ctree.c 24.0 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"
7 8 9 10 11 12

static inline void init_path(struct ctree_path *p)
{
	memset(p, 0, sizeof(*p));
}

13 14 15 16 17 18 19 20 21 22
static void release_path(struct ctree_root *root, struct ctree_path *p)
{
	int i;
	for (i = 0; i < MAX_LEVEL; i++) {
		if (!p->nodes[i])
			break;
		tree_block_release(root, p->nodes[i]);
	}
}

23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96
static inline unsigned int leaf_data_end(struct leaf *leaf)
{
	unsigned int nr = leaf->header.nritems;
	if (nr == 0)
		return ARRAY_SIZE(leaf->data);
	return leaf->items[nr-1].offset;
}

static inline int leaf_free_space(struct leaf *leaf)
{
	int data_end = leaf_data_end(leaf);
	int nritems = leaf->header.nritems;
	char *items_end = (char *)(leaf->items + nritems + 1);
	return (char *)(leaf->data + data_end) - (char *)items_end;
}

int comp_keys(struct key *k1, struct key *k2)
{
	if (k1->objectid > k2->objectid)
		return 1;
	if (k1->objectid < k2->objectid)
		return -1;
	if (k1->flags > k2->flags)
		return 1;
	if (k1->flags < k2->flags)
		return -1;
	if (k1->offset > k2->offset)
		return 1;
	if (k1->offset < k2->offset)
		return -1;
	return 0;
}
int generic_bin_search(char *p, int item_size, struct key *key,
		       int max, int *slot)
{
	int low = 0;
	int high = max;
	int mid;
	int ret;
	struct key *tmp;

	while(low < high) {
		mid = (low + high) / 2;
		tmp = (struct key *)(p + mid * item_size);
		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;
}

int bin_search(struct node *c, struct key *key, int *slot)
{
	if (is_leaf(c->header.flags)) {
		struct leaf *l = (struct leaf *)c;
		return generic_bin_search((void *)l->items, sizeof(struct item),
					  key, c->header.nritems, slot);
	} else {
		return generic_bin_search((void *)c->keys, sizeof(struct key),
					  key, c->header.nritems, slot);
	}
	return -1;
}

int search_slot(struct ctree_root *root, struct key *key, struct ctree_path *p)
{
97 98 99
	struct tree_buffer *b = root->node;
	struct node *c;

100 101 102
	int slot;
	int ret;
	int level;
103 104 105
	b->count++;
	while (b) {
		c = &b->node;
106
		level = node_level(c->header.flags);
107
		p->nodes[level] = b;
108 109 110 111 112
		ret = bin_search(c, key, &slot);
		if (!is_leaf(c->header.flags)) {
			if (ret && slot > 0)
				slot -= 1;
			p->slots[level] = slot;
113
			b = read_tree_block(root, c->blockptrs[slot]);
114 115 116 117 118 119 120 121 122
			continue;
		} else {
			p->slots[level] = slot;
			return ret;
		}
	}
	return -1;
}

123 124 125
static void fixup_low_keys(struct ctree_root *root,
			   struct ctree_path *path, struct key *key,
			   int level)
126 127 128 129
{
	int i;
	/* adjust the pointers going up the tree */
	for (i = level; i < MAX_LEVEL; i++) {
130
		struct node *t;
131
		int tslot = path->slots[i];
132
		if (!path->nodes[i])
133
			break;
134
		t = &path->nodes[i]->node;
135
		memcpy(t->keys + tslot, key, sizeof(*key));
136
		write_tree_block(root, path->nodes[i]);
137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
		if (tslot != 0)
			break;
	}
}

