason 已提交
1196 1197 1198
			btrfs_set_item_offset(leaf->items + i,
					      ioff - data_size);
		}
1199 1200 1201

		/* shift the items */
		memmove(leaf->items + slot + 1, leaf->items + slot,
C
Chris Mason 已提交
1202
		        (nritems - slot) * sizeof(struct btrfs_item));
1203 1204

		/* shift the data */
C
Chris Mason 已提交
1205 1206
		memmove(btrfs_leaf_data(leaf) + data_end - data_size,
			btrfs_leaf_data(leaf) +
1207 1208 1209
		        data_end, old_data - data_end);
		data_end = old_data;
	}
1210
	/* setup the item for the new data */
C
Chris Mason 已提交
1211 1212
	memcpy(&leaf->items[slot].key, &disk_key,
		sizeof(struct btrfs_disk_key));
C
Chris Mason 已提交
1213 1214
	btrfs_set_item_offset(leaf->items + slot, data_end - data_size);
	btrfs_set_item_size(leaf->items + slot, data_size);
1215
	btrfs_set_header_nritems(&leaf->header, nritems + 1);
1216
	mark_buffer_dirty(leaf_buf);
C
Chris Mason 已提交
1217 1218

	ret = 0;
1219
	if (slot == 0)
1220
		ret = fixup_low_keys(trans, root, path, &disk_key, 1);
C
Chris Mason 已提交
1221

C
Chris Mason 已提交
1222
	if (btrfs_leaf_free_space(root, leaf) < 0)
1223
		BUG();
1224
	check_leaf(root, path, 0);
1225
out:
1226 1227 1228 1229 1230 1231 1232
	return ret;
}

/*
 * 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.
 */
1233 1234 1235
int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
		      *root, struct btrfs_key *cpu_key, void *data, u32
		      data_size)
1236 1237 1238 1239 1240 1241
{
	int ret = 0;
	struct btrfs_path path;
	u8 *ptr;

	btrfs_init_path(&path);
1242
	ret = btrfs_insert_empty_item(trans, root, &path, cpu_key, data_size);
1243
	if (!ret) {
C
Chris Mason 已提交
1244 1245
		ptr = btrfs_item_ptr(btrfs_buffer_leaf(path.nodes[0]),
				     path.slots[0], u8);
1246
		memcpy(ptr, data, data_size);
1247
		mark_buffer_dirty(path.nodes[0]);
1248
	}
C
Chris Mason 已提交
1249
	btrfs_release_path(root, &path);
C
Chris Mason 已提交
1250
	return ret;
1251 1252
}

C
Chris Mason 已提交
1253
/*
C
Chris Mason 已提交
1254
 * delete the pointer from a given node.
C
Chris Mason 已提交
1255 1256 1257 1258 1259
 *
 * 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.
 */
1260 1261
static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
		   struct btrfs_path *path, int level, int slot)
1262
{
C
Chris Mason 已提交
1263
	struct btrfs_node *node;
C
Chris Mason 已提交
1264
	struct buffer_head *parent = path->nodes[level];
1265
	u32 nritems;
C
Chris Mason 已提交
1266
	int ret = 0;
1267
	int wret;
1268

C
Chris Mason 已提交
1269
	node = btrfs_buffer_node(parent);
1270
	nritems = btrfs_header_nritems(&node->header);
1271
	if (slot != nritems -1) {
C
Chris Mason 已提交
1272 1273
		memmove(node->ptrs + slot, node->ptrs + slot + 1,
			sizeof(struct btrfs_key_ptr) * (nritems - slot - 1));
1274
	}
1275 1276 1277
	nritems--;
	btrfs_set_header_nritems(&node->header, nritems);
	if (nritems == 0 && parent == root->node) {
C
Chris Mason 已提交
1278 1279
		struct btrfs_header *header = btrfs_buffer_header(root->node);
		BUG_ON(btrfs_header_level(header) != 1);
1280
		/* just turn the root into a leaf and break */
C
Chris Mason 已提交
1281
		btrfs_set_header_level(header, 0);
1282
	} else if (slot == 0) {
1283
		wret = fixup_low_keys(trans, root, path, &node->ptrs[0].key,
C
Chris Mason 已提交
1284
				      level + 1);
C
Chris Mason 已提交
1285 1286
		if (wret)
			ret = wret;
1287
	}
1288
	mark_buffer_dirty(parent);
C
Chris Mason 已提交
1289
	return ret;
1290 1291
}

C
Chris Mason 已提交
1292 1293 1294 1295
/*
 * delete the item at the leaf level in path.  If that empties
 * the leaf, remove it from the tree
 */
1296 1297
int btrfs_del_item(struct btrfs_trans_handle *trans, struct btrfs_root *root,
		   struct btrfs_path *path)
1298 1299
{
	int slot;
C
Chris Mason 已提交
1300
	struct btrfs_leaf *leaf;
C
Chris Mason 已提交
1301
	struct buffer_head *leaf_buf;
1302 1303
	int doff;
	int dsize;
C
Chris Mason 已提交
1304 1305
	int ret = 0;
	int wret;
1306
	u32 nritems;
1307

1308
	leaf_buf = path->nodes[0];
C
Chris Mason 已提交
1309
	leaf = btrfs_buffer_leaf(leaf_buf);
1310
	slot = path->slots[0];
C
Chris Mason 已提交
1311 1312
	doff = btrfs_item_offset(leaf->items + slot);
	dsize = btrfs_item_size(leaf->items + slot);
1313
	nritems = btrfs_header_nritems(&leaf->header);
1314