int __insert_ptr(struct ctree_root *root,
		struct ctree_path *path, struct key *key,
		u64 blocknr, int slot, int level)
{
	struct node *c;
	struct node *lower;
	struct key *lower_key;
	int nritems;
	/* need a new root */
	if (!path->nodes[level]) {
152 153 154
		struct tree_buffer *t;
		t = alloc_free_block(root);
		c = &t->node;
155 156 157
		memset(c, 0, sizeof(c));
		c->header.nritems = 2;
		c->header.flags = node_level(level);
158 159
		c->header.blocknr = t->blocknr;
		lower = &path->nodes[level-1]->node;
160 161 162 163 164 165
		if (is_leaf(lower->header.flags))
			lower_key = &((struct leaf *)lower)->items[0].key;
		else
			lower_key = lower->keys;
		memcpy(c->keys, lower_key, sizeof(struct key));
		memcpy(c->keys + 1, key, sizeof(struct key));
166
		c->blockptrs[0] = path->nodes[level-1]->blocknr;
167
		c->blockptrs[1] = blocknr;
168 169 170 171 172 173
		/* the path has an extra ref to root->node */
		tree_block_release(root, root->node);
		root->node = t;
		t->count++;
		write_tree_block(root, t);
		path->nodes[level] = t;
174 175 176 177 178
		path->slots[level] = 0;
		if (c->keys[1].objectid == 0)
			BUG();
		return 0;
	}
179
	lower = &path->nodes[level]->node;
180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
	nritems = lower->header.nritems;
	if (slot > nritems)
		BUG();
	if (nritems == NODEPTRS_PER_BLOCK)
		BUG();
	if (slot != nritems) {
		memmove(lower->keys + slot + 1, lower->keys + slot,
			(nritems - slot) * sizeof(struct key));
		memmove(lower->blockptrs + slot + 1, lower->blockptrs + slot,
			(nritems - slot) * sizeof(u64));
	}
	memcpy(lower->keys + slot, key, sizeof(struct key));
	lower->blockptrs[slot] = blocknr;
	lower->header.nritems++;
	if (lower->keys[1].objectid == 0)
			BUG();
196
	write_tree_block(root, path->nodes[level]);
197 198 199 200 201 202 203 204 205 206 207
	return 0;
}

int push_node_left(struct ctree_root *root, struct ctree_path *path, int level)
{
	int slot;
	struct node *left;
	struct node *right;
	int push_items = 0;
	int left_nritems;
	int right_nritems;
208 209
	struct tree_buffer *t;
	struct tree_buffer *right_buf;
210 211 212 213 214 215 216

	if (level == MAX_LEVEL - 1 || path->nodes[level + 1] == 0)
		return 1;
	slot = path->slots[level + 1];
	if (slot == 0)
		return 1;

217 218 219 220 221
	t = read_tree_block(root,
		            path->nodes[level + 1]->node.blockptrs[slot - 1]);
	left = &t->node;
	right_buf = path->nodes[level];
	right = &right_buf->node;
222 223 224
	left_nritems = left->header.nritems;
	right_nritems = right->header.nritems;
	push_items = NODEPTRS_PER_BLOCK - (left_nritems + 1);
225 226
	if (push_items <= 0) {
		tree_block_release(root, t);
227
		return 1;
228
	}
229 230 231 232 233 234 235 236 237 238 239 240 241 242 243

	if (right_nritems < push_items)
		push_items = right_nritems;
	memcpy(left->keys + left_nritems, right->keys,
		push_items * sizeof(struct key));
	memcpy(left->blockptrs + left_nritems, right->blockptrs,
		push_items * sizeof(u64));
	memmove(right->keys, right->keys + push_items,
		(right_nritems - push_items) * sizeof(struct key));
	memmove(right->blockptrs, right->blockptrs + push_items,
		(right_nritems - push_items) * sizeof(u64));
	right->header.nritems -= push_items;
	left->header.nritems += push_items;

	/* adjust the pointers going up the tree */
244 245 246 247
	fixup_low_keys(root, path, right->keys, level + 1);

	write_tree_block(root, t);
	write_tree_block(root, right_buf);
248 249 250 251

	/* then fixup the leaf pointer in the path */
	if (path->slots[level] < push_items) {
		path->slots[level] += left_nritems;
252 253
		tree_block_release(root, path->nodes[level]);
		path->nodes[level] = t;
254 255 256
		path->slots[level + 1] -= 1;
	} else {
		path->slots[level] -= push_items;
257
		tree_block_release(root, t);
258 259 260 261 262 263 264
	}
	return 0;
}

int push_node_right(struct ctree_root *root, struct ctree_path *path, int level)
{
	int slot;
265 266
	struct tree_buffer *t;
	struct tree_buffer *src_buffer;
267 268 269 270 271 272 273 274 275 276 277 278
	struct node *dst;
	struct node *src;
	int push_items = 0;
	int dst_nritems;
	int src_nritems;

	if (level == MAX_LEVEL - 1 || path->nodes[level + 1] == 0)
		return 1;
	slot = path->slots[level + 1];
	if (slot == NODEPTRS_PER_BLOCK - 1)
		return 1;

279
	if (slot >= path->nodes[level + 1]->node.header.nritems -1)
280 281
		return 1;

282 283 284 285 286
	t = read_tree_block(root,
			    path->nodes[level + 1]->node.blockptrs[slot + 1]);
	dst = &t->node;
	src_buffer = path->nodes[level];
	src = &src_buffer->node;
287 288 289
	dst_nritems = dst->header.nritems;
	src_nritems = src->header.nritems;
	push_items = NODEPTRS_PER_BLOCK - (dst_nritems + 1);
290 291
	if (push_items <= 0) {
		tree_block_release(root, t);
292
		return 1;
293
	}
294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310

	if (src_nritems < push_items)
		push_items = src_nritems;
	memmove(dst->keys + push_items, dst->keys,
		dst_nritems * sizeof(struct key));
	memcpy(dst->keys, src->keys + src_nritems - push_items,
		push_items * sizeof(struct key));

	memmove(dst->blockptrs + push_items, dst->blockptrs,
		dst_nritems * sizeof(u64));
	memcpy(dst->blockptrs, src->blockptrs + src_nritems - push_items,
		push_items * sizeof(u64));

	src->header.nritems -= push_items;
	dst->header.nritems += push_items;

	/* adjust the pointers going up the tree */
311
	memcpy(path->nodes[level + 1]->node.keys + path->slots[level + 1] + 1,
312
		dst->keys, sizeof(struct key));
313 314 315 316 317

	write_tree_block(root, path->nodes[level + 1]);
	write_tree_block(root, t);
	write_tree_block(root, src_buffer);

318 319 320
	/* then fixup the leaf pointer in the path */
	if (path->slots[level] >= src->header.nritems) {
		path->slots[level] -= src->header.nritems;
321 322
		tree_block_release(root, path->nodes[level]);
		path->nodes[level] = t;
323
		path->slots[level + 1] += 1;
324 325
	} else {
		tree_block_release(root, t);
326 327 328 329 330 331 332 333
	}
	return 0;
}

int insert_ptr(struct ctree_root *root,
		struct ctree_path *path, struct key *key,
		u64 blocknr, int level)
{
334 335
	struct tree_buffer *t = path->nodes[level];
	struct node *c = &path->nodes[level]->node;
336
	struct node *b;
337 338
	struct tree_buffer *b_buffer;
	struct tree_buffer *bal[MAX_LEVEL];
339 340 341 342 343
	int bal_level = level;
	int mid;
	int bal_start = -1;

	memset(bal, 0, ARRAY_SIZE(bal));
344 345
	while(t && t->node.header.nritems == NODEPTRS_PER_BLOCK) {
		c = &t->node;
346 347 348 349 350 351 352 353 354
		if (push_node_left(root, path,
		   node_level(c->header.flags)) == 0)
			break;
		if (push_node_right(root, path,
		   node_level(c->header.flags)) == 0)
			break;
		bal_start = bal_level;
		if (bal_level == MAX_LEVEL - 1)
			BUG();
355 356
		b_buffer = alloc_free_block(root);
		b = &b_buffer->node;
357
		b->header.flags = c->header.flags;
358
		b->header.blocknr = b_buffer->blocknr;
359 360 361 362 363 364 365
		mid = (c->header.nritems + 1) / 2;
		memcpy(b->keys, c->keys + mid,
			(c->header.nritems - mid) * sizeof(struct key));
		memcpy(b->blockptrs, c->blockptrs + mid,
			(c->header.nritems - mid) * sizeof(u64));
		b->header.nritems = c->header.nritems - mid;
		c->header.nritems = mid;
366 367 368 369 370

		write_tree_block(root, t);
		write_tree_block(root, b_buffer);