1315
	if (slot != nritems - 1) {
1316
		int i;
C
Chris Mason 已提交
1317 1318 1319
		int data_end = leaf_data_end(root, leaf);
		memmove(btrfs_leaf_data(leaf) + data_end + dsize,
			btrfs_leaf_data(leaf) + data_end,
1320
			doff - data_end);
C
Chris Mason 已提交
1321
		for (i = slot + 1; i < nritems; i++) {
C
Chris Mason 已提交
1322
			u32 ioff = btrfs_item_offset(leaf->items + i);
C
Chris Mason 已提交
1323 1324
			btrfs_set_item_offset(leaf->items + i, ioff + dsize);
		}
1325
		memmove(leaf->items + slot, leaf->items + slot + 1,
C
Chris Mason 已提交
1326
			sizeof(struct btrfs_item) *
1327
			(nritems - slot - 1));
1328
	}
1329 1330
	btrfs_set_header_nritems(&leaf->header, nritems - 1);
	nritems--;
C
Chris Mason 已提交
1331
	/* delete the leaf if we've emptied it */
1332
	if (nritems == 0) {
1333
		if (leaf_buf == root->node) {
1334
			btrfs_set_header_level(&leaf->header, 0);
1335
		} else {
1336 1337
			clean_tree_block(trans, root, leaf_buf);
			wret = del_ptr(trans, root, path, 1, path->slots[1]);
C
Chris Mason 已提交
1338 1339
			if (wret)
				ret = wret;
1340
			wret = btrfs_free_extent(trans, root,
C
Chris Mason 已提交
1341
						 leaf_buf->b_blocknr, 1, 1);
C
Chris Mason 已提交
1342 1343
			if (wret)
				ret = wret;
1344
		}
1345
	} else {
1346
		int used = leaf_space_used(leaf, 0, nritems);
C
Chris Mason 已提交
1347
		if (slot == 0) {
1348 1349
			wret = fixup_low_keys(trans, root, path,
					      &leaf->items[0].key, 1);
C
Chris Mason 已提交
1350 1351 1352 1353
			if (wret)
				ret = wret;
		}

C
Chris Mason 已提交
1354
		/* delete the leaf if it is mostly empty */
C
Chris Mason 已提交
1355
		if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
1356 1357 1358 1359
			/* push_leaf_left fixes the path.
			 * make sure the path still points to our leaf
			 * for possible call to del_ptr below
			 */
1360
			slot = path->slots[1];
C
Chris Mason 已提交
1361
			get_bh(leaf_buf);
1362
			wret = push_leaf_left(trans, root, path, 1);
C
Chris Mason 已提交
1363 1364
			if (wret < 0)
				ret = wret;
1365
			if (path->nodes[0] == leaf_buf &&
1366
			    btrfs_header_nritems(&leaf->header)) {
1367
				wret = push_leaf_right(trans, root, path, 1);
C
Chris Mason 已提交
1368 1369 1370
				if (wret < 0)
					ret = wret;
			}
1371
			if (btrfs_header_nritems(&leaf->header) == 0) {
C
Chris Mason 已提交
1372
				u64 blocknr = leaf_buf->b_blocknr;
1373 1374
				clean_tree_block(trans, root, leaf_buf);
				wret = del_ptr(trans, root, path, 1, slot);
C
Chris Mason 已提交
1375 1376
				if (wret)
					ret = wret;
C
Chris Mason 已提交
1377
				btrfs_block_release(root, leaf_buf);
1378 1379
				wret = btrfs_free_extent(trans, root, blocknr,
							 1, 1);
C
Chris Mason 已提交
1380 1381
				if (wret)
					ret = wret;
C
Chris Mason 已提交
1382
			} else {
1383
				mark_buffer_dirty(leaf_buf);
C
Chris Mason 已提交
1384
				btrfs_block_release(root, leaf_buf);
1385
			}
1386 1387
		} else {
			mark_buffer_dirty(leaf_buf);
1388 1389
		}
	}
C
Chris Mason 已提交
1390
	return ret;
1391 1392
}

C
Chris Mason 已提交
1393 1394
/*
 * walk up the tree as far as required to find the next leaf.
C
Chris Mason 已提交
1395 1396
 * 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
1397
 */
C
Chris Mason 已提交
1398
int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
1399 1400 1401 1402
{
	int slot;
	int level = 1;
	u64 blocknr;
C
Chris Mason 已提交
1403 1404 1405
	struct buffer_head *c;
	struct btrfs_node *c_node;
	struct buffer_head *next = NULL;
1406

C
Chris Mason 已提交
1407
	while(level < BTRFS_MAX_LEVEL) {
1408
		if (!path->nodes[level])
C
Chris Mason 已提交
1409
			return 1;
1410 1411
		slot = path->slots[level] + 1;
		c = path->nodes[level];
C
Chris Mason 已提交
1412 1413
		c_node = btrfs_buffer_node(c);
		if (slot >= btrfs_header_nritems(&c_node->header)) {
1414 1415 1416
			level++;
			continue;
		}
C
Chris Mason 已提交
1417
		blocknr = btrfs_node_blockptr(c_node, slot);
C
Chris Mason 已提交
1418
		if (next)
C
Chris Mason 已提交
1419
			btrfs_block_release(root, next);
1420 1421 1422 1423 1424 1425 1426
		next = read_tree_block(root, blocknr);
		break;
	}
	path->slots[level] = slot;
	while(1) {
		level--;
		c = path->nodes[level];
C
Chris Mason 已提交
1427
		btrfs_block_release(root, c);
1428 1429 1430 1431
		path->nodes[level] = next;
		path->slots[level] = 0;
		if (!level)
			break;
1432
		next = read_tree_block(root,
C
Chris Mason 已提交
1433
		       btrfs_node_blockptr(btrfs_buffer_node(next), 0));
1434 1435 1436
	}
	return 0;
}