		bal[bal_level] = b_buffer;
371 372 373
		if (bal_level == MAX_LEVEL - 1)
			break;
		bal_level += 1;
374
		t = path->nodes[bal_level];
375 376
	}
	while(bal_start > 0) {
377 378 379
		b_buffer = bal[bal_start];
		c = &path->nodes[bal_start]->node;
		__insert_ptr(root, path, b_buffer->node.keys, b_buffer->blocknr,
380 381 382
				path->slots[bal_start + 1] + 1, bal_start + 1);
		if (path->slots[bal_start] >= c->header.nritems) {
			path->slots[bal_start] -= c->header.nritems;
383 384
			tree_block_release(root, path->nodes[bal_start]);
			path->nodes[bal_start] = b_buffer;
385
			path->slots[bal_start + 1] += 1;
386 387
		} else {
			tree_block_release(root, b_buffer);
388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412
		}
		bal_start--;
		if (!bal[bal_start])
			break;
	}
	return __insert_ptr(root, path, key, blocknr, path->slots[level] + 1,
			    level);
}

int leaf_space_used(struct leaf *l, int start, int nr)
{
	int data_len;
	int end = start + nr - 1;

	if (!nr)
		return 0;
	data_len = l->items[start].offset + l->items[start].size;
	data_len = data_len - l->items[end].offset;
	data_len += sizeof(struct item) * nr;
	return data_len;
}

int push_leaf_left(struct ctree_root *root, struct ctree_path *path,
		   int data_size)
{
413 414 415
	struct tree_buffer *right_buf = path->nodes[0];
	struct leaf *right = &right_buf->leaf;
	struct tree_buffer *t;
416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431
	struct leaf *left;
	int slot;
	int i;
	int free_space;
	int push_space = 0;
	int push_items = 0;
	struct item *item;
	int old_left_nritems;

	slot = path->slots[1];
	if (slot == 0) {
		return 1;
	}
	if (!path->nodes[1]) {
		return 1;
	}
432 433
	t = read_tree_block(root, path->nodes[1]->node.blockptrs[slot - 1]);
	left = &t->leaf;
434 435
	free_space = leaf_free_space(left);
	if (free_space < data_size + sizeof(struct item)) {
436
		tree_block_release(root, t);
437 438 439 440 441 442 443 444 445 446 447 448
		return 1;
	}
	for (i = 0; i < right->header.nritems; i++) {
		item = right->items + i;
		if (path->slots[0] == i)
			push_space += data_size + sizeof(*item);
		if (item->size + sizeof(*item) + push_space > free_space)
			break;
		push_items++;
		push_space += item->size + sizeof(*item);
	}
	if (push_items == 0) {
449
		tree_block_release(root, t);
450 451 452 453 454 455 456 457 458 459
		return 1;
	}
	/* push data from right to left */
	memcpy(left->items + left->header.nritems,
		right->items, push_items * sizeof(struct item));
	push_space = LEAF_DATA_SIZE - right->items[push_items -1].offset;
	memcpy(left->data + leaf_data_end(left) - push_space,
		right->data + right->items[push_items - 1].offset,
		push_space);
	old_left_nritems = left->header.nritems;
460 461
	BUG_ON(old_left_nritems < 0);

462 463 464 465 466 467 468 469 470 471 472 473 474 475
	for(i = old_left_nritems; i < old_left_nritems + push_items; i++) {
		left->items[i].offset -= LEAF_DATA_SIZE -
			left->items[old_left_nritems -1].offset;
	}
	left->header.nritems += push_items;

	/* fixup right node */
	push_space = right->items[push_items-1].offset - leaf_data_end(right);
	memmove(right->data + LEAF_DATA_SIZE - push_space, right->data +
		leaf_data_end(right), push_space);
	memmove(right->items, right->items + push_items,
		(right->header.nritems - push_items) * sizeof(struct item));
	right->header.nritems -= push_items;
	push_space = LEAF_DATA_SIZE;
476

477 478 479 480
	for (i = 0; i < right->header.nritems; i++) {
		right->items[i].offset = push_space - right->items[i].size;
		push_space = right->items[i].offset;
	}
481 482 483 484 485

	write_tree_block(root, t);
	write_tree_block(root, right_buf);

	fixup_low_keys(root, path, &right->items[0].key, 1);
486 487 488 489

	/* then fixup the leaf pointer in the path */
	if (path->slots[0] < push_items) {
		path->slots[0] += old_left_nritems;
490 491
		tree_block_release(root, path->nodes[0]);
		path->nodes[0] = t;
492 493
		path->slots[1] -= 1;
	} else {
494
		tree_block_release(root, t);
495 496
		path->slots[0] -= push_items;
	}
497
	BUG_ON(path->slots[0] < 0);
498 499 500 501 502
	return 0;
}

int split_leaf(struct ctree_root *root, struct ctree_path *path, int data_size)
{
503 504 505 506 507
	struct tree_buffer *l_buf = path->nodes[0];
	struct leaf *l = &l_buf->leaf;
	int nritems;
	int mid;
	int slot;
508
	struct leaf *right;
509
	struct tree_buffer *right_buffer;
510 511 512 513 514 515 516
	int space_needed = data_size + sizeof(struct item);
	int data_copy_size;
	int rt_data_off;
	int i;
	int ret;

	if (push_leaf_left(root, path, data_size) == 0) {
517 518 519 520
		l_buf = path->nodes[0];
		l = &l_buf->leaf;
		if (leaf_free_space(l) >= sizeof(struct item) + data_size)
			return 0;
521
	}
522 523 524 525 526 527 528 529
	slot = path->slots[0];
	nritems = l->header.nritems;
	mid = (nritems + 1)/ 2;

	right_buffer = alloc_free_block(root);
	BUG_ON(!right_buffer);
	BUG_ON(mid == nritems);
	right = &right_buffer->leaf;
530 531 532 533 534 535 536 537 538 539 540
	memset(right, 0, sizeof(*right));
	if (mid <= slot) {
		if (leaf_space_used(l, mid, nritems - mid) + space_needed >
			LEAF_DATA_SIZE)
			BUG();
	} else {
		if (leaf_space_used(l, 0, mid + 1) + space_needed >
			LEAF_DATA_SIZE)
			BUG();
	}
	right->header.nritems = nritems - mid;
541 542
	right->header.blocknr = right_buffer->blocknr;
	right->header.flags = node_level(0);
543 544 545 546 547 548 549 550 551 552 553 554 555
	data_copy_size = l->items[mid].offset + l->items[mid].size -
			 leaf_data_end(l);
	memcpy(right->items, l->items + mid,
	       (nritems - mid) * sizeof(struct item));
	memcpy(right->data + LEAF_DATA_SIZE - data_copy_size,
	       l->data + leaf_data_end(l), data_copy_size);
	rt_data_off = LEAF_DATA_SIZE -
		     (l->items[mid].offset + l->items[mid].size);
	for (i = 0; i < right->header.nritems; i++) {
		right->items[i].offset += rt_data_off;
	}
	l->header.nritems = mid;
	ret = insert_ptr(root, path, &right->items[0].key,
556 557 558 559 560 561
			  right_buffer->blocknr, 1);

	write_tree_block(root, right_buffer);
	write_tree_block(root, l_buf);

	BUG_ON(path->slots[0] != slot);
562
	if (mid <= slot) {
563 564
		tree_block_release(root, path->nodes[0]);
		path->nodes[0] = right_buffer;
565 566
		path->slots[0] -= mid;
		path->slots[1] += 1;
567 568 569
	} else
		tree_block_release(root, right_buffer);
	BUG_ON(path->slots[0] < 0);
570 571 572 573 574 575 576 577
	return ret;
}

int insert_item(struct ctree_root *root, struct key *key,
			  void *data, int data_size)
{
	int ret;
	int slot;
578
	int slot_orig;
579
	struct leaf *leaf;
580
	struct tree_buffer *leaf_buf;
581 582 583 584
	unsigned int nritems;
	unsigned int data_end;
	struct ctree_path path;

585 586 587 588 589 590 591 592 593 594
	if (!root->node) {
		struct tree_buffer *t;
		t = alloc_free_block(root);
		BUG_ON(!t);
		t->node.header.nritems = 0;
		t->node.header.flags = node_level(0);
		t->node.header.blocknr = t->blocknr;
		root->node = t;
		write_tree_block(root, t);
	}
595 596
	init_path(&path);
	ret = search_slot(root, key, &path);
597 598
	if (ret == 0) {
		release_path(root, &path);
599
		return -EEXIST;
600
	}
601

602 603 604 605
	slot_orig = path.slots[0];
	leaf_buf = path.nodes[0];
	leaf = &leaf_buf->leaf;
	if (leaf_free_space(leaf) <  sizeof(struct item) + data_size) {
606
		split_leaf(root, &path, data_size);
607 608 609
		leaf_buf = path.nodes[0];
		leaf = &path.nodes[0]->leaf;
	}
610 611
	nritems = leaf->header.nritems;
	data_end = leaf_data_end(leaf);
612

613 614 615 616
	if (leaf_free_space(leaf) <  sizeof(struct item) + data_size)
		BUG();

	slot = path.slots[0];
617
	BUG_ON(slot < 0);
618
	if (slot == 0)
619
		fixup_low_keys(root, &path, key, 1);
620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645
	if (slot != nritems) {
		int i;
		unsigned int old_data = leaf->items[slot].offset +
					leaf->items[slot].size;

		/*
		 * item0..itemN ... dataN.offset..dataN.size .. data0.size
		 */
		/* first correct the data pointers */
		for (i = slot; i < nritems; i++)
			leaf->items[i].offset -= data_size;

		/* shift the items */
		memmove(leaf->items + slot + 1, leaf->items + slot,
		        (nritems - slot) * sizeof(struct item));

		/* shift the data */
		memmove(leaf->data + data_end - data_size, leaf->data +
		        data_end, old_data - data_end);
		data_end = old_data;
	}
	memcpy(&leaf->items[slot].key, key, sizeof(struct key));
	leaf->items[slot].offset = data_end - data_size;
	leaf->items[slot].size = data_size;
	memcpy(leaf->data + data_end - data_size, data, data_size);
	leaf->header.nritems += 1;
646
	write_tree_block(root, leaf_buf);
647 648
	if (leaf_free_space(leaf) < 0)
		BUG();
649
	release_path(root, &path);
650 651 652 653 654 655
	return 0;
}

int del_ptr(struct ctree_root *root, struct ctree_path *path, int level)
{
	int slot;
656
	struct tree_buffer *t;
657 658 659 660
	struct node *node;
	int nritems;

	while(1) {
661 662
		t = path->nodes[level];
		if (!t)
663
			break;
664
		node = &t->node;
665 666 667 668 669 670 671 672 673 674 675
		slot = path->slots[level];
		nritems = node->header.nritems;

		if (slot != nritems -1) {
			memmove(node->keys + slot, node->keys + slot + 1,
				sizeof(struct key) * (nritems - slot - 1));
			memmove(node->blockptrs + slot,
				node->blockptrs + slot + 1,
				sizeof(u64) * (nritems - slot - 1));
		}
		node->header.nritems--;
676
		write_tree_block(root, t);
677 678 679
		if (node->header.nritems != 0) {
			int tslot;
			if (slot == 0)
680 681
				fixup_low_keys(root, path, node->keys,
					       level + 1);
682
			tslot = path->slots[level+1];
683
			t->count++;
684 685 686 687
			push_node_left(root, path, level);
			if (node->header.nritems) {
				push_node_right(root, path, level);
			}
688 689
			if (node->header.nritems) {
				tree_block_release(root, t);
690
				break;
691 692
			}
			tree_block_release(root, t);
693
			path->slots[level+1] = tslot;
694
		}
695 696 697 698
		if (t == root->node) {
			/* just turn the root into a leaf and break */
			root->node->node.header.flags = node_level(0);
			write_tree_block(root, t);
699 700 701 702 703 704 705 706 707
			break;
		}
		level++;
		if (!path->nodes[level])
			BUG();
	}
	return 0;
}

708
int del_item(struct ctree_root *root, struct ctree_path *path)
709 710 711
{
	int slot;
	struct leaf *leaf;
712
	struct tree_buffer *leaf_buf;
713 714 715
	int doff;
	int dsize;

716 717
	leaf_buf = path->nodes[0];
	leaf = &leaf_buf->leaf;
718
	slot = path->slots[0];
719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735
	doff = leaf->items[slot].offset;
	dsize = leaf->items[slot].size;

	if (slot != leaf->header.nritems - 1) {
		int i;
		int data_end = leaf_data_end(leaf);
		memmove(leaf->data + data_end + dsize,
			leaf->data + data_end,
			doff - data_end);
		for (i = slot + 1; i < leaf->header.nritems; i++)
			leaf->items[i].offset += dsize;
		memmove(leaf->items + slot, leaf->items + slot + 1,
			sizeof(struct item) *
			(leaf->header.nritems - slot - 1));
	}
	leaf->header.nritems -= 1;
	if (leaf->header.nritems == 0) {
736 737 738 739
		if (leaf_buf == root->node) {
			leaf->header.flags = node_level(0);
			write_tree_block(root, leaf_buf);
		} else
740
			del_ptr(root, path, 1);
741 742
	} else {
		if (slot == 0)
743 744
			fixup_low_keys(root, path, &leaf->items[0].key, 1);
		write_tree_block(root, leaf_buf);
745 746 747 748 749 750
		if (leaf_space_used(leaf, 0, leaf->header.nritems) <
		    LEAF_DATA_SIZE / 4) {
			/* push_leaf_left fixes the path.
			 * make sure the path still points to our leaf
			 * for possible call to del_ptr below
			 */
751
			slot = path->slots[1];
752
			leaf_buf->count++;
753
			push_leaf_left(root, path, 1);
754
			if (leaf->header.nritems == 0) {
755 756
				path->slots[1] = slot;
				del_ptr(root, path, 1);
757
			}
758
			tree_block_release(root, leaf_buf);
759 760 761 762 763 764 765 766 767 768
		}
	}
	return 0;
}

void print_leaf(struct leaf *l)
{
	int i;
	int nr = l->header.nritems;
	struct item *item;
769
	printf("leaf %lu total ptrs %d free space %d\n", l->header.blocknr, nr,
770 771 772 773 774 775 776 777 778 779 780 781 782
	       leaf_free_space(l));
	fflush(stdout);
	for (i = 0 ; i < nr ; i++) {
		item = l->items + i;
		printf("\titem %d key (%lu %u %lu) itemoff %d itemsize %d\n",
			i,
			item->key.objectid, item->key.flags, item->key.offset,
			item->offset, item->size);
		fflush(stdout);
		printf("\t\titem data %.*s\n", item->size, l->data+item->offset);
		fflush(stdout);
	}
}
783
void print_tree(struct ctree_root *root, struct tree_buffer *t)
784 785 786
{
	int i;
	int nr;
787
	struct node *c;
788

789
	if (!t)
790
		return;
791
	c = &t->node;
792
	nr = c->header.nritems;
793 794
	if (c->header.blocknr != t->blocknr)
		BUG();
795 796 797 798
	if (is_leaf(c->header.flags)) {
		print_leaf((struct leaf *)c);
		return;
	}
799
	printf("node %lu level %d total ptrs %d free spc %lu\n", t->blocknr,
800 801 802 803
	        node_level(c->header.flags), c->header.nritems,
		NODEPTRS_PER_BLOCK - c->header.nritems);
	fflush(stdout);
	for (i = 0; i < nr; i++) {
804
		printf("\tkey %d (%lu %u %lu) block %lu\n",
805 806 807 808 809 810
		       i,
		       c->keys[i].objectid, c->keys[i].flags, c->keys[i].offset,
		       c->blockptrs[i]);
		fflush(stdout);
	}
	for (i = 0; i < nr; i++) {
811 812 813
		struct tree_buffer *next_buf = read_tree_block(root,
							    c->blockptrs[i]);
		struct node *next = &next_buf->node;
814 815 816 817 818 819
		if (is_leaf(next->header.flags) &&
		    node_level(c->header.flags) != 1)
			BUG();
		if (node_level(next->header.flags) !=
			node_level(c->header.flags) - 1)
			BUG();
820 821
		print_tree(root, next_buf);
		tree_block_release(root, next_buf);
822 823 824 825 826 827 828 829 830 831 832
	}

}

/* for testing only */
int next_key(int i, int max_key) {
	return rand() % max_key;
	// return i;
}

int main() {
833
	struct ctree_root *root;
834
	struct key ins;
835
	struct key last = { (u64)-1, 0, 0};
836 837 838 839
	char *buf;
	int i;
	int num;
	int ret;
840
	int run_size = 1000000;
841 842 843 844
	int max_key = 100000000;
	int tree_size = 0;
	struct ctree_path path;

845 846 847 848
	radix_tree_init();


	root = open_ctree("dbfile");
849 850 851 852 853 854 855 856 857 858 859

	srand(55);
	for (i = 0; i < run_size; i++) {
		buf = malloc(64);
		num = next_key(i, max_key);
		// num = i;
		sprintf(buf, "string-%d", num);
		// printf("insert %d\n", num);
		ins.objectid = num;
		ins.offset = 0;
		ins.flags = 0;
860
		ret = insert_item(root, &ins, buf, strlen(buf));
861 862 863
		if (!ret)
			tree_size++;
	}
864 865 866
	close_ctree(root);
	root = open_ctree("dbfile");
	printf("starting search\n");
867 868 869 870 871
	srand(55);
	for (i = 0; i < run_size; i++) {
		num = next_key(i, max_key);
		ins.objectid = num;
		init_path(&path);
872
		ret = search_slot(root, &ins, &path);
873
		if (ret) {
874
			print_tree(root, root->node);
875 876 877
			printf("unable to find %d\n", num);
			exit(1);
		}
878 879 880 881 882 883 884 885 886
		release_path(root, &path);
	}
	close_ctree(root);
	root = open_ctree("dbfile");
	printf("node %p level %d total ptrs %d free spc %lu\n", root->node,
	        node_level(root->node->node.header.flags),
		root->node->node.header.nritems,
		NODEPTRS_PER_BLOCK - root->node->node.header.nritems);
	printf("all searches good, deleting some items\n");
887 888
	i = 0;
	srand(55);
889 890 891 892
	for (i = 0 ; i < run_size/4; i++) {
		num = next_key(i, max_key);
		ins.objectid = num;
		init_path(&path);
893
		ret = search_slot(root, &ins, &path);
894 895
		if (ret)
			continue;
896
		ret = del_item(root, &path);
897 898
		if (ret != 0)
			BUG();
899
		release_path(root, &path);
900 901 902
		tree_size--;
	}
	srand(128);
903
	for (i = 0; i < run_size; i++) {
904
		buf = malloc(64);
905
		num = next_key(i, max_key);
906
		sprintf(buf, "string-%d", num);
907
		ins.objectid = num;
908
		ret = insert_item(root, &ins, buf, strlen(buf));
909 910 911
		if (!ret)
			tree_size++;
	}
912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929
	close_ctree(root);
	root = open_ctree("dbfile");
	printf("starting search2\n");
	srand(128);
	for (i = 0; i < run_size; i++) {
		num = next_key(i, max_key);
		ins.objectid = num;
		init_path(&path);
		ret = search_slot(root, &ins, &path);
		if (ret) {
			print_tree(root, root->node);
			printf("unable to find %d\n", num);
			exit(1);
		}
		release_path(root, &path);
	}
	printf("starting big long delete run\n");
	while(root->node && root->node->node.header.nritems > 0) {
930 931 932 933
		struct leaf *leaf;
		int slot;
		ins.objectid = (u64)-1;
		init_path(&path);
934
		ret = search_slot(root, &ins, &path);
935 936 937
		if (ret == 0)
			BUG();

938
		leaf = &path.nodes[0]->leaf;
939 940 941 942 943 944
		slot = path.slots[0];
		if (slot != leaf->header.nritems)
			BUG();
		while(path.slots[0] > 0) {
			path.slots[0] -= 1;
			slot = path.slots[0];
945
			leaf = &path.nodes[0]->leaf;
946 947 948 949

			if (comp_keys(&last, &leaf->items[slot].key) <= 0)
				BUG();
			memcpy(&last, &leaf->items[slot].key, sizeof(last));
950 951 952
			ret = del_item(root, &path);
			if (ret != 0) {
				printf("del_item returned %d\n", ret);
953
				BUG();
954
			}
955 956
			tree_size--;
		}
957
		release_path(root, &path);
958
	}
959
	close_ctree(root);
960
	printf("tree size is now %d\n", tree_size);
961 962
	return 0;
}